00:41 mwk: lovely, reading nonexistent res in 4008xx range on NV10 tends to hang the card
00:42 imirkin_: that does sound delightful.
00:43 imirkin_: that's the GeForce 256 right?
00:43 mwk: yep
00:43 imirkin_: twice as good as a Rage 128 :)
00:43 imirkin_: iirc there was some crazy logic to what the 256 meant
00:44 imirkin_: adding bus bit widths with numbers of megabytes of vram, plus other things...
02:27 mwk:wonders what the fuck is going on with the alternative classes on NV10
02:27 mwk: it seems every NV10_whatever class has its evil twin with a different number
02:27 mwk: as in, NV10_CELSIUS is 0x56, and it has an evil twin 0x85
02:27 mwk: it's the first time I see those...
02:28 imirkin_: weird!
02:28 mwk: there's a switch in DEBUG_3 that selects whether the card will accept the usual class IDs and reject the evil ones, or accept evil ones and reject the good ones
02:29 mwk: also, only classes that first appeared on NV10 have evil twins, stuff that was before NV10 or that was added on NV11 doesn't have these
02:30 imirkin_: is there anything different about the evil twins?
02:31 mwk: no idea
02:32 mwk: I haven't submitted any methods on them yet
02:32 mwk: or, for that matter, any other class on NV10
02:32 imirkin_: this was the era when software had local "driver" implementations...
02:32 imirkin_: perhaps they did something special for someone?
02:32 mwk: we'll see
02:33 mwk: maybe 0x85 is NV10_QUADRO_CELSIUS!
02:33 imirkin_: ooh
02:33 imirkin_: i'm guessing you don't have one of those lying around...
02:34 imirkin_: (a Quadro)
02:34 mwk: nope
02:34 mwk: well, not an NV10 one at any rate
02:34 imirkin_: non-nv10 ones are Quadro2 etc
02:35 mwk: FWIW, rules.xml calls the relevant switch bit "ALTARCH"
02:44 mwk: well, that was horrible
02:44 mwk: of course NV10 is different, sigh
02:44 imirkin_: well, NV11 is a diff generation basically
02:44 imirkin_: NV10/NV15/NV1A is one thing
02:45 imirkin_: and NV11/NV17/NV18 are another. and of course NV17 also adds a bunch of stuff.
02:45 mwk: NV15 also adds plenty of its own
02:45 mwk: it's going to be: NV10, NV15, NV11, NV17/NV18/NV1F
02:46 mwk: not sure where NV1A is going to be placed, probably closer to NV11 than NV15
02:46 imirkin_: nv11 added somethign important though... fences? i forget
02:46 mwk: flips
02:46 mwk: as in, the weirdo PGRAPH methods that wait on crtc
02:46 imirkin_: iirc it was fence-related
02:46 mwk: also, semaphores on PFIFO
02:47 imirkin_: yea
02:47 imirkin_: that sounds right
06:20 karthy: How to get reclocking feature working on older Gpu's?
07:09 Guest95932: I read wikipedia DRM article, and its image says userprogram calls DRM modules via ioctl and DRM modules sends commands to VRAM. But I wonder if i can send commands from userspace? like pack commands into a buffer and dma them, execute them.
07:30 gnurou: Guest95932: my understanding is that this is exactly what happens - you build your command buffer from userspace (in system RAM usually, although Nouveau has an option to have them in VRAM - see the vram_pushbuf module option), submit it via an ioctl, and the kernel just programs the DMA hardware to interpret your buffer
08:24 Guest95932: gnurou: oh, I see. thank you.
09:01 janus_: hello people
09:01 janus_: i've been experiencing random crashes with nouveau on old quadro hardware..
09:02 janus_: lenovo t61, kernel 4.6.2-gentoo, NVIDIA Corporation G86M [Quadro NVS 140M] [10de:0429], xf86-video-nouveau-1.0.13(started with 1.0.12)... i use gentoo and posted the same on #gentoo
11:57 pmoreau: skeggsb: Neat! Some additional Pascal work + initial GP102 support. :-)
12:00 mupuf: pmoreau: you are late ;)
12:00 pmoreau: :-D
12:00 pmoreau: Only by a week!
12:01 mupuf: 5 days, let's not overdo it ;0
12:02 pmoreau: I should buy a Pascal one day, but I am waiting for a possible 1080 Ti before doing that.
12:05 mupuf: :o
12:05 mupuf: the 1080 is not expensive-enough?
12:06 mupuf: the cheapest 1050 is at 140e in my nearest shop :D
12:06 mupuf: still too expensive for me :D
12:06 mupuf: and it is out of stock anyway
12:06 mupuf: and this fan issue needs to be adressed first
12:07 mupuf: seriously, what is wrong with the people who designed this table :s
12:13 pmoreau: It’s mainly for gaming, and on 4k, on maximal settings, if possible. I’m hoping as well, that it will have a similar architecture as the NVIDIA Titan X, and as such, have a good support for fp16.
12:14 pmoreau: That fan table seems crazy indeed; I followed a bit your progress in the RE’ing.
12:17 karolherbst: ;D
12:17 karolherbst: pmoreau: I heard you take care of GDDR5X?
12:18 pmoreau: RE’ing it for reclocking?
