05:49 mardikene: i might had been too sarcastic about say mwk, his composed docs in the end are almost fine, definitely unfrair comparison to some amoral terror monkeys that i am used to seeing in my country, those ara arrogant but it's known also very perversely cowardish and around gigantic amount of times more clueless
05:50 mardikene: so i may say, that docs i looked at were suffiecient looking to fix cpu multithreading for instance, so this is a bit of a compliment, i tried to describe also how mesa works, but hw wise the docs had the needed stuff
06:46 mardikene: well i skip all this nightmare at the moment, as it does not belong to this channel, but or1k where the fpu_arith.v is taken from in miaow, is implemented as a typical parallel to serial converter, if you open the file
06:48 mardikene: there is a counter in the middle of states, when you google this it will turn up quickly, meaning typically in verilog small amount of code with 16 instances can be run as 1/16 or less cycles
07:00 mardikene: and as seen from the code, every instance takes 2047bit vector, and there are 64 of them multiplexed, that is 64pointers per SIMD
07:03 mardikene: well 2048 but, it does not try to implement double precision, otherwise it would be two times higher
07:10 mardikene: if you want to view it in another way, that makes 4vectors of 512bits, but the point is the same pointer wise it is 256reg pointers per CU and 64 per SIMD in amd terms
07:17 mardikene: for the real gpu fabbed processor a minor change to the hwacha vliw turns out to be probably more area efficient
07:18 mardikene: we could discuss what the risc-v does , but i yet have not inspected, i just thought to begin with would be easier with miaow which i laid my eyes on
07:18 mardikene: to which
07:28 mardikene: so nvidia has slightly different formation to it's SMX, i will skip that, everything is written in devtalk forums, 192 alus per SMX
07:29 mardikene: https://devtalk.nvidia.com/default/topic/897696/relationship-between-threads-and-gpu-core-units/?offset=3
07:31 mardikene: no one said kepler to be vliw, but the layout or formation sort of indicates it is, vliw is very good, i like it more, but additional developer knowledge is needed for driver developer
07:36 mardikene: well yeah they do not say it, as probably it isn't conventional VLIW in a sense that the lanes do not execute different instructions by default without using pointers, it's what i call an oo pipeline without scoreboarding in hw, they instead have some sched opcodes and syncrhoniation
07:40 mardikene: i'd expect predicates to work similarly in sw and hw there as for vliw though
07:52 mardikene: yeah well i think if you mess up the latency codes, sw can schedule too, by skipping the stalling instruction either with sw predication or sw controlled masked hw one on kepler, allthough sw predication is not tested or referred on the web
08:02 mardikene: let us make one observation like, when you meet a raw hazard, with pointers you have a readback conflict , when you detect this you can apply a sw predicate in the scheduler logic, defnitely i reaveled too much but , let's if you are able to capitlize on that info
08:03 mardikene: this will skip the instruction and assign a new PC to that warp, and you can come back to that instruction with a pointer later as if the in-order pipeline was converted into oo
08:03 mardikene: OoO some mark it so
08:04 mardikene: one do not need a broadasting mask ballot in hw for this, but anything that flips the pointer is fine
08:08 mardikene: that is the theory, that any 1.5ogl complianced gpu is fine for scheduling intelligently, it is because pointers/arrays are allmighty
08:12 mardikene: slight comment yet to go , about your sophisticated clock gating terminology, so what this does with power usage , the calculation is over my head complex, but it will lower it immensively
08:23 mardikene: well i am volunteer, and as my story of life is not really belonging to this channel, and it is so scrappy/crappy that i am embarrassed to write autobiography of this shit, i suggest there is enough work to do, and soon i think let's get down to business of that
08:24 mardikene: currently i have this stuff on hold, until i get responses from some experts if the possible presense of ip allows us to do such things
08:25 mardikene: i am allready sanctioned by monkeys, it'd be not nice if i get into debts on top
08:30 mardikene: that those sort of hacks won't work is impossible, but something held back AMD devs aside from companies strategy, so presense of ip in a form of patent that is filed for someone, could be possible
08:34 mardikene: i have not seen it yet, there is one for intel doing the same end goal, but in totally different way, but there is a private database where some are kept in sort of like carantine or something, depending what the term is in english
08:34 mardikene: it's karantiin in estonian
08:36 mardikene: quarantine
08:38 mardikene: bye
09:02 mardikene: https://patents.google.com/patent/US9547880B2/en heck there was another one too, but both of them are pretty harmless i can't no longer find another one, this was also granted to intel, us patent has a grant
09:05 mardikene: pretty recent 2017, so i have not been able to find with throough research any of the relevant filings yet
