00:03 mooch: mwk: also, the git on 0x04.net doesn't work. it just leads to an nginx installation page
00:06 Lyude: how exactly did you guys figure out which value ranges mmt register values accept? does the GPU just throw an error if you wrote something invalid?
00:07 skeggsb: Lyude: yeah, the class will generally perform some level of error checking (both when setting individual methods, and on some global state when hitting a "launch" method)
00:08 Lyude: ah, I guess I was wondering if and how I could get some hints from the GPU that a register I'm writing to doesn't do anything where I'm trying to write to it
00:09 Lyude: (I am currently working on trying to figure out what we're missing for https://trello.com/c/IAqZXMmt/154-gm200-amd-vertex-shader-layer-viewport-index )
00:10 skeggsb: also, *not* registers :P methods. they're turned (either by a hw class interface, or fw on some engines) into actual register writes eventually, but the userspace interface isn't register-level
00:10 Lyude: ah, I know they're not technically registers but force of habit :P
00:28 mwk: mooch: what git? where is it linked to?
00:31 mooch: 0x04.net links here http://cgit.0x04.net/
00:31 mwk: where from?
00:31 mooch: 0x04.net
00:31 mooch: it redirects here: https://wiki.0x04.net/wiki/Main_Page
00:32 mwk: ok, but where did you find that link?
00:34 mooch: i just typed in 0x04.net into the address bar
00:34 mooch: because i know that's your website
00:37 mwk: ahh, it's on the main page
00:37 mwk: alright, that's strange
00:38 mwk: probably half the links here are broken either way...
00:42 mooch: yeah
03:13 dboyan_: imirkin: You may want to look at https://github.com/envytools/envytools/pull/84 and https://github.com/envytools/envytools/pull/90 if you have time.
03:14 dboyan_: imirkin: The first one is register definition correction, the second is needed my fp64 series
03:54 tstellar: pmoreau: If you want to play around with this, I'm probably not going to have much more time to work on it: https://cgit.freedesktop.org/~tstellar/mesa/log/?h=guda
03:55 tstellar: pmoreau: Biggest missing piece is ptx -> ISA, otherwise there should be enough there to get very simple programs working. It's untested though, so expect bugs.
03:56 tstellar: RSpliet: ^
06:57 pq: ~user@gateway/tor-sasl/snatcher is asking to get unbanned here, can we do that?
07:26 marcheu: pq: yeah he asked me too, not without unbanning all of tor-sasl, including joss
07:26 marcheu: so..
07:52 pq: marcheu, I've been told that one needs to create an account in tor(?) before they can even connect to irc, and claimed that we should bad tor users per account.
07:53 pq: *ban
07:54 marcheu: won't joss create an infinite number of accounts then?
08:00 pq: is it that easy?
08:02 pq: marcheu, hm, maybe we could put ban exceptions on selected freenode registered nicks?
08:02 pq: or something?
08:18 karolherbst: for creating an account for using freenode over tor, you need to first connect via non tor and create it
08:41 pq: mind if I set up a ban exception for this tor-sasl user?
09:04 karolherbst: pq: as long as the user is nice, nobody will complain
09:05 pq: just remember it's a ban exception you need to remove, not a ban to add, if it goes south :-)
09:06 karolherbst: I can't ban anybody anyway
09:06 pq: you can remind me, I'll probably forget myself
09:11 pmoreau: tstellar: Awesome! Thanks a lot :-)
09:11 pmoreau: tstellar: And I am sure our Dutch friends will love the name you picked! ;-) RSpliet
09:14 karolherbst: I get the feeling, that more and more people are working on Nouveau over the time :) I like that or is it just me and it was like that before already
09:21 pmoreau: karolherbst: It is a rMBP. What would you like me to test on it?
09:21 pmoreau: Thought I will most likely not have time to do that before the weekend.
09:22 karolherbst: quite easy: current Nouveau set to highest Pstate -> check PCIe LnkSta (should be 5.0 GT/s), preferable check every Pstate.
09:22 karolherbst: Then do the same for Nvidia
09:22 karolherbst: Nvidia should go to 8.0 GT/s on highest pstate
09:22 karolherbst: then apply my patch: https://lists.freedesktop.org/archives/nouveau/2017-April/027637.html
09:22 karolherbst: with this Nouveau should do the same Nvidia does
09:23 karolherbst: ohhh wait, this patch is wrong
09:23 karolherbst: meh
09:23 karolherbst: I will update the path later this day
09:23 karolherbst: case 0x21: is missing and it will also result in 2_5
09:28 pmoreau: Ok, will make a note to check that
09:29 karolherbst: awesome, thanks :)
09:33 pmoreau: tstellar: Not sure if NV ISA is also stored in the application binary, but I think the kernels’ executables can be found in the binary (at least when using nvcc) depending on the compile flags.
09:54 dboyan_: pmoreau: I guess you need to do recompile if the kernel binary doesn't match current architecture
10:00 RSpliet: pmoreau: guda? gouda! :-P
10:09 mac-: hey guys
10:13 mac-: I'm looking now on Feature Matrix and Video Acceleration Matrix
10:13 mac-: and from FM I see that nv40 has DONE status for VDPAU/XvMC
10:14 mac-: but when I look at VAM I see that VP1 (which part of is nv40 series) has only N/As and TODO statuses on everything ?
10:17 mac-: then I'm little confused
10:19 mac-: I have quite old PC system which has AGP and PCI slots and I wish to get some video card which has video acceleration support to play videos there
10:21 karolherbst: mac-: there are also modern PCI cards
10:21 mac-: any examples ?
10:22 karolherbst: checking
10:22 mac-: sure
10:22 mac-: take your time
10:22 mac-: currently I've got FX5200 but it is unsupported or have nothing implemented
10:23 karolherbst: GeForce GT 710
10:23 karolherbst: ohh wait, is that PCIe as well?
