12:53 zmike: eric_engestrom DavidHeidelberg: is there some way we can create global skips/flakes files for zink? things like spec@arb_timer_query@query gl_timestamp flake on every single job and listing it in every file is a bit tedious
14:40 robclark: zmike: toml file can list multiple skips files if you want a global one
15:12 zmike: robclark: oh right
15:18 dj-death: is there someone who looks after the angle build on CI?
15:18 dj-death: I think the build we're running is from July 2023
15:19 dj-death: exposing some new extension on Anv is apparently hitting asserts in Angle
15:19 dj-death: but I can't reproduce them on a more recent version of Angle
15:32 daniels: DavidHeidelberg: ^ can we update angle please?
15:52 jfalempe: tzimmermann: I've tested fbdev-shmem with mgag200, and it has no measurable impact on hwlatency.
15:52 tzimmermann: ok
15:53 tzimmermann: thanks for testing this
15:55 jfalempe: I'm testing on another server (hpe xl260), and base latency is a bit higher ~80us
15:57 tzimmermann: ok
15:57 tzimmermann: btw, the idea with the bmc bridge is likely not gonna fly
15:58 jfalempe: and disabling Write-Combine makes it lower than 10us. I will check with the ioburst workaround if that's similar. But I know at least on 2 other servers where it makes a difference.
15:58 jfalempe: ok so, we'll probably do the same as AST ?
15:59 tzimmermann: i think i might be able to handle this problem entirely within the probe helpers.
16:37 DavidHeidelberg: dj-death: daniels: I'll check, updating angle usually take sometimes, because of needed patching.
16:38 DavidHeidelberg: *some time
20:03 karolherbst: mareko: any idea why LLVM inserts vec4 stores here? https://gist.githubusercontent.com/karolherbst/ec33b4f1cdecf7946f5aeaee0029a767/raw/3f05e4608aebc46bc032f87ff04e8cd0b55e1185/gistfile1.txt
20:03 karolherbst: like the last one "store <4 x i16> %53, ptr addrspace(1) %85, align 8"
20:04 karolherbst: especially as it writes undef values to a location neither touched by the nir, nor the llvm later
20:05 karolherbst: also... what is llvm doing here anyway, it all looks very suspicous
20:06 karolherbst: (maybe I should just add the missing bits to aco I'd need)
21:09 DottorLeo: hi!