00:36 fdobridge_: <g​fxstrand> Okay, I think I got 3D compressed working. Just need to wait for my desktop to become available to CTS it.
00:36 fdobridge_: <g​fxstrand> 1-line bug in NIL
01:00 fdobridge_: <k​arolherbst🐧🦀> the best bugs
01:58 withraindeer: It seems i got a brain freeze , do not know how to approach into a full compiler coding, you drive me nuts, however i seem to be getting only numbers to work both access and queue jumping , i do not know how linking could be done at all, i imagine there is some numbering or block annotation encoding spec, but i am not familiar with any of that.
02:30 gaunitar: i think, they can fuck off with their stinky cambodia and my dads, in other words my million dollars, in the end i just make sure they end up dead, it was not possible to live their same way as in estonia, everyone harassed, i will be killing off many of you too as soon as the more extensive war starts
02:31 fdobridge_: <k​arolherbst🐧🦀> @gfxstrand fyi (as you were involved in that API pointer mess as well) https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27376 😄 I think it's way nicer now
04:13 fdobridge_: <a​irlied> @gfxstrand btw I just completed a full turing 2070 run with your branch + the zeroing fix, just ran ./deqp-vk, got 146 fails
04:17 fdobridge_: <g​fxstrand> What CTS branch?
04:17 fdobridge_: <a​irlied> vulkan-cts-
04:17 fdobridge_: <a​irlied> with that fix on top
04:17 fdobridge_: <a​irlied> just generating the csv file to see what failed now
04:19 fdobridge_: <g​fxstrand> Hrm...
04:19 fdobridge_: <g​fxstrand> Do you have 27365?
04:19 fdobridge_: <a​irlied> I think some of the dEQP-VK-experimental snuck in though
04:20 fdobridge_: <a​irlied> I should just use mustpass and do it again 🙂
04:20 fdobridge_: <g​fxstrand> And are any of those devices lost?
04:20 fdobridge_: <g​fxstrand> Do you have !27365? (edited)
04:21 fdobridge_: <a​irlied> one is unknown extension VK_KHR_index_type_uint8
04:23 fdobridge_: <a​irlied> wow the csv conversion is slow on a full run 😛
04:29 fdobridge_: <a​irlied> yeah all seem experimental apart from one
04:29 fdobridge_: <a​irlied> dEQP-VK.info.device_extensions is that one
04:30 fdobridge_: <g​fxstrand> The only way I've been able to reproduce it is to do an actual conformance run as per the CTS instructions.
04:30 fdobridge_: <g​fxstrand> Unfortunately, it reproduces on every incantation of an actual conformance run I've attempted
04:31 fdobridge_: <g​fxstrand> If you run with 8 shards, it breaks in shard 4
04:31 fdobridge_: <a​irlied> I did an unsharded run
04:31 fdobridge_: <g​fxstrand> Yeah but you didn't use the mustpass list
04:31 fdobridge_: <g​fxstrand> You have to use the mustpass list
04:31 fdobridge_: <a​irlied> I'll kick it off again using mustpass
04:32 fdobridge_: <a​irlied> my office aircon needs a workout
04:32 fdobridge_: <g​fxstrand> I haven't actually tried with a single shard
04:32 fdobridge_: <g​fxstrand> They've all been sharded runs
04:33 fdobridge_: <g​fxstrand> Which have even more special instructions because there's an every-shard mustpass too
04:39 fdobridge_: <g​fxstrand> I wonder if I bought a 2nd Ampere if the CTS would go 2x as fast....
04:40 fdobridge_: <g​fxstrand> I mean normally it's pretty compiler-bound but I'm doing 18 thread runs so it may be bottleknecking on the GPU
04:54 fdobridge_: <g​fxstrand> 1/3 of the way in and it looks good so far
05:08 fdobridge_: <r​edsheep> What's the next big priority after formal 1.3 conformance? Is it vkd3d support, or 2024 roadmap, or performance, or something else?
