00:40 Hazematman: you might need to put quotes around "developer" :P
01:45 alyssa: Hazematman: "developers developers developers developers developers"
01:46 Hazematman: 😂 even better
02:46 CounterPillow: I think the Windows® way, where overcommit doesn't exist, to deal with malloc failure is to sleep for an arbitrary amount of time and then try again. That's because in that time, the system may have increased the size of the page file, and the allocation will succeed.
03:09 alyssa: CounterPillow: Why don't I hate this?
03:10 karolherbst: systemd-swap does support an increasing swap file btw
03:11 CounterPillow: maybe you have not yet realised that every application needs to implement this behaviour everywhere it might allocate with some exponential backoff, finely tuned waits, and a maximum number of tries, instead of the kernel just doing this for you, which sounds pretty horribl :P
03:11 CounterPillow: I don't think any application can ever just go "ah, guess I'll go on without this memory, it's fine"
03:18 orowith2os: karolherbst: at runtime? like, can it be dynamically resized without interrupting apps at all?
03:18 karolherbst: I think it depends how quickly stuff is eating RAM
03:18 karolherbst: *on
03:19 orowith2os: mainly I'm wondering if everything stays in the same actual swapfile or if it needs to get booted or moved to another one
03:19 orowith2os: would be nice to have zswap with like, 8GB of compression and a swapfile that's basically nil until it's actually needed :melt
03:19 orowith2os: s/:melt/🫠/
03:19 karolherbst: ohh yeah so no...
03:20 karolherbst: for resizeable swap systemd is using multiple files or so
03:20 karolherbst: there is a chunked swap file thing
03:20 HdkR: zram+backing store is nearly that, but it's kind of terrible
03:20 karolherbst: inside `/etc/systemd/swap.conf`
03:20 karolherbst: `swapfc`
03:20 HdkR: zswap has no LRU model for dumping stuff in to the backing store, which is kind of eeeh
03:20 orowith2os: HdkR: zram + backing store, for lack of a better word, fucking sucks 🫠
03:20 HdkR: yeppers
03:21 karolherbst: I run zram only...
03:21 karolherbst: but I also have 64GiB of RAM...
03:21 karolherbst: (and 64GiB of zram swap)
03:21 HdkR: zram without backing store is great although
03:21 orowith2os: I quite like the idea of zswap as a cache before the actual swap disk, but it technically functions perfectly fine without using the disk and without using the compression
03:21 HdkR: Yea, that's kind of the ideal
03:21 karolherbst: my thinking is... the amount of swap file that would matter on my 64GiB system would take like... 25% of my disc storage 🙃
03:21 HdkR: zswap and zram need to mingle and make the best of both
03:22 orowith2os: karolherbst: I have 8gb of zram with 16gb of physical memory, and don't really use it, but it would be nice for when my tabs get out of control :)
03:22 orowith2os: yeah, that's why a dynamic swapfile would be best
03:22 karolherbst: heh.. yeah.. I need to compile llvm
03:22 HdkR: Frame ships an 8GB zram as well :P
03:22 orowith2os: IIRC macos kind of just dumps things into part of the disk that's unused so the disk can still actually be used without bothering anything
03:23 orowith2os: like zswap does
03:23 orowith2os: but encrypted
03:23 orowith2os: god, how I would love to see encrypted hibernation become a thing 🙏
03:24 orowith2os: okay wait, hold up a minute.
03:25 orowith2os: karolherbst: apparently macOS does something like systemd-swap does too 🙃
03:25 orowith2os: or rather, systemd-swap does what macOS does.
03:25 orowith2os: multiple swap files.
03:25 orowith2os: I love it.
03:25 orowith2os: I love everything.
03:26 alyssa: HdkR: im very curious what people are using frame for in practice these days
03:27 orowith2os: Frame as in Steam Frame or Framework?
03:27 orowith2os: (or other)
03:27 HdkR: alyssa: I would hope primarily gaming, but it's not a bad little ARM dev platform, just weird shaped.
