01:28airlied: dakr, Lyude : just sent two patches on top of skeggsb gsp work to fix builds on other architectures
01:50fdobridge: <conan_kudo> huh is Lyude here?
01:52fdobridge: <conan_kudo> oh I guess on the other side of the bridge
17:23fdobridge: <karolherbst🐧🦀> anybody wants to help a user to get GSP up and running with fedora on that bug? https://gitlab.freedesktop.org/drm/nouveau/-/issues/269 @airlied you have a copr right?
17:23fdobridge: <karolherbst🐧🦀> would appreciate it if you could leave the infos on how to test it
17:25AndrewR: currently running proprietary driver on Salix 15.0 live x64 - it was a chore to compile ;)
17:26AndrewR: but I can confirm my card on hw level have a) hvenc part (even if only for h264) and b) can run luxcorerender 2.6 via openCL 3.0 as provided by proprietary driver. kernel compilation a bit slow - few minutes, and virt memory goes up to 60GB - no way it can be 32-bit ...
17:27karolherbst: uhh
17:27karolherbst: impressive
17:28karolherbst: sure you are using their driver for CL? :D
17:30AndrewR: karolherbst, https://pastebin.com/V6ixds9G
17:31karolherbst: impressive
17:31karolherbst: I know that mesa has issues compiling luxmark, because "function calls" but nvidia?
17:33AndrewR: well, I guess their compiler kinda good? cache directory has some ptx files ....
17:33karolherbst: uhm.. it shouldn't take minutes to compile though
17:35AndrewR: karolherbst, [LuxRays][630.097] [GammaCorrectionPlugin] Compiling kernels
17:36AndrewR: karolherbst, 2.6 uses this megakernel design for making blender viewport more responsive ...
17:37AndrewR: karolherbst, https://pasteboard.co/n5hX7R8sIoVg.png
17:38karolherbst: ohhh
17:38karolherbst: I see
17:38AndrewR: karolherbst, but gpu load monitoring kinda busted :P
17:38karolherbst: I should try 2.6 as well
18:57AndrewR: https://www.x.org/wiki/SummerOfCodeIdeas/ - I wonder how big is "Kepler Accelerated Video Encoding" task :}
18:58karolherbst: big
18:59AndrewR: sad :} but basics should include mmiotrace-enabled kernel or something heavier like qemu vm with pcie virtualization?
19:00karolherbst: Mary has a new thing to intercept push buffers, so maybe that helps
19:04AndrewR: karolherbst, is this tool public? sorry I skipped few years due to "get a new dog " adventure
19:04Mary: I only intercept gpfifo push buffers, might need to hook something else for nvdec/nvenc I think
19:04Mary: AndrewR: https://gitlab.freedesktop.org/marysaka/envyhooks/
19:05fdobridge: <mohamexiety> this is the same issue I had...
19:05fdobridge: <mohamexiety> well, have
19:07fdobridge: <mohamexiety> given their GPU is a RTX 3060 LHR I think this aids my theory that the firmware we got from NVIDIA doesn't really work for LHR cards
19:07fdobridge: <karolherbst🐧🦀> yeah... wouldn't surprise me
19:08fdobridge: <mohamexiety> I still haven't installed the new GSP kernels and all that so I am afraid to give instructions
19:08fdobridge: <karolherbst🐧🦀> heh 😄
19:08fdobridge: <karolherbst🐧🦀> you should try that
19:08fdobridge: <karolherbst🐧🦀> but we also ping somebody from Nvidia about that bug with more details, just that we don't really have many details here
19:08fdobridge: <mohamexiety> but yeah, Dave has a copr and there's this https://patchwork.freedesktop.org/series/123876/
19:08AndrewR: 'nv-ioctl.h' file not found
19:09fdobridge: <mohamexiety> yeah I just resigned myself back to the 1030 for now. stability and all that haha
19:09karolherbst: Mary: one needs to use the header files from the currently running kernel, right?
19:10karolherbst: uhm...
19:10karolherbst: currently installed driver
19:10fdobridge: <mohamexiety> I can help with bringing more details but I don't know where to start. the card just refuses to boot with that same `nouveau 0000:01:00.0: gr: DISPATCH 80000001 [INJECTED_BUNDLE_ERROR]` error unless on GSP.
19:11fdobridge: <karolherbst🐧🦀> yeah...
