00:01fdobridge: <gfxstrand> What's the test list?
01:28fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and bood kernel from drm-misc-next, though I am not sure what to do with nvidia drivers yet. Looks like I need to compile them for new kernel, though it does no sence if I am volontiring to wok on NVK xD
01:29fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and bood kernel from drm-misc-next, though I am not sure what to do with nvidia drivers yet. Looks like I need to compile them for new kernel, though it does no since if I am volunteering to wok on NVK xD (edited)
01:29fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and boot kernel from drm-misc-next, though I am not sure what to do with nvidia drivers yet. Looks like I need to compile them for new kernel, though it does no since if I am volunteering to wok on NVK xD (edited)
01:29fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and boot kernel from drm-misc-next, though I am not sure what to do with nvidia drivers yet. Looks like I need to compile them for new kernel, though it does no since if I am volunteering to wok on NVK xD
01:29fdobridge: <maba_kalox> What would be the next step? (edited)
01:29fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and boot kernel from drm-misc-next, though I am not sure what to do with nvidia propritary drivers yet. Looks like I need to compile them for new kernel, though it does no since if I am volunteering to wok on NVK xD
01:29fdobridge: <maba_kalox> What would be the next step? (edited)
01:30fdobridge: <gfxstrand> Yeah, you don't need the property driver kernel.
01:30fdobridge: <maba_kalox> @gfxstrand Hi! I am back with update, I was able to compile and boot kernel from drm-misc-next, though I am not sure what to do with nvidia propritary drivers yet. Looks like I need to compile them for new kernel, though it does not make sense, if I am volunteering to wok on NVK xD
01:30fdobridge: <maba_kalox> What would be the next step? (edited)
01:57fdobridge: <gfxstrand> You'll also need to build Mesa with NVK enabled. Don't install system-wide, though. Install to a random folder somewhere (I use `$MESA_DIR/_install`.) You can set `VK_ICD_FILENAMES=$INSTALL_DIR/share/vulkan/icd.d/nouveau_icd.x86_64.json` which will make the Vulkan loader load NVK as the only Vulkan driver.
02:13fdobridge: <esdrastarsis> gsp ampere firmware is used for ampere, hopper and ada: https://github.com/NVIDIA/open-gpu-kernel-modules/blob/12c0739352fbfd58cba4444c5636a8d49a73f813/kernel-open/common/inc/nv-firmware.h#L87
04:02fdobridge: <mhenning> If you use `meson devenv`, you don't need to install at all, which is nice
04:06fdobridge: <gfxstrand> Yeah, definitely an option. My setup predates devenv by a good 5 years or more but whatever. 🤷🏻♀️
04:07fdobridge: <gfxstrand> `devenv` is probably the better choice for new devs.
04:07fdobridge: <gfxstrand> Yeah, definitely an option. My setup predates `devenv` by a good 5 years or more but whatever. 🤷🏻♀️ (edited)
04:45fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I wonder if anyone has tried the Ampere GSP firmware on a consumer Ada GPU though
05:26fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> What is the FMC firmware for? I see it for GH100 :nouveau:
06:35fdobridge: <georgeouzou> Here it is
06:35fdobridge: <georgeouzou> https://cdn.discordapp.com/attachments/1034184951790305330/1139809396092837938/devloss.txt
06:53fdobridge: <dadschoorse> aco was merged without vgpr spilling
13:52fdobridge: <conan_kudo> my account is so old that it predates my current nickname, so it's not so obvious anymore 😛
15:02fdobridge: <gfxstrand> RB'd on the ML. If some of our flake problems really are page migration bugs, this should help. The thing I didn't like about that theory is how the hell am I running out of memory on a 12GB card. Every every allocation is 2MB, though... Yeah, that just might do it. I'm doing another CTS run now to see how it shakes out. It already looks faster.