05:15 fdobridge_: <a​irlied> build an nvidia based steam deck
05:27 fdobridge_: <r​edsheep> I bet you really could with a GA10B... I'm not *quite* enough of a hardware guru to make that work though 😄
05:30 fdobridge_: <g​fxstrand> I'll probably spend some time on performance before digging into D3D12.
05:30 fdobridge_: <g​fxstrand> Also, we should spend more time on Zink.
05:31 fdobridge_: <r​edsheep> Oh yeah that would be good. I assume the nastiness around WSI will make for some serious difficulties getting zink to run a compositor though...
05:32 fdobridge_: <a​irlied> what nastiness around WSI?
05:32 fdobridge_: <r​edsheep> It has to be disabled in the CTS for it to pass, right? Is that unrelated?
05:32 fdobridge_: <a​irlied> I doubt one wierd cts/sync interaction is going to be the problem stopping zink running a compositor
05:32 fdobridge_: <a​irlied> it's passing on my machine
05:33 fdobridge_: <g​fxstrand> Yeah, I think that interaction is pretty weird. I want to figure out out but dang.
05:33 fdobridge_: <a​irlied> I'm going to guess @gfxstrand faster processor is triggering a race
05:33 fdobridge_: <g​fxstrand> I've got my methods for triage of that sort of thing. I need to try again.
05:34 fdobridge_: <a​irlied> like I wouldn't be shocked that there are even harder to find races in the nouveau fence code 😛
05:34 fdobridge_: <a​irlied> though I'd expect more like the fence hangs and timeout than the mmu error
05:35 fdobridge_: <a​irlied> it would be good to look at the wierd depth issue with zink and stk
05:35 fdobridge_: <a​irlied> but it might be worth finishing the depth exts before then so we know cts has covered all it can
05:36 fdobridge_: <r​edsheep> Right, hopefully now that CTS mostly works it will be easier to know that stuff like performance changes aren't messing things up right?
05:36 fdobridge_: <r​edsheep> I assume that's most of the point of the CTS
05:37 fdobridge_: <r​edsheep> Does it get ran during CI once a driver is conformant?
05:37 fdobridge_: <a​irlied> not unless someone has a bunch of nvidia hardware we don't know about
05:38 fdobridge_: <g​fxstrand> Yup! That's exactly why I work CTS-first. I can effectively regression test now. The CTS is far from perfect but it's really quite good.
05:38 HdkR: NVIDIA based SteamDeck using a Tegra SoC </s>
05:38 fdobridge_: <g​fxstrand> I need to talk to the Valve folks about enabling it again.
05:39 fdobridge_: <r​edsheep> Is the issue actually just lack of nvidia hardware to run CI on? Honestly I wouldn't mind buying and shipping off one basic card from each gen if that meant this stuff getting tested better
05:39 fdobridge_: <r​edsheep> Cound probably get one of each gen post turing for less than half what I spent on my current gpu lol
05:41 fdobridge_: <r​edsheep> If that means this card working well sooner I would do it
05:41 fdobridge_: <a​irlied> lack of hardware, and places to host it, and someone to manage it all
05:41 fdobridge_: <a​irlied> the first is probably the easiest to solve
05:41 fdobridge_: <r​edsheep> Ah. Yeah that's harder.
05:43 fdobridge_: <g​fxstrand> The Valve folks are ready to host it and they've got an Ampere to run it on. I just need to talk to @mupuf and eric_engstrom about turning it on.
05:43 fdobridge_: <a​irlied> yeah it's just if we have enough hw to scale to nightly or what % of cts we can get away with
05:43 mupuf: gfxstrand: yeah, the ga106 is coming back very soon
05:44 mupuf: Eric has been asking me to replug it for a while
05:45 mupuf: Actually, I could replug it today and simply remove the polaris10 since we have more of these in the kws farm
05:45 fdobridge_: <g​fxstrand> Full CTS on my box is 1.5 hours or so. If we want 15 min runs, we're looking at probably running any 1/8th of the CTS.