10:23 karolherbst: dunno
10:23 karolherbst: or prefer an AMD card with PCI
10:23 karolherbst: there is stuff newer than NV40 though
10:26 karolherbst: GeForuce 210 for sure, which is GT216
10:26 karolherbst: or GT 430 which is GF108
10:27 karolherbst: there is also a GT 630 which is a GF108 as well
10:28 karolherbst: mac-: but usually you should prefer AMD GPUs if you have to buy something new and don't care about supporting Nouveau :p We are wroking on making the driver stablier and everything else, but you should get better open source support from AMD
10:30 mac-: new ?
10:30 karolherbst: well, if you buy something
10:30 mac-: I wish to buy sth quite old but with opensource support
10:30 koz_: To totally contradict what karolherbst said: nouveau has been great to me and has very kind devs.
10:30 karolherbst: true, the devs are kind :3
10:30 koz_: And my Kepler is still serving me well, and I still have room to grow into another better one.
10:30 karolherbst: like I would say anything else
10:31 koz_: (once it stops costing the Earth)
10:31 mac-: the system I'm talking about is Sempron 2200+ with AGPx8 :]
10:31 karolherbst: ohh AGP
10:31 karolherbst: well
10:31 pmoreau: dboyan_: Right, but that’s why they also ship the PTX in the application binary, to my understanding.
10:31 mac-: AGP or PCI 32bit
10:31 karolherbst: mac-: simple: there are newer PCI cards than there are AGP cards
10:32 karolherbst: koz_: and don't get me wrong. for most users it's quite good and so on, just AMD provides better support, that's all
10:32 mac-: ok I don't care about 3D performance and etc, I rely on video acceleration only tbh, as I know it is VDPAU hardware support
10:32 karolherbst: if the user decides to take the risk and want to help out in supporting Nouveau, then they can go ahaed
10:32 karolherbst: mac-: then any Fermi based NVidia GPU should do
10:32 karolherbst: preferable Kepler, but I doubt there is any PCI version
10:32 mac-: AMD has better support ? are you kiddin me? they drop support quickly
10:33 karolherbst: for older GPUs?
10:33 koz_: I have *never* gotten anything but pain out of AMD anything.
10:33 karolherbst: okay, can't say for older ones
10:33 koz_: GPU, CPU, whatever - it's been one parade of fail.
10:33 koz_: s/one/one long/
10:33 mac-: I have E-350 based netbook, no more support for it built-in Radeoin HD6310
10:33 karolherbst: :D
10:33 karolherbst: crazy
10:33 karolherbst: with the open drivers?
10:33 koz_: karolherbst: I've hit *init firmware* bugs with AMD.
10:33 koz_: On more than one occasion.
10:33 mac-: opensource drivers are sth separate than AMD drivers
10:33 karolherbst: koz_: the open driver?
10:33 koz_: karolherbst: CPU.
10:33 karolherbst: yeah, I am only talking about open drivers
10:34 karolherbst: ohhhhhh
10:34 karolherbst: well
10:34 karolherbst: I only talk about the GPUs :p
10:34 koz_: The GPUs either ran like hot garbage or not at all.
10:34 karolherbst: well with Nouveau it's not much better
10:34 mac-: yeah thank to opensource drivers I can still work on the netbook, but I've lost some poer management features
10:34 karolherbst: we don't really reduce power consumption
10:34 koz_: karolherbst: It was for me.
10:34 karolherbst: so under Nouveau Nvidia GPUs drain more power
10:34 koz_: karolherbst: A device that doesn't work drains zero power. :P
10:34 karolherbst: !
10:35 karolherbst: see
10:35 karolherbst: mac-: sad :(
10:35 karolherbst: mac-: well, modern chips should be well supported as well, but I don't really know the details for older GPUs
10:35 koz_: But yeah, go with a Nouveau-supported card - I've yet to see any personal experience to recommend AMD anything.
10:35 koz_: s/see/have/
10:36 karolherbst: koz_: for modern high end cards, AMD is to be prefered, I think we can settle with this :p
10:36 koz_: karolherbst: I can't really say personally - I've never needed that kinda firepower.
10:36 karolherbst: they provide better performance
10:36 koz_: That being said, I'm not trying to drive a AAA game on a 4K monitor at max settings.
10:36 karolherbst: which is important to most people with modern and fast cards
10:36 koz_: So perhaps my use case is somewhat different.
10:36 karolherbst: yeah, true
10:37 karolherbst: I can't talk about stability either
10:37 pmoreau: RSpliet: ;-)
10:37 karolherbst: but their devs get paid
10:37 karolherbst: Nouveau devs.... are not except one
10:37 koz_: Yeah, sadly.
10:37 RSpliet: karolherbst: don't forget the pay for the GSoC students ;-)
10:37 karolherbst: !
10:37 koz_: There needs to be an eccentric millionaire who loves Nouveau.
10:37 karolherbst: true
10:38 karolherbst: mhhhh
10:38 karolherbst: good thought
10:38 koz_: Or a *billionaire* to just buy the company and give equal stock to all the core Nouveau devs.
10:39 RSpliet: imirkin_: speaking of big GPUs by the way, here's the one that didn't fit in my PC http://www.ascendtech.us/mmASC/Images/456139001-02.jpg
10:40 RSpliet: PCIe x16 connector for scale
10:40 karolherbst: buy a better case :p
10:40 RSpliet: karolherbst: weren't we *just* talking about not getting paid?
10:40 mac-: like VLB on '90s :D
10:40 karolherbst: a case costs like 30€ roughly
10:41 karolherbst: and if you choose the wrong one to begin with, it's your fault :p
10:41 RSpliet: karolherbst: here's the monitor I use
10:41 RSpliet: https://cdn.globalauctionplatform.com/99aea2d3-8b69-4797-b772-a4fa01003f6b/ca01dacd-0477-45d6-b1bc-ce8adedd7326/original.jpg
10:41 RSpliet: it's 15 years old
10:41 RSpliet: not because I want to
10:41 pmoreau: Who needs a case anyway! :-p
10:41 koz_: pmoreau: Yep, naked computer = best computer.
