08:50mripard: tzimmermann: FTR, I think I addressed all your comments on the create state series, are you ok if it gets merged?
08:50mripard: (i'm not going to do it today, probably early next week)
08:50tzimmermann: mripard, sure go ahead
08:50mripard: thanks :)
08:51tzimmermann: mripard, it's that new v6 revision?
08:52mripard: yes
08:52tzimmermann: let me just r-b that first patch. IIRC that was still missing
08:53mripard: you had comments on patch 16 too
09:05tzimmermann: it's good
10:29bbrezillon: mripard, tzimmermann: gentle ping on the -rc5 -> drm-misc-next backmerge (unless I missed it, the gem_lru fix is not in drm-misc-next)
10:30tzimmermann: sima, airlied ^
10:30tzimmermann: please update drm-next to rc5 so it can be backmerged
10:39tzimmermann: airlied_ ^
11:05bbrezillon: beware of the silent confict
11:06bbrezillon: resolution provided here https://lore.kernel.org/dri-devel/20260518165225.145175b1@fedora/
12:34Lynne: are there any plans for implementing coop matrices on RDNA2?
12:34Lynne: pendingchaos: ^
12:35pendingchaos: WMMA instructions don't exist on RDNA2
12:40Lynne: oh, rip, I thought they were added with raytracing instructions
12:41Lynne: no matrices, no atomic floats, first gen raytracing, must have been a simpler time back then
12:55glehmann: we could try to do something with v_dot2c_f32_f16/v_dot4c_i32_i8 and DPP
12:55MoeIcenowy: For https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41737 , now I wonder whether wlroots should be fixed instead (by exposing the real renderer's node as zwp_linux_dmabuf_v2 default device)
12:55MoeIcenowy: although in case any compositor implements 2D hardware acceleration on non-3D-capable hardware the problem still arise
12:56glehmann: but without WMMA it probably won't match the performance that applications expect
15:16f_: MoeIcenowy: from what I understand it's not just wlroots?
20:39airlied: pendingchaos: with that transpose change did you see any difference in any benchmarks?
20:39pendingchaos: I didn't run any benchmarks
20:39pendingchaos: fossil-db changes with cts tests looked good though
20:42airlied: okay, I'll see if I can spot any differences on the coop mat perf test
21:29airlied: pendingchaos: interesting drops code size in the shaders, but still don't get more throughput. mem bw bound completely