05:47 mupuf: Yeah, don't aim for pre-merge, just post merge and manual testing
05:48 mupuf: I can't speak for Valve, but if NVK becomes a viable option for gaming, I'm sure we could host more hardware
05:48 mupuf: And get you to pre-merge testing
05:53 mupuf: Oh, and we have 5950X CPUs, so CTS is much faster than that! On radv that be closer to 55minutes
06:29 fdobridge_: <a​irlied> @gfxstrand oh also we still have s8 to figure out
06:33 fdobridge_: <a​irlied> parallel piglit on zink also seems to take some mmu faults
07:36 fdobridge_: <!​DodoNVK (she) 🇱🇹> So when FH4 starts working I'll able to run it at 1080p60?
07:41 fdobridge_: <!​DodoNVK (she) 🇱🇹> https://mstdn.games/@ReverseModule/111847905246233171
08:35 fdobridge_: <d​jdeath3483> Is someone looking at VK_KHR_dynamic_rendering_local_read on NVK?
08:37 fdobridge_: <d​jdeath3483> I kind of messed up the runtime bits : https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27380
08:41 i509vcb: redsheep: I spent a bit of time just reading around nouveau to get an idea what I'd need to do for ga10b, hopefully not painful but I don't know yet lol
09:47 fdobridge_: <p​rop_energy_ball> Why wouldn't NV work on Wine 9?
11:09 fdobridge_: <S​id> can't relate, nv and wine 9 working fine for me
11:10 fdobridge_: <S​id> 550 driver on GTX turing too
11:29 fdobridge_: <a​irlied> @gfxstrand okay the single shard full run has one fail, the uint8 extension one above
11:29 fdobridge_: <a​irlied> if we revert turning that ext on, I expect I could generate a full qpa with wsi if you want it
12:52 fdobridge_: <g​fxstrand> I've honestly have mostly ignored that one.
12:53 fdobridge_: <g​fxstrand> Interesting. Unfortunately, I can't submit 1-shard because it'll OOM the qpa file on 32-bit. 🤡
12:54 fdobridge_: <!​DodoNVK (she) 🇱🇹> What even is a shard?
12:54 fdobridge_: <g​fxstrand> I think if we update to 1.3.8, they've reduced the log file size enough that it should fit again but then we have to fix all the new tests and test bugs.
12:55 fdobridge_: <d​jdeath3483> Alright, I'll try to get some attention from radv
12:55 fdobridge_: <g​fxstrand> The Vulkan CTS is so large that we often break it into "shards". So it'll run like every 8th test or something.
12:56 fdobridge_: <g​fxstrand> And then the next shard is every 8th test plus one and so forth.
12:57 fdobridge_: <g​fxstrand> The exact algorithm doesn't matter. The important part is that it deterministically breaks the CTS into more manageable pieces.
12:57 fdobridge_: <!​DodoNVK (she) 🇱🇹> So 1-shard is a full CTS test, right?
12:58 fdobridge_: <g​fxstrand> Yes
12:58 fdobridge_: <g​fxstrand> A 4-shard run is also the full CTS, just run as 4 different pieces.
12:58 fdobridge_: <g​fxstrand> Well, depending on how sloppy one is with their English. 🙃
12:59 fdobridge_: <!​DodoNVK (she) 🇱🇹> Does it help with running CTS on under-powered hardware?
13:04 fdobridge_: <k​arolherbst🐧🦀> yes, as you can run a different shard on each runner
13:04 fdobridge_: <k​arolherbst🐧🦀> so you can just scale through having more hardware
13:07 fdobridge_: <m​ohamexiety> that's interesting, I thought for conformance you _needed_ it to be single threaded. by running multiple shards you can have some sort of concurrency and speed things up a bit
13:31 fdobridge_: <g​fxstrand> You're allowed to run up to 16 shards for a conformance run.
