00:21JustinHitla: anyone here ?
00:22imirkin: 208 people, looking at the chan list
00:33mwk: alright, methods up to 0x4c0400 covered by tests, except for the IFC stepping part
00:34imirkin: i barely know anything about nv1
00:34imirkin: i just remember it has that funny interpolation?
00:34imirkin: or was it that it didn't use triangles?
00:35mwk: yeah, the "textured quad" bit
00:35mwk: complete madness
00:35imirkin: i assume you still have a live one
00:36imirkin: last one out there in the wild? :)
00:36mwk: I heard ajax has one or two
00:36mwk: but then I heard that a long time ago :p
00:37imirkin: back when nv4 was coming out? heh
00:45mwk: methods up to 0x4d0400 covered :)
00:49mwk: of course, 0x4d0400 is the worst method there is, "render a texel of a crazy quadratic quad"
00:49imirkin: how bad can it be
00:49imirkin: just some underlying formula right?
00:50mwk: there's a whole state machine involved
00:51mwk: and there's a *lot* of cases
00:51mwk: first off, you may be rendering a border texel, which needs an extra connecting triangle rendered
00:52mwk: and this is handled differently for each of the 4 borders
00:52imirkin: sounds ... involved
00:53mwk: second, if this is the first texel rendered, you need to start up the interpolation machine
00:53mwk: third, if this is the last texel of a line, you have to reset the machine to the first texel of the next line
00:53mwk: and otherwise, step it to the next texel
00:54mwk: and all these things come in two versions, linear and quadratic
00:55imirkin: why did they do this again?
00:56mwk: quadratic curves look beautiful on demos!
01:02mwk: eh, screw texquads, I'll do IFC step first
02:13agrecascino: nouveau 0000:12:00.0: DRM: GART: 1048576 MiB
02:13agrecascino: i think i found an issue
02:14agrecascino: also it isn't a gt 520
02:14agrecascino: it seems to be a gt 210
02:14imirkin: what's the issue?
02:14imirkin: 1TB gart size is standard
02:14agrecascino: imirkin, the probe failure
02:14imirkin: 40-bit of access to sysmem
02:15imirkin: obviously things would fail horribly if it ever tried to really use that much
02:15imirkin: but that's the max, yeah
02:15imirkin: GT 210 is probably a GT218?
02:15agrecascino: i'm stupid
02:15agrecascino: nouveau says it's a g98
02:16imirkin: ah ok. G98 also makes sense
02:16imirkin: (they dicked around with the marketing names quite a bit)
02:16imirkin: anyways, would be interesting to see your full dmesg on pastebin or in a fd.o bug
02:19agrecascino: imirkin, having an issue with finding out how to add kernel options with SILO
02:20imirkin: or you can set the default level in the kconfig
02:23imirkin: also you can add cmdline options in the kconfig
02:28agrecascino: imirkin, but that would require rebuilding the kernel?
02:29imirkin: indeed it would
02:29imirkin: i assumed that wasn't a problem.
02:29agrecascino: imirkin, it isn't, but my ultrasparc t2 isn't _THAT_ fast
02:30imirkin: the incremental build from changing one of those options shoudl be ~instantaneous
02:36agrecascino: iomem=relaxed doesn't fix anything, btw
02:39agrecascino: it's nouveau.trace=debug? right?
02:39imirkin: other way
02:40imirkin: didn't think the iomem would do anything...
02:47agrecascino: console is now entirely nouveau
02:50imirkin: trace spews a lot of stuff... get the dmesg and post it
03:03imirkin: i'm having trouble diagnosing where it's dying exactly
03:23imirkin: no way that still applies
03:23imirkin: i think benh gave up on that work
03:24benh: what work ?