03:27 HdkR: Steam Frame yea
03:27 orowith2os: ah
03:27 orowith2os: (also I may look at getting Bazzite to ship systemd-swap with zswap now 🙃)
03:27 orowith2os: I'll bother Kyle about it later.
03:28 orowith2os: It's super interesting and I'm sure I can get it into testing too.
03:29 alyssa: HdkR: that's fair
03:30 alyssa: i guess i'm also curious on the split between VR & non-VR games
03:30 CounterPillow: there's probably someone out there using it to do their taxes
03:31 HdkR: alyssa: It's the everything machine, it's hard to tell. Seems like people are enjoying it both ways at least :D
03:31 karolherbst: VR tax simulator
03:31 karolherbst: sounds like a game Germans would enjoy
03:33 HdkR: Run all those simulator games standalone on it
03:33 dwfreed: karolherbst: orowith2os: note that systemd-swap is not part of systemd itself, it is a third party thing
03:33 karolherbst: you'll see me again in two years
03:34 orowith2os: dwfreed: uhhhh.... systemd-swap is https://github.com/nefelim4ag/systemd-swap, right?
03:34 dwfreed: yeah
03:34 HdkR: Karol got stuck in the vr-universe again. Pull them out
03:34 orowith2os: well, that's joyful.
03:34 orowith2os: I was really looking forward to something like macOS had. It would have made my life so much easier.
03:34 karolherbst: orowith2os: I'm sure that one is outdated
03:35 karolherbst: systemd has its as part of core
03:35 dwfreed: are you sure? I could find no reference to it
03:36 dwfreed: particularly, there is no documentation for swap.conf
03:36 karolherbst: huh...
03:36 karolherbst: maybe you are right actually...
03:37 orowith2os: hate everything
03:38 dwfreed: There is a maintained fork of that repo according to GitHub; they rewrote it in rust: https://github.com/biglinux/biglinux-systemd-swap
03:38 orowith2os: no longer hate everything
03:39 karolherbst: mhhh curious
03:39 dwfreed: note there is an AGENTS.md that contains information that would be useful to LLMs. I'll let you make your own assessment on that.
03:40 dwfreed: (vs an AGENTS.md that tells LLMs to go away)
03:40 orowith2os: their issue tracker is disabled, and no contributors list...
03:40 dwfreed: helpful...
03:40 HdkR: Beeg Linux is watching
03:41 orowith2os: wording and whatnot (and unrelated changes in commits) makes me want to jump to the conclusion of being AI 🙃
03:56 Company: did they copy that agents.md or did they have to talk some LLM out of adding a GTK dep to a swap deamon?
04:09 CounterPillow: You don't want to know the answer to that
04:43 dwfreed: HdkR: Factorio runs on it now, that's all that matters :D
04:50 HdkR: dwfreed: Factorio in VR so you never need to leave the automation
04:52 dwfreed: The Factory Must Grow!
04:54 dwfreed: The Matrix won't be a completely simulated world, you'll just be able to play whatever video games you want all day long
04:57 HdkR: Seems like a fine trade-off
04:59 dwfreed: Yeah, I'd be fine with that
06:08 CounterPillow: if I wanted to see ever expanding pollution I'd just look out the window
06:10 dwfreed: laughcry.gif
06:10 dwfreed: https://media.giphy.com/media/IV2DvTArF5b9u/giphy.gif this is what I meant
09:49 dolphin: airlied, sima: looks like no drm-intel-next-fixes PR this week, DIM didn't pick up anything
13:03 alyssa: zmike: I had a dream last night involving a future millenium in which Princess Celestia censors information about big-endian GPUs to avoid the truth about their world getting out
13:03 alyssa: which means you get to review https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/44974
13:04 zmike: !?
13:04 zmike: you were actually able to pierce the veil of truth?!
13:05 HdkR: The little endian must flow \o/!