13:32 fdobridge_: <g​fxstrand> That's still like a whole day on phones but at least it's tractable.
13:49 fdobridge_: <e​sdrastarsis> NVK 5% slower, but the resolution is different 🐸
13:53 fdobridge_: <S​id> from what I've seen in my own gaming, NVK is anywhere between abysmal perf to ~75% of proprietary driver
13:53 fdobridge_: <S​id> same settings used
13:54 fdobridge_: <S​id> some games still don't launch because of some kernel issues
13:57 fdobridge_: <S​id> and some don't launch because of missing extensions that dxvk presumably needs
14:05 fdobridge_: <!​DodoNVK (she) 🇱🇹> What games?
14:06 fdobridge_: <S​id> outer wilds doesn't launch, that's the only dx11 one I've seen so far
14:07 fdobridge_: <!​DodoNVK (she) 🇱🇹> What does it do?
14:07 fdobridge_: <S​id> black screen
14:07 fdobridge_: <S​id> no mangohud either
14:07 fdobridge_: <S​id> nothing in dmesg
14:08 fdobridge_: <S​id> and, well, you already know the situation with vkd3d-proton :>
14:09 fdobridge_: <!​DodoNVK (she) 🇱🇹> Outer Wilds is a D3D11 game though
14:09 fdobridge_: <!​DodoNVK (she) 🇱🇹> Can you get the Proton (?) logs?
14:11 fdobridge_: <!​DodoNVK (she) 🇱🇹> And DXVK logs (if they still exist on Proton)
14:11 fdobridge_: <S​id> I know, just mentioning it for `reasons why some games don't work`
14:12 fdobridge_: <S​id> I can get the logs in some time yes
15:10 fdobridge_: <g​fxstrand> At this point, if D3D11 games are failing to start or misgendered, we should be filing bugs with logs and traces (when appropriate).
15:10 fdobridge_: <r​hed0x> misgendered 🙃
15:13 fdobridge_: <k​arolherbst🐧🦀> :ferrisUpsideDown:
15:14 fdobridge_: <g​fxstrand> At this point, if D3D11 games are failing to start or misrendered, we should be filing bugs with logs and traces (when appropriate). (edited)
15:15 fdobridge_: <g​fxstrand> I mean...
15:39 fdobridge_: <g​fxstrand> Yeah... That one's going to be annoying, isn't it?
15:41 fdobridge_: <g​fxstrand> 3D compressed is merged now. That should help Zink. IDK if the bug i fixed affects anything else. I kinda don't think so but it may.
15:45 fdobridge_: <h​untercz122> Lethal Company wanted 3d compressed textures
15:45 fdobridge_: <h​untercz122> one patch less nice
15:49 fdobridge_: <g​fxstrand> Cool
15:53 fdobridge_: <k​arolherbst🐧🦀> oh wow.. rust is almost at 1% in mesa :ferrisUpsideDown:
15:53 fdobridge_: <k​arolherbst🐧🦀> (behind assembly which is at 1.1%)
15:55 fdobridge_: <S​id> ok, a quick reboot to grab outer wilds logs
15:55 fdobridge_: <S​id> then back to nvidia gaming
15:55 fdobridge_: <S​id> because I wanna stream my game to frens e-e
15:56 fdobridge_: <t​riang3l> Would VK_NV_texture_compression_vtc theoretically be somehow compatible with Vulkan?
15:57 fdobridge_: <t​riang3l> Would VK_NV_texture_compression_vtc theoretically be somehow compatible with Vulkan's definitions of textures? (edited)
16:03 fdobridge_: <g​fxstrand> Probably, though IDK who actually uses it.
16:04 fdobridge_: <g​fxstrand> There are also 3D ASTC formats but NVIDIA doesn't support ASTC except on Tegra.
