10:13karolherbst: this damn regs...
10:14karolherbst: I think I need a nvf1 to play around a bit
10:14karolherbst: because for whatever reasons the gk208 is different again and then the gm are also different
10:16martm_: hi hakzsam_ https://github.com/hakzsam/mesa/commit/314b7349afdb0cb4d5bc1f084525df83bfbf240c are you still missing something that blocks the merge of images to mesa master?
10:17martm_: hakzsam_: did you get all dEQP tests to pass after this diff mentioned?
10:18karolherbst: mupuf: I think the nvidia partition is a little messed up... gdb: error while loading shared libraries: libpython3.4m.so.1.0: cannot open shared object file: No such file or directory
10:19mupuf: karolherbst: hmm, give me a sec
10:20mupuf: doing the updates now
10:20mupuf: I may have done a partial update, which is not supported
10:21karolherbst: and the libnouveau stuff is somewhat broken... no idea if that was me with my branch or not.. :D
10:23karolherbst: a thread crashes.. should be some kind of worker thread
10:25martm_: hakzsam_: you gave the status on irc , i forgat which date it was, judging by the comment i think, the only possible blocker could be 3d arrays, as it seems to be something different then cube arrays
10:26martm_: deqp-gles31/functional/shaders/opaque_type_indexing/sampler/const_literal/compute/sampler3d: fail
10:26martm_: i mean this test is probably the only question if that passes , if it does then probably images are done
10:27karolherbst: mupuf: but in the end I expect that it works on gk208 just at it is, for some reasons gk110 is different or nvidia does something differently, but I can't remember what exactly anymore. I think in the end we really need one of those for reator :/
10:34hakzsam_: martm_, I'm working on images support yeah, should be ready soon for kepler at least
10:34hakzsam_: I still need to fix some things before sending the series
10:34RSpliet: karolherbst:random funfact, GK110 whitepaper lists 16 threadblocks max per SMX (but 2048 max threads? that's not 192*16 but 128*16, wtf? :-D)
10:35RSpliet: that implies that ideally the GPR count for a shader is 21
10:35karolherbst: those are weird
10:35martm_: hakzsam_: i have one kepler card, so i can test too, but i am at my grannies at the moment, i do not recall what the codename was at the moment, but it's gt730
10:35mupuf: karolherbst: someone proposed to send us GPUs, maybe we can ask for one GK110 :D
10:35RSpliet: just in case you're interested how far you could go with scheduling :-P
10:35karolherbst: mupuf: :)
10:36martm_: hakzsam_: when you are ready, then i update my mesa and test if those tests pass on my card too
10:36karolherbst: RSpliet: funny though, the spec state 2880 and 15 SMX'
10:36hakzsam_: martm_, fine by me, but I can test too :)
10:36RSpliet: 15*192 == 2880, that's not funny, that's the truth :-P
10:36karolherbst: max threads
10:37hakzsam_: martm_, I'll let you know when it's ready anyway
10:38karolherbst: RSpliet: maybe the don't need more
10:38karolherbst: or it was simplier to build the chip this way
10:38RSpliet: karolherbst: I'm sure there's a rationale behind the limitations of 2048 threads max, 16 thread blocks max
10:39karolherbst: mhh they seems to be more efficient than the gk104 though
10:39RSpliet: but the interesting thing is the numbers
10:39karolherbst: how is maxwell here?
10:39RSpliet: in the case you wanted to revisit instruction scheduling, there's a clear target now for kepler
10:40karolherbst: mupuf: did pacman fail?
10:40karolherbst: RSpliet: yeah well, I think I won't get this far until next year anyway
10:40RSpliet: karolherbst: fair enough :-)
10:40karolherbst: I want to add dynamic reclocking before that :D
10:42RSpliet: just sth to keep in mind. I think aiming for a specific reg count (whether it's 21 or 32 - but stretching to the next "bucket" when it's infeasible) is algorithmically easier than trying to go as low as possible
10:43karolherbst: but yeah, as low as possible is really tricky though
10:43karolherbst: I don't know really, I think this needs a lot of time to get into the stuff
10:44karolherbst: I even tried to schedule isntructions first which free the most live values and still got a higher reg count in the end
10:44karolherbst: so I think in the end it doesn't matter because we have to look at the whole thing somewhat
10:44RSpliet: yeah, it's not an easy problem!