10:41 karolherbst: RSpliet: well, you need to find a new job inside the EU anyway :p
10:41 RSpliet: pmoreau: okay you have a point there
10:42 koz_: Also, my *whole room* smells of pizza.
10:42 mac-: http://gordogato.com/oscommerce/catalog/images/zm_vlb_videocard_1.jpg
10:42 mac-: :D
10:42 koz_: (and not because I ate one recently, which is especially unfortunate)
10:42 karolherbst: pizza smells make computers faster, everybody knows that
10:42 pmoreau: :-D
10:42 koz_: karolherbst: Maybe I should order like, a hundred pizzas. Then my research code will finish in less than five days.
10:43 karolherbst: koz_: no, the max effect isn't that high actually, to improve perf any further, you also need LEDs
10:43 koz_: karolherbst: Pizza and LEDs are the keys to performance, got it.
10:43 pmoreau: koz_: Next step, write a grant request for buying pizzas! :-D
10:44 karolherbst: koz_: and the computer has to be below ground level for maximun effect
10:44 koz_: Tech news: NVIDIA releases GeForce PizzaX1000. With added LEDs. Performance outstrips the Titan Black tenfold. People astounded.
10:44 koz_: (and of course, it runs signed firmware)
10:44 karolherbst: please keep sunlight away as well
10:45 pmoreau: koz_: Not signed firmware, signed pepperoni.
10:45 RSpliet: does pizza abbreviate to PX?
10:45 koz_: pmoreau: So you can't eat it without NVIDIA's signing key?
10:45 koz_: RSpliet: I guess so.
10:45 RSpliet: that explains the power of the Drive PX2
10:45 pmoreau: Exactly!
10:45 koz_: pmoreau: This is the disruptive future of the next pizza startup - DRM on toppings.
10:46 pmoreau: Damn…
10:46 koz_: It *will* happen, I'm pretty sure.
10:46 pmoreau: And if you are ready to pay more, you can have an early access to the pizza?
10:46 koz_: pmoreau: And there'll be a GameStop special.
10:47 koz_: (and a Collector's Edition too)
10:47 koz_: (where you get blue pepperoni or something)
10:47 pmoreau: Digital edition: to make use of your 3Dprinter
10:48 koz_: I've eaten 3D-printed food.
10:48 koz_: (chocolate cookies)
10:48 pmoreau: How was it?
10:48 koz_: pmoreau: Basically just like a non-3D-printed chocolate cookie.
10:49 pmoreau: disappointing! :-D
10:49 koz_: Although to be honest, I could totally see the use of 3D food printers for really fiddly dishes.
10:49 koz_: (like those really fancy decorated pastries or something)
11:07 tstellar: pmoreau: The function extract_binary() I think will give you the actual ISA.
11:09 pmoreau: tstellar: Thanks for the pointer. I’ll try to play with it a bit over the weekend. :-)
11:34 mac-: hm
11:34 mac-: I've just realized that I have nvs280 PCI card
11:34 mac-: but it is old and unsupported as well :/
11:44 haagch: gnurou: do you work on other parts of the mainline tegra stack too? e.g. cpufreq seems to still be disabled on tegra k1
11:53 karolherbst: no, he doesn't
12:12 RSpliet: haagch: gnurou does not work on any part of tegra
12:17 haagch: Not anymore?
12:18 RSpliet: haagch: he notified us of his resignation last friday
12:19 haagch: I did not know that. Does anyone at nvidia take it up?
12:20 karolherbst: who knows
12:21 karolherbst: no IRC communication no merges :p
12:53 imirkin: mac-: nv40 should give you MPEG2 acceleration via XvMC
12:55 imirkin: mac-: the first nvidia chip with competent h.264 video decoding accel was G84
13:11 mac-: imirkin: hm, but this is pci-e only
13:15 imirkin: mac-: no, they make PCI versions of many of the modern cards
13:15 imirkin: they're a bit rare, but definitely out there
13:23 karolherbst: mac-: As I said, the most modern I found are Fermi Cards, and they are quite enough for video acceleration
14:11 pmoreau: Eh, just got a graphics programming book with a good chunk of it dedicated to VGA programming… :-D
14:26 Satchelboi_: karolherbst: Up and ready to start on some work! Did you find any patch ideas to work on that can help me for the next ones?
14:28 karolherbst: Satchelboi_: I have a lot for the community bonding phase, but nothing for the first patch
14:28 karolherbst: I think something super easy is the best here
14:28 karolherbst: the better stuff we can do then
14:28 mac-_: hm, I took a look at Radon family of cards but there video acceleration is also only on PCI-e models
14:28 mac-_: and AGP ones are quite expensive
14:32 karolherbst: Satchelboi_: ohh wait, I should learn to read more carefully, sorry for that
14:32 karolherbst: Satchelboi_: so yeah, I have a few ideas what you could after
14:32 imirkin_: pmoreau: the black book? :)
14:32 Satchelboi_: karolherbst: Its fine. Im looking back through the logs right now for the recommendations that were made the other day too
14:33 pmoreau: imirkin_: yup :-)
14:33 karolherbst: Satchelboi_: but in my opinion, community bonding should start as soon as possible :p and being on IRC actively is a good first step already!
14:33 imirkin_: i remember i had that... so long ago... i wonder if it's still at my parents house or something
14:34 pmoreau: I have to go through it at some point, see how it worked back then
14:34 imirkin_: it was mostly beyond me when i tried to make my way through it... maybe 13 or 14?
14:35 Satchelboi_: karolherbst: I'm happy to work on that too. I just need to make sure I don't screw up the first assignment I have
14:35 karolherbst: yes
14:35 karolherbst: you can always ask for suggestions in IRC though
14:37 imirkin_: Marex: manio: mrcooper sent a patch for your xinerama issues... can you guys give it a whirl?