03:24benh: sorry, haven't followed
03:24benh: ah yes
03:24agrecascino: skeggsb, apparently was also working on it
03:24benh: fixing 64k would be handy for G5 users ;-)
03:24agrecascino: benh, nouveau doesn't even function on sparc
03:26agrecascino: is their some way i can fix the console atleast
03:32imirkin: agrecascino: you can probably get cpu console to work with nouveau.noaccel=1 nouveau.nofbaccel=1
03:33agrecascino: imirkin, alright
03:36imirkin: this will disable all manner of acceleration, it'll basically just be a framebuffer. but you'll be able to switch resolutions :)
03:37agrecascino: and it might not crash with a bus error
03:37imirkin: yeah, those unaligned accesses are annoying
03:37imirkin: i had a patch in that one bug to fix at least some of them
03:38imirkin: https://bugs.freedesktop.org/show_bug.cgi?id=96836#c14 and the next comment
03:39imirkin: but i think the unaligned accesses are emulated, the kernel just screams bloody murder about them.
03:52agrecascino: imirkin, still fails
04:44JustinHitla: so can nouveau do 3d ?
04:48imirkin: not 100% sure what you mean by '3d', but probably?
04:48imirkin: it can't make things appear on your desk
04:49JustinHitla: I mean last time I checked it says nouveau can accelerate 2D but not 3D, how is it now ?
04:50imirkin: what gpu do you have?
04:50imirkin: also, when was the last time you checked?
04:50JustinHitla: Riva TNT 16MB
04:51JustinHitla: imirkin: last time I checked I don't remember 4 or 5 years ago
04:51imirkin: should work as well as it did 4-5 years ago. afaik most of the hw-supported functions are available
04:51imirkin: you should get GL 1.3 or so
04:52imirkin: the main thing that jumps out in my mind as non-functional is glClipPlane()
04:52imirkin: and 3d textures, for which there's no hw support
04:52imirkin: otherwise should work. not that it gets a ton of testing, to be fair.
04:53JustinHitla: imirkin: I tryed nouveau back then, because Nvidia stopped supporting Riva TNT and I couldn't install old drivers on new linux kernel so I tryed nouveau back then but I remember people said in IRC nouveau can't do 3D acceleration
04:53imirkin: ok. well it should be able to now. but note that this 3d accel won't do you too much good
04:53JustinHitla: but about 2 years ago I did upgrade and now I have Radeon x300, so I don't need nouveau, it only supports Nvidia cards, right ?
04:54imirkin: it's a very old gpu, with very limited functionality
04:54imirkin: nouveau is for nvidia gpu's, yes.
04:54JustinHitla: you mean Radeon x300 a very old gpu ? you don't say
04:55imirkin: i mean riva tnt is a very old gpu
04:55imirkin: much older than that radeon
04:55JustinHitla: I know
04:55JustinHitla: I used that PC for 14.5 years
04:55imirkin: so i wouldn't expect magic to happen from nouveau
04:55imirkin: it should support all of the hw functionality though
04:56JustinHitla: so why nouveau ? aren't there enough drivers ?
04:56JustinHitla: there is another one open source driver for nvidia cards, no ?
04:57imirkin: there's nv, which is nvidia's unmaintained 2d xorg driver, and there's nvidiafb, which is a fbdev driver for pre-G80 gpu's (actually it might support G80 and only stop support at fermi, i forget)
04:58JustinHitla: imirkin: do you play games on linux using nouveau ?
04:59imirkin: rarely. mostly just when debugging issues people report.
04:59JustinHitla: so does it allows to play many games ? crysis ? starcraft ?
04:59JustinHitla: minecraft ?
05:01imirkin: any game that conforms to opengl standards, sure
05:01JustinHitla: so how is performance of nouveau
05:01JustinHitla: and why don't you rename it to something more easy to remember ?
05:02imirkin: what's difficult to remember about it?
05:02JustinHitla: imirkin: are you french ?
05:02JustinHitla: is it a french word ?
05:02JustinHitla: french is difficult
05:02JustinHitla: imirkin: do you know french ?
05:02imirkin: have you looked at english?