10:44karolherbst: or at least at an entire BB at once
10:44karolherbst: and order while knowing each instruction
10:45karolherbst: in the end the target shouldn't be to increase the gpc count or decrease it
10:45RSpliet: but I'm sure there's heuristics for this in graph theory
10:45karolherbst: but more to fill up all the stalls with fitting isntructions
10:45karolherbst: yeah well
10:45karolherbst: this is a gpu though
10:46RSpliet: no, reducing GPC count is the most effective way of filling up stalls
10:46mupuf: karolherbst: you may reboot reator, everything is up to dat
10:46karolherbst: mupuf: well gdb works now :D
10:46karolherbst: RSpliet: how so?
10:46mupuf: yeah, but please reboot because it also upgraded the nvidia driver
10:46mupuf: and the kernel
10:46karolherbst: nvidia puts a lot of instructions between the use and the consumer of a value
10:46karolherbst: sometimes even 30 or 40 instructions
10:47RSpliet: because the SMX round-robins through the threads. If one thread stalls, it picks the next thread
10:47RSpliet: (either switch every insn or when a stall is detected, not sure about the impl. details)
10:47karolherbst: but usually it is more than 10 instructions between the write and read
10:47mupuf: RSpliet: maybe this has a cost too
10:47RSpliet: mupuf: unlikely
10:48karolherbst: well at least nvidia does fill up with instructions
10:48karolherbst: should be a reason to do so
10:48RSpliet: karolherbst: sure, but you've seen the effects of reducing the GPR count just over a boundary (so you had one extra thread in flight)
10:49RSpliet: mupuf: ARM Mali for instance switch threads every cycle
10:49karolherbst: but just one more thread might be actually slower than more effective threads
10:49RSpliet: remember that all data is already in the registers, there's no load/store from mem :-)
10:49karolherbst: doesn't matter
10:50karolherbst: I once scheduled randomly without changing gpc count and got a big varriance in performance
10:50mupuf: RSpliet: intel has something like that too
10:50karolherbst: it was more than -+5%
10:51RSpliet: mupuf: I assume every GPU does something similar
10:51mupuf: quite probable, hiding memory accesses is super importan
10:51RSpliet: yeah, throughput is more important than latency for individual threads :-)
10:52karolherbst: subdev->device->fb was null
10:54karolherbst: yeah I don't want to reclock in libnouveau
10:56mupuf: ah ah
10:57karolherbst: well this was in nvkm_pstate_prog with pstate = -1
10:58martm_: so after we reach to high opengl levels, to conquer all graphics apps, i don't think we'd need opencl, rather i'd go for dx11.3 and vulkan and dx12, i never read vulkan spec, hopefully karolherbst: can do that work:)? but i think at some point i'll for directx 11.3
11:01karolherbst: mupuf: thanks for updating
11:01mupuf: karolherbst: YW
11:03martm_: i'll go, i was frustrated and anxious to get to dx using a binary code, after i've educated myself, i think it's better off to implement support ontop of open source drivers
11:04karolherbst: mupuf: nvidia-settings got whiped out
11:05karolherbst: ohh well
11:05karolherbst: nvidia-smi seems to be good enough
11:05karolherbst: I hope
11:05karolherbst: or not
11:05mupuf: karolherbst: it got split into another package
11:06mupuf: installed it
11:06martm_: btw: of course it's not me, who has been spamming the netowork with highly vulnerable and dangerous subjects, i have stayd with my nick those days around
11:06martm_: have used joss sometimes too with those numbers in the end..but no other nicks user by me
11:09karolherbst: okay, I have no idea if it is this gk208, but this seems fine
11:09karolherbst: yeah I really think I need to toy on a gk110
11:12karolherbst: so back to crashing my own kernel today
11:21karolherbst: this is getting painful now...
11:21martm_: i'll log on from home later, thanks, bye.