14:48 dboyan: imirkin_, how do I set an predicate input to an instruction as true (PT in nvdisasm)
14:49 dboyan: imirkin: ^
14:59 Satchelboi_: karolherbst: How about I try changing some constants and see what happens like you recommended the other day?
15:03 Marex: imirkin_: I saw it
15:04 Marex: imirkin_: mrcooper is Michel Danzer the weirdo AMD guy from Japan ? :)
15:04 imirkin_: s/weirdo//, i think the rest is accurate :)
15:04 imirkin_: (only 90% sure he's in Japan, but it does sound familiar)
15:05 Marex: imirkin_: hehehe
15:06 dboyan: imirkin_: did you see my question above?
15:06 imirkin_: dboyan: i have now. hexchat crashed on me, but i see it in the log.
15:06 imirkin_: dboyan: P7 == PT
15:06 karolherbst: Satchelboi_: go ahead, just be sure to not spend too much time into this. If you let say find a constant which breaks stuff within 1-2 hours, that's fine. Or you just find a file where too many magic numbers are used and try to figure out what it means and give a proper name or so
15:07 Marex: imirkin_: quasselirc or irssi on a server wfm :)
15:07 imirkin_: (just as R255/R63 == RZ, depending on ISA)
15:07 karolherbst: Satchelboi_: but yeah, I would prefer something like this over whitespace fixes
15:07 karolherbst: Satchelboi_: just keep the time aspect in mind
15:07 imirkin_: Marex: first time i've seen hexchat crash. in over a decade of use of it and xchat.
15:07 karolherbst: we can do something like this also later
15:08 Satchelboi_: karolherbst: Little less of a low hanging fruit. I'll work on it now then and try to have it submitted for review by saturday, after running it by you.
15:08 karolherbst: Satchelboi_: like you should post a patch for to the Mailing List this weekend as the latest, otherwise reviewing might take too long
15:09 Satchelboi_: I was told it could take up to a week, so saturday is a week from the deadline.
15:10 karolherbst: Satchelboi_: yeah
15:11 karolherbst: so, it would be good if you post a patch on Saturday then
15:11 karolherbst: Satchelboi_: CC me and I will (try to) review it on Sunday, but I have time on Sunday, so chances are high I would review it
15:12 karolherbst: *will
15:12 karolherbst: over the week is bad, cause I have to work then
15:13 Satchelboi_: karolherbst: I'll note to do that too then. I can't imagine it will have a ton that needs to be reviewed in it either
15:13 karolherbst: Satchelboi_: currently I would have to do that fully in my spare time, just that you know, but I still try to respond to anything within the next 36 hours
15:13 dboyan: imirkin_: then is it correct to use FILE_PREDICATE new_LValue and set its storage as reg number 7 to get PT?
15:13 imirkin_: dboyan: probably not. what are you trying to do?
15:13 Marex: imirkin_: whatever works for you, I switched from xchat to irssi in screen on a server
15:14 imirkin_: dboyan: you don't have to model the whole instruction if you don't want to...
15:14 imirkin_: dboyan: or is this for the vote result?
15:15 dboyan: imirkin_, I'm trying to implement readFirstInvocationARB, which uses a ballot(true), and it should be like "vote any $r 0x1 0x1"
15:15 imirkin_: rrrright
15:15 Satchelboi_: karolherbst: Don't worry about it too much if you can't end up getting to it. Preferably I'll have it done before its even saturday
15:15 imirkin_: so from a modelling standpoint
15:16 imirkin_: what we want is to be able to feed it a "true" value
15:16 imirkin_: and then use the LoadPropagation to insert it there.
15:16 karolherbst: Satchelboi_: and there are others who can review it as well :p
15:16 imirkin_: ultimately it should be an immediate
15:16 dboyan: imirkin_, so i just need a OP_SET there?
15:16 imirkin_: to make your life simpler, you can just feed it the immediate directly
15:16 karolherbst: Satchelboi_: ping imirkin_ if in doubt as well, he merges most of the nouveau patches
15:17 imirkin_: and then have a thing in the emitter which emits PT if there's an immediate == 1 there
15:17 imirkin_: and emits !PT when it's 0
15:17 karolherbst: Satchelboi_: and he lives in the same timezone as well, I think
15:17 imirkin_: and asserts when it's something else :)
15:17 dboyan: imirkin_: That's a lot simpler, I guess
15:18 Satchelboi_: karolherbst: I can ping him and someone else if it's at a bad time for you
15:18 imirkin_: that way, when the next gen comes out with 16 predicates instead of 8, it'll keep on working.
15:18 imirkin_: Satchelboi_: remind me what you're working on and/or trying to do in the short term?
15:19 Satchelboi_: imirkin_: Right now Im just trying to put together a trivial patch and submit it as a competency test. It looks like that patch will be changing constants and fixing anything that could break as a result
15:19 imirkin_: erm ... which constants?
15:21 Satchelboi_: imirkin_: I'm, uh, not really sure yet. The general guideline was to find some somewhere, change them around and see if anything happens, then patch it if it does.
15:21 Satchelboi_: Im honestly a little lost on where to start
15:21 imirkin_: ok, well, randomly changing stuff around for the sake of changing stuff isn't the way to go.
15:22 imirkin_: which GPU(s) do you have access to?
15:23 karolherbst: imirkin_, Satchelboi_ : I would prefer that you could change something inside nvc0/nvc0_video.h
15:23 karolherbst: ohh wait, there are no constants
15:23 karolherbst: ,eh
15:23 imirkin_: erm ...
15:24 Satchelboi_: imirkin_: At this moment, nothing nvidia based. But if I need something I can bug Lyude and borrow one of hers for a weekend potentially.
15:24 imirkin_: most of the logic is in nouveau/nouveau_vp3_*
15:24 imirkin_: as it's shared by all the vp3+ stuff
15:24 Lyude: yeah, I'm willing to get you dedicated hardware as well if needed
15:24 Lyude: anything I can do to help out students is A-OK with me
15:24 imirkin_: (vp2 lives alone in its own sad little world in nv50/nv84_*)
15:25 imirkin_: Satchelboi_: not necessarily. just wanted to know what you had easy access to.
