01:28fdobridge: <rinlovesyou> I may have finally seen this in action
01:29fdobridge: <rinlovesyou> MSAA in VRChat (in vr) is causing the game to drop to like 2fps in some situations on proprietary haha
01:30fdobridge: <rinlovesyou> Amd & even windows users report fps drops with that game's msaa, but nothing on the level of Nvidia proprietary
04:05fdobridge: <nanokatze> I'm not sure that has any relation whatsoever because vrc is d3d11 and in d3d11 you can't have multisampled uavs (= storage images)
04:09fdobridge: <nanokatze> indeed, it's only a thing since d3d12 sm 6.7
05:16fdobridge: <rinlovesyou> Hm i see
05:16fdobridge: <rinlovesyou> It is still disagreeing with the Nvidia driver specifically
07:19fdobridge: <airlied> @gfxstrand you know much about coop matrix bits on nvidia?
07:34dwfreed: random: since fdobridge seems to be a discord bridge, y'all might be interested in this: https://github.com/openttd/dibridge/
07:38fdobridge: <magic_rb.> Id also recommend, which is a more complec setup, bridging through a private matrix room (since matrix bridging got dropped). Heisenbridge and the mautrix discord bridge are really good
07:39dwfreed: The matrix bridge to IRC sucks in many ways
07:39fdobridge: <magic_rb.> There are more than one
07:39Sid127: there's like
07:39Sid127: 2
07:40fdobridge: <magic_rb.> Ive used heisenbridge for a year now it works great
07:40Sid127: there's heisenbridge and there's the appservice one
07:40fdobridge: <magic_rb.> Yeah i havent tested the appservice one
07:40Sid127: and anyway, current system works .-.
07:40fdobridge: <magic_rb.> Yep
07:41Sid127: @magic_rb. the appservice one is meant to be run network-wide, not by a single user
07:41dwfreed: The current system doesn't work when someone spams the discord channel and I kill fdobridge automatically >.>
07:41fdobridge: <magic_rb.> I go through discord directly now :) so like, i dont notice issues
07:42dwfreed: I sit in the openttd channel, which is where dibridge was originally made for, and I have no issues with it
07:42Sid127: dwfreed: the problem is really with people posting multi-line messages and then edit those a few times e-e
07:42fdobridge: <magic_rb.> Also, funnily enough, mautrix discord keeps changing your name, Sid to dwfeed and back wtf
07:42fdobridge: <magic_rb.> Wait are these two people talking
07:42fdobridge: <magic_rb.> Wtf
07:42dwfreed: yes
07:42Sid127: @magic_rb. that's because dwfreed and I are two different users
07:42fdobridge: <magic_rb.> 🤣🤣🤣
07:42dwfreed: Sid127: dibridge does not bridge edits
07:43Sid127: yeah, I did see
07:43fdobridge: <magic_rb.> Just as i say mautrix discord works great, wtf
07:43dwfreed: lol
07:43fdobridge: <magic_rb.> Oh but youre both irc, ffs
07:43Sid127: this isn't mautrix discord I don't think
07:43Sid127: we don't have a matrix bridge atm
07:43fdobridge: <magic_rb.> I do :)
07:44Sid127: ¯\_(ツ)_/¯
07:44fdobridge: <magic_rb.> I think it doesnt diferentiate between the different names for the webhook, thats a bug for sure
07:44dwfreed: dunno that I've seen any heisenbridge bridges to OFTC; all the matrix bridges I've seen on OFTC are the appservice, because they're run by the respective homeservers
07:44Sid127: you might also know me as @tiredchiku
07:44Sid127: yeah, same
07:44fdobridge: <magic_rb.> heisenbridge can also work in relay mode or whatever its called apparently
07:45dwfreed: the value of dibridge is that everybody can appear to be their own user on both irc and discord
07:45Sid127: I wonder how that's done
07:45Sid127: on irc, easy
07:45Sid127: on discord? sounds jank
07:46dwfreed: on the discord side, bots are allowed to pretend to be multiple different users
07:46dwfreed: they'll get the [bot] tag
07:46Sid127: so webhooks
07:46dwfreed: yeah
07:46Sid127: the same thing the current bridge does
07:46Sid127: .-.
07:46dwfreed: the real value is on the IRC side, where all the discord users are their own connection too
07:47Sid127: I don't see value there tbh, it just makes it a bit more confusing
07:47dwfreed: nah, it's less confusing than a single bridge user
07:47Sid127: how are IRC query buffers to those fake connections handled?