11:21karolherbst: I guess I have to add a mutex around the entire clocking state in nvkm_clk
11:22karolherbst: otherwise we get situations like this: therm daemon triggers reclockin while pstate is changed by used
11:22karolherbst: => fun
14:00Newwww: Hello! I have a problem wit reclocking cat: /sys/class/drm/card0/device/pstate: No such file or directory
14:03karolherbst: Newwww: debugs
14:03karolherbst: it is in /sys/kernel/debug/dri/0 most likely
14:03karolherbst: or 1 if you have a laptop
14:04karolherbst: Newwww: or you forgot to boot with nouveau.pstate=1
14:04karolherbst: Newwww: depends on your kernel version actually
14:04Newwww: I use Linux pc 4.6.0-1-MANJARO
14:04karolherbst: yeah, then it is inside debugfs
14:04karolherbst: the change was mode in 4.5
14:04Newwww: [root@pc doaxan]# ls /sys/kernel/debug/dri/0/pstate ls: cannot access '/sys/kernel/debug/dri/0/pstate': No such file or directory
14:05karolherbst: Newwww: check if debugfs is mounted
14:05Newwww: Need write in terminal sudo debugfs&
14:05karolherbst: ohh yeha, you need to be root to read in debugfs
14:05karolherbst: Newwww: mount| grep debugfs
14:06Newwww: debugfs on /sys/kernel/debug type debugfs (rw,relatime)
14:06karolherbst: if this prints nothing you have to mount it first
14:06karolherbst: Newwww: do you have a laptop or dual gpu setup in any way?
14:06Newwww: laptop with nvidia optimus? intel hd4000 + nvidia gtx 660m
14:06karolherbst: then it is 1
14:06karolherbst: as I said
14:06karolherbst: not 0
14:06karolherbst: 0 is the intel one
14:07Newwww: Thank you!
14:10Newwww: Wow! echo 0f > /sys/kernel/debug/dri/1/pstate very fast!
14:10karolherbst: course it is :p
14:11karolherbst: but there is a chance it crashes the gpu
14:12Newwww: crashes due to high temp?
14:13karolherbst: Newwww: no, low voltage, other stuff
14:17karolherbst: Newwww: well when the gpu crashes for you, I have some experimental stuff to fix that. I just need to finish the work with the most annoying parts before me...
14:26Newwww: hmm, DRI_PRIME=1 '/home/doaxan/WarThunder/launcher' dont work(
14:26Newwww: *** Error in `./aces': double free or corruption (fasttop): 0x00007f8ad8669030 ***
14:26Newwww: Maybe i need update mesa?
14:26Newwww: To git version?
14:29karolherbst: worth a shot?
14:29karolherbst: but maybe it is the game doing stupid stuff
14:30Newwww: ======= Backtrace: ========= /usr/lib/libc.so.6(+0x6f364)[0x7f96ef334364] /usr/lib/libc.so.6(+0x74d96)[0x7f96ef339d96] /usr/lib/libc.so.6(+0x7557e)[0x7f96ef33a57e] /usr/lib/xorg/modules/dri/nouveau_dri.so(+0x43034f)[0x7f96e85f834f] /usr/lib/xorg/modules/dri/nouveau_dri.so(+0x43058e)[0x7f96e85f858e] /usr/lib/xorg/modules/dri/nouveau_dri.so(+0x430771)[0x7f96e85f8771] /usr/lib/xorg/modules/dri/nouveau_dri.so(+0x430878)[0x7f96e85f8878]
14:32karolherbst: yeah well maybe there is a corner case nouveau doesn't handle that well
14:33karolherbst: Newwww: you should try out newest mesa
14:36Newwwww: Laptop freezes after i launch a game with low settings :p
14:42karolherbst: Newwwww: mhh the laptop?