15:25 imirkin_: so that i might recommend a task
15:25 karolherbst: Lyude: good to know. I will (most likely) take over the GSoC mentoring for Satchelboi_
15:26 imirkin_: Satchelboi_: can you self-evaluate your C and C++ knowledge on a scale of 1 to 5?
15:26 Lyude: karolherbst: cool! I would have volunteered to be a mentor this year but me and Satchelboi_ know eachother IRL, so I didn't think it'd be fair for me to grade them
15:27 Satchelboi_: I'd put my C at a solid 2.5-3, but I haven't really done any C++ work before. Most of my programming comes over from microcontroller use
15:27 imirkin_: like 8051 or something?
15:27 karolherbst: Lyude: you could Co-mentor :p
15:28 karolherbst: Lyude: yeah, and I was already a little worried, that I would take away that mentoring part from you, cause it sounded a little like you would like to do that
15:30 Lyude: hehe, it's no problem. plus I never actually signed up as a mentor, so
15:30 Lyude: i just like helping out people getting involved, but i'd be fine with co-mentoring yeah
15:31 imirkin_: dboyan: http://docs.nvidia.com/cuda/parallel-thread-execution/index.html#special-registers-clock
15:31 karolherbst: Lyude: okay, nice, because it would be my first mentoring as well, and I would gladly accept any help here
15:31 Lyude: i should note I have also never mentored in the past either :P
15:32 Satchelboi_: karolherbst: Lyude can give me a punch or two if I fall behind on anything too :P
15:32 karolherbst: !
15:32 karolherbst: good idea
15:33 dboyan: imirkin_: so you think there is a way to read out the whole 64-bit clock?
15:33 imirkin_: dboyan: well, at least in PTX-land, it's just the 2 regs...
15:33 karolherbst: Satchelboi_, Lyude: I guess it wouldn't be any problems for you to come to the XDC this year?
15:33 imirkin_: dboyan: and if you look below, globaltimer is a NS-based clock
15:34 imirkin_: isn't XDC in silicon valley?
15:34 Lyude: karolherbst: yeah I'm already planning on coming
15:34 karolherbst: googleplex, yes
15:34 imirkin_: that's a bit of a walk from boston...
15:34 Lyude: hehe
15:34 Satchelboi_: karolherbst: I'd have to see if I can save enough to make it out there
15:34 karolherbst: Satchelboi_: .... :P by then you got the money from google already
15:34 Lyude: Satchelboi_: if you pass GSOC X.org will sponsor your travel fees
15:34 Lyude: at least we used to
15:35 Lyude: I'd imagine that's still the case though
15:35 karolherbst: yeah, I think as well
15:35 Satchelboi_: karolherbst: Yes, and many other bills to pay off haha
15:35 karolherbst: I heard rumours about that there is money for that kind of stuff
15:35 karolherbst: Satchelboi_: you would also give a presentation
15:35 Satchelboi_: karolherbst: It's an excellent thing I love presenting then!
15:35 dboyan: imirkin_: yeah. But I guess the ptx get lowered into 32-bit reads
15:36 imirkin_: dboyan: probably. you can always check with ptxas :)
15:36 karolherbst: Satchelboi_: see, and I think presenters have also special rights for getting money for traveling and so on
15:36 imirkin_: dboyan: you can look at how ptxgen works in envytools
15:36 karolherbst: Satchelboi_: it's end of September, so enough time for everything
15:36 Satchelboi_: karolherbst: I'll keep a close eye on it and how the funding will work, especially as it gets closer.
15:36 dboyan: imirkin_: I know it's somewhere in ptxgen
15:37 karolherbst: yeah, details for later
15:37 dboyan: imirkin_: Do you know what the predicate output of SHFL means?
15:37 imirkin_: dboyan: not a clue.
15:38 imirkin_: does blob use it?
15:38 dboyan: imirkin_: The first 0x1 in "shfl idx 0x1 $r1 $r0 0x2 0x1f", I guess this means "not using it"
15:39 imirkin_: dboyan: yep
15:39 Satchelboi_: karolherbst: All fun stuff aside, I still need to pass this first step, and I'm not sure where a good file to start working on constants is
15:39 imirkin_: Satchelboi_: i'm tryign to think of something good
15:39 karolherbst: Satchelboi_: ask imirkin_ , hey knows the mesa part best (especially for video :p )
15:40 dboyan: imirkin_: And I wrote some code, went on testing on my gk208, and finally found there wasn't a emitSHFL for gk110 :/
15:40 imirkin_: dboyan: well, should be easy to write
15:40 Satchelboi_: imirkin_: That's fine, just poke me if you think of something that would work
15:41 karolherbst: Satchelboi_: do you have everything prepared already? Like building mesa and anything else?
15:41 dboyan: imirkin_: yeah, but I guess I also have to set the predicate output to 7 gm107's in emitSHFL
15:41 karolherbst: being able to run something through your own built mesa, etc..
15:41 imirkin_: dboyan: probably yea
15:41 imirkin_: dboyan: probably already done for you
15:42 imirkin_: dboyan: emitPRED (0x30); -- that just emits a PT there
15:42 dboyan: yep
15:42 imirkin_: dunno what this is - emitField(0x22, 13, 0x1c03);
15:42 Satchelboi_: karolherbst: I spent yesterday setting up envytools, I have mesa downloaded, but I just need to build the package
15:43 Satchelboi_: I'm running this all on a virtual machine, so its a bit slow.
15:43 imirkin_: dboyan: probably needs some more work with nvdisasm. i dunno if anything uses it atm.
