03:03 airlied[d]: https://github.com/tinygrad/tinygrad/blob/d7ca048022490032fb6f7e9c204a9109da09cb0b/extra/nv_gpu_driver/clcec0qmd.h RE
04:08 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1375686901473738842/image.png?ex=683297c4&is=68314644&hm=fa29cd3be52bdeaecefeb6cc02acdce82d2f046f136fc4515cbbf3da50bf840c&
04:08 mangodev[d]: i want to throw someone
04:08 mangodev[d]: not anyone here, just someone
04:09 mangodev[d]: probably someone at mozilla because AAAAAAAAAAAAAAAA
04:12 mangodev[d]: best i can do https://crash-stats.mozilla.org/report/index/99eb6fa9-9c3b-44ee-9eec-0fb4e0250524#tab-rawdump
04:12 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1375687933251682396/image.png?ex=683298ba&is=6831473a&hm=71c09a3390c8decc3e862d531f988554cf1163b347d08f472c7f4785afde6799&
04:12 mangodev[d]: at least it gives an instruction
04:14 mhenning[d]: that looks like mesa symbols are missing
04:45 gfxstrand[d]: airlied[d]: Nice! Sadly, doesn't seem to have any information I didn't already know. 😢
04:52 mangodev[d]: mhenning[d]: but i'm using a debug build :(
14:30 mangodev[d]: i'm curious
14:30 mangodev[d]: what does https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35141 exactly do? is it a bugfix for an issue caused by the postpass scheduler?
15:55 mhenning[d]: mangodev[d]: basically, yes. right now the scheduler is allowed to reorder some things that it probably shouldn't (although I don't think it will actually make a difference in practice). That commit prevents it from reordering them
15:55 mangodev[d]: mhenning[d]: ah
15:56 mangodev[d]: while you're here
15:56 mangodev[d]: great job on the second regression fix in a row!
15:56 mangodev[d]: back to back
15:56 mangodev[d]: i already feel major improvement from the last optimization, can't believe there's already another
15:57 mhenning[d]: aww thanks
15:58 mangodev[d]: just curious
15:58 mangodev[d]: what's the practical difference between `Spills/Fills from memory` versus `Spills/Fills to reg`?
15:58 mhenning[d]: the spill_values one isn't technically a regression fix because it never worked in the first place but yeah, should improve perf a bit more
15:58 mangodev[d]: i've noticed that from/to reg has helped with texture reads a ton
16:00 mhenning[d]: from memory goes between the main register file and general memory. these are generally pretty slow
16:00 mangodev[d]: the pr from yesterday made anti-aliasing stop sucking as much on nvk anymore
16:01 mangodev[d]: before, just enabling fxaa would drop me from 80fps to 50fps
16:01 mangodev[d]: spectacle's lag is far more bearable too
16:01 mhenning[d]: from reg goes between two different types of registers (nvidia has a bunch of different register files). these tend to be a bit faster
16:03 mhenning[d]: mangodev[d]: yeah, that pr specifically fixed some issues that happen with texture ops. It's possible we're handling some of those cases a lot better
16:20 gfxstrand[d]: gfxstrand[d]: Just pushed a couple more
16:20 gfxstrand[d]: Kepler A doesn't have 64-bit atomics and that created a bit of havoc
16:20 gfxstrand[d]: I think MSAA is the only thing missing now
16:21 kar1m0[d]: Guys I installed the drivers on my old laptop
16:21 kar1m0[d]: I can test the drivers now
16:22 kar1m0[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1375871509800943726/20250524_191607.jpg?ex=683343b2&is=6831f232&hm=4a2e0c16aa5b9f7b28ae10479d1bb51700e42c529a09f79cc0f6832855fb07c6&
16:22 kar1m0[d]: Do you have any specific suggestions on what I should test?
16:23 gfxstrand[d]: gfxstrand[d]: And only R32G32B32A32 4x and 8x. Everything else is fine. Weird...
17:01 snowycoder[d]: kar1m0[d]: That's a KeplerB, the main branch is a bit behind, but textures and basic compute stuff should be working (expect it to be slow though)
17:04 kar1m0[d]: snowycoder[d]: it's better than nothing
17:04 kar1m0[d]: also I will test some games
17:04 kar1m0[d]: with mangohud
17:04 kar1m0[d]: and send the results here
17:05 snowycoder[d]: Oh god, tell me if works :3
17:05 snowycoder[d]: I've only run conformance tests on it
17:07 kar1m0[d]: I installed the nouveau drivers on manjaro so I don't know if I have the latest version of them
17:07 kar1m0[d]: but I will test it regardless
17:09 kar1m0[d]: is there any other software besides mangohud that I can borrow results from?