09:09 mardikene: well estonians have betrayded me round about in anything, three surgeries while being under narcosis, fucked up, drugged in the bars and assaulted from back, head concussion, stalked and ontop diagnosed, you can expect any violations from those but competetively most are named as abortion leftovers by me, cause they can compete only by amorally breaking the rules, meaning they in legal life are entire underperforming crap
09:12 mardikene: it appears vast amount of them are born with huge braindamage, not useful to interact with such
09:17 mardikene: if your dream is not to be an amoral cowardish scammer and senseless picketer using senseless scammed mass psychosis phenomen. i suggest taking talks about me with reasonable scepticism, it is fair to use force when such come to scam
09:24 mardikene: back days well airlied advised me to get new friends, this is not easy, this things was accounted with spreading false positives to communities by professional scammers here, my new friends especially female friends do not tolerate that kind of picketations around me
09:26 mardikene: as much as i want to get rid of those dudes that sucked into this game as professional agents, it is very hard, while they diagnose me with mental illness as a backupplan, and tell i am antisocial and paranoid about people stalking me
09:26 mardikene: no i am not, that was the agents backupplan to present that kind of false positives to society
10:42 mardikene: so provided i am a total average guy, logically what makes the picketers to be? probably total outsiders right, provided i was also hansom, than what can be their backup, this is where the real paranoia starts to happen in both ways, for all average and very talented humans the most important thing is timing in sports
10:43 mardikene: meaning that an average guy can get his body to support the punch, while very talented guys are in sort of a suspension, the transfers of energy is divided more equally in the body, and inertia is not seen where they get it from, it is not obvious
10:46 mardikene: this is how for instance michael maze plays table tennis, he is like floating on suspension, cause he was and is the biggest talent in europe in the sports, all he needs is to time the shot correctly, but chinease put the body behind the shot, and for me to knock out any bullyer i need to get the body
10:46 mardikene: behind the punch also, as anett kontaveit doing the cross, he needs is legs to support the cross backhand
10:50 mardikene: the anology is like japanease houses, when you build them to suspension the eggs in the basket do not break, but those average guys when something is cleverly done to them, they fall down cause of the earthquake at least momentarily, you should not be afraid of anyone in estonia as long as you being average enough time the punch correctly and get your body behind the punch
10:50 mardikene: no one is going to survive that even if there is a block ahead in bare knuckles fight
10:53 mardikene: so even though i have been cracked up with surgeries, they read my stuff, drugged me and injured me, i have enough resources to defend myself still present
10:54 mardikene: it is sophisticated cause the weaker nasty agents are not moral to be punished, but ...
10:55 mardikene: and also the more talented ones are not moral to be punished
10:59 mardikene: it is easy to be a talent or retard, but it is not easy to stay on the average, and here i am teaching you how to defend yourself, with same injuries other people have died that they caused me though
10:59 mardikene: but i stayd on the average level without dying
10:59 mardikene: which is as complex as it can get
11:03 mardikene: for instance looking at the pictures, your biggest talent is RSpliet, i must be thinking how would he feel, if i were to think his dangerous maybe he can land a punch and to prevent that i orginise a surgery to him to injure one
11:03 mardikene: i feel like it is pathetic i can't do that
11:06 mardikene: but the both classes are around me to think it is justified when a hansom looking average guy can be dangerous to retards or top seeds
11:07 mardikene: hence they fixed the results and i got 10times done this crank stuff on me
11:30 mardikene: well yeah the biggest hurdle is to take the step to break through to professional career of something, progress to the higher class then the average, if you had that capability, not many want to see that to happen
11:35 mardikene: for some reason to me, that step was infinite, it almost seemed that the sociaty did not accept a person like me to compete with one of the worlds best, but earlier there were guys saying the same, maybe the welfare of our country makes the envy to happen
11:43 mardikene: and painfully enough the conclusion to paul keres issues and some wrestlers except palusalu and lots of others who brought that up, seems to originate from friendly fire and envy in that poor country which has not enough social program to keep people happy
11:52 mardikene: i go now, and i really commented on palusalus stuff, he was lucky enough to have brothers of almost same strength and came out as similar talents, otherwise without their defence he would had ended in the institution and death like raimond valgre
11:52 mardikene: in other words he would had been killed by friendly fire and envy in our country like raimond was
11:52 mardikene: bye
12:52 sigod: is there any way to stop a CTXSW_TIMEOUT error on a gtx660 without installing the non-free firmware
12:57 imirkin: did using the blob ctxsw fw help your situation?
