02:05imirkin: some genius clearly names my files...
02:05imirkin: o well.
02:10curro: gregory38, glennk, imirkin: here's the branch that exposes MESA_shader_framebuffer_fetch if you guys are interested:
02:11curro: basically the only thing missing from mesa master is the extension table entries and a couple of glapi changes... EXT_shader_framebuffer_fetch is exposed already but as imirkin said it's ES-only, the MESA extensions should work on desktop GL as well
02:11imirkin: curro: what does it say about ordering? i think gregory38 was interested in a more limited "critical section"
02:11imirkin: i.e. more ARB_shader_interlock_whatever
02:12curro: imirkin: it's unspecified, the extension doesn't force the whole fragment shader to run in order -- only the portion of the shader that's necessarily data-dependent on the result of a fetch from the framebuffer will be serialized with respect to previous overlapping invocations for obvious reasons (otherwise coherency would buy you nothing)
05:50mlankhorst: skeggsb: did you test https://patchwork.freedesktop.org/patch/132029/ ?
08:44gregory38: curro: thanks for the info.
09:25dibblego: Hi. Why is nouveau 1.0.13-2 crashing on archlinux with the following message https://gist.github.com/anonymous/bca179b9a5f96fd5591c56b8db0730a6 which results in a log out?
14:25RSpliet: gnurou: sorry to pesker you, but you seem to be closest to the source. do you have any news on firmwares?
16:30Lyude: just confirming here: the quadros have some significant difference from the normal geforce GPUs other then just the closed blob drivers right?
16:31imirkin: not that i'm aware
16:31imirkin: i mean, the label on them says "quadro" instead of "geforce". depends on measure of significance?
16:32Lyude: So I'd be able to use geforce GPUs to reproduce issues people are seeing on the quadros from the same generation and vice versa
16:32imirkin: [that said, various hw features were only made available on quadros, like the vsync 3-pin din connector used for the shutter glasses]
16:33imirkin: should be, yeah
16:33imirkin: depends on the issue
16:33imirkin: some issues are that something's wrong in a vbios (or we interpret it incorrectly), in which case you need that specific board (not just the same marketing model)
16:34imirkin: what's the issue btw? just so you don't waste a bunch of time on an issue that we're all well aware of in here?
16:43Lyude: imirkin: none specifically, I'm just making sure we've got all of the GPUs available here that we need and I haven't worked much with nouveau
16:43Lyude: also- I do still want to work on it eventually! sorry I got too sidetracked by other things to work on the stuff you all suggested
16:45imirkin: generally speaking, one gpu per generation is sufficient to work on most issues
16:45imirkin: you don't need full chip coverage
16:45Lyude: perfect, that's exactly what I needed to know
16:46imirkin: as you move earlier and earlier in time, this becomes less true
16:46imirkin: actually, i guess usually families are split into 2 one way or another
16:46imirkin: maxwell1 vs maxwell2
16:46imirkin: kepler1 vs kepler2
16:47imirkin: nvc0 fermi vs nvd0 fermi (diff display units)
16:47imirkin: gt21x vs g8x/g9x (not to mention g80 as being a total weirdo)
16:47Lyude: what would the marketing names be for each one?
16:47imirkin: marketing names don't map well to chips
16:47Lyude: or would they not actually be useful
16:47Lyude: doesn't seem like anyone's do
16:47imirkin: that'd be much too simple
16:47imirkin: that's a rough guide though
16:47Lyude: cool, thanks for the info
16:48imirkin: i think nvidia's sitting on a huge pile of GF108's though
16:48imirkin: so they've just been reselling them as updated GT x10's for a while
16:48imirkin: GT 610, 710 are all GF108's potentially
16:48imirkin: and they played the same game with G84 vs GT218
16:49imirkin: either way, what you need are chips, not marketing names. but when you go on ebay, you see marketing names :)
16:50Lyude: yeah, it's the same thing with amd and older intel
17:02vileperson: Good Morning...
17:11vileperson: A few months ago I was pointed to a patch for xf86-video-nouveau-1.0.12 that had to do with multipl CRTCs, drmmode, and "dirtytracking", that allowed me to have the correct number of CRTCs to use "xrandr" to and get all 3 of my external monitors working...
17:13vileperson: My question for you all is, Did this patch get rolled into the release of Nouveau-1.0.13-2?
17:14ploy: any decent support for nvidia 970 gtx 4gb?
17:14ploy: would i have to enable nouveau as a moudle?
17:17pmoreau: ploy: No decent support for that one, see https://bugs.freedesktop.org/show_bug.cgi?id=94990
17:17ploy: hmm, seems to work fine on gentoo though
17:18imirkin_: ploy: none.
17:18imirkin_: GTX 970 with 4GB of vram won't load with nouveau
17:18imirkin_: oh, unless you don't have the nvidia firmware. in which case you'll get modesetting without accel.
17:19imirkin_: vileperson: yes. https://cgit.freedesktop.org/nouveau/xf86-video-nouveau/log/ - note the patch from me right after 1.0.12 but before 1.0.13
17:20ploy: then what explain gentoo working with gtx 970 without nvidia binary?
