00:11 tleydxdy: I usually use radeontop -b to select gpus, but that's with multiple dedicated gpu
00:12 tleydxdy: it may be that some apps are rendering on the igpu and getting copied to the dedicated gpu for output?
00:12 tleydxdy: aka PRIME
01:51 Lynne: "/usr/lib/x86_64-linux-gnu/libvulkan_nouveau.so supports Vulkan 1.3, but only supports loader interface version 4."
01:51 Lynne: it tried
02:00 Lynne: also that kernel fork doesn't seem able to load the gsp firmware for ad10x devices yet, it's just an empty stub
06:42 Lynne: hmm, I don't see it loading any firmware, is it relying on the firmware already existing on the GPU (if nvidia GPUs do that at all?)
06:43 Lynne: nope, it's just not finished yet, well, it's still progress, I'm sure in a couple of hundred lines it'll work
14:43 penguin42: airlied: I see your tinygrad article; I've got OpenCL profile patches for rusticl in a pull request but sitll being worked on; also if your kernels use any vector load/stores there's an outstanding patch to turn them on in rusticl
17:09 airlied: penguin42: full profile support for radeonsi already has an open MR, and my branch has nir vectorisation in it
17:16 penguin42: airlied: This is the profiling on the rusticl side; https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24101
17:17 penguin42: airlied: The vectorise patch I've got was one from Marek; https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24101
17:17 penguin42: no, bad c/p
17:17 penguin42: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9139#note_1945109 better
21:44 Company: apparently on AMD I get glitches if I draw my icon textures instanced, instead of 1 vkCmdDraw() per icon: https://i.imgur.com/K687t7S.mp4
21:45 Company: first icon (the hamburger) is glitchfree, but the 2nd and 3rd one get glitches - unless I do the mouse hovering where only one icon gets redrawn, then it works fine
21:48 Company: I should maybe mention that I'm using EXT_descriptor_indexing, so I have all my textures in a huge array and just index them
21:49 Company: now, how do I turn that into a useful bug report?
22:15 dottedmag: Is this comment wrong? https://elixir.bootlin.com/linux/v6.4.2/source/drivers/gpu/drm/drm_mode_object.c#L418
22:34 dottedmag: Also, is GAMMA_LUT[,_SIZE] CRTC property a new (atomic-compatible) version of DRM_IOCTL_MODE_{GET,SET}GAMMA?
22:59 zamundaaa[m]: dottedmag: I assume you're referring to the fact that the comment talks about one property, instead of all of them?
22:59 dottedmag: zamundaaa[m]: yes
23:00 zamundaaa[m]: Yeah that is wrong
23:00 dottedmag: ty
23:00 zamundaaa[m]: About the gamma properties, that is indeed the atomic version of the ioctl
23:01 dottedmag: thanks again