07:47dwfreed: you can't query them
07:47Sid127: and what if I connect both ways :D
07:48fdobridge: <Sid> like I do all the time
07:48dwfreed: if your username collides, the discord user will get a [d] added
07:48dwfreed: to the irc nick
07:48Sid127: I see
07:48Sid127: well, hm
07:48Sid127: you'll have to convince karolherbst about it, I don't mind it either ways
07:53Sid127: hmm
07:53Sid127: it seems a bit annoying to maintain
07:53Sid127: since we also bridge #rusticl and #zink
07:54Sid127: "This server is very limited, as in: it only bridges a single Discord channel with a single IRC channel. If you want to bridge multiple, you will have to run more than one server."
07:54Sid127: so we'd have to spin up 3 discord bots to bridge the three channels
08:00dwfreed: That's not the worst thing; they really can use the same IPv6 range for IRC, that would not be an issue
08:01dwfreed: It could probably be extended to handle multiple channels; it's just OpenTTD only needed it for one channel on each side
08:04HdkR: Oh cool, I didn't know the bot could masquerade as a different user. I'm surprised that's allowed
08:05dwfreed: Do believe it requires special permissions (mentioned on the readme)
12:16karolherbst: if a new bot improves the bridging I'm all for it
12:16karolherbst: but how does that work with authentication? Though that might not really matter
12:17dwfreed: it doesn't, but if the channel is +R, you can exempt it with a matching +I
12:18dwfreed: that's one thing fdobridge has over this, is as it's one connection, it can identify and talk through +M
12:18karolherbst: right
12:18karolherbst: I _think_ this can be worked around in oftc with a SSL client cert
12:18karolherbst: lemme try something
12:19dwfreed: have to be using a nick on the account to be considered identified on OFTC
12:19karolherbst_: sooo
12:19some_bot: du
12:19some_bot: test
12:19some_bot: yeah..
12:19some_bot: SSL cert works
12:20karolherbst: dwfreed: so I have two connections using the same SSL cert and whatever the nick all connections have all permissions
12:20karolherbst: and you get auto identified anyway
12:20dwfreed: that's because karolherbst_ was linked to your account
12:20karolherbst: nah, with SSL cert connection it doesn't matter
12:20karolherbst: I changed the nick
12:20dwfreed: it does
12:20karolherbst: it doesn't
12:20dwfreed: you opped before you changed nick
12:20dwfreed: the op overrides +M
12:21karolherbst: sure.. but you establish the connection with the "original" nick and then you can switch to whatever
12:22dwfreed: I still doubt it works the way you're thinking
12:22karolherbst: oh huh.. no, you are right
12:22karolherbst: that's annoying...
12:22dwfreed: yeah
12:22dwfreed: you *could* go with +R instead of +M
12:23dwfreed: +R is can't join unless identified; +M is can't speak; +R can be overridden with a matching +I (which takes a mask like a ban would); +M cannot be overridden in a similar fashion
12:23karolherbst: mhhh
12:24karolherbst: yeah.. +R might be good enough here
12:24dwfreed: so then you'd set an +I that matches the bridge's v6 range, and it would be able to join the puppets without issue
12:24karolherbst: but does each user get a new connection or how does the masquerading work?
12:24dwfreed: new connection, yeah
12:25karolherbst: so does it join each nick automatically?
12:25dwfreed: yeah
12:25dwfreed: a connection to irc is created when they first speak; if they go offline in discord for a while, it'll disconnect them from IRC
12:25karolherbst: so in theory it could join with it's account and then change the nick?
12:25dwfreed: yeah, but that'd be noisy, and wouldn't work for +M
12:25karolherbst: ahh
12:25karolherbst: right
12:26karolherbst: discord has strong moderation features, so I'm totally not against giving the bot speak rights for all its connections
12:27dwfreed: oh, I guess another bot (or a client with a script) that is opped on IRC could give +v to the bridge users on join; that'd also override +M
12:27dwfreed: either or
12:27karolherbst: yeah...
12:27karolherbst: it might make sense to give the bot a normal interface in IRC anyway
12:28karolherbst: so you can interact/configure it or something
12:33karolherbst: dwfreed: but yeah, thanks for pointing out the bot, definetly something I want to toy around with
12:33dwfreed: feel free to ask me if you need help with anything
13:11fdobridge: <gfxstrand> Little to none but @marysaka was going to look into it
13:12fdobridge: <marysaka> I'm currently on it
13:13fdobridge: <marysaka> Still doing some researches but from what I'm currently seeing it should be hopefully simple
13:15fdobridge: <Sid> *Morgan Freeman voice: "Little did she know, it was not in fact simple"*
13:16fdobridge: <marysaka> me with the mesh shaders local invocation shenanigans
13:18fdobridge: <Sid> I wonder if Morgan Freeman or one of the video game trailer dudes are on Cameo
13:19fdobridge: <Sid> would pay to get a recording of them saying "little did they know, it was not easy"
13:24fdobridge: <gfxstrand> Hopefully it'll be easier than Mesh
13:25fdobridge: <karolherbst🐧🦀> is it more than just new instructions?