16:07 fdobridge_: <t​riang3l> The font in Tales of Vesperia on the PS3 :blobcatnotlikethis:
16:07 fdobridge_: <g​fxstrand> Fun...
16:07 fdobridge_: <g​fxstrand> I don't think Vulkan has a format for it, though, so it would need an extension. 🙃
16:08 fdobridge_: <n​anokatze> you know es 1.x has paletted formats? which break lots of assumptions pretty significantly
16:08 fdobridge_: <n​anokatze> you know es 1.x has paletted formats? which break lots of things you get used to assuming about sane formats pretty significantly (edited)
16:08 fdobridge_: <n​anokatze> e.g. you can't consider each mip in isolation
16:08 fdobridge_: <n​anokatze> because the palette is at the start of the image and shared with all mips
16:09 fdobridge_: <n​anokatze> I'm sure you could shove absolutely random stuff like that into the api if you really wanted but hopefully nobody thinks of doing that
16:09 fdobridge_: <t​riang3l> VTC is just 2 or 4 layers of S3TC interleaved though, so probably not as important as ASTC
16:09 fdobridge_: <n​anokatze> I'm sure you could shove absolutely random crazy stuff like that into the api if you really wanted but hopefully nobody thinks of doing that (edited)
16:10 fdobridge_: <n​anokatze> does current hw even support that or why'd you want an ext for that
16:10 fdobridge_: <n​anokatze> does current hw even support that or why'd you even think about an ext for that (edited)
16:11 fdobridge_: <t​riang3l> It does :happy_gears: http://opengl.gpuinfo.org/listreports.php?extension=GL_NV_texture_compression_vtc
16:12 fdobridge_: <t​riang3l> Although I'm not sure if it's supported in hardware or just on pack/unpack
16:12 fdobridge_: <n​anokatze> and are you sure that's not just a thing they do on upload
16:13 fdobridge_: <g​fxstrand> I'm not sure how. There doesn't seem to be a format for it. There's probably a bit somewhere.
16:13 fdobridge_: <n​anokatze> and are you sure that's not just a thing they implement in copies instead of there being actual hw support (edited)
16:13 fdobridge_: <g​fxstrand> That's entirely possible. Just swizzle it differently and do copies with shaders.
16:14 fdobridge_: <t​riang3l> I think from the Vulkan perspective the palette theoretically can be considered an aspect
16:14 fdobridge_: <g​fxstrand> Vulkan doesn't have pallette textures.
16:14 fdobridge_: <t​riang3l> Would be interesting to play with textures the PS2 way
16:15 fdobridge_: <n​anokatze> maybe, I'd rather not think about it
16:16 fdobridge_: <t​riang3l> while using hardware anisotropic filtering (though the PS2 didn't have it) :cursedgears:
16:33 fdobridge_: <S​id> got sidetracked
16:33 fdobridge_: <S​id> updating my installation of nvk beforr grabbing logs
16:52 fdobridge_: <S​id> so one of the commits fixed outer wilds
16:52 fdobridge_: <S​id> but it borks if you alt-tab :D
16:53 fdobridge_: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1202295647710232616/steam-753640.log?ex=65ccf013&is=65ba7b13&hm=f1b820fffb9f00504cfe26a3863b2c29ad6977db144c0c38155bf87ec2cebee3&
16:53 fdobridge_: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1202295648033190008/OuterWilds_d3d11.log?ex=65ccf014&is=65ba7b14&hm=21beed0406222b35cf8404c9bcb67c577ee956a3e9f7e84a70c8aeeaa4f7cf48&
16:53 fdobridge_: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1202295648372654100/OuterWilds_dxgi.log?ex=65ccf014&is=65ba7b14&hm=0ec5727fde85912171d35c3dce3367d34efcedd343064b579ec731d6b5a7d81f&
17:05 fdobridge_: <S​id> oh, I see...
17:06 fdobridge_: <S​id> when I alt tab..
