04:03 mareko: soreau: does this fix the blend issue? https://pastebin.com/raw/ZKsPdi49
04:13 kode54: that patch looks incomplete, as it leaves on the conditions that are being false'd out
04:14 mareko: that's the intent
07:13 soreau: mareko: no sadly, I'm still able to reproduce the bug with that patch
09:14 jani: I want to make 'get_maintainer.pl -f drivers/gpu/drm/drm_bridge.c' *only* return the drm bridge maintainers, but --pattern-depth 1 already includes drm misc maintainers too. anyone have any idea how to make get_maintainer.pl return the most specific MAINTAINERS entry, and only that?
09:24 tjaalton: dcbaker: how far are we from having 23.2.0 out? ubuntu 23.10 will release in under four weeks, and it's getting pretty late (again) to get a new mesa in
11:30 vsyrjala: mlankhorst: tzimmermann: can i get an ack for this https://lists.freedesktop.org/archives/dri-devel/2023-September/422649.html ?
11:30 vsyrjala: mripard not around it seems...
11:31 tzimmermann: vsyrjala, go ahead
11:31 vsyrjala: ta
13:19 karolherbst: daniels: any thoughts on my loader changes to make it possible to load zink per render node? https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24839/diffs?commit_id=7c1d49db17e4d4e94928eeeae4e7e2e3e71c7f13
16:07 karolherbst: ahhh.. how can I make nir_copy_prop stop creating vec8/16 :'(
16:08 karolherbst: though maybe I should just split all vec8/16...
16:12 karolherbst: ehh wait.. this is my fault
16:16 HdkR: karolherbst: Delete all vec8/vec16, they aren't real :)
16:16 karolherbst: uhhhh
16:17 karolherbst: yeah... but... we don't have a proper way of doing that :D
16:25 cwabbott: markco: your change to add a wrap for lua has the unfortunate effect that even if an older version of lua is there (like 5.3) it'll download and build lua 5.4 which is completely unnecessary
16:25 cwabbott: and for me it promptly fails because I don't have some builddep
16:26 cwabbott: I thought the point is that wraps are supposed to be fallbacks, but that's not what's happening here
16:28 cwabbott: (and it doesn't even properly check for the dependency either, just fails compiling with a missing header)
16:28 cwabbott: dcbaker: ^
16:29 zmike: I have to filter all those patches out of my git repo every time I update since they don't even build for me 🤕
16:30 markco: I'll fix the issue with lua only falling back after checking for all versions
16:31 markco: In general, it seems like wraps are way shoddier than expected, I'm having to fix quite a few things upstream in wrapdb
16:37 markco: zmike: What issues are you running into? Is it mainly the libarchive issue?
16:37 zmike: it's the conflicting type decls
16:37 markco: In what file?
16:37 zmike: uhhh
16:37 markco: libarchive config.h?
16:38 zmike: yeah
16:38 zmike: that
16:39 markco: Right, currently waiting on a merge in wrapdb to fix that: https://github.com/mesonbuild/wrapdb/pull/1171
16:56 alyssa: ...since when does mesa depend on lua?
16:59 zmike: I rip van winkled last week and then just now I updated and it was there
17:00 ccr: grep indicates that only freedreno depends on lua
17:01 dcbaker: alyssa: freedreno tests
17:03 ccr: "y u no Python"
17:06 dcbaker: cwabbott: the lua thing is super annoying. I think we can at least solve the falling back when it shouldn’t, I’ll write a patch for that. You can add —wrap-mode= to disable wraps altogether. I need to look at why they still get enabled for required false though
17:27 dcbaker: cwabbott: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25251 should fix the case of falling back when 5.3 is installed but 5.4 isn't
17:33 dcbaker: mareko: I reviewed that MR for you, but I'm not really a committer in the wrapdb, so it'll have to wait for someone to come by make the release
18:01 markco: Thanks, that's fine. I'll wait for someone to merge it then update it on the Mesa side.
18:53 robclark: sima (or anyone that knows drm/i915): does intel_context map 1:1 to drm_file?
18:55 sima: robclark, nope
18:56 robclark: what does it map to? I assume everything per drm_file (or pipe_screen) maps to a single gpu virtual address space?
18:57 sima: also no
18:58 robclark: does iris really have to track and allocate a va per pipe_context?
18:58 sima: that's maybe a bit much
18:58 sima: robclark, ok I read up enough that I wont mix up all the structs
19:00 sima: so there's i915_address_space, which you can have any number of and look up through file_priv->vm_xa
19:00 sima: then there's i915_gem_proto_context, which is just an extremely entertaining uapi retcon, you can ignore
19:01 sima: then there's the i915_gem_context, which is the "context" uapi object which ties everything together
19:01 sima: you can find these through file_priv->context_xa
19:01 sima: there's a ctx->vm
19:02 sima: and there's ctx->engines, which eventually point at intel_context, which is the per hw engine actual hardware context
19:02 sima: it's a bit ... organically evolved
19:15 alyssa: dcbaker: ...yeah, why not Python?
19:17 dcbaker: alyssa:🤣
19:32 alyssa: i mean
19:43 HdkR: I mean, it's not the worst to integrate and extending it to have custom module imports isn't too terrible...
19:44 HdkR: And you get the benefit of writing stuff in python instead of Lua, which is a plus :)
19:53 HdkR: (Biased because all my scripts are poorly written python or terribly written bash scripts)
19:56 DavidHeidelberg[m]: I have this "huge" definition moving for CI, tell me what do you think: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25254/
20:07 robclark: alyssa: idk if python has improved since then but at least at the time lua was _far_ easier to integrate into .c app
20:08 robclark: most other scripting langs are designed as a stand alone thing, not something to embed
20:09 HdkR: https://docs.python.org/3/extending/embedding.html#very-high-level-embedding :>
20:10 HdkR: Calling a function and passing arguments might be more of a PITA I guess
20:11 robclark: yeah, I mean what I use it for needs to tie back to rnndec based register/packet decoding
20:12 robclark: lua was an easy and light weight way to achieve that.. it's a pretty nice way to add scripting for extension to an app
20:16 alyssa: robclark: oh, yeah thats fair
20:17 alyssa: i'd be tempted to invert things and have python call into c since ctypes is straightforward to work with but yeah i see why you went lua
20:26 robclark: it's more of a bolt on, and I believe completely optional, feature.. although we do have some tests for it so I guess if you want to run all the tests you need lua
20:27 alyssa: sure
22:16 Lynne: any way to capture a vk trace from a program that doesn't draw?
23:26 Company: Lynne: if you figure out a good way, tell me
23:30 Lynne: figured something out, but rgp is useless, just tells me the obvious - my compute dispatch takes years
23:31 Lynne: doesn't tell me why my very riced vullkan algorithm is slower than a naive opencl implementation
23:38 Company: that's prettymuch always what happens, isn't it?
23:38 Company: somebody invent perf for GPUs - or flamegraphs or so
23:41 Lynne: gpu crash, should've tested on my other machine...
23:43 Lynne: why do I like the red team so much when they handle crashes like it's 1995?
23:44 psykose: who doesn't
23:46 Lynne: nvidia
23:46 Lynne: intel
23:46 Lynne: not sure about apple, but probably
23:50 DemiMarie: Does virtio-GPU support cross-VM buffer sharing? I’d like to have VM A run a fullscreen nested Wayland compositor, with VMs B and C rendering through VM A.
23:52 DemiMarie: Qubes OS already supports this with X11 and its bespoke GUI protocol, but that protocol uses software rendering.