13:05 ccr: "Meanwhile, in the lair of the BBEG .."
13:07 daniels: Assisted-by: sed
13:09 alyssa: I can't tell if this is a serious ask or not
13:09 daniels: no lol
13:17 Ermine: Assisted-by: coffee
13:20 daniels: reminds me I need to grab a new LiIon pack for my grinder
13:21 HdkR: I use the Niche Zero. Is good stuff, I'd recommend
13:22 daniels: nack
13:22 daniels: I do a large V60 pot at a time, so single-dose is out
13:23 daniels: the Timemore Go is great, except that the LiIon is ... not the best
13:28 HdkR: ah, I'm an espresso lad so single-dose is what I need. Fair enough :D
13:31 soreau: either way, it's always too early to grep unless you've had your coffee
13:38 MoeIcenowy: daniels: Some project of my daily work has Assisted-by in the PR comment template, and I was trying to do something like
13:38 MoeIcenowy: by: GNU sed x.x.x
13:38 MoeIcenowy: however I never had the chance to open a PR with only trivial replacement
13:40 HdkR: Assisted-by: neovim
13:40 alyssa: Co-authored-by: nano
13:43 daniels: I've never used GPT-5-nano personally
13:46 Hazematman: Assisted-by: My parents without whom I wouldn't be here, all my teachers who never gave up on me, all the community members who listen to my dumb questions, and my dog who gives meaning to all of this
13:51 alyssa: Hazematman: woof woof
13:54 Hazematman: 🐶
13:59 sima: daniels, and here I thought you've gone into car restoration or why exactly do you need a new lion pack for your angle grinder ...
14:00 MoeIcenowy: simon-perretta-img_: I am listening to your talk at XDC, and will abusing LR then block implementing proper function call?
14:00 MoeIcenowy: (well I think only a few drivers in Mesa are currently implementing proper function call, but it might then be useful for rusticl?
14:01 zmike: mareko: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37420 https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/44948
14:01 MoeIcenowy: OT: the main developer of NCNN complains about the computation performance of IMG DDK Vulkan (because NCNN implements Vulkan acceleration) and she says DDK OpenCL is better than DDK Vulkan
14:02 MoeIcenowy: (and with her vkpeak benchmark, Mesa PowerVR outperforms DDK Vulkan in some tests
14:05 simon-perretta-img: MoeIcenowy: I've got a branch that does "function calls" (calling into a common shader helper library) and the LR just gets backed up/restored before/after subgroup ops to take care of that, so no breakage
14:06 MoeIcenowy: simon-perretta-img: glad to hear this
14:06 MoeIcenowy: well is the helper library here usclib?
14:09 simon-perretta-img: Nah, something else; just an experiment to move really verbose ops (like lowered barriers, atomic emu, etc.) into like a shader... BIOS...? that can be called into with the function call ops
14:10 MoeIcenowy: well sounds like these are currently part of usclib (and inlined when compiling the shader)
14:12 simon-perretta-img: Oh I see what you mean, yeah
14:12 MoeIcenowy: but isn't a call on GPUs very heavy?
14:14 simon-perretta-img: Not necessarily, but even if it were, these ops in particular are already orders of magnitude heavier, since e.g. for barriers we have to serialise slots (thread/sub groups) and deschedule each one
14:18 MoeIcenowy: well I must agree this...
14:18 simon-perretta-img: So you wouldn't want to do this sort of thing for ops that are already lean, or where you'd need to overallocate/spill temps to be able to implement them, but it lets you reduce shader bloat in the cases I mentioned above
14:23 MoeIcenowy: BTW could anyone familiar with LLVM take a look at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/44521 ? It seems that someone other than me is also bited by this in #16445
14:40 alyssa: simon-perretta-img: is that a good idea?
14:40 alyssa: wrong answers only
14:41 karolherbst: all I will say here is that I'm a big fan of backend compilers having proper support for function calls
14:42 pac85: karolherbst: what would use them?