19:11fdobridge: <karolherbst🐧🦀> so I think the GR firmware might need to be updated
19:11fdobridge: <karolherbst🐧🦀> or the code constructing the golden image
19:11fdobridge: <karolherbst🐧🦀> *context
19:13AndrewR: after symlinking few things run into fatal error: 'nv_escape.h' file not found
19:14AndrewR: pre. files lived in /usr/src/nvidia-470.199.02/common/inc
19:17Mary: karolherbst: yeah but I use the headers from the open source one
19:18Mary: so if it's too old might not have the right headers...
19:18karolherbst: ahh, fair
19:18Mary: tho might be possible to guess the needed structs them maybe...?
19:18karolherbst: though I think closed and open source should be ioctl compatible anyway
19:19karolherbst: AndrewR: just throw at it until it builds or something
19:19Mary: karolherbst: yeah they are but I'm pretty sure there is no stable api here tho between different versions
19:19karolherbst: right
19:20karolherbst: it has to match the current drivers version
19:23AndrewR: well, in my case new (530?) binary driver probably will simply not work, because GK208B
19:24karolherbst: I meant current as in currently installed one
19:24karolherbst: but yeah.. not sure what to do if you need something older...
19:24karolherbst: but anyway
19:25karolherbst: the header from the blob driver should be good enough
19:25AndrewR: 470.199.02 ... I dloaded open-gpu-kernel-modules but still same err?
19:25Lyude: conan_kudo: yes I'm here :), but I don't really use discord much
19:25AndrewR: I mean I symlinked it into /usr/include
19:25karolherbst: well..
19:26karolherbst: I suspect you'll have to nodify Marys build stuff
19:26AndrewR: in rust, lol ...
19:26karolherbst: AndrewR: https://gitlab.freedesktop.org/marysaka/envyhooks/-/blob/master/wrapper.h?ref_type=heads
19:27karolherbst: and probably some cargo whatever flag to deal with it inside the build.rs https://gitlab.freedesktop.org/marysaka/envyhooks/-/blob/master/build.rs?ref_type=heads#L26
19:27karolherbst: like those include paths
19:28karolherbst: I'm sure you can figure it out with Mary :)
19:30fdobridge: <gfxstrand> ```
19:30fdobridge: <gfxstrand> git submodule init
19:30fdobridge: <gfxstrand> git submodule update
19:30fdobridge: <gfxstrand> ```
19:30AndrewR: it was dangling symlink ...cp continues
19:31fdobridge: <karolherbst🐧🦀> sure, but version 470.199 is needed 🙂
19:31fdobridge: <gfxstrand> Oh... That's problematic.
19:31fdobridge: <karolherbst🐧🦀> yep
19:33AndrewR: gfxstrand, thanks
19:33AndrewR: error[E0554]
19:34AndrewR: rustc 1.70 ...
19:35karolherbst: anyway, I'm sure the tool needs more work to make it work with pre open source nvidia drivers
19:35karolherbst: AndrewR: well.. "rustup install stable"
19:43AndrewR: stable-x86_64-unknown-linux-gnu installed - rustc 1.73.0 (cc66ad468 2023-10-03
19:43AndrewR: guess I just need to find to where ...
19:48AndrewR: still error[E0554]: `#![feature]` may not be used on the stable release channel
19:48AndrewR: even if I exported export PATH=${PATH}:/root/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/bin/
20:07Mary: you need nightly because of c_variadic AndrewR
20:07Mary: I thought the rust-toolchain file would be enough to grab the right toolchain oh well... really need to write a readme later
20:07karolherbst: ohh yeah.. that's a nightly thing :') "rustup install nightly" then
20:11Mary: really need to find a way to get ride of that requirement somehow
20:12karolherbst: I doubt there is one...
20:12AndrewR: done!
20:12karolherbst: Mary: you could write those function in C :D
20:12AndrewR: does this binary works without rust night toolchain installed?
20:14AndrewR: 1.1G with debug, lol
20:14AndrewR: too good I got 16 Gb ram ..
20:16AndrewR: [nvenc_h264 @ 0x3f4cf40] dl_fn->cuda_dl->cuInit(0) failed -> CUDA_ERROR_INVALID_VALUE: invalid argument
20:16AndrewR: so I guess it does not work with this driver and ffmpeg
20:17karolherbst: yeah.. you need headers specific to your driver, and then some of the uvm stuff might need more work. something something
20:18karolherbst: the tool is still like heavy WIP
20:18karolherbst: I also have some UVM based valgrind-mmt stuff around, but that's even more messy to use
20:26AndrewR: well, thanks anyway, lets hope this setup will live for some more time, I saved most compiled components to hdd
22:42airlied: TimurTabi, dakr : finally got back to it and got a newer firmware to run
23:43dakr: airlied, that's awesome!