04:27Dark-Show: Had a recent run in with a GLXBadFBConfig issue that I've debugged, turns out it was "user" error on my part. I was using the github.com/mesa3d/mesa which hasn't been updated in a few months. Switching to the gitlab fixed all my issues. Might help someone save some debug time.
04:28Dark-Show: Is the github mirror dead? Should it be removed?
09:51linkmauve: Asahi people (marcan, alyssa, lina, perhaps I’m missing some), did you end up anywhere with notch support in DRM? I’m playing with a phone which has annoying round corners which userland doesn’t take into account, plus the front camera is in a notch.
09:58jannau: linkmauve: no, we currently only render below the notch. tracking issue is https://gitlab.freedesktop.org/wayland/wayland-protocols/-/issues/87 currently unavaible due to fdo gitlab upgrade
10:06linkmauve: Thanks, I’ll have a look once the upgrade is done. :)
10:08An0num0us: I have been browsing compiler/glsl/glsl_to_nir.cpp and on line 166-169 saw if (!param->type->is_vector() || !param->type->is_scalar()) {. Is that intended?
10:08An0num0us: unsupported = true;
10:08An0num0us: return visit_stop;
10:08An0num0us: }
10:08An0num0us: sorry, new to IRC
10:12An0num0us: Essentially, it seems that the condition marks the parameter as unsupported regardless of the type class, which seems off to me and there is no comment explaining this. On line 177 the return type is being checked with !glsl_type_is_vector_or_scalar
11:56FireBurn: Is gitlab down just now?
11:59jannau: FireBurn: yes, see #freedesktop
12:02daniels: expect it to be down all day
12:02HdkR: It's really not merge day today
12:04karolherbst: An0num0us: if gitlab isn't down I'd suggest opening an MR
12:04karolherbst: but if you think the change is fine you can always submit it
12:04karolherbst: some of those helpers got added later and not all code got ported over
12:05karolherbst: although
12:05karolherbst: the original code kinda looks wrong :D
12:08karolherbst: An0num0us: anyway.. once gitlab is up again, you can either file an issue or suggest a change. I don't really know what the intention here was but a patter of "!condition || !other_condition" is kinda an odd thing to see as this most certainly evaluates to always true
12:09karolherbst: especially as is_scalar checks for vector_elements == 1 && ... and is_vector checks for vector_elements > 1 & ...
12:09karolherbst: *&&
12:25Manas1: hi
12:28Manas1: I am new to drm and wanted to know how can I get started with it. Like I dont have any experience with gpu programming but have worked a lot in C. Are there any particular resources I should look in order to get started with it?
13:17emersion: Manas1: the KMS docs have an "external links" section with tutorials talks etc
13:19emersion: https://dri.freedesktop.org/docs/drm/gpu/introduction.html#external-references
16:19DavidHeidelberg[m]: airlied: how did you run tinygrad? I did `GPU=1 ./examples/llama.py --prompt "hello!"` it show me it uses GPU, but I see 0 utilization of GPU and utilization of CPU
17:09randman: gfxstrand, it's that dma controller gives cisc capabilities to the offload engine, listen when you lay out pc-relative addressing such way, that first you make symmetric answer sets, and compute the aggregate of answers, then you do same with masks, then you have indexed answer sets, so printf is done on dma and also all the io engine stuff, like sound , most of network, and event loop is reconstructed on the cpu, compiler just puts things so that
17:09randman: outputs such as gpio is handled by dma, you are jason ekstrand i think, i am about to lift a biggest trophy in programming i know what i talk about, you need no locking semaphores and such in anywhere for gpu or cpu, so you are trying to fight too complex war, it is only needed in the kernel, but likely those bits are mature. so i only tried to help you, maybe you overcomplicate things, though they say hard in training and easier in the battle, i
17:09randman: can not see where such paradigm could be useful in practice. It took me some time after youth talent to be someone again, at 40 there is another trophy i will lift. 11 i lift a rare trophy by winning youth tournament multi-country without losing a set, after this i got injured and fell out, but i fought back to hang in there, and i know what i talk about.
19:07airlied: DavidHeidelberg[m]: does it crash or anything? I just ran it into a gpu timeout here, need to reboot with the command line to turn them off
19:10DavidHeidelberg[m]: airlied: first mistake was not habing Mesa installed, w/ installed mesa it's finally not using pocl
19:11DavidHeidelberg[m]: I didn't realized that Debian server setup has no mesa...
19:16airlied: I think mine just crashing the gpu, but I don't have the version of rusticl I used before, I'll have to rebuild it
19:24DavidHeidelberg[m]: What is weird that tx6800 seems similar in performance terms as integrated intel...
19:26airlied: I should try on integrated
19:32airlied: seems to be mostly just crashing my gpu
19:36DavidHeidelberg[m]: I'm using debian trixie 23.1.6-1 and on nightly ci-deb-repo it also worked
19:37DavidHeidelberg[m]: (but cannot re-download and re-test, since FDO is down)
19:40airlied: I'll hopefully have a 7900 of some sort this week
20:00airlied: running with DEBUG= set doesn't seem as crashy