05:40 Lynne: can we turn on unifiedImageLayoutsVideo for gfx10?
05:41 Lynne: I just need that, I don't care about the general unifiedImageLayouts
05:41 Lynne: mareko: ^
07:24 glehmann: Lynne: > If VK_KHR_unified_image_layouts is supported, unifiedImageLayouts must be supported
07:24 glehmann: no
07:25 Lynne: no fun allowed
08:06 simon-perretta-img: I'm trying to debug a failing windows-msvc job that has warnings in a generated python->C file, but the build artifacts only seem to contain a .gitignore and the meson log; is there a way to grab the full build output/files?
08:09 glehmann: simon-perretta-img: if it's !42748, the reason is pretty simple. by allowing tuint, you also allow tuint64 sources, so msvc complains about the src & (1 << bit) when src is 64bit
08:10 glehmann: explicitly casting src to uint32_t before the & should prevent the warning I think
08:12 simon-perretta-img: glehmann: Makes sense, I'll give that a shot - many thanks
08:54 simon-perretta-img: Worked perfectly, thanks again!
11:08 dviola: mareko: in that case I wonder if I should reopen this: https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15102 -- I didn't realize drm native context was a different thing when I closed it
11:11 dviola: I'm still able to reproduce that issue with virgl, but not with llvmpipe or drm native context
11:19 dviola: I'll reopen it
11:35 mareko: native context is preferred if it's supported
11:40 dviola: right
11:40 dviola: it works great
14:20 dliviu: Hi, dim push is failing to merge drm-tip, looks like some amd changes in amdgpu_dm.c are causing it. Should I ignore it or try to follow the instructions in conflict-resolution.rst?
18:53 karolherbst: do we have a pass to optimize OOB scratch accesses to undef or something?
18:54 karolherbst: I e.g. have scratch: 16 and "div 8 %23 = @load_scratch_nv (%9, %22 (0x0)) (base=128, access=readonly, align_mul=1, align_offset=0)" which I know is always OOB, so I could eliminate that thing and replace it with undef, no?
18:56 karolherbst: though I guess it could overflow, mhh
19:11 karolherbst: mhhh I just found this: (('f2u32', ('fmul', ('frcp', ('u2f32', a)), 'b(is_ult_0x4f800000)')), 0)...
19:13 karolherbst: ehh maybe I need to adjust the pattern
22:21 karolherbst: uhm.. fmin/fmax require to return nan when one input is nan, right?
22:26 karolherbst: ohh it returns the other? mhh interesting