11:14 karolherbst: somebody want to do a quick review of llvm-21 compat patches? Touches clc_helpers.cpp https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35399
11:15 dj-death: karolherbst: done
11:15 karolherbst: thanks!
11:32 bluetail: karolherbst why you close so many PRs? https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/?sort=created_date&state=closed&first_page_size=20
11:43 mattst88: bluetail: https://gitlab.freedesktop.org/xorg/xserver/-/work_items/1822#note_2951449
11:52 bluetail: Wow. That really clears things up. The not-so-well-informed making up things and really bad-mouthing karol
12:12 daniels: and sure enough, https://github.com/X11Libre/xserver/pull/61
12:13 karolherbst: yeah well.. comes with the job
12:14 bluetail: I read that x11libre was dead before it was created. Wondering if it holds true.
12:17 karolherbst: only time will tell
12:26 Company: I think xlibre will be as successful as goneme
12:27 feaneron: goneme? gnome?
12:27 Company: feaneron: https://www.osnews.com/story/7751/ali-akcaagac-attempts-to-fork-gnome/
12:27 feaneron: hah
12:29 karolherbst: that's like 20 years ago, no wonder I never heard of it
12:42 Company: it's really hard to find stuff about it, not even archive.org has that
12:42 Company: https://tech.slashdot.org/story/04/07/25/144244/project-goneme-fixes-perceived-gnome-ui-errors does exist though
12:42 Company: 20 years and nothing has changed...
12:43 llyyr: wow that X11Libre repo is like a honeypot for finding people mindbroken by the existence of CoCs
13:04 jani: mripard: dianders: thanks for the reviews/acks/testing on https://lore.kernel.org/r/cover.1749199013.git.jani.nikula@intel.com, much appreciated! do you think dianders' Tested-by is enough for merging? do we need more testing on a DT system or are we good to go?
13:06 jani: (I was planning on merging it all via drm-misc, as we can get a backmerge fairly quickly at this point of the development cycle)
13:24 tintou: zmike: Hi there! Do you have any further comment regarding https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35091 ? 🙂
13:28 zmike: uhhhh
13:29 zmike: I guess not
13:30 tintou: Thank you :)
15:12 glehmann: does anyone know why we use ieq(fabs(x), inf) for OpIsInf instead of feq!(fabs(x), inf)?
15:12 glehmann: I assume most backends only have abs on feq, not ieq
15:17 karolherbst: could nan interfer?
15:18 karolherbst: though for nan this should be false...
15:19 karolherbst: but yeah, nvidia has the abs modifier only on float instructions
15:21 karolherbst: glehmann: fcca6a83cddf
15:22 karolherbst: though we kinda fixed unordered comparisons since then, so maybe it's fine?
15:22 glehmann: feq is clearly defined to be ordered now
15:23 karolherbst: yeah.. I think we can use feq these days now
15:23 glehmann: we should make both the feq and the fabs exact though, and force unset the ninf/nnan flags
15:36 glehmann: huh, +0.63% instructions in MHW... this is going to be quite the rabbit hole
15:38 dianders: jani: I think it should be OK. It's not like anything else is affected other than switching "OF" calls to "firmware" calls and my testing showed that worked OK on DT systems. I kinda doubt you'll get anyone else to test before landing anyway.
15:54 glehmann: okay, the rabbit hole wasn't that deep: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35437
22:19 robclark: karolherbst: how do you test cl_khr_image2d_from_buffer? Looks like cl cts tests require cl_ext_image_requirements_info as well?
22:20 karolherbst: nope, it's a bit hidden, let me check it again
22:20 robclark: thx
22:20 karolherbst: robclark: ./build/test_conformance/images/kernel_read_write/test_image_streams 2D read CL_R CL_SIGNED_INT8 CL_FILTER_NEAREST CL_ADDRESS_CLAMP_TO_EDGE UNNORMALIZED
22:20 karolherbst: you'll see it gets executed as 2D and 2D from buffer
22:21 robclark: ok.. let me try that
22:22 karolherbst: _but_ in theory I could implement support for this extension in the frontend thanks to zmike :D
22:22 robclark: perfect.. and I guess I can just try different formats to make sure no bytes vs pixels unit screwups
22:22 karolherbst: right
22:22 robclark: ok, perfect
22:22 karolherbst: apparently you can prime export a buffer and then uhm prime import it as a 2D thing and....
22:22 karolherbst: but I think I prefer drivers to implement it natively :D
22:24 zmike: coward