14:42martm: Newww: what game is it, i'd always like to see some list of failing games , i susepct that radeonsi is the running the most games ...i'd help to hunt some bugs down, if someone points me what is the failing game list
14:42karolherbst: mhhh it shouldn't
14:44martm: Newwww: that is quite not enough of a backtrace that would help, you would need to step through the game
14:45Newwwww: This game warthunder
14:45Newwwww: I tried install mesa-git
14:45martm: it's very complex to hunt that bug i belive, give you'r card from lspci too
14:46Newwwww: Game load, i see load screen and laptop freezes
14:46martm: Newwww: when you get a backtrace like this, it almost shows, that there is a fault in the cpu part of the stuff
14:47Newwwww: Sorry i have bad internet
14:47Newwwww: And poor englisg :)
14:48martm: Newwww: mesa is quite big, normally i have not tried to debug multithreaded code, and i don't even know today how to do that, there probably is a way
14:49martm: but mesa is almost as big as llvm ... so that is quite heavy steps you need to make
14:50martm: i did it so that i set a breakpoint to the code, at some address , and then step one line forward
14:50martm: and wait for the game to crash
14:52Newwwww: Ok, can i help?
14:53martm: yes, you can, but i think i can tomorrow try to run that game, thanks for the info...
14:54martm: or what is that some commercial game, i think normally they give a gltrace for that
14:54martm: apitrace or something
15:00martm: Newwww: unfortunently it seems that i can not access this game, maybe even some others are interested to get an apitrace of your crash
15:00martm: i would but i don't participate yet in the mailing lists, that is more trouble then good i belive
15:01martm: i think open a bug, and attach an apitrace there , kinda shirnk it down where the crash happens
15:03Newwwww: Cannot access?
15:04martm: Newwwww: well is that commercial game that costs something, cause i don't want to play a pirate, neither want to pay anything to get the access to that game
15:04Newwwww: This game is free
15:05martm: Newwwww: aah fabulous!!!
15:05Newwwww: You dont need to pay
15:05martm: this is really good, thanks, i will try
15:07Newwwww: Compile mesa-git...its so long...:)
15:08martm: i have to do that too, i have a tree on my harddrive, kinda need to do checkout or update or something
15:12kiljacken: Hey, I'm having problems with a region fault / timeouts during loading of the new gm204 signed firmware, does anybody here know about that?
15:12martm: Newwwww: ok i read now, what type of challanges did you complete when it crashed?
15:15Newwwww: martm: hmm, i download client, on intel card launcher is work, on DRI_PRIME=1 is black window, but button play is work, game running, i see load screen, logging and laptop freezing, hard shutdown help only
15:16Newwwww: Freeze laptop or crash game depends on graphics settings
15:16martm: allright, i am downloading, never used DRI_PRIME though, it only crashes while using prime?
15:16martm: ahah, ok...
15:17Newwwww: martm: is there another way ? :))
15:18Newwwww: To dont use dri prime?
15:18martm: yeah..i belive so, that probably you need to switch from bios to one of the adapters
15:18Newwwww: Bios switching to only intel or both
15:18martm: really i am not very big expert on it, they called that as muxed graphics, they have some tool for this too probably under linux
15:19Newwwww: You mean use only nouveau? Hmm, its diffucult, but i tried
15:19martm: Newwwww: to the card you want to test that on, we currently ara talking about probably crash in nouveaus driver!
15:19martm: so switch to nvidias card
15:20Newwwww: Yes i mean nouveau of course
15:20martm: Newwwwww: yeah
15:21martm: i am downloading that game...i need to wait for 2hours more
15:21martm: 3to be exact now, cause it seems to be 3gb
15:22martm: i have a shitty 5megs down pipe in second
15:23martm: divided by 8 it's around 700kb/s
15:24Newwwww: DRI_PRIME works like bumblebee, use nvidia only i want this
15:24martm: Yeah it does...
15:25martm: what it does, someone can correct me, but it's a driver way to offload the computation to another gpu, and show it on another output
15:25martm: i've inspected the patchset very long time ago, kinda almost liked the idea though
15:26martm: so you can configure stuff, via xrandr, i have also nvidia and hd4400 so i could in theoory even try
15:26martm: dunno what the reverse prime was for, i knew but i forgat
15:27martm: mux is like a multiplexer, it can be controlled in software, i dunno this some acpi stuff prolly
15:27martm: airlied sometime ago talked about this, sometimes two graphics are wired over one output
15:28martm: controlling the mux via acpi or something, you control which graphics are connected to it
15:30Newwewwwww: Ok, i tried xrandr
15:31martm: i think it really depends how the designers have put it together, is it with mux or without, and how many outputs and stuff
15:32Newwewwwww: although I would like to use something like bumblebee
15:32Newwewwwww: Because laptop on nvidia is so noize
15:33Newwewwwwew: Stupid internet :))
15:33Newwewwwwew: (Newwewwwww) although I would like to use something like bumblebee
15:33Newwewwwwew: (Newwewwwww) Because laptop on nvidia is so noize
15:38martm: ouh dear, some serious netowrk problems, i belive New socket would be freed allready. muhahahaaa!