15:43 karolherbst: Satchelboi_: will be enough fun already :p
15:43 mac-_: but please tell me
15:43 karolherbst: Satchelboi_: any reason why you do this in a virtuel machine? Using your real GPU there is kind of.... difficult
15:43 mac-_: why on https://nouveau.freedesktop.org/wiki/FeatureMatrix/ NV40 has status DONE when on https://nouveau.freedesktop.org/wiki/VideoAcceleration/ NV40 doesn't have anything as DONE ?
15:44 dboyan: imirkin_: okay I'll check it out later
15:44 karolherbst: mac-_: never forget that people can be lazy as well
15:44 Satchelboi_: karolherbst: It was for ease of use testing right now, since I don't have any dedicated machine I can test on atm. I'm going to set up dual boot on my desktop this weekend to help
15:44 imirkin_: mac-_: VPE2 works fine on nv4x.
15:44 karolherbst: Satchelboi_: yeah, please do that. You will need a dedicated machine you can work on
15:44 imirkin_: [ok, there's a semi-recently introduced but that makes nv40-nv43 not work, but that's being fixed]
15:45 imirkin_: bug*
15:45 Satchelboi_: karolherbst: That part should take no time at all either, I've done that many times
15:45 karolherbst: Satchelboi_: also, before I forget to ask that later: I hope you have no other plans during your GSoC work periods
15:45 imirkin_: mac-_: the situation is a little confusing as a single card has multiple engines, in varying states of support. i tried to make the videoaccel page as accurate as possible, but in the process probably made it difficult to understand.
15:46 Satchelboi_: karolherbst: If you mean other jobs, then I don't have anything. I refrained from looking for any since I wasn't sure what the time requirements are
15:46 karolherbst: Satchelboi_: time requiernment is: 40 hour weeks
15:46 mac-_: imirkin_: I have nv34 (FX5200) and nv86 (nvs285)
15:46 karolherbst: it's basically a full time job
15:47 imirkin_: mac-_: the G86 should be much better supported on almost every level of everything involved.
15:47 Satchelboi_: karolherbst: I can meet that right now, so that's not a worry
15:47 karolherbst: Satchelboi_: and usually, proposals are rejected, if the student has internships or anything like this at the same time
15:47 imirkin_: mac-_: however my records indicate NVS 285 is actually a NV44, not a G86...
15:47 mac-_: imirkin_: then nv34 doesn't support almost nothing in hardware in contect of videoaccel, when nv86 should ?
15:47 karolherbst: Satchelboi_: yeah, just wanted to clear this up in advance
15:48 karolherbst: Satchelboi_: no worries, it would be just stupid if something comes up later I wasn't aware of
15:48 mac-_: imirkin_: yeah sorry, I have also nvs290 which is nv86 but I haven't had time to put it in the mainboard and I use there nvs285 so far
15:48 imirkin_: mac-_: NV34 technically has VPE2 support. however when i last tried it (many years ago), the system hung. i have no clue why, but i just ended up disabling support for it in mesa and moved on with life.
15:48 Satchelboi_: karolherbst: I may be taking a night class to catch up on some credits here, but that wouldn't intrude on hours in any way.
15:48 karolherbst: Satchelboi_: that's fine, as long as there is a progress I can see over the days :p
15:49 mac-_: imirkin_: I'm atlking about nv34 only because I have FX5200 in ancient Sempron 2200+ machine
15:49 imirkin_: mac-_: the G86 will be much better supported. for vdec, it uses VP2, which should get you full-stream H.264 accel.
15:49 karolherbst: Satchelboi_: something like "I have to learn for that exam, and I have to pass this" -> not so good
15:49 imirkin_: [once you extract the firmware, as indicated on the videoaccel wiki page]
15:49 mac-_: yeah I noticed it
15:49 mac-_: :]
15:50 imirkin_: mac-_: what kinds of videos are you looking to decode though? the format is what matters most...
15:50 imirkin_: mac-_: for example, G86 won't do "MPEG4", only H.264 (i know it's mpeg4 as well, but it's a different "part" of the spec)
15:50 mac-_: I use mvp with vo vdpau (as I know) which requiers OpenGL 2.1 support
15:50 imirkin_: and afaik the VP1 accel engine was never useful on any hardware it was shipped on
15:51 karolherbst: Satchelboi_: but it would be still fine for me if you would say like "this week I can only do like 70%, because of $insert_valid_reason_here, but I can make up for that in the next two weeks"
15:51 Satchelboi_: karolherbst: There's still plenty of time in the day after work for any studies. I'm still not sure if I'm even going to need it or not
15:51 imirkin_: not sure what mvp is... maybe you meant mpv? that won't work with nouveau. use mplayer.
15:51 mac-_: yep mpv
15:51 mac-_: :]
15:51 imirkin_: mac-_: for nv4x, you can use mplayer -vo xvmc foo.mpeg
15:51 Satchelboi_: karolherbst: The only planned downtime I have for this summer is already in the proposal too
15:51 imirkin_: [assuming you have it all set up properly]
15:52 karolherbst: Satchelboi_: well, when in doubts: you'll need the time to work on your GSoC project
15:52 mac-_: what about nv34, any acceleration there ?
15:52 imirkin_: mac-_: the hw has it, but it's not exposed by mesa
15:52 imirkin_: mac-_: it'd be the same xvmc thing.
15:52 mac-_: so bad
15:52 Satchelboi_: karolherbst: That's the priority right now
15:52 imirkin_: vdpau isn't well-suited for IDCT/mocomp accel engines
15:52 karolherbst: Satchelboi_: okay, this trip end of june(begin of july
15:53 Satchelboi_: karolherbst: Yeah, that's the only planned vacation I have
15:53 karolherbst: let me see
15:54 karolherbst: Satchelboi_: 26th of june is the first evulation phase
15:55 Satchelboi_: karolherbst: I saw that. I will have a laptop with me and can take some side time if it's needed, or I can do any evaluation work in the first two days if I can
15:56 karolherbst: Satchelboi_: I think it is okay, as long as there is constant progress. I am reading the mentoring guide at home and I might look up especially for things like this and we will probably discuss something like that.