17:14 snowycoder[d]: Benchmarks? I don't really know, I'm new here
17:21 tiredchiku[d]: dxvk has a hud that can give you some info but that's about it
17:22 tiredchiku[d]: hw monitoring isn't much of a thing on nouveau yet
17:22 tiredchiku[d]: mangohud still works for framerates/times though
18:03 snowycoder[d]: tiredchiku[d]: That could be great though
18:04 kar1m0[d]: I updated nvidia drivers today and my dgpu isn't work mmm nvidia
18:04 kar1m0[d]: how I love it when my 2000$ gaming laptop can't do anything without nvidia drivers
18:20 kar1m0[d]: snowycoder[d]: tiredchiku[d] sorry for the ping but so far I have tried making the witcher 3 run but I don't even get the vulkan shaders compiling screen
18:20 tiredchiku[d]: what distro
18:20 tiredchiku[d]: and what gpu
18:21 kar1m0[d]: manjaro and geforce 920m
18:21 tiredchiku[d]: kepler..
18:21 kar1m0[d]: can I somehow see some sort of logs?
18:21 tiredchiku[d]: do you have vulkan-nouveau and lib32-vulkan-nouveau installed?
18:22 tiredchiku[d]: assuming you're trying to run NVK
18:22 kar1m0[d]: I have installed all of the drivers I am pretty sure at least what the system required me to
18:23 tiredchiku[d]: unfortunately that tells me nothing .-.
18:23 tiredchiku[d]: I'm not familiar with manjaro's hw detection and driver recommendation system
18:23 tiredchiku[d]: could you quickly run `pacman -Q | grep -iE 'nvidia|nouveau'`
18:24 tiredchiku[d]: and share the output
18:25 kar1m0[d]: One second
18:25 kar1m0[d]: Okay I managed to boot the witcher 3 on minimal settings
18:26 kar1m0[d]: But
18:26 kar1m0[d]: It's very slow
18:26 kar1m0[d]: And I get about 8fps
18:26 tiredchiku[d]: is this the enhanced edition of witcher 3, or the old one
18:26 kar1m0[d]: Unfortunately mangohud doesn't display any information about the gpu usage
18:26 kar1m0[d]: tiredchiku[d]: Enhanced
18:26 tiredchiku[d]: that's kinda expected then, no?
18:27 kar1m0[d]: Okay I think it might run on the igpu
18:27 snowycoder[d]: Did you use any flags? it might be using the old compiler.
18:27 snowycoder[d]: If it is, the new driver will be even slower for now😂
18:27 tiredchiku[d]: either way, could be software rendering, or the iGPU yeah
18:27 kar1m0[d]: Because I see it is used to 95%
18:27 kar1m0[d]: snowycoder[d]: Nope I did not compile a thing
18:27 tiredchiku[d]: snowycoder[d]: or that
18:27 snowycoder[d]: (I'm not sure that the new compiler is already released though)
18:27 tiredchiku[d]: don't think it is
18:27 kar1m0[d]: No vulkan shaders compilation
18:28 kar1m0[d]: My dgpu is seen but there is no info about it
18:28 kar1m0[d]: Nor does it use it in the witcher 3
18:28 tiredchiku[d]: tiredchiku[d]: ^
18:28 snowycoder[d]: vulkan for kepler cards is disabled unless you start programs with `NVK_I_WANT_A_BROKEN_VULKAN_DRIVER=1`
18:28 tiredchiku[d]: no info about dgpu does confirm nouveau kernel driver though, yes
18:30 kar1m0[d]: tiredchiku[d]: pacman -Q | grep -iE 'nvidia|nouveau'
18:30 kar1m0[d]: lib32-vulkan-nouveau 1:25.0.5-1
18:30 kar1m0[d]: mhwd-nvidia 570.144-3
18:30 kar1m0[d]: mhwd-nvidia-390xx 390.157-20
18:30 kar1m0[d]: mhwd-nvidia-470xx 470.256.02-12
18:30 kar1m0[d]: nvidia-driver-assistant 0.21.51.03-1
18:30 kar1m0[d]: vulkan-nouveau 1:25.0.5-1
18:30 kar1m0[d]: xf86-video-nouveau 1.0.18-1
18:30 tiredchiku[d]: oh lord
18:31 tiredchiku[d]: oh
18:31 tiredchiku[d]: oh mhwd is the installer script
18:31 kar1m0[d]: I booted with open source drivers