13:06 sigod: imirkin, no i installed it but the firmware didnt load even though it was enabled in /etc/modprobe.d
13:06 sigod: im using parabola
13:07 imirkin: sigod: ok, well step 1 is to determine if it helps at all
13:07 sigod: but it didnt even load
13:07 sigod: in dmesg
13:07 sigod: it said non-free firmware loading disabled
13:08 imirkin: ah
13:08 skeggsb:loves how these "freedom" distros limit freedom of choice ;)
13:08 imirkin: i think what it meant to say "loading non-free firmware from hard drive disabled. from on-board rom is fine"
13:09 sigod: ha
13:09 imirkin: anyways, feel free to give it a shot with a kernel that respects user choice
13:09 sigod: someone at parabola said that i could set nouveau to 2d acceleration
13:10 sigod: and slowly turn on 3d settings?
13:10 sigod: is that possible
13:10 imirkin: if they're the ones telling you this, ask them
13:10 imirkin: i'm not sure what "2d acceleration" is vs "3d acceleration" ...
13:10 imirkin: esp in the context of ... context switches
13:11 sigod: oh ok
13:11 sigod: so whether its 2d or 3d wont change things
13:12 imirkin: maybe, maybe not. doubt that the graph class in use makes a big difference. but we haven't a clue what's causing the issue, so ...
13:13 sigod: is there any way to diagnose without the firmware?
13:15 imirkin: blob firmware isn't a way of diagnosing, really
13:15 imirkin: more like ensuring that you have the same non-identified issue as a bunch of other people
13:16 imirkin: i suspect we're missing some bit of error handling in there somewhere
13:18 sigod: sorry i just froze again
13:21 sigod: what would nouveau.noaccel=1 kernel parameter do
13:22 imirkin: disable acceleration
13:23 imirkin: but still leave fbdev/fbcon accelerated
13:24 skeggsb: it disables it completely
13:24 imirkin: skeggsb: iirc that's nofbaccel or whatever
13:24 imirkin: i.e. they're separately controlled
13:25 skeggsb: noaccel does both, nofbaccel just disables fbcon
13:25 imirkin: oh, hm, ok
13:25 imirkin: that was not my recollection. perhaps i had it wrong, or something changed.
13:27 sigod: so where else would the error handling be?
13:28 imirkin: sigod: let me save you a little time ... you won't solve this issue. if you can get the blob firmware to work for you, great. otherwise get a different board.
13:29 sigod: ok i just dont like to be defeated so easily
13:30 imirkin: really you just want to get an amd board and enjoy the open-source love
13:31 sigod: amd gpu?
13:31 skeggsb: how would those work on that distro? they have firmware too
13:31 karolherbst: they don't
13:31 sigod: intel is the best
13:31 karolherbst: newer intel won't work as well
13:31 imirkin: skeggsb: idiotic distros are easily replaced
13:31 imirkin: esp their kernels
13:31 karolherbst: or newer wifi or ethernet or all kind of stuff
13:32 sigod: i like parabola
13:32 imirkin: great. enjoy it then :)
13:32 sigod: lol
13:33 sigod: it works fine this is the only issue
13:33 imirkin: the logic that causes one to refuse loading firmware from disk to be uploaded to hardware is, invariably, foolish.
13:33 karolherbst: imirkin: well, always depends on what your thread model is ;) on board firmware isn't "upgradeable" so you can't just put new functionalities in them
13:34 imirkin: i don't think they're doing it for the threat model
13:34 imirkin: if "threat model" was anywhere on the list, then disable gpu drivers outright
13:34 karolherbst: well, it isn't that bad with open source firmware, no?
13:35 karolherbst: but right, hardware can always have functioanlities you are not aware of
13:35 sigod: is someone writing free firmware for this?
13:35 karolherbst: for what?
