07:41AndrewR: ..because bugs doesn't fix itself (yet?) - can anyon throw any dea what I can enable in mesa source for tracking down diff between good and bad state (?) ?
07:41AndrewR: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9582 -this
09:18AndrewR: huh ... I was pleased to read this https://blogs.gnome.org/tbernard/2022/10/10/post-collapse-computing-2/ - but I guess idea a bit slow at being adopted ....
15:47fdobridge: <esdrastarsis> Minecraft 1.20.2 on nouveau opengl and zink
15:47fdobridge: <esdrastarsis> https://cdn.discordapp.com/attachments/1034184951790305330/1181260572281544704/20231204_12h41m58s_grim.jpeg?ex=658069a3&is=656df4a3&hm=a203923d7e8dcdcad8a558eb9b3519f10c701e04aeb719c1ac13a316f6093460&
15:47fdobridge: <esdrastarsis> https://cdn.discordapp.com/attachments/1034184951790305330/1181260572768092202/20231204_12h45m35s_grim.jpeg?ex=658069a3&is=656df4a3&hm=8b2caed666cc50edbbcd654548de0b418f8c9938773784ea99e89b2e9aab6495&
15:58fdobridge: <karolherbst🐧🦀> now make it crash :ferrisUpsideDown:
15:58fdobridge: <karolherbst🐧🦀> but you might also want to try out electron based apps like discord on top of zink as I think there are multi-threading bugs and I'm sure zink hits them
16:07fdobridge: <esdrastarsis> I think it needs vulkanMemoryModel, but its working
16:07fdobridge: <esdrastarsis> `Unsupported SPIR-V capability: SpvCapabilityVulkanMemoryModelDeviceScope (5346)`
16:07fdobridge: <karolherbst🐧🦀> I see
16:08fdobridge: <karolherbst🐧🦀> I know that there are use-after-free in zink for driver internal sub-allocations
16:08fdobridge: <karolherbst🐧🦀> not sure if GL is able to hit those or not
16:08fdobridge: <karolherbst🐧🦀> CL does
17:11fdobridge: <orowith2os> @gfxstrand any thoughts on raytracing with NVK?
17:11fdobridge: <orowith2os> mainly ooc
17:12fdobridge: <orowith2os> not sure if you've even poked at it yet :v
17:14fdobridge: <gfxstrand> Multi-threading bugs in what? NVK? I doubt it.
17:15fdobridge: <karolherbst🐧🦀> in zink
17:15fdobridge: <gfxstrand> Oh, sure. That I can believe easily.
17:15fdobridge: <karolherbst🐧🦀> I just don't know if zink hits them in GL 🙂
17:15fdobridge: <pixelcluster> there have been first strokes at acceleration structure building at least IIRC?
17:15fdobridge: <gfxstrand> That's on the fairly short list.
17:15fdobridge: <karolherbst🐧🦀> and if we go "zink only for GL" the question is if we want to have them fixed quickly or not 😄
17:15fdobridge: <gfxstrand> Yeah, @konstantinseurer has been playing with it.
17:16fdobridge: <pixelcluster> shader calls in NAK :frog_demon:
17:16fdobridge: <karolherbst🐧🦀> will that even be needed?
17:16fdobridge: <pixelcluster> sort of I guess
17:16fdobridge: <gfxstrand> RT-style shader calls are (or at least should be) easy
17:16fdobridge: <pixelcluster> I mean there's nir_lower_shader_calls which takes care of nonuniform calls
17:16fdobridge: <gfxstrand> The usual non-inlined functions is a bit harder.
17:16fdobridge: <pixelcluster> I mean there's nir_lower_shader_calls which takes care of nonuniformity in calls etc (edited)
17:17fdobridge: <konstantinseurer> I can already build nvidia acceleration structures, but I'm putting of reverse engineering the shader side
17:17fdobridge: <konstantinseurer> there are a bunch of new opcodes
17:18fdobridge: <karolherbst🐧🦀> are they known?
