03:48 mareko: only 1 powerpoint? unbelivable
07:15 MrCooper: there's a Silent Bob joke in there somewhere, having trouble finding it though
10:32 MTCoster: I've just applied a series to drm-misc-next, but I need the first two patches to also get picked into -fixes. Can I just do that myself or should I be asking someone?
12:47 pinchartl: MTCoster: generally speaking, patches should never be merged into multiple branches as separate commits
17:44 alyssa: has anyone benchmarked _mesa_hash_table against other hash-table impls?
17:45 alyssa: it's consistently showing up in flame graphs but I'm.. unsure if hash-tables are intrinsically this slow or we're doing something wrong
17:45 zmike: I've found ht/set to be extremely slow too
17:47 mareko: it should be O(1) if there are no collisions
17:47 alyssa: unfotunately, 1 billion is O(1)
17:47 zmike: it's basically unusable in hotpaths
17:49 mareko: what you probably mean is that 1+ million lookups/s is slow, though even a dumb array may be too slow in that case
17:49 zmike: I've had to switch to sparse array in a number of cases because set/array was unusably slow
17:49 mareko: at that point the optimization goal is to minimize x86 instructions, jumps, and L1 cache misses
17:50 alyssa: I mean that the constant factor on mesa_hash_table is enormous and I'm not sure whether the constant factor on competing impls is noticeably smaller
17:55 alyssa: zmike: is util/sparse_array noticeably better than hash_table_u64..?
17:55 alyssa: (I don't really know how it works, only that it has a bunch of atomics that I don't care for.)
17:55 mareko: I haven't found sparse_array to be better
17:56 zmike: alyssa: yes extremely
17:56 zmike: I cut like 2ms off frametime by switching
17:58 mareko: I don't see zink using it
17:58 zmike: this wasn't in zink
17:59 zmike: but zink bottlenecks on hashtable/set usage so it's on my list to evaluate
17:59 mareko: it must be a secret project then
18:00 zmike: it is
18:00 zmike: I regrettably cannot say more about it at this time
18:00 mareko: zink2
18:01 zmike: would almost certainly be easier and saner
18:16 mareko: or the long-awaited softpipe rewrite
18:17 zmike: damn you figured me out
18:32 dcbaker: Vulkan -> OpenGL translator, so we can run shiny new software on the hardware we already have, since none of us can afford RAM
18:33 zmike: I should start taking notes...
18:55 ccr: call it kniz?
19:03 dcbaker: I was thinking "stink"
19:24 pinchartl: dcbaker: I'll dig up my voodoo 2
19:24 dcbaker: pinchartl: I still have a beloved laptop with a S3 savage :D
19:59 dviola: zmike: sorry to bug you again, but I got news in #radeon that the apitrace https://x0.at/QhlB.xz is working
19:59 dviola: for doom 3 bfg: https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15223
19:59 zmike: cool
19:59 zmike: I'll check it out on zink at some point
20:00 dviola: ty
20:05 dviola: in my case, this is what I see: https://x0.at/QTa3.png
20:06 dviola: for others, it works with allow_glsl_builtin_variable_redeclaration=true force_glsl_extensions_warn=true glretrace
20:37 alyssa: update: I got rid of the hash map and now it fast
20:39 zmike: brrrrr
20:39 anholt: daniels: should I see those s3 changes you made in helm-gitlab-deployment main?
21:28 airlied: Company: just ran gtk4-demo with vvl enabled "(gtk4-demo:3335933): Gdk-WARNING **: 07:25:18.242: Vulkan: Validation: vkCmdDraw(): the descriptor [VkDescriptorSet 0xcb00000000cb, Set 0, Binding 0, Index 1, variable "GSK_TEXTURE0_ARRAY"] is being used in draw but has never been updated via vkUpdateDescriptorSets() or a similar call.
21:28 airlied: The Vulkan spec states: Descriptors in each bound descriptor set, specified via vkCmdBindDescriptorSets, must be valid if they are accessed as described by descriptor validity by the VkPipeline bound to the pipeline bind point used by this command and the bound VkPipeline was not created with VK_PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT
21:28 airlied: (https://docs.vulkan.org/spec/latest/chapters/drawing.html#VUID-vkCmdDraw-None-08114)
21:28 airlied: this is on fedora 43
21:40 Company: oh, F43
21:41 Company: I was trying to repro on F44
21:42 Company: airlied: I'm pretty sure I fixed that, but it's entirely possible that the fix only made F44/4.22, not F43/4.20
21:43 Company: (it's fallout from me trying to make the spirv usable with spriv-cross so I can get DX working, but that never ended up working)
22:09 daniels: anholt: _jedi hand wave_ yes
22:10 anholt: gosh i must have needed to refresh that tab or something, huh.
22:37 anholt: daniels: hmm, "FAIL steam-d3d8/DukeForever_unkown_dx8_unknown_unknown_none.trace: upload failed: upload to https://s3.freedesktop.org/artifacts/anholt/mesa/mesa-tracie-results/: HTTP 403 Forbidden"