17:20imirkin_: you probably don't have the firmware installed
17:20imirkin_: which means it doesn't try to start accel up
17:20imirkin_: so "all is well" (kinda)
17:20imirkin_: except you don't get any sort of accel, naturally
17:21ploy: so in order for me to work, i have to uninstall nvidia frimware?
17:21ploy: *for the card to work
17:21vileperson: imirkin_: O.K. That makes sense. I was wondering how my xranr scripts were still working after Arch gave me 1.0.13
17:21imirkin_: ploy: for some definitions of work, yes.
17:22ploy: how can i remove the firmware? is it a package or?
17:22ploy: *should it be a package
17:22imirkin_: the files in question would be in /lib/firmware/nvidia
17:22imirkin_: just move that directory out of the way
17:26ploy: when is nvidia going to pitch in for support with this card?
17:26imirkin_: i would assume never
17:26imirkin_: and they're unwilling to release sufficient docs for proper nouveau support of it
17:27imirkin_: basically it has a weird memory layout that we're not 100% sure how to properly detect or do with once detected
17:27imirkin_: i wholeheartedly recommend getting an amd board if you're looking for good open-source graphics support
17:29Grotesque: [NV84] system freeze when try to resume from S3, with modprobe.blacklist=nouveau (without this parameter is kernel panic). Netconsole log http://pastebin.com/PpeCfxEQ
17:33ploy: imirkin_: its not that i absolutely need open-source support, but everything is compiled against musl
17:33ploy: on my sys
17:34imirkin_: i don't really care about your reasons... just pointing out the current state of affairs :)
17:34imirkin_: [and my expectations about the future]
17:36ploy: imirkin_: well thanks
17:37ploy: i do have the gtx 740
17:37ploy: 4gb edition
17:37ploy: which worked well with nouveau
17:37imirkin_: yeah, kepler works mostly ok
17:37imirkin_: and with kernel 4.10 you should even be able to reclock it
17:37imirkin_: there's lots of nice things about nouveau. but with maxwell2, nvidia kinda dropped the big one
17:38ploy: gtx 970 is a weird card i guess
17:39imirkin_: well, GTX 970 4GB plain doesn't work with accel right now (only one in the GM20x series)
17:39imirkin_: however nvidia has to release a nouveau-friendly PMU in order for us to be able to control the fan, etc on all the other GM20x's. they have not, and in my estimation are unlikely to.
17:40ploy: i thought they responded to linus's middle finger video
17:40ploy: and are working harder with kernel devs
17:41ploy: was in 2012 though
17:41imirkin_: well, there are developers who are improving open-source support of the tegra gpu's
17:42imirkin_: and they're generally nice guys and i believe act as something akin to our advocates within the company. but they don't exactly have a lot of executive power.
17:42ploy: i see
17:45ploy: ill just rm my 970 and rpelace with gt 740
19:24Lyude: imirkin_: so maxwell1, maxwell2, kepler1 and kepler2 all have significant differences? as well, do you know what ccards fall into each category? I don't see anything on the codenames page on fdo to that I can use to tell
19:25imirkin_: Lyude: maxwell2 = GM20x, maxwell1 = GM10x, kepler2 = GK110 + GK20x, kepler1 = GK10x
19:25Lyude: imirkin_: cool, thanks
19:26imirkin_: fermi is mostly the same, but GF119 has a different display unit than the rest of the GPUs (which may or may not matter to what you're hoping to achieve)
19:27imirkin_: there can be additional subtle differences, which might matter again depending on what you're trying to do
19:27imirkin_: if you want to mess around with reclocking, then obviously different memory types are important to have as well
19:28imirkin_: but that's mostly a numbers game - just get a bunch of boards and pound on your keyboard until they all work
19:37Lyude: imirkin_: ah, most of what I'll be doing is just testing to make sure things work
19:44imirkin_: Lyude: well, as with everything, depends on the precise things you want to test, and how thorough you want to be :)
19:44imirkin_: if you want to be extremely thorough, get every board ever released by every manufacturer
19:44imirkin_: it then scales down from there :)
19:44Lyude: hehe, I want to be as thorough as I can without going overboard
19:45Lyude: luckily it's not my money being spent :3
19:45imirkin_: oh, not to mention getting every display ever made, and a giant switch in between ;)
19:45imirkin_: or, to be safe, just get one of each display for each board in existence
19:46airlied: just program chamelium to be broken like every display :-P
19:46imirkin_: but yeah. on a serious note, i think having a fermi1, fermi2, kepler1, kepler2, maxwell1, maxwell2, pascal1 should cover most of your testing
19:46imirkin_: both on the 3d end and on the display end
19:46imirkin_: it gets a bit more subtle with tesla if you're futzing with display things
19:47imirkin_: (where fermi2 is defined as GF119 btw, since getting a GF117 will be rather annoying and it doesn't have a display anyways)
20:05Lyude: Would reverse engineering power management for >kepler still be an option as a GSoC project? I might have actually met someone who would be interested in doing it
20:06Lyude: they might also be interested in other nouveau stuff, trying to get them in here
20:06imirkin_: doing what specifically?