17:02martm: i don't quite understand, the biggest question almost like seems to be, when you see me again doing sports, i quit like 15years ago, my father was one my way, i probably can do another year, when he's moved out of the way
17:03martm: because it/he does not move from out of the way with his own will
17:04martm: this one has screwd me so bad, that i can't even hypothesize, how well i could had competed with other worlds sportsmen
17:06martm: i've commented on this michael maze stuff, due to the difficulties in my life, i did not have that kinda of backhand in my games reportoir, i would had never developed something so beatiful has his one, but was possible to develope competively as effecient
17:06martm: if my father was not on my way with his kill off attempts
17:16martm: it was never easy to screw me that deep out of the sports, but it was consistently and by closest possible source, and to be honest i even think that my dad hating sportsmen so deep, that he has killed even more talented ones before
17:19martm: he has been incapable of acheiving his goals in life by normal means, a probable possibility that his mother boozed, he is very angry due to his body parts that he was given at birth, and he found his way to pass his anger and realize his stuff on me
17:20martm: by way of blocking every possible chanche that i had
17:22martm: you could not ask for better results that i had like 15 or 20years ago, but so it went, i was not allowed to advance any further
17:24martm: take it as you want, lots of estonians have complained an actions against theirs..to capture it so close from the father, noone has experienced that..i don't know what the fuck is wrong with that quassimodo
17:25martm: i'd say just suicidal terrorist blocked my way of advancinf any further
17:27martm: i know i score heavy in most of the events that i do, cause i am strong man, but the stuff to get me to excile like my grannie said, was too big to handle to deal with sports overall at high level
17:44martm: imirkin: don't trip it man, i love russian sportsmen they are the best what ever happened to me, my dad is some kind of abhasian terrorist, where as i look like kaukasian, i am white he is some sort more darker version
17:50martm: this kinda thing i say it publicly that russians have said that this guy sits, and is not allowed to do anything is crap, russians are moral
17:51martm: this is only one suicidal terrorist and couple of his abortion leftovers who want you to think like that
17:59martm: i just downloaded the game, it seems to run..but i can not log on. as i downloaded the stuff via facebook
18:00martm: first game i have ever tried during some years...seems beatiful
18:01martm: beautiful.. can log on, but i think it runs here nicely
18:21martm: i never get to the game, it throws some js errors, surely those are not related to nouveau ones
18:21martm: tried to log on
19:00martm: i had so many glib errors and stuff, nothing has been my tested or developed, there are too many errors
19:00martm: but i do not see anything graphics related to blame you here
19:06martm: those days around i gotta do some choises , and i haven't been actively participated in anything...i gotta write pieces of code, network stuff, its just i have not practised
19:06martm: i sorta gave up on it, when they picketed decades on the run, or in a row or something
19:11martm: https://www.facebook.com/profile.php?id=100010527543118 this is joss after 10years of hiding i put my account up again
19:24martm: as i am a white person, and according to my information those are a defiency, i dunno what the fuck are those chipsies and such terrorists wining about, everything is racistic against those right?
19:25martm: well guess what those violate my rights too, and i have only russians and germans to take
19:28martm: chipsies have like all arabian countries
19:28martm: all fucking their black hair brothers
19:29martm: i only have my fellow russians and germans and couple of other nordic countries
19:32pkolaczk: Everything works great, but I have a minor problem setting my external display as primary with xrandr (using nouveau with nvidia optimus). Is it the right channel to ask?
