01:57 mareko: tarceri: is it required to remove NIR variables of dead input/outputs? Since my nir_opt_varyings pass works with load/store intrinsics, it doesn't touch any NIR variables, so I wonder if I have to do anything with them.
13:25 DavidHeidelberg[m]: someone with god-like bash trap skills? I need review https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21715
13:26 DavidHeidelberg[m]: It fixes the issue, I'm only worried about side effects
14:34 DavidHeidelberg[m]: For CI, can we gather DEQP flakes as a Pokemon cards? :D
14:35 ccr: gotta catch 'em all?
14:35 DavidHeidelberg[m]: You could play common flake for lower points or put on the table flake with flake once a year ✌️
14:38 DavidHeidelberg[m]: eric_engestrom: Btw. did RPI farms encountered some issues yesterday? seeing bootloader fails in the flakes https://gitlab.freedesktop.org/mesa/mesa/-/issues/8436
18:48 DemiMarie: mi6x3m: what happens if you build with `-fno-strict-aliasing`? Is Mesa built with that flag?
19:02 alyssa: nir_ishl_imm doesn't match the (SM5) definition of nir_ishl
19:02 alyssa: this seems screwy
20:01 mareko: alyssa: what's the SM5 definitino?
20:13 alyssa: mareko: shift is masked by the number of bits
20:13 alyssa: e.g. (1 << 34) = (1 << 2) = 4 in nir
20:14 alyssa: (for 32-bit "1")
20:14 alyssa: (this is what direct3d requires and what most hardware does)
20:14 alyssa: but nir_ishl_imm(1, 34) = 0, contradicting
20:15 mareko: yes, that's wrong
20:16 alyssa: cmarcelo: and lionel (don't think he's here) -- you two added that in !13739, any chance you remember why?
20:16 mareko: ishr is correct though
20:16 alyssa: yep
20:16 alyssa: confusing in't it
20:19 Lynne: I got a report that descriptor buffers are broken-ish on the steam deck, would anyone like to take a look at it?
20:20 Lynne: the code is for ffmpeg and still hasn't been merged, which is why there isn't a bug report yet
20:20 dj-death: alyssa: copied from ishr_imm I suppose
20:21 mareko: dj-death: no, ishr_imm is correct
20:23 alyssa: dj-death: oh hello yes that's you ok
20:23 alyssa: I was trying to tab complete lland and coming up empty :)
20:24 dj-death: lol
20:25 dj-death: looks like a rebase fail then
20:25 dj-death: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13739#note_1174931
20:25 dj-death: the Rb version was right
20:26 dj-death: I think it got messed up with the nir_ior_imm in the other patch
22:46 Ristovski: is there a way to spoof the GPU vendor id for OpenGL?
22:47 Ristovski: ah, force_gl_vendor :)