17:18fdobridge: <karolherbst🐧🦀> like..
17:18fdobridge: <karolherbst🐧🦀> does nvdisasm know them?
17:18fdobridge: <konstantinseurer> nvdisasm doesn't disassemble them
17:18fdobridge: <karolherbst🐧🦀> pain
17:18fdobridge: <karolherbst🐧🦀> but also figures...
17:18fdobridge: <karolherbst🐧🦀> the ISA docs I have also don't contain some mesh shader relevant instructions either
17:18fdobridge: <karolherbst🐧🦀> ...
17:19fdobridge: <konstantinseurer> Are there references to RT or opcodes?
17:19fdobridge: <karolherbst🐧🦀> no
17:19fdobridge: <karolherbst🐧🦀> I suspect they removed it all
17:19fdobridge: <orowith2os> late addition: both in-hardware and emulated
17:19fdobridge: <karolherbst🐧🦀> they also removed mesh/ray related bits from the class headers
17:20fdobridge: <prop_energy_ball> Yes, the open NV headers are limited to pascal featureset on newer GPUs fwiu
17:20fdobridge: <marysaka> yeah but at least mesh shaders instructions does disassemble fine at yeast...
17:20fdobridge: <marysaka> yeah but at least mesh shaders instructions does disassemble fine at least... (edited)
17:20fdobridge: <karolherbst🐧🦀> :blobcatnotlikethis:
17:22fdobridge: <gfxstrand> I'm sure we can figure them out. I'm guessing that NV RT looks a lot like Intel in that there's probably call shader and trace ray instructions.
17:22fdobridge: <karolherbst🐧🦀> yeah.. wouldn't surprise me
17:22fdobridge: <gfxstrand> And a bunch of one-off stuff like terminate raay
17:22fdobridge: <gfxstrand> And a bunch of one-off stuff like terminate ray (edited)
17:23fdobridge: <gfxstrand> And then some sysvals we have to find.
17:23fdobridge: <konstantinseurer> And also quite a bit of shader code. It looks similar to what amd is doing.
17:23fdobridge: <orowith2os> rewrite it in Rust, shrimple as
17:23fdobridge: <prop_energy_ball> You could also consider asking. I was visiting NV last week and this was brought up and the response was "someone should open an issue about that"
17:23fdobridge: <karolherbst🐧🦀> yeah....
17:23fdobridge: <karolherbst🐧🦀> so I was asking about tracking requests on github and the response I got was "please don't"
17:24fdobridge: <karolherbst🐧🦀> well.. anyway...
17:24fdobridge: <karolherbst🐧🦀> my last request also doesn't go anywhere
17:24fdobridge: <orowith2os> this all sounds like it'll be a Fun Time To Work On
17:24fdobridge: <karolherbst🐧🦀> it's always easy, they generally say "just make a request" and then you don't get anything for two years 🥲
17:24fdobridge: <prop_energy_ball> I also got this cool ruler
17:24fdobridge: <prop_energy_ball> https://cdn.discordapp.com/attachments/1034184951790305330/1181284942932607056/2023-12-04-17-24-07-974.jpg?ex=65808055&is=656e0b55&hm=7f2570de6df5c595eb334b185eb9a44432d79a6cb30314d42144894e6bff48b8&
17:24fdobridge: <prop_energy_ball> https://cdn.discordapp.com/attachments/1034184951790305330/1181284943431741450/2023-12-04-17-24-13-401.jpg?ex=65808055&is=656e0b55&hm=a4dab281de1a85d11329065e0f07330f35af87339aadc18b206e2c470b4c1038&
17:24fdobridge: <karolherbst🐧🦀> yeah, that one is cool
17:24fdobridge: <karolherbst🐧🦀> is it an updated one?
17:24fdobridge: <karolherbst🐧🦀> mine is from.. uhh...