19:38pkolaczk: I'm getting this:
19:38pkolaczk: X Error of failed request: BadMatch (invalid parameter attributes)
19:38pkolaczk: Major opcode of failed request: 140 (RANDR)
19:38pkolaczk: Minor opcode of failed request: 30 (RRSetOutputPrimary)
19:38pkolaczk: Serial number of failed request: 45
19:38pkolaczk: Current serial number in output stream: 47
19:39martm: ok ok, lets not forget about england, australia and united states, where there are some of my fellows yet
19:39pkolaczk: after invoking xrandr --output DP-1-1 --primary
19:39martm: we are outpowered by black, and my father terrorist new that
19:44martm: while we can have negotiations with russians or germans who are closer here, to manslaughter terrorists, we have to make choises, and my dad is top of the list
19:46martm: two hacks and the head is gone like united states miliatary said, we'd need to do their style of stuff
19:47martm: two hacks means , a kneee to the back, and round the head with nife boom
19:47martm: head sliced off
19:51martm: yeah so that terrorist will be talked faced to face 10th of april
19:51martm: wether he will be slaughtered or not
19:51karolherbst: pkolaczk: ohh the external display is wired to your intel or the nvidia gpu?
19:51martm: russians are basically have always been at my side
19:52karolherbst: RSpliet: ohh I have his 780 Ti vbios, I know the reasons
19:52karolherbst: and I already fixed them :D
19:53pkolaczk: the external display is wired to the nvidia chip
19:53pkolaczk: but it does work
19:53pkolaczk: the only problem is I cannot make it primary
19:54karolherbst: no idea where the issue might be
19:54pkolaczk: it does even automatically detect connecting the external display and powers up the nvidia chip - and this is purely awesome - better than the proprietary driver
19:54martm: and btw. i did not like the hate against jews, so it's definitely not me spamming this
19:54martm: it was hitlers big mistake thinking that jews are the evil and laizy clever bastards
19:55pkolaczk: I don't need the final solution, but maybe some hints how to debug?
19:55martm: hitler had some nobel ideas too, but jews were not at fault at all
19:56karolherbst: pkolaczk: no clue. You could ask in the radeon channel if somebody has the same issues with a intel/amd combination
19:57martm: hitler insanitly became evident when he wanted to kill all the jews and invade into russia
19:57pkolaczk: karolherbst: but first - is it supposed to work at all?
19:58karolherbst: no idea
20:00martm: i.e the one that he thought hating retards , was just justified, making a white sociaty and power, was also justified
20:04martm: talking about hitler the most influenced people in 20th century, he seemed to have another flood, about gays, which did not look that bad, but though gays are harmless, they think hitler was jelous
20:04martm: i.e was gay himself
20:06martm: we had a sportsmen, berlin 1932... who lived to see adolf hitler being dissapointed
20:06martm: kristjan palusalu, one of my greatest real dads in life
20:07martm: i think hornfischer was the oponent back then, which made adolf to leave
20:07martm: kristjan rolled him over
20:12martm: there are lot of them to be called those days around too, and i have signalled most of them, that i have trouble in life where terrorists bother me, white sportsmen hopefully will try to save me
20:31Newwwww: Hi again, i tried install PRIME but i get black screen. (Nvidia optimus, intel hd4000+gtx660m)
20:40Newwwww: Warthunder worked the guy who tried to download game?
20:56martm: well yeah as you see this is my identity i haven't got anything to be afraid off this thing, how i was held in mental insitution will go to supreme court, which handles it
20:56martm: and basically those who made false statements in court, will get their sentence
20:57martm: i don't give a shit what is logged or not, that is my true opinion what i expressed
21:04martm: you do remember that during the court of law, no way that false statements could be presented, you remember why italian court imprisoned two persons for 18years in prison
21:04martm: only cause they made false statements
21:04martm: the murder of meredith kercher case
21:05martm: and it is idetintifed that to keep me in mental institution my family made 10lies about me in purpose
21:05martm: of course they receive a sentence cause of that
21:13martm: so what i am trying to acheive, is that terrorists don't lock me up, it ends with restraining order to them all, plus a penalty a tarif to pay
22:01karolherbst: NV_shader_thread_group might be of some use in performance analysis maybe