05:11 airlied: JoshuaAshton: btw my latest encode branch should probably also handle the get encoded headers late
08:06 HdkR: karolherbst: Yea, c++23 is giving us those types, which are at least namespaced so it won't hit the same problem
08:07 airlied: karolherbst: you mentioned llvm18 brokeness, https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27133 should fix it
08:07 karolherbst: cool
08:13 karolherbst: airlied: getting a "test_bruteforce: /home/kherbst/git/llvm-project/llvm/include/llvm/Support/Casting.h:578: decltype(auto) llvm::cast(From*) [with To = ConstantInt; From = Constant]: Assertion `isa<To>(Val) && "cast<Ty>() argument of incompatible type!"' failed. Aborted (core dumped)" now
08:38 pinchartl: karolherbst: doesn't C have _Float{16,32,64,128} ?
08:40 pinchartl:just learned that nvidia has a 19-bit floating point format
08:40 karolherbst: not sure if it's part of the C spec
08:40 karolherbst: mhh.. looks like it's part of a "extended float type" spec on top of C
08:41 pinchartl: it's part of https://www.open-std.org/jtc1/sc22/wg14/www/docs/n1945.pdf
08:42 pinchartl: https://en.cppreference.com/w/c/language/history
08:42 CounterPillow: Nothing says "open standard" like a big copyright notice at the beginning disallowing reproduction and utilisation
08:45 karolherbst: pinchartl: yeah, so not part of any C core standard
08:45 ccr: CounterPillow, even more "open" are the standards which you can't get a copy of without paying :P
08:46 mripard: I believe those are called industry standards
08:51 pinchartl: the best ones are the ones you have to pay for to only then realize that nobody implements them
08:59 CounterPillow: Since I do live in Switzerland it'd be devilishly affordable for me to send ISO a snail mail letter to ask for written permission to save a copy of this document to my local drive, and then also send them subsequent letters each day asking them for permission to run my daily backups, as ISO intended.
09:02 karolherbst: research publishing is scam, and everybody knows it :P
09:14 CounterPillow: You're technically allowed free viewing of any ISO standard through your country's local standards body, but at least here in Switzerland that means travelling to one specific viewing room in the country where you get to look at it with your eyes only and get to only make handwritten notes ;)
09:16 MrCooper: where's that viewing room located, out of curiosity?
09:17 CounterPillow: For me it's in Winterthur if I remember correctly, which is quite close to me
09:18 ccr: probably some place like this :P - https://www.youtube.com/watch?v=o2ObCoCm61s
09:53 tomeu: how did one get around this?
09:53 tomeu: Fetching http://zlib.net/fossils/zlib-1.2.13.tar.gz
09:53 tomeu: urllib.error.HTTPError: HTTP Error 415: Unsupported Media Type
09:54 tomeu: https://gitlab.freedesktop.org/tomeu/mesa/-/jobs/53892312
09:54 tomeu: downloading from my desktop is fine, so I wonder if it isn't due to some CDN, or a proxy wherever that runner is, ...
10:02 tomeu: woah, so downloading from firefox works, but when running external/fetch_sources.py here, I get 403 instead...
10:03 tomeu: hmm, it alternates between 415 and 403
10:05 tomeu: now a captcha popped up... "zlib.net is protected by Imunify360"
10:05 daniels: ugh ...
10:05 daniels: looks like we'll need a new source then
10:08 tomeu: will look a bit more at why it is only failing via http.requests
10:14 tomeu: daniels: looks like downloading it previously with wget and placing it in /VK-GL-CTS/external/zlib/packages/ should cause the script to skip the download
10:15 daniels: yeah strange, curl works to download as well
10:30 tomeu: daniels: will be testing that workaround
10:30 tomeu: btw, I suspect that HTTP Error 415: Unsupported Media Type is due to the captcha page being delivered instead of the tarball
10:43 MrCooper: mripard: FWIW, some KMS drivers allow enabling the CRTC but no plane, so "An active CRTC is guaranteed to have an active plane" isn't true
10:58 javierm: MrCooper: an active plane is one that's visible? Because I only see the struct drm_crtc_state to have an active field, not struct drm_plane_state
11:00 tomeu: daniels: actually, curl is also broken in the runner, as it downloads the captcha page as well
11:01 tomeu: maybe we should just accept that zlib.net is not a reliable place from where to download stuff
11:01 MrCooper: javierm: not 100% on the details, but IIRC: a plane is enabled if its CRTC_ID property has a non-0 value, otherwise it's not
11:02 daniels: tomeu: yeah, maybe using the Debian archive is a better idea
11:03 MrCooper: javierm: anyway, some IGT tests try to disable all planes while leaving the CRTC active, amdgpu has to explicitly reject that
11:05 javierm: MrCooper: yes, I think internally that's the "visible" field in the plane state and that's why tzimmermann suggested the drm_for_each_primary_visible_plane()
11:06 tomeu: daniels: are you thinking of http://deb.debian.org/debian/pool/main/z/zlib/zlib_1.2.13.dfsg.orig.tar.bz2 ?