13:35 skeggsb: we already have it, it just doesn't work on this one particular board
13:35 karolherbst: nouveau has only free firwmare
13:36 karolherbst: well
13:36 karolherbst: kind of
13:36 sigod: we are talking about the nouveau-fw
13:36 karolherbst: yeah.. but you don't need really
13:36 karolherbst: and for you it isn't used anyway
13:36 sigod: well my card is crashing without it i think
13:36 karolherbst: it should make no difference
13:37 karolherbst: or well
13:37 karolherbst: except the blob stuff is loaded
13:37 karolherbst: then it does
13:37 sigod: what's the CTXSW_TIMEOUT error then
13:37 karolherbst: probably a bug inside our firmware
13:37 sigod: when it isnt loaded
13:38 karolherbst: but ours is shipped with the kernel
13:38 skeggsb: sigod: the other thing you were asked to try was loading *nvidia's* firmware, ours is built-in
13:38 sigod: yea i know
13:39 sigod: so if i wanted to find the bug where would i start?
13:39 sigod: imirkin says give up
13:39 skeggsb: https://github.com/skeggsb/nouveau/tree/master/drm/nouveau/nvkm/engine/gr/fuc
13:39 skeggsb: here
13:40 karolherbst: skeggsb: in theory we could validate all values and cause an interrupt/break in the firmware, no?
13:40 karolherbst: that would be a nice way to track it down
13:41 skeggsb: i haven't the feintest idea what could be wrong. it's not something obvious
13:41 karolherbst: huh
13:41 karolherbst: skeggsb: since when is our firmware GPL-2?
13:41 imirkin: since some jackass went around adding SPDX headers to all files
13:41 karolherbst: beh
13:41 skeggsb: yep, what imirkin said
13:41 karolherbst: that's illegal you know...
13:41 karolherbst: violation of license and all
13:42 karolherbst: or well
13:42 karolherbst: meh... annoying
13:42 karolherbst: of course it is allowed for the source inside the kernel, because that's all GPL-2 as a total thing anyway
13:42 karolherbst: but out of tree stuff still isn't
13:43 sigod: i really want to help on this one
13:43 sigod: i have time as well
13:45 karolherbst: skeggsb, imirkin: seems like that was scripted and I think we kind of have to fix that, because as it seems the license header shall be used for enforcing the license
13:46 imirkin: sigod: you mentioned that programming wasn't in your current skillset
13:47 imirkin: this is ... advanced programming.
13:47 sigod: ok so i have to learn to program...
13:47 karolherbst: skeggsb: well maybe it helps to validate each value against something we expect and maybe in rare cases we get unexpected values, like bits set, we didn't... or something. Dunno
13:47 karolherbst: but maybe we just put somthing somewhere the hw doesn't like
13:48 skeggsb: i suspect we're missing some magic wait or register write (there's tons of them in the ucode already)
13:48 karolherbst: mhh
13:48 karolherbst: yeah
13:48 karolherbst: might be
13:48 skeggsb: it's going to be a game of "see what nvidia do that we don't and try random things until it works"
13:48 skeggsb: painful
13:48 karolherbst: yes
13:48 skeggsb: and hard
13:48 karolherbst: do you have such a gpu?
13:49 skeggsb: no, mine works
13:49 karolherbst: meh :(
13:49 skeggsb: i have a gtx660, yes.
13:49 imirkin: usually gpu's in skeggsb's hands don't stay non-working for too long
13:49 karolherbst: all of mine also just survive a complete piglit run
13:49 karolherbst: but mhh
13:49 karolherbst: I didn't sed them as a desktop GPU
13:49 skeggsb: i compared all the obvious differences the fw handles against the trace you sent me, and there aren't any
13:49 karolherbst: ahh
13:49 karolherbst: well
13:49 karolherbst: there has to be something :p
13:50 skeggsb: you have access to the hw ;)
13:50 karolherbst: right...
13:50 karolherbst: I forgot about it
13:50 karolherbst: has to wait until next week though
13:50 karolherbst: holiday tomorrow and then PTO until next week
13:50 imirkin: karolherbst: you have an affected gtx660?
13:50 karolherbst: imirkin: I guess...