05:03imirkin: english is way less regular than french
05:03imirkin: (not that french is some beacon of regularity in the language world)
05:03JustinHitla: why not to rename it to "nova" ?
05:03imirkin: why rename?
05:04JustinHitla: or even "NoVa"
05:04JustinHitla: as reference to "nv" driver
05:05JustinHitla: you won't understand
05:05imirkin: i guess not.
05:07imirkin: or perhaps as a reference to northern virginia...
08:00karolherbst: mhh, I like when my system freezes and nothing is inside pstore :/
11:44hakzsam: imirkin, bin/polygon-offset also fails on gm206 :/
11:44hakzsam: I think it's an other issue
14:49imirkin: hakzsam: did you fix the MS stuff?
14:50kloofy: hello, allmighty ducks of nouveau:)!!
14:52kloofy: minor question coming through, final days of me on the channel, maybe someone could even try to answer
14:52kloofy: is it possible for different warps to enter different interrupt handlers in parallel?
14:56kloofy: karolherbst: especially that question is directed to you, old buddy old pal:), actually i'm pretty much afraid the same way as you are that interrupts are slow, can you confirm this stuff too?
14:59hakzsam: imirkin, nope, I didn't work since yesterday and you, did you fix the bgra stugf? :)
14:59imirkin: hakzsam: nope, i did not.
15:01kloofy: i finally landed my nose into code, in pretty harch manner, and indeed the hypthesis about relative indirect addressing did not work the way i expected, maybe even vadimg made some sense , however this can be uglily worked out perhaps with interrupts+vop3 relative addressing, overall it's incredible how bad the radeon ISA sucks
15:07kloofy: imo overall as the ISA is so horribly bad there, it infact leaves a whole lot burden to carry for programmers indeed, but enough of rant, it's probably one of those days when i no longer get answers anymore
15:10imirkin: hakzsam: feel free to test this out for fixing MS on the GM20x: http://hastebin.com/ofebumeyob.avrasm
15:11hakzsam: you don't re-use get_sample_position()?
15:12imirkin: we can fix it later if this works
15:12imirkin: if it doesn't, i'd rather have something more easily hackable
15:15imirkin: hakzsam: there's a break missing in case 4, btw
15:15hakzsam: and two compilation erros, fixed lcoally ;)
15:16imirkin: and it's GM200_3D_CLASS
15:16imirkin: aaand i messed up the class_3d thing
15:17hakzsam: your stuff seems to work
15:17imirkin: that fixes MS?
15:18hakzsam: I did somehow the same patch yesterday
15:18imirkin: but it didn't quite work?
15:18imirkin: well, now you can try to figure out why :)
15:19hakzsam: it worked :)
15:19kloofy: allright, give an enthusiastic aplause for the loon, i think perhaps the time is mature, to invest a little in my health if possible, and leave those spamming scenes!
15:19imirkin: why didn't you send it out?
15:19hakzsam: because I did something else
15:19imirkin: ah. i thought you were still having trouble with it =/
15:20hakzsam: it worked for MS 2 and 4 IIRC
15:20hakzsam: anyways, I will send out a patch later today :)
16:50imirkin_: pmoreau, hakzsam - i think this gives us some concrete steps to work towards
16:51hakzsam: oh nice
16:51imirkin_: which are more manageable than "implement all of vulkan in one shot" :)
17:10imirkin_: interesting. https://www.opengl.org/registry/specs/NV/gpu_multicast.txt
17:12imirkin_: that's roughly in line with how SLI works
17:24imirkin_: mwk: how far did you get with SLI? iirc you worked the linking protocol... what about how to actually make use of it for bigger-better-faster memory accesses?
17:43karolherbst: imirkin_: as you might know, mupuf has a 690 now, so everybody can work on SLI now anytime
17:43mupuf: not received yet
17:43karolherbst: sad :/
17:43karolherbst: but it is coming!