11:06 daniels: tomeu: yeah, exactly
11:07 javierm: MrCooper: and you are correct I think that an active CRTC can have no active (visible) planes
11:07 CounterPillow: Somebody should probably tell zlib.net about this
11:10 MrCooper: javierm mripard: looking at amdgpu code, the FB_ID property needs to be non-0 as well
11:12 tomeu: CounterPillow: will drop them an email
12:22 karolherbst: jenatali: I'm thinking of bumping the LLVM req to 14 or even 15 for clc... using anything below 14 is broken anyway
12:23 karolherbst: and 15 gives us fast compile times, though I can see why that's not as important
12:24 karolherbst: mhh.. llvm-15 was released sep 2022, llvm-14 in march 2022
12:24 karolherbst: atm we depend on 10+ which was released in March 2020
12:25 karolherbst: radeonsi needs 15+ anyway
12:25 karolherbst: (intel_clc 13, clover 11 and apparently gallium still supports 5+?)
13:02 alyssa: karolherbst: please bump to llvm-23+ for the spir-v backend bug fixes
13:02 karolherbst: 23?
13:02 alyssa: :p
13:02 karolherbst: :P
13:06 jenatali: karolherbst: ack, we're using 15 atm so that sounds fine
13:10 karolherbst: jenatali: I still wonder about 14 vs 15... 14+ allows us to get rid of 2 ifs.. 15+ 5 additional ones
13:10 karolherbst: maybe I do it in two commits and the ones wanting to build with llvm-14 can just revert it
13:10 jenatali: Up to you, doesn't impact me
13:11 karolherbst: okay
13:11 karolherbst: CI seems to be on 15 as well
13:24 mripard: MrCooper: I didn't know that, thanks :)
13:24 mripard: MrCooper: out of curiosity, which driver do you have in mind?
13:25 javierm: mripard: he mentioned amdgpu
13:29 MrCooper: actually I suspect most drivers allow this (some of which possibly unintentionally :), with a few exceptions
13:29 MrCooper: such as amdgpu
13:32 karolherbst: done... I got to remove more code than I thought :D
14:19 sima: MrCooper, part of the reasons for the simple kms helpers was that way too many drivers failed to reject the config if the primary plane was disabled :-)
14:19 sima: it's a very common bug :-/
14:27 jenatali: karolherbst: not pushing r-b tags? Are we dropping that req for common code?
14:27 karolherbst: uhh.. good question
14:28 jenatali: IMO we should so I don't care, but I suspect it was unintentional:)
14:28 karolherbst: yeah.. not really thinking about it too much honestly. I just mostly do it if I need to push anyway
14:32 karolherbst: I also still need to merge 25568 but I think I've waited long enough for others to complain about forcing fPIC :D
16:37 javierm: sima: I failed to parse your comment, do you mean that drivers that allow to have an active CRTC with its primary plane disabled are buggy or that drivers that don't allow that are?
16:42 sima: javierm, they're only buggy if they allow it but the hw/driver doesn't actually support it
16:42 sima: so neither
16:42 javierm: sima: got it
16:43 javierm: and that explains why I failed to parse your comment. It makes sense to me now :)
20:22 JoshuaAshton: airlied: Cool! Sorry for late response, was ill yesterday and didn't get out of bed e-e
22:19 gfxstrand: Does anyone else have a "real" 32-bit imad?
22:19 gfxstrand: I know a few people have imad24 or something
22:20 gfxstrand: Ooh, NIR has an iadd3 already. :D
23:42 airlied: karolherbst: on that brute force fail any info on what test it was, wimpy mode?