17:06 fdobridge_: <S​id> ```
17:06 fdobridge_: <S​id> Jan 31 22:32:48 strogg kernel: Asynchronous wait on fence drm_sched:nouveau_sched:61ac timed out (hint:submit_notify [i915])
17:06 fdobridge_: <S​id> Jan 31 22:32:49 strogg kernel: nouveau 0000:01:00.0: gsp: mmu fault queued
17:06 fdobridge_: <S​id> Jan 31 22:32:49 strogg kernel: nouveau 0000:01:00.0: gsp: rc engn:00000001 chid:24 type:31 scope:1 part:233
17:06 fdobridge_: <S​id> Jan 31 22:32:49 strogg kernel: nouveau 0000:01:00.0: fifo:001001:0003:0018:[OuterWilds.exe[9893]] errored - disabling channel
17:06 fdobridge_: <S​id> Jan 31 22:32:49 strogg kernel: nouveau 0000:01:00.0: OuterWilds.exe[9893]: channel 24 killed!
17:06 fdobridge_: <S​id> ```
17:06 fdobridge_: <S​id> I also had a nvkm related RIP in the log that shows on screen when shutting down (systemd) but journalctl does not record that far
17:10 fdobridge_: <!​DodoNVK (she) 🇱🇹> So will you create a R600 backend for PCSX2?
17:15 fdobridge_: <t​riang3l> It doesn't have paletted textures… let alone nice ways to express PS2's blending capabilities
19:25 fdobridge_: <a​irlied> @gfxstrand can you give me the shard script you are using? Btw I assume your wsi bugs aren't just 32 bit only
19:27 fdobridge_: <g​fxstrand> No, they repro 64-bit
19:28 fdobridge_: <g​fxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1202334609636077618/run-conformance.sh?ex=65cd145d&is=65ba9f5d&hm=41c382313eee69438682cfe159b87b9568423474c46e6af7f7b1e13751adc963&
19:48 fdobridge_: <a​irlied> @zmike. [28820/28820] skip: 1841, pass: 26358, warn: 11, fail: 577, crash: 31 one at a time zink on piglit completed
19:48 fdobridge_: <a​irlied> nvk/zink
19:57 fdobridge_: <g​fxstrand> Ooh! Rust 1.77 added `offset_of()`!
19:57 fdobridge_: <g​fxstrand> Ooh! Rust 1.77 added `offset_of!()` (edited)
19:57 fdobridge_: <g​fxstrand> We needed that sooooo bad
19:58 fdobridge_: <z​mike.> Maybe sounds about right... Tried running it again earlier today but my new kernel has zero hang recovery so I gave up pretty quick
19:59 fdobridge_: <a​irlied> when run in parallel it seems to hit a fence crash
19:59 fdobridge_: <a​irlied> fence race, where lots of processes get stuck
19:59 fdobridge_: <a​irlied> hang recovery should be fine though
20:01 fdobridge_: <z​mike.> 🤔
20:01 fdobridge_: <z​mike.> I didn't look too closely
20:01 fdobridge_: <z​mike.> I had a ticket open for piglit fails
20:04 fdobridge_: <z​mike.> I keep hoping 6.7 will hit f39 soon so I can do real testing
20:04 fdobridge_: <z​mike.> Alas
20:09 arpeas: Hello, I'm using nouveau on a GeForce 2 Go (NV11). I had to recompile the Slackware 15 kernel to even get the module to load, and everything works except going into suspend, which whites the display out.
20:09 arpeas: When resuming, the system displays garbage barely reminiscent of what the image *should* be.
20:10 arpeas: In framebuffer the text is all jumbled up, but displayed at a proper resolution.
20:11 fdobridge_: <a​irlied> https://people.freedesktop.org/~airlied/piglit/nvk/ is the results
20:12 fdobridge_: <a​irlied> the ubo ones are probably the biggest standout
20:13 arpeas: I should mention that this also occurs on a PowerBook G4, with a GeForce 4 420 Go.
