Dave Airlie

Linux kernel hackers, Distinguished Engineer @ Red Hat

Public Key

npub19ehxktytzlyvtk9upqk9vvq4rsvg6t4uw8p776krt5rwsqv0pnnqmv6wxr

NIP-05 Address

airlied_at_fosstodon.org@mostr.pub

Profile Code

nprofile1qqszumnt9j930jx9mz7qstzkxq23cxyd9678rsl0dtp46phgqx8seespz4mhxue69uhhyetvv9ujumt0wd68ytnsw43qzxthwden5te0wfjkccte9eeks6t5vehhycm99ehkuegpmtydc

Publishing to

relay.mostr.pub

relay.shitforce.one

nostr.coinfundit.com

Author Public Key

npub19ehxktytzlyvtk9upqk9vvq4rsvg6t4uw8p776krt5rwsqv0pnnqmv6wxr

Show more details

Last Notes

2025-04-24 02:19:04

"So you'll need a vmap for folios to support this use case."
this isn't where to day was meant to lead.

2025-04-24 00:08:02

https://lkml.iu.edu/0709.2/1709.html
I went down a well of what if? folios today, and ended up in AGP code, and even though I'm probably AGP maintainer, I'd forgotten in 2007 I might have understood some of this stuff.

2025-04-23 21:44:52

https://lore.kernel.org/dri-devel/20250423214321.100440-1-airlied@gmail.com/T/#t
You should always worry when a maintainer writes patches.

2025-04-11 22:18:00

https://camps.aptaracorp.com/ACM_PMS/PMS/ACM/HCDS25/10/13a8f7c0-0a7e-11f0-ada9-16bb50361d1f/OUT/hcds25-10.html
don't cross the streams!

2025-04-01 20:11:00

- reply

@nprofile…ntmq for LLM it seems like you have to hand craft layers to offload which is definitely beyond the effort anyone wants to put in that isn't the vendor

2025-04-01 19:37:35

- reply

@nprofile…ntmq has anyone worked that out yet?

2025-03-27 21:29:00

the Astro's Playroom GPU song might have to precede my entrance to all rooms, I'm considering just making unboxing videos of that song and opening the 5080 box until I get copy striked.

2025-03-16 06:42:11

- reply

@nprofile…vklh where you hoping for mothra?

2025-03-11 08:24:18

- reply

@nprofile…8p6q I'm sure NVIDIA GPU shader engines share some C64 history 😛

2025-03-11 06:02:00

- reply

turns out it wasn't a missing barrier, I just needed to add some nops.

2025-03-11 04:44:41

going insane and missing barriers have a lot in common

2025-03-05 02:39:14

Looking like a rough couple of days coming towards the area!
https://cdn.fosstodon.org/media_attachments/files/114/107/503/589/657/203/original/080a4d3984e8871b.png

2025-03-04 22:00:19

- reply

@nprofile…vklh you've triggered my someone mentioned Ultrix, now i have to find my ultrix source code CDs again.

2025-02-27 22:16:48

"If you're going to use cmake, you might want a guide. I do have a bit of experience in that arena."
"If you consider failure experience."
"I consider experience, experience.

2025-02-27 00:23:37

building the compilers for NPUs has burned more power than NPUs will ever save in real life 🙂

2025-02-26 22:09:11

- reply

@nprofile…pyv2 bots just DoSing that bar by continuously asking for drinks but not consuming them

2025-02-26 22:08:02

bash: cmake: command not found...
ah that new computer smell, I hate to tranish it

2025-02-26 10:32:12

- reply

@nprofile…vvsn I don't think he is leaving the kernel either, he maintains lots of stuff and will probably retreat more as rust encroaches. I'm kinda reminded of the north and south facing Zax and the world just routes around them.

2025-02-24 05:21:44

oh look a YAK, the Vulkan CTS test suite has a lot of tests that allocate their readback buffer for CPU access using HostVisible only, which ends up in VRAM on a bunch of drivers, and is slow all crap. WHY!!!

2025-02-21 10:52:31

- reply

@nprofile…xmfv @nprofile…htxq I really don't get the point of that, the kernel has never had a long term viability plan, it has evolved, but you mention rust and suddenly everyone cares about alpha and mips32, multiple compilers, 20 year stability, stuff the C side grew over 30 years

2025-02-21 10:48:59

- reply

@nprofile…htxq I had a lot of Grandpa Simpson/old man shakes fist at cloud feels, in my day we used memory unsafe languages and C++ was the future and we liked it! Definitely a lot of off point wandering, maybe next time Linus will save us the journey 🙂

2025-02-21 03:04:25

https://lore.kernel.org/rust-for-linux/CAHk-=wgLbz1Bm8QhmJ4dJGSmTuV5w_R0Gwvg5kHrYr4Ko9dUHQ@mail.gmail.com/ we good now? Can we just write some more code?

2025-02-19 10:06:27

- reply

@nprofile…8p6q quick where's my "stop trying to make C++ happen" mean girls meme

2025-02-17 05:27:36

Where's all the commentary and speculation for good kernel rust stories? https://www.phoronix.com/news/Linux-6.14-Faux-Bus-Merged

2025-02-17 05:27:36

Where's all the commentary and speculation for good kernel rust stories? https://www.phoronix.com/news/Linux-6.14-Faux-Bus-Merged

2025-02-13 11:09:44

- reply

@nprofile…vvsn cat eating security key and then having to be held against NFC scanners would also be bad

2025-02-06 20:29:34

- reply

@nprofile…lv5y thankfully most people understood GMM wasn't useful for anything 🙂

2025-02-06 20:01:28

- reply

@nprofile…lv5y I hear daily we should enable NPUs on Linux and I'm in the corner screaming TO DO WHAT! Like if you have a very specifically tuned model that is sized for the NPU you have maybe we can talk, then someone says LLM and I return to screaming 🙂

2025-02-06 19:49:31

https://cdn.fosstodon.org/media_attachments/files/113/958/673/949/621/850/original/1241ec517e9a504c.jpg

2025-02-06 19:48:41

https://imgflip.com/i/9jbniq

2025-01-24 23:54:15

Saw my first waymo in L.A. today, it's kinda "pale green pants with nobody inside them" eerie

2023-10-21 22:59:56

- reply

@npub1tjg…gye3 @npub1cm7…32y8 when it comes to asking someone to adopt your software all they have is the UI to go by, and gerritt for codereview was a huge turn off, like if the concern is new.programmers don't want to use archaic badly designed systems, moving to gerritt won't move in the right direction

2023-10-21 19:37:35

- reply

@npub1tjg…gye3 @npub1cm7…32y8 but gerritt really sucked, like I still use it for one or two projects and it's universally awful at the UI level. I can't imagine why any project outside of Google would use it. I use gitlab everyday and would love to see the kernel move towards it. But hosting a gitlab takes a lot of resources, and kernel sucks at finding tools development because nobody will agree on a toolset unless Linus dictated it and so many maintainers OCD their processes.