17:24fdobridge: <karolherbst🐧🦀> old
17:25fdobridge: <orowith2os> oh @karolherbst were #zink :angy:
17:25fdobridge: <prop_energy_ball> No idea! Whats the newest chip yours has? :)
17:25fdobridge: <karolherbst🐧🦀> GP104
17:25fdobridge: <prop_energy_ball> zmike is on sabbatical atm 🐸
17:26fdobridge: <prop_energy_ball> Same
17:26fdobridge: <karolherbst🐧🦀> `02/19` is on mine
17:26fdobridge: <karolherbst🐧🦀> it's near the holes on one side
17:26fdobridge: <karolherbst🐧🦀> besides the Made in usa
17:27fdobridge: <prop_energy_ball> 29/17
17:27fdobridge: <karolherbst🐧🦀> mhh
17:27fdobridge: <karolherbst🐧🦀> and here I thought it would be month/year
17:27fdobridge: <orowith2os> REing NVIDIA goodies 🐸
17:28fdobridge: <karolherbst🐧🦀> but yeah.. mine looks exactly the same
17:29fdobridge: <prop_energy_ball> 1080 was the last real GPU before we got e=mc^2 +AI going on anyway
17:29fdobridge: <prop_energy_ball> https://cdn.discordapp.com/attachments/1034184951790305330/1181286102724788284/20231204_164304.jpg?ex=6580816a&is=656e0c6a&hm=1b1ec53e28fd0ad965c4b2f192f5a6255473bc5d4794a65fac89ec875313ba1a&
17:29fdobridge: <gfxstrand> WTH?!?
17:29fdobridge: <gfxstrand> That's not the way equations work.
17:29fdobridge: <gfxstrand> Or science
17:29fdobridge: <gfxstrand> Or anything, really.
17:31fdobridge: <karolherbst🐧🦀> pain
17:31fdobridge: <airlied> I made a req for cons raster which we already know, the response was along the lines of we had a lot of trouble releasing what we did, don't expect too much 🙂
17:31fdobridge: <karolherbst🐧🦀> welcome to business bs
17:32fdobridge: <airlied> I expect we need to create extra class headers for ourselves alongside the NVIDIA ones
17:32fdobridge: <airlied> For the RE bits
17:32fdobridge: <marysaka> That's what I'm currently doing for mesh shaders attm
17:32fdobridge: <marysaka> That's what I'm currently doing for mesh shaders atm (edited)
17:33fdobridge: <marysaka> It would be nice to mix the official headers and the extra headers when we parse them
17:33fdobridge: <!DodoNVK (she) 🇱🇹> The real nouveau experience 🧑💻
17:36fdobridge: <airlied> Yes that was the idea. That we write our headers the same way and parse both at once
17:36fdobridge: <airlied> Both for additional values and new classes
17:37fdobridge: <airlied> Probably should get the cons raster MR to do that first
17:39fdobridge: <marysaka> sure, I still have a bit to implement/fix on mesh shaders (hopefully will be done before the end of the year but will see)
17:59fdobridge: <mohamexiety> yooo I didn't know they still do these. want!
17:59fdobridge: <gfxstrand> @airlied @marysaka https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26507
17:59fdobridge: <gfxstrand> That should help us move in the right direction.
18:08fdobridge: <airlied> Looks like a good first step
18:37fdobridge: <gfxstrand> We can bikeshed the name of the mixin folder late.r
18:37fdobridge: <gfxstrand> We can bikeshed the name of the mixin folder later (edited)
19:21publicface: You have some AI algorithm that learns so: the more sense you make the more it sanctions. I am the mastermind, I pleasure my soul by giving honor to myself, call me simply master or Mart the Conqueror, Mart the King or Majesty Highness Mart!
19:44publicface: don't trip it , you are way over my skill set in linux base drivers (i studied those much later), i have many job offers too though, so i am fine, the compiler is just very simple if you follow the simplest banking paradigm. The system is good to be honest, since i already have the compilers in all license forms that i need. It takes me no effort, i got through of that already