13:25fdobridge: <gfxstrand> I don't thinkso
13:25fdobridge: <gfxstrand> I don't think so (edited)
13:25fdobridge: <karolherbst🐧🦀> I do have the docs for those tho I think :ferrisUpsideDown: At least if that's the matrix multiplication ones
13:33fdobridge: <marysaka> It's new instructions from what I saw, nothing too fancy, just need to place everything in the appropriate packed way, still need to check if SPH changed a bit or not
13:35fdobridge: <marysaka> and some new fancy LDSM to load a matrix from shared memory
13:36fdobridge: <karolherbst🐧🦀> LDSM?
13:37fdobridge: <karolherbst🐧🦀> oh indeed
13:37dwfreed: karolherbst: are there any other channels our "friend" likes to talk in that I'm not already in that you know of?
13:37fdobridge: <karolherbst🐧🦀> LDSM is Turing+
13:38karolherbst: dwfreed: mhh?
13:38dwfreed: the guy that likes to spew random garbage
13:38karolherbst: ahh
13:38karolherbst: dunno, kinda all GPU/fdo related ones
13:38dwfreed: I'm in most of them already
13:39fdobridge: <karolherbst🐧🦀> @marysaka anyway, I have the docs on all of that :ferrisUpsideDown:
13:39fdobridge: <karolherbst🐧🦀> it's really not complicated
13:39fdobridge: <marysaka> sure, might help a bit but seems not complicated
13:39fdobridge: <marysaka> yeah
13:40fdobridge: <karolherbst🐧🦀> LDSM is pretty straightforward
13:40fdobridge: <karolherbst🐧🦀> though it helps to know which threads loads what 😄
13:40fdobridge: <karolherbst🐧🦀> well..
13:40fdobridge: <marysaka> I think it's mostly going to be me figuring out the encoding and seeing how to lower nicely for NAK
13:40fdobridge: <karolherbst🐧🦀> not even that, because all threads must be active anyway
13:41fdobridge: <marysaka> makes sense yeah, I think I will do everything without LDSM at first and add it after it's mostly done
13:41fdobridge: <karolherbst🐧🦀> within a warp I think
13:41fdobridge: <karolherbst🐧🦀> all threads in a warp have to be active for all those instructions
13:41fdobridge: <marysaka> yeah the matrix is spread across all invocations
13:42fdobridge: <karolherbst🐧🦀> anyway, looks fun
13:42fdobridge: <marysaka> the only thing I'm not sure is the layout per thread
13:42fdobridge: <karolherbst🐧🦀> ahh yeah.. I know the layout per thread 😛
13:43fdobridge: <marysaka> it seems like a row (so always 4 registers holding the values) but didn't tested all combinations yet
14:12fdobridge: <mhenning> there's some information on the tensor cores in section 4.3 of https://arxiv.org/pdf/1804.06826.pdf
14:28fdobridge: <marysaka> PTX docs also have some info from what I just saw <https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#register-fragments>
19:22Lyude: airlied: did that patch make things any more stable with MST on your end by the way?
20:07airlied: Lyude: nope, still fails, not sure if it fails in exact same way
20:08Lyude: gotcha, if you could check if it's the same I'd appreciate it - otherwise i can try to look more into this, although I might need to try a few setups since I don't think I'm seeing any problems with my old MST 1.2 hub
20:11Lyude: airlied, karolherbst, dakr - btw, csoriano had poked me about getting a write-up of the current status of nouveau with GSP. Do we think that the old nouveau status page would be a good place for this?
20:14karolherbst: Lyude: uhh.. probably? I kinda updated it for GSP support, but it's not really fine grained and it's all in the details
20:15airlied: Lyude: attached the newer logs to issue 339
20:15airlied: Lyude: do you have any of the Dell 4k monitors with the two panels?
20:28Lyude: airlied: you don't mean the P24Q or whatever from a while ago do you?
20:29Lyude: P2415Qb
20:30airlied: if it presents as two MST panels to do 4K@60, yes that one
20:32Lyude: ahhhh ok. that one is very uh, special
20:32Lyude: but iirc I think someone from intel managed to get it working
20:33Lyude: I know I mostly stopped using it in my tests because it just kept hitting so many issues that I never ran into at all on other hubs
20:33Lyude: I will see if I can reproduce the problem here then using that display