00:02salawat: I've noticed Nvidia tends to not offer anything until you guys already know it; sometimes nailing down a concise, articulable description understandable by those less in touch with the nitty gritty of cryptography results in a bit more support and awareness.
00:02salawat: that's my naive hope anyway.
01:02imirkin: that's definitely the majority of the docs, but sometimes they publish stuff we actually need
05:27damo22: salawat: i found a windows registry key called RMSkipDHKey which i am guessing refers to some kind of diffie hellman key exchange in an old 340 driver
08:32pmoreau: imirkin: Re “have you had any trouble with atomics on nv50?” I don’t think I even tried those yet.
08:33pmoreau: I’ll look around for some blob traces; I think I might have a couple of them, but unsure.
08:45pmoreau: imirkin: Last week was really busy so didn’t have any time, but it should be better going forward (well as easy as working towards a paper deadline is, but at least I won’t have additional stuff on top of it) so I can spend some time on Nouveau this upcoming weekend and looking at patches.
16:04imirkin: pmoreau: don't worry abou tit
16:04imirkin: pmoreau: i worked it out
16:04imirkin: the issue was with g[$r63]
16:05pmoreau: I saw it later on as I was scrolling through the logs.
16:05imirkin: i'm working on the image load/store stuff, it's surprisingly annoying to pack stuff
16:05imirkin: at least without resorting to tons of format-specific code
16:05pmoreau: So is there a reg in the end which is truly 0?
16:06imirkin: the problem is just with g[] indirection with $r i think
16:06imirkin: you can't use an out-of-bounds register
16:06imirkin: $r63 is fine elsewhere in the shader
16:06imirkin: (i think, will need to double-check, for now i'm just not using it in compute)
16:07pmoreau: Mmh, okay
16:10karolherbst: to load at 0x0 sounds like a bad idea anyway :p
16:11imirkin: karolherbst: well it's relative to a base address
16:11karolherbst: ahhh
16:11imirkin: so it's basically the first "element" of the buffer
16:12HdkR: Could also be the case of r63 behaves in fun/interesting ways on old hardware :P
16:14imirkin: HdkR: well, it behaves normally in other shaders on this hw
16:14imirkin: but note that on tesla, r63 isn't necessarily 0
16:14imirkin: it's just that if the gpr count < 63, it works out to 0
16:14imirkin: no reason we couldn't use r62 or t45
16:14imirkin: r45*
16:14imirkin: except that 63 is the cut-off between "small" and "large" registers
16:14imirkin: if we RA > 63 regs, then we start using r127
16:17HdkR: ah
16:18imirkin: (short encodings must use small regs, as do encodings that have imm's in them)
17:33zocker: hi!
17:34zocker: i'm thinking about buying a geforce 1030 gt card for a htpc
17:34zocker: would this be usable with the current nouveau driver?
17:34zocker: (https://geizhals.de/asus-geforce-gt-1030-low-profile-silent-90yv0at0-m0na00-a1625262.html)
17:34imirkin: define 'usable'
17:34imirkin: it will light up your monitor(s)
17:35zocker: watching movies, kodi, maybe some retro gaming?
17:35imirkin: i wouldn't advise it.
17:35imirkin: there's no hw video decode
17:35imirkin: which is probably desirable if you plan on watching high-bitrate of hevc/etc format movies
17:36imirkin: there is no reclocking, so you're stuck at the lowest memory/shader/etc clocks, so not exactly ideal for playing video games
17:36imirkin: (although if 'retro' is like arcade, that doesn't really benefit from accel in the first place, so that'd be fine actually)
17:36zocker: yes, i read about that
17:37imirkin: but if retro is like playing games from the ancient times of 2018, then you're going to have a bad time
17:37zocker: so my use case would be watching movies at 1080p and maybe play some old console games with mednafen
17:37HdkR: imirkin: Arcade systems can have fairly powerful bit of GPUs in them if you're considering like...1995+ :P
17:38imirkin: HdkR: they might, but the emulators don't use GL to emulate those powerful GPUs
17:38zocker: i would rather buy something with an amd gpu on it, but apperently they dont care that much about low profile/low power cards
17:38imirkin: zocker: so ... i can tell you that i have one plugged in right now
17:38imirkin: (GT 1030)
17:38imirkin: probably not the precise same model you pointed to, but likely similar
17:38imirkin: i have 2x 1200x1920 monitors
17:39imirkin: i can play 1080p videos just fine
17:39HdkR: imirkin: Naomi and Triforce both use GL/Vulkan for their GPU emulation :)
17:39imirkin: with my not-exactly-new Core i7-920 CPU
17:39zocker: so as long as my cpu is powerful enough i should be fine regarding video?
17:40zocker: i plan to connect it to a 1080p projector so that would be all i need
17:42imirkin: yeah, i think you should be fine.
17:42imirkin: kodi might be problematic, but if you convince it to not use any sort of accel at all, you'll be fine
17:42imirkin: might even be fine with GL, dunno, you can try
17:43zocker: how feature complete is the GL support? i mean, apart from the reclocking issue
17:43imirkin: feature-wise, we have all of GL 4.5 and almost all of GL 4.6
17:44imirkin: it's all fairly conformant, although due to some stability issues we can't actually completely a full "CTS" run
17:44zocker: hmmm... what are these stability issues?