15:56 karolherbst: Satchelboi_: how much time could you spend daily on your laptop working on stuff?
15:57 Satchelboi_: I could set aside a good 3 hours each day for it. It's for a convention, so I'm sure there will be downtime I can sit at and work
15:58 karolherbst: 3 hours each day? so 15 hours in total?
15:58 karolherbst: ohh I forget the weekend
15:58 karolherbst: 21 hours in total then?
15:58 Satchelboi_: Actually, I can probably do a solid 8 on two of those days since I'm going to be on a buss for 10 hours.
15:58 karolherbst: and what kinf od convention if I may ask?
15:59 Satchelboi_: karolherbst: It's a, uh, furry convention in Pittsburg
16:03 karolherbst: Satchelboi_: okay, let's do it like this: we will have some kind of daily "status meeting" or every other day or something like this
16:03 lesderid: what's the status of reclocking on NVC0? I want to use nouveau, but running at the default clocks is not really an option
16:04 Satchelboi_: karolherbst: I can have time set for that, no problem
16:04 karolherbst: and if you say, you will be able to also work in this time period and I see that it works out like this, it's fine from my side. I may most likely still try to get an advice from somebody else regarding this
16:04 karolherbst: lesderid: it's kind of improving, ask RSpliet
16:04 RSpliet: lesderid: it's WIP, in my spare time which is not that plentiful
16:04 lesderid: is there anything I can do to help, e.g. test?
16:05 RSpliet: lesderid: I think at this point no
16:05 karolherbst: Satchelboi_: but it's good that you are honest about this, so thanks for that
16:05 RSpliet: I had a limited success recently, but not release-worthy or even beta-test worthy yet
16:05 imirkin_: RSpliet: well, it kinda works on 1 card right? :)
16:06 RSpliet: imirkin_: low two perflvls on one GDDR5 card. With Bens display fix I can flicker-free switch between the two reliably :-)
16:06 Satchelboi_: karolherbst: Especially with evaluation phase being that same weekend, I'll keep contact open and work on whatever is needed for that. I'm not sure really what needs to be done during evaluation
16:07 imirkin_: RSpliet: great, so you're like 99% done :)
16:07 imirkin_: er wait, i meant 0.99% :p
16:08 RSpliet: imirkin_: 99‰
16:12 Satchelboi_: karolherbst: I have to run to a lecture for a bit now. I'll be back in about an hour and a half if you'd like to keep discussing anything
16:12 karolherbst: okay, thanks for letting me know
17:43 karolherbst: Lyude: by the way, can you assure, that the Nvidia you want to give to Satchelboi_ works under Nouveau reliably enough to work on stuff?
17:44 Lyude: karolherbst: all my cards work perfectly afaik, the only weird one is the GTX970 with the 4GB of ram
17:44 karolherbst: Lyude: which one maxwell ones do you have?
17:44 Lyude: but there are some patches for making that work en-route for stable
17:45 karolherbst: it would be good to have one GM206 one and a non GM206 one
17:45 karolherbst: *a
17:45 Lyude: karolherbst: Just the 970 and a 750Ti
17:45 Lyude: i keep forgetting the 750Ti is a maxwell
17:46 Lyude: i'm also planning on looking to see if I can find some cheap ones on ebay or secondhand
17:46 karolherbst: okay, the 750 TI is good enough for VP6
17:46 imirkin_: 750 is probably preferable
17:46 karolherbst: GM206 are usually cheap
17:46 karolherbst: yes
17:46 imirkin_: as it won't have any of the signature stuff
17:46 karolherbst: no signed firmware crap
17:46 Lyude: huh? I thought the GM206 had that
17:46 karolherbst: 750 ti is maxwell1
17:46 karolherbst: aka GM107
17:47 Lyude: oh, misread your message
17:47 karolherbst: I see
17:47 karolherbst: GM206 is something like 950/960 GTX
17:47 karolherbst: but there is also a GM204 960 GTX
17:48 karolherbst: VP7 is basically VP6 + HEVC and VP9
17:49 karolherbst: hum
17:49 karolherbst: VP6 does HVEC on the host and the GPU
17:49 karolherbst: messy
17:50 imirkin_: start small. just get the 750 to work.
17:50 karolherbst: yeah
17:51 imirkin_: and i'm not just saying that because i have one ;)
17:51 karolherbst: :D
18:18 pmoreau: karolherbst: There is really a 960 GTX sporting a GM204? That would be crazy to have some 960s using eGM204 and others using GM206.
18:18 pmoreau: s/eGM20/GM20
18:20 karolherbst: pmoreau: well, it wouldn't be the first time that a model has different chipsets, never investigated though in depth about the 960
18:21 pmoreau: But usually they span several architectures, not within the same architecture, don’t they?
18:21 karolherbst: sometimes they do
18:21 RSpliet: pmoreau: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_600_series look at the 630 and 640
18:21 RSpliet: and note how there's 630s with DDR3 and with GDDR5
18:22 pmoreau: Right
18:23 RSpliet: if at NVIDIA you shout to marketing "ey yo, we've got some spare G?XYZ in stock, what should we call them", marketing throws a 9-sided dice to fill the gap between the current major and the trailing 0
18:23 pmoreau: But it’s 630/640: GF1xx,GK1xx,GK2xx, not something like 630: GF1xx,GF1yy,GF1zz
18:25 pmoreau: Those G?XYZ are going to be from the previous gen, not the current one. As if they are from the current one, just keep selling the product as is. (Though they did just release a new Titan, using the same chipset (though all cores are available this time.)
18:27 pmoreau: I guess I don’t have the feeling/skills for doing marketing.
18:27 RSpliet: you need a very advanced RNG to do marketing
18:27 RSpliet: and a fine markov chain
18:31 Satchelboi_: imirkin_: Any luck with ideas for which files I should be looking at constants for?