13:51 karolherbst: somebody from the office noticed that bug and gave it to me to make some traces for skeggsb
13:51 karolherbst: I just forgot about it
13:51 karolherbst: skeggsb: if you don't have any ideas, I could start to dig something up
13:52 karolherbst: maybe I get lucky
14:44 imirkin_: karolherbst: just send it to ben and tell him you bet he can't fix it :)
14:48 karolherbst: :D
15:33 juri_: I have a 660 here, as well, if someone needs me to run envytools stuff on it.
15:33 imirkin_: does it work with nouveau?
15:34 juri_: it's reported to have graphical artifacts... but i'm not using X11 on that machine, so the report is second hand.
15:34 juri_: the in-kernel driver is working on it, however. text mode works.
15:34 imirkin_: it = that same board, or some other 660?
15:34 imirkin_: with just text mode, not a ton of context switching that goes on
15:35 juri_:nods.
15:35 juri_: i've tested the board that x11 being graphically artifacty with the text mode.
15:36 juri_: gv-n66toc-2gd rev 1.0. it will be tomorrow before i can fire X up on it, however.
15:37 juri_: text mode works without either of the pci-e power plugs plugged in.
20:52 nikolasc: hi
20:53 nikolasc: I'm using a gtx 1060 with nouveau in a gentoo 4.14-39 kernel but fan speed is noisy, that doesn't happens with the nvidia driver
20:53 imirkin_: correct.
20:54 HdkR: does missing pmu firmwares cause the fan to crank to the max?
20:54 imirkin_: we appreciate that you have a choice in GPU vendors. you appear to have chosen poorly.
20:55 imirkin_: HdkR: it's technically not missing
20:55 HdkR: sorry, not under a license for redistributing
20:55 HdkR: :p
20:55 imirkin_: i mean, it's there and loaded
20:55 imirkin_: https://www.youtube.com/watch?v=BoekVaXdA3U
20:56 imirkin_: it just doesn't control the fan in a conservative way
20:56 nikolasc: any help?
20:56 imirkin_: sure. cancel your order of $2B worth of nvidia hardware.
20:57 imirkin_: [sorry if i may sound unhelpful. there's just not a ton to be done here.]
21:02 HdkR: imirkin_: Are you saying that fan control could be implemented without relying on the blob fairy to come and save the day?
21:03 HdkR: :)
21:05 imirkin_: not to my knowledge
21:05 imirkin_: but canceling $2B orders could move nvidia
21:06 imirkin_: (you know, probably even the $1B...)
21:06 imirkin_: therm is only accessible in high-secure mode
21:07 HdkR: Highly secure fan control. Nailed it
21:07 HdkR: Next I'm going to rig a panel removal detection switch to the GPU and have it combust immediately upon tampering
21:22 reds: o/
21:25 reds: i've got a 7300GT on c2d E7400 + Void Linux. Card works fine, but i'd love to get three-head setup. I can get the two outputs on the card working just fine, and I thought about connecting the third screen to the iGPU, it doesn't work properly tho. I can execute xrandr --setprovideroutputsource, but then I won't get anything but garbled mess and a mouse pointer on the third screen. Any ideas/help would be
21:25 reds: greatly appreciated!
21:25 imirkin_: 7300GT only has 2 CRTC's, so you can only get 2 outputs
21:25 imirkin_: no matter how many connectors it might have
21:26 imirkin_: also 7300GT + prime pretty much won't work
21:26 imirkin_: sorry :(
21:26 reds: i think that you didn't understand me correctly
21:26 imirkin_: i understood you.
21:26 imirkin_: you have a second gpu
21:26 imirkin_: and want to offload one of its outputs
21:26 reds: two connectors on the 7300GT + one on the integrated
21:27 imirkin_: this is called "PRIME"
21:27 reds: oh
21:27 imirkin_: (or "reverse prime"
21:27 imirkin_: coz ... optimus
21:27 imirkin_: transformers
21:27 imirkin_: etc
21:27 reds: out of curiosity, why won't it work? is it a hardware, or software related issue?
21:27 imirkin_: unfortunately the 7300GT's DMA capabilities aren't ideal for this
21:27 reds: ah
21:27 imirkin_: i suspect software could be made to make it go
21:28 reds: ok, thanks for your time!
21:28 imirkin_: but it's not in that state right now
21:28 imirkin_: a G80+ GPU is more likely to function in this scenario
21:28 imirkin_: a kepler+ gpu will actually let you use up to 4 heads on a single GPU