07:26 Daanct12: greetings, does anyone know why this happens on mesa when i turn on fog? https://echo.danctnix.org:7269/danct12/fog.mp4
07:29 Daanct12: this is code snippet https://echo.danctnix.org:7269/danct12/snippet.txt
07:30 Daanct12: doesn't happen with the proprietary driver, and i did check with LIBGL_ALWAYS_SOFTWARE
07:30 Daanct12: doesn't happen with the nvidia proprietary driver, and i did check with LIBGL_ALWAYS_SOFTWARE
09:43 karolherbst: jenatali: I have "dEQP-VK.glsl.builtin.precision_double.mix.compute.vec3" failing here with the ffma rework: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/98997768 is there any good way to look into this locally? Do we have something like drm-shim for dozen, where I could at least check out what nir it generates?
12:16 jenatali: karolherbst: not without windows unfortunately. I'll try to look today
12:29 karolherbst: thanks!
15:32 jenatali: karolherbst: WARP bug - you can add it as a known failure
15:33 jenatali: Pretty sure the version that's currently in CI is before I made it use hardware FMA (when available, i.e. AVX) and I guess the software version's got some bugs in handling around infs/nans/denorms
15:33 jenatali: But passes through dzn on real GPU hardware and on an updated WARP. I should try an updated WARP through CI, it's been a couple years...
16:45 karolherbst: jenatali, thanks for checking!
16:45 karolherbst: jenatali: oh btw.. heard about the CL 3.1 announcement? 🙃 I don't know if you were in the loop or not
16:45 jenatali: I only noticed through a PR to the CL ICD loader repo lol
16:46 jenatali: I'm out of the loop on CL these days
16:46 karolherbst: heh.. I think the "biggest" feature is subgroup support though not sure if you already support it or not
16:46 jenatali: I don't even remember, probably not
16:46 karolherbst: heh...
16:47 karolherbst: it's a good release, but nothing really difficult to implement, especially in mesa: https://registry.khronos.org/OpenCL/specs/unified/html/OpenCL_API.html#_summary_of_changes_from_opencl_3_0_to_opencl_3_1
16:49 jenatali: Unfortunately I haven't had any real time to work on CL, ever since Qualcomm decided they were going to ship their own CL drivers for the X1 (and now X2)
16:49 karolherbst: ohh, I see
16:49 jenatali: Same for dzn tbh
16:49 karolherbst: maybe I should look into making rusticl run on top of dzn...
17:14 robclark: karolherbst: ig v2026-03-25-00 is the right tag for 3.1 cts?
17:40 karolherbst: robclark: I don't think a tag was created yet...
17:41 karolherbst: there is a private branch, not sure why that hasn't been merged yet..
17:43 karolherbst: repo is here: https://gitlab.khronos.org/opencl/conformance-tests
17:53 robclark: karolherbst: ahh, ok.. lmk when there is a public branch (not urgent, working on other things atm)...
17:56 karolherbst: robclark: will do, I already created an MR to add all the necessary runtime bits at least: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41358
17:56 robclark: 👍