17:43imirkin_: karolherbst: yeah, but as usual, i want to minimize the quantity of work i have to do
17:43karolherbst: imirkin_: but that extension makes sense, because nvidia said, they don't want to provide SLI profiles anymore
17:44imirkin_: i don't blame them
17:44imirkin_: it's ~impossible to figure it out in the driver
17:44karolherbst: some "gamers" are really fed up about this already (that they don't get profiles for a 3x or 4x SLI setup)
17:44karolherbst: on pascal
17:45imirkin_: well, i don't see anything preventing that ext from working on all SLI-capable gpu's (G80+)
17:45mupuf: karolherbst: I am sure they figured out that noone did it anyway
17:45karolherbst: the performance benefit was questionable from the start anyway
17:45imirkin_: however i'm not 100% sure that the cross-sli-slave sync stuff is available there. but perhaps it is.
17:45karolherbst: both cores appear as their own pci devices
17:45karolherbst: each with their own vbios
17:46mupuf: gl_spirv.txt --> Cool, it got ratified! I checked it out a month ago and liked it :)
17:46karolherbst: (which indeed differes)
17:46imirkin_: i'm talking about it working all the way back to G80 :)
17:46karolherbst: I see
17:48karolherbst: mupuf: maybe at some point we should talk about unsurance for all those gpus and share the bill somewhat :p especially with the titan and the 690 the value of all your gpus should have doubled now :D (or is there already something done and I don't know about that?)
17:48imirkin_: unclear whether the mask is doable on G80. definitely on GF100+
17:48mupuf: karolherbst: /me should just list them already
17:48mupuf: my home insurance should cover stuff like this
17:49karolherbst: I am always wondering if there is some limit or so
17:52karolherbst: mupuf: not that they thinkg at some point, that you do something commercially, and you have to explain to them, why you just love nvidia gpus and why you are a big collector and why you just need those nvidia gpus :p
17:53mupuf: oh, right
18:01mupuf:should get the 690 soon
18:01mupuf: it apparently is in Finland now
18:15Tom^: what gcc flags is used when building nouvea in the uh drm folder?
18:15imirkin_: whatever you tell it...
18:15Tom^: it isnt fetching some KFLAGS from the kernel or such?
18:16imirkin_: it's using the kernel's build mechanics
18:16imirkin_: so however you configure your kernel is however it will build it
18:20Tom^: imirkin_: know where those are defined? im bit curious what this current running kernel got built with.
18:21karolherbst: Tom^: kernel config file of your distribution :p
18:22karolherbst: usually also in /proc/config.gz
18:22Tom^: karolherbst: yea the thing is im running a self compiled one. and im not sure what i used on it xD
18:22karolherbst: I think the highest the kernel does is to enable core2 opts
18:22Tom^: yes yes, im such an irresponsible user.
18:23imirkin_: Tom^: look at the makefile - should be pretty clear how it works
18:23imirkin_: you can also build with V=1 which will show you all the exact commands
18:27Tom^: imirkin_: ah thanks
18:29JustinHitla: you can't catch me
18:34Calinou: I have my 1080 now btw
18:34Calinou: runs great
18:59Tom^: karolherbst: i guess power sensors still is a bit wrong?
18:59Tom^: 63.45W at full load :p
19:48karolherbst: Tom^: do you have my patch?
19:51karolherbst: Tom^: https://github.com/karolherbst/nouveau/commit/0469626778c7b0f74c195476a3639a0dd0ff1890
19:51karolherbst: skeggsb: for you too :p ^^
19:52Tom^: probably not
19:52karolherbst: ohh, I didn't send that patch out to the ML? odd
19:53karolherbst: I am sure I did
20:06agrecascino: skeggsb, i call upon the gods to deliver you to me
21:10mwk: what the fuck
21:11mwk: NV01's IFC actually starts rendering a new image immediately to the right of the first if you submit more data...
