00:40 fdobridge_: <g​fxstrand> Okay, I've got it working with hacks, both in the kernel and userspace
00:45 fdobridge_: <g​fxstrand> I think there are three problems:
00:45 fdobridge_: <g​fxstrand> 1. For some reason GL on Kepler isn't handling linear input images properly. They're fine on Maxwell and Turing but not Kepler. It treats them as tiled or screws up the stride or something. (When I say fine on Maxwell, I mean 750 TI.)
00:45 fdobridge_: <g​fxstrand> 2. For tiled, we aren't properly communicating the tiling information via the BO because, thanks to VM_BIND, all our BOs are now linear. This *should* be fixable with modifiers once we get those hooked up.
00:45 fdobridge_: <g​fxstrand> 3. We *can't* set the tiling information because the kernel rejects all BOs with tiling information when you're in VM_BIND mode. While it's true that the tiling information is ignored on bind, it's still necessary so that the BO gets properly bound by the GL driver.
00:45 fdobridge_: <g​fxstrand> @airlied @karolherbst ^^
00:48 fdobridge_: <g​fxstrand> I suspect 3 can't be fixed, even with modifiers. Not unless nouveau GL is going to start VM_BINDing imported BOs to override the linear tiling we set on them.
00:48 fdobridge_: <g​fxstrand> I've got a kernel patch which lets me allocate tiled things again. No idea if it breaks anything subtle.
00:56 fdobridge_: <g​fxstrand> Yeah, I think my kernel patch is bogus.
00:57 fdobridge_: <g​fxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1205678887032463410/0001-nouveau-Allow-allocating-tiled-buffers-in-VM_BIND-mo.patch?ex=65d93ef7&is=65c6c9f7&hm=3311683431cff7c4edbe0fc624113f7bb9294df79dd513fa555dee604ea7f068&
00:58 fdobridge_: <g​fxstrand> But it's good enough to try things out
00:58 fdobridge_: <k​arolherbst🐧🦀> maybe something wrong with the tex header?
00:58 fdobridge_: <k​arolherbst🐧🦀> they are different on kepler after all
00:59 fdobridge_: <g​fxstrand> Yeah, that's my theory
00:59 fdobridge_: <g​fxstrand> Unfortunately, debugging this mess requires figuring out how to run your compositor on a debug Mesa build. 😭
00:59 fdobridge_: <g​fxstrand> Maybe I can debug in Weston?
01:00 fdobridge_: <k​arolherbst🐧🦀> systemd override and then hack up LD_LIBRARY_PATH and stuff?
01:01 fdobridge_: <k​arolherbst🐧🦀> but apparently "some" compositor devs told me to "just install it system-wide" when I asked about that 🤷
01:01 i509vcb: switch to another tty, enter a mesa devenv and spawn the server from the cli is probably what I'd do
01:02 i509vcb: although some compositors are painful to start that way
01:02 fdobridge_: <k​arolherbst🐧🦀> it doesn't work
01:02 fdobridge_: <k​arolherbst🐧🦀> yeah
01:02 fdobridge_: <k​arolherbst🐧🦀> logind and such
01:03 fdobridge_: <t​om3026> i havent read up on it yet but i had this idea a while back of pci passthrough the dgpu in the laptop to a VM and test things out there heh
01:06 i509vcb: it might be a configuration issue, I've never had trouble spawning a compositor from a tty, especially smithay based ones.
01:06 i509vcb: although by default your user isn't really setup for that
01:07 fdobridge_: <k​arolherbst🐧🦀> well.. it's always a different thing for compositors/desktop handling multi-user setups
01:07 fdobridge_: <k​arolherbst🐧🦀> but yeah...
06:45 fdobridge_: <!​DodoNVK (she) 🇱🇹> "I don't want to end up as a kernel maintainer" - Faith, 2022
16:10 fdobridge_: <g​fxstrand> Yup. This is why I throw Dave bad patches and let him write the correct ones. He does the same thing to me with NVK.
18:28 phodius: hi
18:29 phodius: can someone ping me back i dont see any msgs
18:29 Sid127: phodius: yes your connection is fine
18:29 phodius: cool niceone
18:35 phodius: ive upgraded to ubuntu 24.04 which include Mesa 23.3.3-1ubuntu2 i tried installing mesa 24.0 which installed to /usr/local/lib etc but it caused a tty switching issue where it would automically switch over so i had to uninstall it, how do i tell if NVK is installed correctly?
18:35 phodius: im on 6.7 kernel on a rtx 2070 mobile any pointers would be great...
18:37 Sid127: phodius: unless you compile mesa manually, no distro bundles nvk with their mesa package as of now
18:38 phodius: i did compile manaually for 24.0 but maybe it conflicts with the package in the /usr/lib etc when i installed it to /usr/local etc
18:39 phodius: do i have to install it it to /usr/libs etc and overwrite the systems package?
18:41 Sid127: no, you can install it anywhere, as long as you point the program you want to run to the correct ICD
18:43 phodius: so how would i do that for wayfire? export VK_ICD_FILENAMES=/path/to/your/vulkan_icd.json when i run wayfire in the terminal
18:53 phodius: anyway thanks for the help i just do some research onto icds
18:53 Sid127: that is correct, yes
19:37 phodius: im getting ERROR: [Loader Message] Code 0 : vkCreateInstance: Found no drivers! Cannot create Vulkan instance. This problem is often caused by a faulty installation of the Vulkan driver or attempting to use a GPU that does not support Vulkan. ERROR at ./vulkaninfo/vulkaninfo.h:688:vkCreateInstance failed with ERROR_INCOMPATIBLE_DRIVER
19:39 phodius: do i need to update the nouveau driver as well
20:05 phodius: do i have to install the properitory graphics driver as well it says ] Code 0 : loader_scanned_icd_add: Could not get 'vkCreateInstance' via 'vk_icdGetInstanceProcAddr' for ICD libGLX_nvidia.so.0
20:26 fdobridge_: <a​verne> https://github.com/NVIDIA/open-gpu-doc/commit/2ef75cf8d6cc6311625c772171834c465e0319d0
20:29 fdobridge_: <a​verne> i'm surprised that they added drm-related fields for nvdec
20:46 phodius: which icd do i use?
22:18 fdobridge_: <a​irlied> Oh wow they released them, I asked months ago! @dwlsalmeida ^ av1 headers
22:22 fdobridge_: <a​verne> in addition to this, i noticed that they renamed a couple fields that were marked "reserved" before (cycle_counts in particular
22:23 fdobridge_: <a​verne> also thet new nvjpg_dec structure is completely incompatible with the previous one (the old one was accurate for the tegra x1, at least)
22:23 fdobridge_: <a​verne> also the new nvjpg_dec structure is completely incompatible with the previous one (the old one was accurate for the tegra x1, at least) (edited)