15:04fdobridge: <gfxstrand> RB'd on the ML. If some of our flake problems really are page migration bugs, this should help. The thing I didn't like about that theory is "How the hell am I running out of memory on a 12GB card?" Every every allocation is 2MB, though... Yeah, that just might do it. I'm doing another CTS run now to see how it shakes out. It already looks faster. (edited)
15:32fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Is it needed for Vulkan? The description mentions OpenGL
17:43fdobridge: <gfxstrand> Yes, very much so.
17:44fdobridge: <gfxstrand> Without it, fragment shaders will randomly not work or even hang.
17:44fdobridge: <gfxstrand> But only on Maxwell and earlier, I think.
17:44fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So I have added that patch to my local kernel
17:49fdobridge: <gfxstrand> FYI: I try to keep my NVK branch up to date: https://gitlab.freedesktop.org/gfxstrand/linux/-/commits/nvk/
17:49fdobridge: <gfxstrand> That's just whatever I'm testing with at the time.
18:17fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> But anyway I'm recompiling the kernel with some of the fixes applied
18:26fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> And the kernel build succeeded 🐧
18:28AndrewR: karolherbst, hello ... sorry for bothering you, but apparently somewhere between 6.1 (tested 6.144) and 6.3.9 up to yesterday linus' git nouveau can't light up my VGA monitor on GF710 (edid eaten by vga cable, but it works in earlier kerns and setting it on kernel cmd line doesn not help). Any ideas for debug?
18:30AndrewR: karolherbst, more accurately, vga framebuffer stays frozen while system still boots, there is no normal modeswitch ... like in 6.1.44
18:33karolherbst: mhhh..... I think there was something and I already sent out fixes I think..
18:34karolherbst: AndrewR: mind testing the `drm/nouveau/disp: Revert a NULL check inside nouveau_connector_get_modes` patch?
18:34karolherbst: https://cgit.freedesktop.org/drm/drm-misc/commit/drivers/gpu/drm/nouveau?h=drm-misc-fixes&id=d5712cd22b9cf109fded1b7f178f4c1888c8b84b
18:35karolherbst: AndrewR: ^^
18:35AndrewR: karolherbst, I think I just pulled this into my copy before building ...:(
18:36karolherbst: ahh... it's already in linus tree.. mhh
18:37karolherbst: then I don't really know.. maybe reverting ea293f823a8805735d9e00124df81a8f448ed1ae, 2b5d1c29f6c4cb19369ef92881465e5ede75f4ef and 752a281032b2d6f4564be827e082bde6f7d2fd4f? But that's not on 6.3.9...
18:37karolherbst: guess you might have to bisect then or make sure you actually tested with the fix
18:42AndrewR: affected kernel says "Cannot find any crtc or sizes" - may be I can add/activate few debug lines in place where this message is generated? I googled and found few AMD boards had similar symptoms ...
18:44AndrewR: ah, /drivers/gpu/drm/drm_fb_helper.c
18:52karolherbst: yeah.. ignore that message
18:52karolherbst: but maybe it does not query connectors or something silly...
18:52karolherbst: anyway
18:53karolherbst: probably better to bisect and figure out what broke it
18:55AndrewR: karolherbst, maybe I missed some config switch? I see there was rework of fbcon emulation ...
18:55karolherbst: potentially?
18:55karolherbst: does your distros config work?
18:56AndrewR: karolherbst, , not ried yet ...
18:57AndrewR: karolherbst, Slackware still on 6.1 (even in -current)
19:03karolherbst: mhh, yeah then no idea. You could check for new config options or something, but it might also be a proper regression... hard to tell anyway
19:03AndrewR: karolherbst, CONFIG_DRM_FBDEV_EMULATION=y at least ...
19:06fdobridge: <gfxstrand> `Pass: 403071, Fail: 497, Crash: 14, Skip: 1729474, Timeout: 2, Flake: 312, Duration: 1:06:31`
19:06fdobridge: <gfxstrand> So, seemed to help runtime a good bit and reduce flakes by around 200.