21:12mwk: I guess that's what the SIZE_OUT thing is for, to disable this... feature
21:12mwk:wonders if newer GPUs work the same way
21:22agrecascino: mwk, IFC?
21:34karolherbst: that makes actually more sense than my thought: "if condititonal"
21:35imirkin_: precursor to SIFC presumably
21:35imirkin_: (scaled image from cpu)
21:41mwk: imirkin_: stretched image from cpu
21:41mwk: SIFM is the scaled one :p
21:41mwk: yeah, IFC is a poor version of SIFC
21:41imirkin_: stretched... scaled... all the same :p
21:41imirkin_: SIFM is from memory though
21:42imirkin_: i.e. from pushbuf vs from vram
21:42mwk: SIFM scales, SIFC stretches
21:42imirkin_: what's the diff?
21:42mwk: there is actually a difference
21:42mwk: SIFM can have bilinear filtering, SIFC is ugly
21:42imirkin_: ah ok
21:43mwk: because SIFC basically renders every incoming texel as a solid rectangle on screen
21:43mwk: while SIFM works as a single primitive
21:44mwk: presumably because it's easy to re-read memory on SIFM for neighbouring pixels, while you can't rewind the pushbuf for SIFC
21:44imirkin_: not without a staging area
21:45imirkin_: which presumably defeats the purpose of doing SIFC in the first place
21:46mwk: on NV1, IFC is basically a funny machine used as a frontend to POINT rasterization
21:47mwk: NV3 SIFC probably does the same with RECT rasterization and a much more complex machine
21:47agrecascino: i imagine SIFC would be fine if you were doubling things
22:02pmoreau: imirkin_: Yup, I saw the OpenGL extension for SPIR-V was out! Haven’t looked at it yet.
22:09karolherbst: is it a real ARB extension which will be implemented by everybody, or just nvidia going ahead and is hoping everybody else will add it too?
22:09imirkin_: karolherbst: it's an ARB extension, which means it got voted on and whatnot
22:09karolherbst: but that still doesn't mean everyody will have it, right?
22:09imirkin_: of course so was GL_ARB_shading_language_include
22:10imirkin_: no, extensions are just that - extensions
22:10imirkin_: however they could well announce GL 4.6 or 5.0 or whatever
22:10imirkin_: which will have that functionality be a part of the spec
22:10karolherbst: I know
22:10karolherbst: okay, so it is kind of sure, that that spir-v stuff will get added if there is a new spec?
22:11imirkin_: i don't think anyone here knows
22:11imirkin_: and if they did, they wouldn't be able to tell you
22:12karolherbst: at least I know, that the one eon dev would be happy about that spir-v extension being used by everybody :p
22:38pmoreau: They did advertise an update of the red book titled "OpenGL 4.5 with SPIR-V", coming out in Sept. IIRC
22:39agrecascino: pmoreau, wait
22:39agrecascino: pmoreau, they are actually making another one, that isn't about vulkan
22:40pmoreau: Another red book? yes, there are two books coming in Sept. – Oct. One about Vulkan, one about OpenGL
22:50karolherbst: agrecascino: why shouldn't they make another one which isn't related to vulkan? there are no plans to discontinue opengl afaik
22:50agrecascino: karolherbst, it seems everything thinks it's that way :\
22:50karolherbst: which is untrue
22:50karolherbst: not every indie game dev wants to mess with vulkan for example
22:50orbea: vulkan is still experimental, no?
22:51karolherbst: orbea: not the point
22:51karolherbst: vulkan isn't intented to use for "easy" and "simple" things
22:51karolherbst: for that you should use opengl
22:51orbea: good to know
22:51karolherbst: with vulkan you get more control, but you have to code a lot
22:52agrecascino: karolherbst, i swear every time I hear someone talking about lower level graphics APIs they think it's a solution to everything
22:53imirkin_: that's why programming in assembly is so popular and no one uses languages like python
22:53imirkin_: oh wait. i might have that backwards...