18:31 kar1m0[d]: not sure about nvidia there
18:31 kar1m0[d]: and what it is
18:32 tiredchiku[d]: okay, might just be a case of needing the env var snowy mentioned then
18:32 kar1m0[d]: no idea what ur talking about but if you 2 know it's enough
18:32 kar1m0[d]: still no idea why my igpu is used though
18:32 kar1m0[d]: instead of the dgpu
18:33 tiredchiku[d]: NVK does not load by default on your GPU because it is too old
18:33 tiredchiku[d]: NVK being the open source vulkan driver
18:33 kar1m0[d]: how do I fix it
18:33 kar1m0[d]: is it fixable at all
18:33 tiredchiku[d]: snowycoder[d]: ^
18:33 snowycoder[d]: Honestly, I think it's a bit early to run games, but you can try:
18:33 snowycoder[d]: `NVK_I_WANT_A_BROKEN_VULKAN_DRIVER=1 vulkaninfo`
18:33 snowycoder[d]: to see if basic vulkan works
18:33 kar1m0[d]: snowycoder[d]: I will try to open he witcher 3 with it again then
18:34 kar1m0[d]: I will share the results here somehow
18:35 snowycoder[d]: That only tells you the vulkan extensions/features that our driver exposes
18:38 kar1m0[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1375905846139162624/20250524_213721.jpg?ex=683363ac&is=6832122c&hm=56ae13658950372d5903c423998a7fdb1162e29718c889a6825d3a7e4b2cc19a&
18:38 kar1m0[d]: It still runs on the igpu
18:38 kar1m0[d]: And no info about the dgpu nor was there a vulkan shaders compilation...
18:40 kar1m0[d]: when I run nvtop it also shows me only the igpu
18:40 kar1m0[d]: for some reason my dgpu is seen but isn't used
18:43 snowycoder[d]: Our vulkan driver for kepler isn't ready yet, I don't think it will run games and I'm not sure is present in the main mesa releases, sorry
18:44 snowycoder[d]: If you're hard-fixed on making it work you'd need quite a bit from technical knowledge, but I'd suggest to wait a bit. (don't worry, it's coming)
18:45 snowycoder[d]: *of
18:45 kar1m0[d]: are there any other tests that I can run?
18:45 kar1m0[d]: snowycoder[d]: alright I'll wait
19:29 snowycoder[d]: Hold on, KeplerB passes all texture tests even with `nak: Fix a perf regression in tex lowering` in the tree, texture source alignment isn't necessary even on kepler.
19:46 mhenning[d]: snowycoder[d]: That patch changes code that only runs on maxwell+
19:46 mhenning[d]: look at the outer if statement
19:55 snowycoder[d]: mhenning[d]: Oh sorry then, I misinterpreted some comments on the commit's MR
20:40 kar1m0[d]: I don't mind waiting:)
20:40 kar1m0[d]: It's not my main laptop
20:40 kar1m0[d]: So I don't really use it
20:40 kar1m0[d]: Only for tests such as drivers
20:40 kar1m0[d]: Because I can't do that on my main laptop
21:31 snowycoder[d]: gfxstrand[d]: I was debugging sm32 fadd and sm20 fadd seems bugged too: the long-immediate legalization copies to register if `self.saturate` is true, so in encoding it must be false, but it does `assert(self.saturate)`.
21:31 snowycoder[d]: Ugh we should really add hw_tests for float ops.
21:34 snowycoder[d]: (sorry for the pings, I'm trying to find a KeplerA so I can test those too)
22:37 gfxstrand[d]: No worries. Just go ahead and fix SM20
22:38 gfxstrand[d]: And if you want a Kepler A, they're like $25 on eBay with free shipping.
22:38 gfxstrand[d]: And if that's a problem just DM me and tell me where to send it.
22:39 gfxstrand[d]: Assuming you have something to plug it into
22:46 snowycoder[d]: I think a friend of mine has one he doesn't use, I need to check