17:45imirkin: hangs, etc
17:45imirkin: good times
17:45zocker: hmm
17:45zocker: doesnt sound so good :D
17:47imirkin: it's not
17:47imirkin: the more GL stuff you do, the more you're likely to hit them
17:47imirkin: CTS is a long-running process which exercises the full GL capabilities (it's the Conformance Test Suite)
17:47imirkin: and somewhere we fall short
17:48imirkin: every individual test does pass though
17:49zocker: i see
17:50imirkin: it's quite rare that i get hangs
17:51imirkin: and doing nouveau development greatly increases the odds you get one...
17:51zocker: i can imagine ;-)
17:52imirkin: i probably end up in a hang situation every couple months? but i'm also careful to avoid things which cause hangs.
17:52zocker: like not using any GL stuff at all?
17:52imirkin: nah, i use GL stuff
17:52imirkin: i just don't use pointless GL stuff
17:53ullbeking: i'm looking here: https://nouveau.freedesktop.org/FeatureMatrix.html as i'm speccing out a home server, and i wonder, how do i find out what size pci-e slot each card takes (x8, x16, etc)?
17:53imirkin: i use chrome (and i force-disable the gpu blacklist, so it's using gpu accel)
17:53imirkin: ullbeking: many (non-x16) slots in motherboards have a hole in the back
17:53imirkin: which allows you to plug e.g. a x16 card into a smaller slot
17:54zocker: imirkin: can you do 1080p60 youtube in chrome with it?
17:54imirkin: however there are power requirements on x16 slots which aren't necessarily available with smaller slots
17:54imirkin: zocker: dunno. happy to click a link and find out though.
17:54ullbeking: but plugging an x16 card into a slot that doesn't have enough pci-e lanes.... errr.....
17:54imirkin: ullbeking: it's fine
17:54imirkin: all cards should be able to deal with that
17:55imirkin: you obviously lose some perf :)
17:55imirkin: the lanes aren't just there for fun
17:55ullbeking: oh, it's about performance
17:55imirkin: ullbeking: maor lanes = maor perf
17:55ullbeking: you don't lose functionality?
17:55imirkin: depends how you define functionality
17:55imirkin: you lose bandwidth to CPU memory
17:56imirkin: so if the functionality is xx gb/s to system ram, then yes, you lose functionality
17:56ullbeking: that's a non-functional requirement, to be specific
17:56ullbeking: ;-)
17:56zocker: imirkin: how about that one? https://www.youtube.com/watch?v=XfIibTBaoMM
17:56zocker: :D
17:56imirkin: zocker: also how do i find out if it's working?
17:56ullbeking: imirkin: i know the board i want, but it only has x8 and x4 slots
17:57imirkin: ullbeking: probably not a top-of-the-line GPU then
17:57ullbeking: it's a server board, so maybe a bad choice
17:57imirkin: ullbeking: the tricky bit is that x16 slots guarantee 75W of power
17:57imirkin: while x8 and lower guarantee only 25W
17:57zocker: imirkin: open in chrome and check if the playback is smooth etc?
17:58imirkin: zocker: seems fine
17:58ullbeking: thanks imirkin
17:58imirkin: zocker: i'm looking at "stats for nerds", and it lists a handful of dropped frames, but it's not a number that's going up
17:58imirkin: probably when i was resizing the browser
17:59zocker: sounds good
17:59ullbeking: i want to support the nouveau project, but my family also need to be able to watch their movies and tv shows
17:59imirkin: zocker: note that i'm (a) using chromium and (b) have force-enabled gpu usage for accel
18:00imirkin: zocker: and a i7-920 CPU. not exactly the world's fastest, but also decent.
18:01zocker: imirkin: should be fine for my usecase then
18:01zocker: imirkin: thank you for checking! :-)
18:02zocker: i'm primarily going to use mpv anyways, but that should just fine as well i guess
18:02imirkin: zocker: ftr, this is what i have plugged in: 02:00.0 VGA compatible controller [0300]: NVIDIA Corporation GP108 [GeForce GT 1030] [10de:1d01] (rev a1)
18:04imirkin: Subsystem: Dell GP108 [GeForce GT 1030] [1028:11c7]
18:05zocker: how much vram? 2GB?
18:07imirkin: yep
18:07zocker: alright... i think i'll give it a shot
18:15zocker: imirkin: https://www.youtube.com/watch?v=ATAWWxtcxjo
18:15zocker: that one is supposed to be 1080p60 at max quality
18:15zocker: any frame drops with that one?
18:15imirkin: well, quality diff at 1080p@60 is just going to be differences in cpu usage levels
18:16imirkin: yes, i get consistent frame drops
18:16imirkin: like a few per second
18:16zocker: ok, i guess because of cpu load?
18:16imirkin: i would think so
18:16imirkin: note that i also have 2x rotated monitors
18:16imirkin: so that involves some potentially unnecessary copies
18:17zocker: so few frames dropped, but not a total desaster?
18:18imirkin: not total.
18:18imirkin: and if you have a newer-than-2010 cpu, then perhaps it could keep up
18:19imirkin: side note: i didn't realize it was such a horrible movie
18:19zocker: hehe :)
18:19zocker: first thing i found when i searched for "1080p60"
18:20imirkin: youtube is going to start suggesting me some weird videos after this...
18:20zocker: sorry for that
18:20imirkin: no worries ;)
18:20zocker: better delete your cookies
18:20zocker: hehe