18:31 pmoreau: RSpliet: :-)
18:32 karolherbst: Satchelboi_: I came up with another plan: for the first assignment you really should do something which can't backfire, because the task takes too long
18:33 Satchelboi_: karolherbst: What would be something that can't backfire?
18:33 karolherbst: Satchelboi_: there is an alternative though despite whitespace fixes: documentation. Allthough that also might look like a boring/troublesome task, it's good enough for now
18:33 karolherbst: in addition to that
18:34 karolherbst: we could think of something more complex/challenging where you would have much more time to actually work on that
18:34 karolherbst: let's say until the 18th
18:34 Satchelboi_: Documentation is fine with me, just point me towards what needs to be documented
18:34 karolherbst: and this could be something like: trivial bug fixing or add a small feature, or something else
18:36 karolherbst: Satchelboi_: you could document some video code. Most of the nouveau code isn't documentat at all, but maybe imirkin_ has a nice suggestion. I am currently looking through the code, maybe something jumps into my eyes or so
18:37 Satchelboi_: I was browsing through the whole nouveau package yesterday, there's so much in there I'm not really sure where to go haha
18:37 karolherbst: the kernel module or mesa?
18:37 Satchelboi_: Mesa, sorry
18:40 karolherbst: when in doubt, you could look through codegen/nv50_ir_peephole.cpp there are a lot of compile optimisations, which are quite small, and undocumented. And you can usually understand those without need a lot of background information (eg: // a * 1 = a). But maybe imirkin_ knows something nice from the video code
18:41 Satchelboi_: karolherbst: My second lecture for today just got cancelled too, so I can trying and get that dual boot config running today now
18:41 karolherbst: Satchelboi_: or you just read through the files containing *video* and if you don't understand something due to bad documentation, figure it out and write the missing bits
18:41 karolherbst: Satchelboi_: sounds good.
18:42 karolherbst: Satchelboi_: until will you have your last lecture for this term?
18:42 karolherbst: *until when
18:43 Satchelboi_: karolherbst: My classes will be done after the first week in may. Lectures will be until the last week of april, then it'll be finals the week after.
18:46 karolherbst: Satchelboi_: okay, good. Regarding your planned vacation we came up with another idea: You could start a week earlier
18:47 karolherbst: that would mean at the 23th instead of 30th and then you don't have to work on stuff during your vacation
18:48 Satchelboi_: karolherbst: Start a week earlier on the evaluation?
18:48 karolherbst: no, the work period
18:49 karolherbst: we just move the entire start of the project one week earler, so that you have one week free for your vacation
18:49 karolherbst: officially the work period starts 30th may, so you could start at the 23th of may
18:50 Satchelboi_: karolherbst: Ohhh, alright I see what you're saying now.
18:52 karolherbst: if there is nothing blocking this on your end, we could just go with this
18:53 Satchelboi_: I'll be free by then, no worries
18:53 karolherbst: nice
18:59 karolherbst: Satchelboi_, Lyude, mupuf: fyi I will be on vacation for 7-8 days between 25th May and 5th June. If none of you, or somebody else could take care of Satchelboi_ in this time, I can also cancel this.
18:59 imirkin_: i can act as backup as necessary
18:59 mupuf: imirkin_: thanks
18:59 karolherbst: nice, thanks
18:59 mupuf: karolherbst: that's the first week, right?
19:00 karolherbst: basically, yes
19:00 karolherbst: I will be away later though
19:00 karolherbst: more starting at 29th
19:00 karolherbst: will be in barcelona with family during this
19:00 karolherbst: I hope I will have enough time to respond though
19:02 Satchelboi_: karolhebst: You can set up a task list to do then too for me while you're gone, I can follow down that, ask questions here and document daily what progress is made
19:02 karolherbst: you'll need this anyway
19:02 karolherbst: most likely
19:03 karolherbst: you can always ask question to anybody here, and you also should
19:03 karolherbst: it's not like there will be only me, but the entire community as well
19:04 karolherbst: mupuf: I guess going to sha2017 from 4th until 8th is out of question then, especially because it's so close to the end
19:05 karolherbst: (or Satchelboi_ comes and we work together on stuff there :p )
19:05 mupuf: ah ah
19:05 mupuf: karolherbst: dude, being a mentor does not mean 24/7 contact
19:05 Satchelboi_: karolherbst: Will $50 be enough to get there?
19:05 mupuf: an hour a day max
19:06 karolherbst: Satchelboi_: more like a plan trip to the netherlands + 250$
19:06 mupuf: in average, otherwise you are not mentoring the student, you are doing his job ;)
19:06 karolherbst: mupuf: k, I didn't plan for sha yet, so
19:06 karolherbst: :D
19:06 karolherbst: true
19:06 karolherbst: Satchelboi_: okay so, it would be nice if you are able to come as well :p
19:06 karolherbst: allthough you wouldn't have much from sha then
19:07 karolherbst: oh well
19:07 mupuf: that does not sound like a good plan to me
19:07 mupuf: :s
19:07 karolherbst: I know
19:07 mupuf: you are being too enthusiastic :D
19:07 karolherbst: !
19:07 Satchelboi_: karolherbst: Haha don't worry about that, I'm just playing around
19:07 mupuf: karolherbst: time flies
19:08 Satchelboi_: karolherbst: I almost forgot, do you need other points of contact to reach me if I'm not here?
19:08 mupuf: We'll have XDC to meet ;)
19:09 karolherbst: yes
19:09 karolherbst: Satchelboi_: email makes sense
19:10 Satchelboi_: Perfect. That was in my proposal, so you have that already
19:23 Lyude: karolherbst: sgtm, I should be here around then
19:42 karolherbst: Lyude: nice, thanks
22:38 mooch2: mwk: i need assistance
22:38 mooch2: i have nt4's nv3 drivers not crashing, but it hangs before doing anything
22:39 mooch2: i can give you a log, if you'd like
22:47 Satchelboi: karolherbst: Fedora workstation is up and running so I have a dedicated system for work now :)