20:18AndrewR: karolherbst, it works! I need to force connector like: video=VGA-1:1440x900@60 video=VGA-1:e drm.edid_firmware=edid/1440x900_60.00.bin
20:18AndrewR: VGA-1:e was the key ...
21:07karolherbst: AndrewR: ohh, you use the video argument? You probably shouldn't, unless there is a reason to keep it these days
21:13AndrewR: karolherbst, my vga cable eat edid from monitor ... but it was working with default 1024x768 mode with radeon IGP and nouveau in earlier kernels
21:14karolherbst: pain
21:19fdobridge: <mhenning> Does nvk's support for clip/cull work correctly? The gl driver forces the generation of user clip planes in a shader based on pipe_rasterizer_state->clip_plane_enable but it doesn't look like nvk ever does this, which strikes me as odd.
21:21diogenes_oftc: karolherbst, your name is again mentioned on youtube: https://youtu.be/q8AYOdi_ZU4?t=103
21:21karolherbst: yeah, I've seen
21:22diogenes_oftc: i didn't know you work for RHEL.
21:22karolherbst: I kinda thought that this part is known
21:23Ermine: Easy to tell if you look at the email address
21:23diogenes_oftc: so NVK is a thing now.
21:23Ermine: It was merged recently
21:40fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Does NVK mean Nvidia VulKan or Nouveau VulKan? :triangle_nvk:
21:56fdobridge: <airlied> Ambiguous
22:24fdobridge: <mohamexiety> yes
22:54fdobridge: <gfxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1140055710797922394/image0.gif
22:56fdobridge: <gfxstrand> I think so. Vulkan doesn't have old-school user clip planes. Just clip/cull distances.
23:06fdobridge: <gfxstrand> The ambiguity is actually intentional. Nouveau has a bit of a bad reputation these days because... Well... It kinda sucks. 🫤 I mean no disrespect to anyone who works on it when I say that but it's also true. By not explicitly branding NVK nouveau, I was hoping to make a new thing people would be excited about and leave behind the baggage the nouveau name brings with it, at least in the public eye. It still lives inside src/nouveau and th
23:08fdobridge: <gfxstrand> *Phoronix flame wars aside...
23:08fdobridge: <gfxstrand> The ambiguity is actually intentional. Nouveau has a bit of a bad reputation these days because... Well... It kinda sucks. 🫤 I mean no disrespect to anyone who works on it when I say that but it's also true. By not explicitly branding NVK nouveau, I was hoping to make a new thing people would be excited about and leave behind the baggage the nouveau name brings with it, at least in the public eye. It still lives inside src/nouveau and th
23:08fdobridge: <gfxstrand> Reality is that it can mean either.
23:08fdobridge: <gfxstrand> I was calling it Nouvulkan for a little while
23:12fdobridge: <karolherbst🐧🦀> yeah, I doubt anybody (well.., probably) will deny there are many bad things about nouveau, and inhereting all that mess doesn't really make it easy improving things anyway
23:14fdobridge: <karolherbst🐧🦀> Nouveau, at least on the GL side is certainly at the point where one needs to start from scratch, but doing that is hard if there is legacy code to maintain 🥲
23:17fdobridge: <karolherbst🐧🦀> anyway... if we want to ditch part of the gl driver and make nouveau zink only, we have to figure out if we want ot support fermi in gl or make `nvc0` to be a fermi only driver (or figure out how to integrate it into `nv50`, but the push buffer format changed...)
23:27fdobridge: <mhenning> ah, you're right - the vulkan spec says "A shader must write a single clip distance for each enabled clip half-space" which removes the corner case I was worried about
23:32fdobridge: <gfxstrand> Ditching the marketing is very different from ditching the code. 😅
23:33fdobridge: <karolherbst🐧🦀> yeah....
23:33fdobridge: <gfxstrand> Sorry if that makes me sound like a corporate type but psychology is real. 😕
23:33fdobridge: <karolherbst🐧🦀> nah, I totally agree