12:18 karolherbst: sure
12:18 pmoreau: uhhh…
12:18 karolherbst: we can't do anything about other things anyway
12:19 karolherbst: fan is pmu domain
12:19 pmoreau: Got OpenCL on my plate! Sorry, I can’t :-D
12:19 karolherbst: :D
12:19 karolherbst: well, you thought of buying a GDDR5x gpu
12:19 pmoreau:slowly backs off to the exit
12:19 karolherbst: your fault :p
12:19 karolherbst: I am quite sure the GDDR5 stuff will be quite sane and not that different actually
12:20 karolherbst: looked nice enough
12:20 pmoreau: You’ll most likely get access to it though. I still have to work that out, but I’ll have a look around Christmas at making everything available.
12:21 karolherbst: nice
12:21 pmoreau: I could give you access to the box already, but I have nothing to have it boot on demand yet, so it’s not that convenient.
12:21 karolherbst: I only have traces with GDDR5X sadly :(
12:21 karolherbst: a trace would be enough
12:21 karolherbst: and also if skeggsb would push the maxwell pdisplay rnndb stuff :p
12:21 karolherbst: uhh
12:21 karolherbst: pfb
12:21 karolherbst: *pascal
12:22 pmoreau: So a trace of what would you need, if you already have one with GDDR5X? A Pascal with regular GDDR5?
12:23 karolherbst: yes
12:23 karolherbst: should also look into the engine stuff at some point...
12:23 karolherbst: but this also looked sane
12:23 karolherbst: the only _real_ problem should be the vbios
12:23 karolherbst: and I don't really like guessing everything
12:24 pmoreau: Right…
12:25 karolherbst: pmoreau: they doubled the P bit table size ;)
12:25 karolherbst: basically
12:25 karolherbst: ohh and a vbios is also good
12:25 pmoreau: 2x more of the same data?
12:25 karolherbst: I only have garbage vbios files for odd reasons
12:26 karolherbst: pmoreau: well, more like they droped the old ones and made completly new ones
12:26 karolherbst: I think only the votlage table is the same?
12:26 karolherbst: I don't plan on getting anything working next year, so pascal reclocking has to wait until 2018 at least
12:29 pmoreau: I’ll try to get the VBIOS from this one, but I would need to reboot with the iomem relax thing.
12:30 karolherbst: would be awesome :)
14:33 imirkin: uh oh. https://reviews.llvm.org/D26602 - i think we might have code that does something like that. hakzsam_ - care to check it out?
14:34 imirkin: i have faint recollections of writing such logic
14:34 imirkin: maybe i was smart and only did it for int's?
14:34 imirkin: seems unlikely
14:34 hakzsam_: yes, but I can't right now
14:38 imirkin: ah no. looks like i only did it for SHL + ADD, not MUL + ADD =]
14:39 imirkin: that was prescient of me...
14:44 karolherbst: imirkin: wouldn't we do fadd(x, 1) -> mov(x) anyway?
14:44 karolherbst: ohh add
14:45 karolherbst: hum
14:50 imirkin: only if x == infinity ;)
14:50 karolherbst: :O
14:51 karolherbst: allthough I am wondering if nvidia is able to predict the right value ranges to enable such an opt
14:51 imirkin: a decent compiler ought to
14:51 karolherbst: mhh
14:52 karolherbst: would be a good project to make the nouveau ones smarter in this regard
14:52 imirkin: yep, a VRP would be nice
14:52 karolherbst: I would even try to help the volunteer out on that :D
14:53 karolherbst: imirkin: wanna to come and visit hamburg between christmas and new years? I totally forgot if I got any answers from you, allthough I am quite sure you won't want to come anyway
14:54 imirkin: no thanks ;)
14:54 karolherbst: :(
14:54 karolherbst: but it is the 33c3!
14:55 karolherbst: just saying, the next year it will be even harder to get tickets
14:58 imirkin: it's a bit of a walk from here... also i have other plans.
15:00 karolherbst: mwk: you are my only hope now
15:04 Lekensteyn: when does the ticket sale start? Last year tickets were sold out before I could get them
15:05 imirkin: sounds like karolherbst has some tickets for you :)
15:05 karolherbst: ;D
15:05 karolherbst: Lekensteyn: https://tickets.events.ccc.de/33c3/
15:05 karolherbst: Lekensteyn: first round was sold out after 30 minutes
15:05 karolherbst: just saying
15:06 karolherbst: webpage felt like a big DDOS
15:07 karolherbst: Lekensteyn: I will try to buy at least 2 anyway
15:07 Lekensteyn: heh, need at least 3 tickets for me and some friends :o
15:08 karolherbst: :D
15:08 karolherbst: I see
15:09 karolherbst: Lekensteyn: I won'T try to buy +3 though
15:11 Lekensteyn: have you tried to buy in the first round?
15:13 mwk: karolherbst: talking to the management about it...
15:14 karolherbst: mwk: yay
15:15 mwk: that may take some time
15:15 karolherbst: mwk: uhh
15:15 karolherbst: mwk: there isn't much left though
15:15 karolherbst: 25.11. is the last chance to buy one
15:15 karolherbst: and then chances are rather slim to actually get one
15:15 karolherbst: I will try to buy two though
15:17 karolherbst: mwk: but if I get one I also plan to do a nouveau assembly there
15:17 karolherbst: should be fun
15:17 karolherbst: some are already interested
23:10 mwk: of course, attempting to scan non-exposed Celsius state results in a hard hang
23:10 mwk: this is going to be annoying.
23:12 mupuf: mwk: the fun begins
23:12 mupuf: Celcius is nv20?
23:14 mwk: NV10
23:19 mupuf: off by one ... generation :D