14:42 karolherbst: rusticl mainly
14:42 alyssa: chonky opencl kernels
14:42 alyssa: ray tracing
14:42 pac85: Ok I see
14:42 karolherbst: *side eye-s this 10 Million SSA nir shader*
14:42 pac85: radv can do something that's function calls but might be specificly only exposed for rt
14:42 pac85: but not sure
14:43 karolherbst: yeah.. I'm happily ignoring it, because it's like a single application I know off that really blows up when inlining everything
14:43 karolherbst: well.. it doesn't blow up actually, it just takes hours to compile
14:44 pac85: makes sense
14:44 pac85: So I guess you'd also need euristics to decide what would be functions and what would be inlined
14:45 karolherbst: there is some basic support for function calls in llvmpipe there, but it also blows randomly on images or subgroup operations
14:45 karolherbst: yes
14:45 pac85: lol
14:45 karolherbst: we kinda have that already
14:45 pac85: in nir?
14:45 karolherbst: yes
14:45 pac85: oh I see
14:45 pac85: cool
14:47 karolherbst: it's not great, but there is a "nir_function_can_inline"
14:47 karolherbst: and "driver_functions" as a nir shader options
14:51 pac85: yeah that seems fairly minimal
18:19 simon-perretta-img: alyssa: Rogue hw engineers (pun unintended) apparently thought that sw having to roll their own barrier instruction was a good idea, so the idea bar is... somewhere...
18:23 HdkR: That's as bad as me needing to roll my own split-lock implementation :D
19:12 CounterPillow: gotta somehow prove your professor was right when they said the primitive of the atomic register is enough to build all concurrency primitives
19:21 Hazematman: I'm waiting for a GPU ISA that just uses NAND for all its arithmetic :P
19:34 alyssa: Hazematman: NANDvidia
19:39 Hazematman: Damn that name is too good 😂
19:44 HdkR: It's almost too close to how some people pronounce it navidia.
21:06 mareko: can we stop using Rb and instead click Approve in Gitlab?
21:07 karolherbst: some drivers/parts already do that kinda informally
21:08 mareko: I mean formally and for everything
21:08 zmike: rb
21:09 jenatali: rb
21:09 karolherbst: yeah.. I don't mind. I think if we really want that to be documented, marge bot could also add a "approved by" to the commit
21:10 jenatali: It adds part-of, I don't see it as a big deal to click the link to see who approved it
21:10 karolherbst: fair
21:11 karolherbst: though I do think there is some value to see it directly in the git log and it would be almost free to add
21:12 zmike: surely we wouldn't trust a machine to add a human's name to a commit message ?
21:12 dcbaker: Only if they are also a fictional character
21:13 zmike: what if it frames me for breaking gallium again
21:13 HdkR: That's a chance I'm willing to take.
21:13 HdkR: rb
21:13 karolherbst: as long as it only frames you, I don't mind
21:14 dcbaker: I'll just have marge add zmike's r-b to every commit
21:14 karolherbst: finally the lack of review problem is solved
21:15 dcbaker: It's like the good ol' days when you could just slap "trivial" on a patch and push it to origin/master :D
21:19 alyssa: Previous discussion: https://gitlab.freedesktop.org/mesa/mesa/-/work_items/11640
21:19 alyssa: dcbaker: git rm -r src/nouveau/vulkan && git commit -am "Trivial." && git push -f origin master
21:20 dcbaker: Don't threaten karolherbst with a good time
21:20 karolherbst: it has already been decided as the first governance decision
21:23 dcbaker: to delete nouveau, or to end the use of r-bs?
21:23 karolherbst: to delete nvk specifically
21:23 karolherbst: didn't you watch the talk?
21:23 dcbaker: no
21:24 dcbaker: They don't let me watch it on the clock, and I don't have time to watch youtube otherwise :D
21:24 karolherbst: sad
21:24 dcbaker: Yeah, being an adult sucks
21:25 karolherbst: so glad that I'm not one
21:25 dcbaker: 10/10 recomend you never become one
21:25 dcbaker: the only thing worse than being an adult is being a manager
21:25 mareko: wait, don't you play games after work?