22:54karolherbst: agrecascino: the golden hammer never works :p
23:33imirkin_: gnurou: could also be that 4GB happens to be a popular amount :)
23:33imirkin_: lots of confounding variables
23:34gnurou: imirkin_: true, but this is troubling... and I think I remember my boards only had 2GB (need to check that at the office though)
23:34imirkin_: also i'm moderately sure there are GTX 960's with 4GB of vram, but no one's complained about those
23:35imirkin_: but ... again, there are fun overlaps of popularity, etc
23:36imirkin_: gnurou: i have to imagine it shouldn't be too difficult for you to get your hands on such a board...
23:36imirkin_: i kinda envision nvidia offices to be swimming in GPU's... heh
23:36gnurou: imirkin_: should not indeed
23:36pmoreau: I think my 960 has 4GB of VRAM, and I do not have that error.
23:36pmoreau: I’ll just check
23:36imirkin_: office door won't close coz of too many gpu's :)
23:37imirkin_: [i guess that's more like ben and dave's office...]
23:39pmoreau: Hum, no, only 2GB. Nevermind
23:40imirkin_: well, i def see GTX 960's with 4GB on newegg.
23:41pmoreau: I recall seeing 960s with 4GB when I bought mine. Can’t remember why I picked a 2GB instead of a 4GB though.
23:41imirkin_: $50? :)
23:42pmoreau: For 50$, I would have taken the VRAM upgrade! :-D
23:43pmoreau: Must have been something else. Oh well, it’s not like I’m really using it anyway.
23:43imirkin_: either way, it's a lot to pay for the privilege of developing graphics drivers for free
23:44pmoreau: That one was for work purposes, cause I needed a Maxwell card.
23:45imirkin_: shoulda gotten the Titan X if work was picking up the tab...
23:46karolherbst: does anybody of you do something _not_ graphics related for work by the way? :D
23:46pmoreau: karolherbst: Nop
23:47karolherbst: imirkin_: ohh, right, no clue what you do :p
23:47imirkin_: me neither
23:47imirkin_: something with computers, i think
23:47imirkin_: not sure
23:48karolherbst: do you know word? do you have to work with it?
23:48imirkin_: i know many words.
23:48karolherbst: I see
23:50karolherbst: tried to do stupid jokes with those other applications, but none is worth the time to write them :/ they should have used better words for that
23:50karolherbst: *came up with
23:51imirkin_: wow, those gtx 1060's are selling like hotcakes
23:51imirkin_: all sold out on newegg
23:51karolherbst: the 1060 is like as fast as a 980 or something...
23:52imirkin_: huh. so is the 1080
23:52karolherbst: no, the 1080 is much faster :p
23:52agrecascino: i'm going to get my 670 till the end of time ;)
23:52imirkin_: i mean the 1080 is also all sold out
23:53karolherbst: but that is on purpose
23:53imirkin_: and 1070. wow.
23:53imirkin_: i wonder if that's supply issues or popularity
23:53imirkin_: [actually there's like one model of 1070 and 1080 available. but most are sold out.]
23:54karolherbst: I think for the 1070/1080 it is due to the limited production, or at least that's the reason you usually hear about :p
23:55imirkin_: lol. RX 480 also sold out
23:57agrecascino: do people using linux ever buy nvidia cards by their own volition?
23:57chithead: could have bought a rx 480 between 2:02am and 2:08am from amazon https://www.nowinstock.net/computers/videocards/amd/rx480/
23:58imirkin_: i don't get the "must always have the fastest latest everything" mentality
23:58imirkin_: such consumerism...
23:58karolherbst: imirkin_: you also don't use modern DEs :p
23:58imirkin_: nope, i do not.
23:58imirkin_: i don't use any DE in fact
23:58imirkin_: wtf do i want desktop icons for
23:59karolherbst: well, at some point you will be forced to go with the time and then it is too late for you
23:59imirkin_: however i don't think that'll happen for another decade or so