20:13 fdobridge_: <a​irlied> the sparse ones are wierd, like zink is unreffing a bunch of resources, and submitting more works causes a vm fault, need to dig in a bit more, the tests pass then blow up in the teardown
20:14 fdobridge_: <g​fxstrand> The ANV people are having trouble with Zink sparse as well, BTW.
20:16 fdobridge_: <a​irlied> yeah I think zink sparse is a bit sparsely tested 😛
20:16 fdobridge_: <d​jdeath3483> Yeah that should be fixable I think
20:16 fdobridge_: <d​jdeath3483> just need to round mapping ranges to a page
20:33 fdobridge_: <p​homes_> Baldurs gate 3 crashes in nvk_heap_upload when it is calling nvk_heap_alloc_locked when the heap is already at 256MB. I should try with the DMA engine patches
20:33 fdobridge_: <g​fxstrand> Yeah, that needs the DMA patches.
20:34 fdobridge_: <g​fxstrand> I'm trying to fix them up right now. Hitting some weird valgrind errors
20:43 fdobridge_: <g​fxstrand> Latest version (just pushed) is looking pretty good 5m into the CTS run
20:51 fdobridge_: <g​fxstrand> 10 min, no fails
20:51 fdobridge_: <p​homes_> BG3 is different with the DMA patches. Now I get ./src/nouveau/vulkan/nvk_cmd_pool.c:29: VK_ERROR_OUT_OF_DEVICE_MEMORY
20:52 fdobridge_: <g​fxstrand> Ugh... That might be running out of BAR, too.
20:56 fdobridge_: <g​fxstrand> We're setting `NOUVEAU_WS_BO_GART`, though...
21:02 Lyude: https://paste.centos.org/view/2b8a0cc4 hm, managed to finally get rvkms compiling with the asahi DRM rust bindings - but unfortunately it seems like we're still crashing immediately on driver load
21:02 Lyude: with a uh, very obtuse dmesg to boot :(
21:03 Lyude: https://gitlab.freedesktop.org/lyudess/linux/-/commits/rvkms (updated work btw!)
21:04 Lyude: airlied, dakr ^ btw. this looks basically the same as the message I was getting before using the bindings so I'm wondering if maybe there's something silly I'm missing that's needed to get rust kernel modules to load properly?
21:07 dakr: Lyude, try adding ibt=off to the kernel command line
21:08 Lyude: had a feeling it'd be something like that :P, giving that a shot now
21:09 Lyude: YES wooooo
21:09 Lyude: rvkms loads!!
21:09 dakr: great!
21:13 Lyude: now to figure out why it's not calling probe :P
21:15 Lyude: oh I think I see, I'm going to have to hack stuff a bit more so all of this happens in the real module init. presumably asahi's module_platform_driver! is likely relying on some other kernel mechanism to initiate a probe when it finds an agx gpu
21:29 fdobridge_: <S​id> does B0 refer to the BAR?
21:30 fdobridge_: <a​irlied> No
21:30 fdobridge_: <S​id> ok just making sure I understood it right
22:21 fdobridge_: <!​DodoNVK (she) 🇱🇹> Literally Buffer Object
22:22 Lyude: oh hey someone on mastodon just linked me to bindings for doing exactly what I need
22:26 Lyude: https://mairacanal.github.io/rust-for-vgem/
23:06 airlied: Lyude: maira is on #dri-devel usually
23:06 Lyude: yeah I've messged them before :), going to go through their work and see what I can pull in
23:08 Lyude: btw airlied: sometime this week I'd like to actually write up a plan for some of the plans that we have around the modesetting driver for nova and post it on the relevant MLs if you think that's a good idea
23:10 airlied: Lyude: yup would be good
23:25 fdobridge_: <p​homes_> I tested a few other games with the DMA engine patches. They worked well both before and after the DMA