20:06imirkin_: either way, i'm not the best person to talk to re PM
20:06imirkin_: mupuf or karol would be better
20:06Lyude: that's what we're trying to figure out, they've definitely gotthe skillset required for working on this
20:07imirkin_: and are a college student?
20:34RSpliet: Lydde: Talk to me in about 2,5 hours
20:34RSpliet: or with karolherbst earlier ;-)
20:34Lyude: sure thing! I think they're still in a lecture right now anyway
20:34imirkin_: pfft. overrated.
20:34RSpliet: They can e-mail the mailinglist as well
20:51karolherbst: what's the problem?
20:51imirkin_: read logs from 30s before you joined
20:53karolherbst: I might would have somebody who would like do to one to, so somebody else would have to take care of him/her
20:53karolherbst: and there are still doztens of pm topics left
22:10Lyude: karolherbst: ^ that's the gsoc student
22:10Lyude: also hey Satchelboi_
22:12Satchelboi_: Hello ^^
22:12Lyude: s/gsoc student/student interested in gsoc/
22:12Lyude: fyi, people also brought up that it might be a good idea to get on mailing lists
22:13Satchelboi_: Certainly doesn't hurt.
22:13karolherbst: trello is a good place to start
22:13karolherbst: there are many open things there
22:13karolherbst: if something interests you, you could work on that for a gsoc project in theory
22:14Satchelboi_: I'll browse through quick and see if there's anything I can be of service to
22:15Lyude: remember too it's okay to pick something you don't know a whole ton about, the most important part of GSoC is that you learn
22:17Satchelboi_: I could try working on the maxwell shader delay fixes
22:18karolherbst: mhh, interesting topic
22:18karolherbst: here is my idea for a gsoc project regarding this: write a tool, which is able to figure all latencies between instruction out itself for a given ISA
22:19karolherbst: and tries to detect "groups" of instructions by itself as well
22:19hakzsam: karolherbst: he is talking about adding maxwell control codes to the DDX
22:19hakzsam: it's on my "todolist" though but I don't have time :)
22:19karolherbst: but this wouldn't be enough work for a gsoc project, would it?
22:19hakzsam: definitely not
22:19hakzsam: but a good simple task for a newcomer
22:20karolherbst: for warming up for a gsoc project it would be okay
22:20karolherbst: I was already thinking about what a good gsoc project would be :D
22:20Satchelboi_: Some warming up for me wouldn't hurt at all, believe me.
22:20hakzsam: the instruction scheduling pass is one
22:21hakzsam: (I mean a good gsoc project)
22:21karolherbst: this as well
22:21karolherbst: but I would say that figuring out the instruction latencies is a bit more important so I would suggest this before the scheduling pass
22:22hakzsam: well, as I already said we know them on maxwell at least :)
22:23hakzsam: Satchelboi_: if you are interested for working on maxwell control codes https://github.com/NervanaSystems/maxas
22:23hakzsam: and you can ask me if you need more info (I added them in our compiler)
22:25Satchelboi_: I'll take a look at them right now
22:25karolherbst: at least this is mesa stuff, so I am not needed as a mentor here :D
22:26hakzsam: and in the compiler :)
22:26hakzsam: not for the DDX though, but same deal
22:26hakzsam: Satchelboi_: DDX is here https://cgit.freedesktop.org/nouveau/xf86-video-nouveau/
22:26hakzsam: look for "st 0x0"
22:27karolherbst: moving the compiler into a seperate project would be also a good thing
22:27hakzsam: (the default very bad control code) :)
22:27hakzsam: for what? vk?
22:27hakzsam: I think skeggsb is just going to copy&paste for now
22:27karolherbst: and be able to compile stuff for the ddx as well
22:28karolherbst: we shouldn't have two copies of the code
22:28karolherbst: would end up in something terrible
22:28hakzsam: ddx uses envyas IIRC
22:28imirkin_:recommends talking to /me about any potential compiler-related projects
22:28hakzsam: via some makefiles
22:28karolherbst: yeah, but it isn't optimized in any way
22:28hakzsam: imirkin_: ahah, yep! except maxwell sched codes :p
22:29imirkin_: hakzsam: the sched code stuff is in a pretty good place now though, no?
22:29hakzsam: it should
22:29Satchelboi_: It actually may help if it's not well optimized, since the steps are generally easier to read. Useful for a beginner like me.
22:30imirkin_: however i'd also recommend having some modicum of compiler experience before undertaking a compiler-related GSoC
22:31karolherbst: Satchelboi_: yeah, but a compiler could take the code and optimize it
22:31hakzsam: DDX code is optimized by hands, isn't imirkin_?
22:32imirkin_: yeah, but optimizing that is a 1-day job, not exactly GSoC material
22:33hakzsam: writing the intrusction scheduling pass should take enough time for a GSoC
22:33imirkin_: if not two :)