21:28 mareko: you can play them with your kids, some parents do ;)
21:28 dcbaker: Lolololol. Only when @kayden invites me to play BG3
21:29 mareko: you can also have your kids farm for you in an MMO
21:29 dcbaker: my kids only want to play ipad games
21:30 dcbaker: and I only play steamdeck games
21:30 mareko: they can be re-trained ;)
21:30 zmike: thanks for your support
21:30 dcbaker: Thanks for making a linux gaming machine that Just Works™
21:31 alyssa: dcbaker: buy them a steam frame
21:31 alyssa: i'm probably not a shill or anything
21:31alyssa:glances frantically at the FEX commit logs
21:32 HdkR: :)
21:33 HdkR: It's one of the better pieces of hardware for SVE development! </shill>
21:33 karolherbst: my steamdeck is my favorite piece of gaming hardware
21:34 alyssa: HdkR: did frame ship with SVE in the end?
21:34 alyssa: (SVE-256 i assume and not SVE-512?)
21:34 alyssa: i've only ever had NEON hardware, SVE is make-believe and can't hurt me
21:35 HdkR: alyssa: SVE-128, it indeed shipped
21:35 alyssa: Neat :p
21:35 orowith2os: karolherbst: steam deck my beloved ❤️
21:35 orowith2os: if they came out with a slightly newer one that was a bit more efficient that I could trade in I would, but damn if it isn't good already
21:35 orowith2os: aging like fine wine 🥲
21:41 karolherbst: well that's the OLED version
21:41 DragoonAethis: HdkR: can I get free FEX support here when the Frame arrives? :D
21:41 HdkR: :thonk:
21:42 HdkR: Would be better to go to the IRC channel or official Discord server
21:42 DragoonAethis: just joking here ;p
21:43 HdkR: I'm sure Mike and the Turnip devs would love the complaining although
21:44 zmike: zink has no bugs
21:45 HdkR: Intentional features only
21:45 DragoonAethis: random black flicker in Shipbreaker is the intended display of the vastness of space
21:55 orowith2os: karolherbst: er, newer CPU and the like 🙃
21:55 orowith2os: but even now it's pretty amazing
21:55 orowith2os: I would really love to see a more switch-like format, where I can detach the controllers
21:55 karolherbst: the OLED one has a die shrink, and LPDDR5 or something
21:55 karolherbst: or was it LPDDR5 vs LPDDR5X?
21:56 karolherbst: ahh yeah that was it
21:56 karolherbst: the fans are also better
21:56 karolherbst: I have both
21:56 karolherbst: the OLED is better
21:57 orowith2os: yeah, I have the OLED
21:57 karolherbst: ahh...
22:10 mareko: I've taught my static analyzer short commands like ":rh", meaning "review HEAD"; long prompts are so 2025
22:14 zmike: mine is currently optimizing a breakfast shake for both performance and robustness
22:22 orowith2os: you know what it can't optimize? your mom.
22:23 orowith2os: (yay for cringe jokes :P)
22:27 zmike: yo momma so fat a claude max subscription ain't enough to come up with a diet plan
22:32 kisak: Can we ... not ... deal with the unsolvable mother problems?
22:34 kisak: Sorry, knee-jerk. Carry on.
22:44 orowith2os: kisak: ooc are you the same kisak I see wrangling Valve's issue trackers on github?
23:19 Company: austriancoder: with your gles 3 stuff being done, I assume GTK4 works fine now? Is it fast enough or is there anything we could look into to improve things?
23:50 Hazematman: I'm excited for the day my kid is old enough to play games lol, right now it's like work, clean up the house, sleep, repeat
23:53 orowith2os: you mean you don't want to make another ipad kid? :P
23:55 orowith2os: you're never too young to be an ipad kid.