06:11mareko: FYI, gitlab times out
11:35K900: zmike is zink-vaapi still a thing that is planned?
12:24alyssa: gfxstrand: what happened with map_mem_placed tests?
12:24alyssa: I see an unmerged MR and an unanswered CTS ticket..?
12:24alyssa: ^crucible
12:25alyssa: I'm just kinda scared having no CI coverage for this rn
12:32zmike: K900: yes, we usually work on it at the end of the year though so probably not any time soon
12:32K900: Thanks
13:57Lynne: why are mesh shaders so tied into renderpasses?
13:59Lynne: I don't even want any vertex/fragment shader legacy crap, not when I can just send ray queries via a compute shader and draw manually
14:00alyssa: what
14:01zmike: WHAT
14:02HdkR: wha?!
14:02alyssa: quoi
14:04Lynne: pretty sure it's allowed, I've seen compute use ray queries before
14:05Lynne: so if I just construct geometry via a mesh shader, then bvh, I can call ray queries in compute and fill the screen via imageStores
14:05alyssa: perf on mobile will be trash
14:05Lynne: mobile has raytracing and mesh shaders?
14:05alyssa: yes
14:06alyssa: but also mesh is for the rasterization pipeline not RT
14:06alyssa: I don't understand how they would even interact
14:06alyssa: i guess you could use a ray query in a mesh shader :clown:
14:07zmike: you could dgc a mesh shader which writes to a storage buffer containing commands for dgc rt
14:07alyssa: galaxy brain
14:08Lynne: can't I call vkCmdBuildAccelerationStructuresKHR() after the geometry's been built?
14:08alyssa: the purpose of mesh shaders is to give a faster path to feed the rasterization hardware
14:09alyssa: so I don't know how RT fits in
14:10Lynne: its a neat way of constructing vertices, and raytracing takes vertices to generate a structure?
15:11Lynne: seems to be working so far
15:13Lynne: to give some context, this is just for fun, it won't get ran on anything mobile (not like android speaks vulkan imho), and yes, I am going out of my way to avoid any and all boring rasterization pipelines
18:13alyssa: bikeshed: what directory name (and where in tree) would a hypothetical common geom/tess/xfb/pipeline-stats/indirect-draws/primitive-restart/maybe-mesh? emulation library go
18:16zmike: src/alyssa
18:17alyssa: zmike: what a callout
18:17HdkR: src/emulators_are_us/
19:08mareko: alyssa: I'm thinking of transforming simple GS into VS with an option to pass data between invocations of the same input primitive, but executing 1 invocation per output primitive
19:08mareko: per output vertex I mean
19:16alyssa: mareko: that seems pretty different from what my code does - I don't know how I would pass data between invocations
19:16alyssa: I guess AMD hw gives more guarantees about subgroup ops and things than the vulkan spec does
20:10mareko: it's a mesh shader workgroup (with shared memory), but the number of invocations is equal to the maximum number invocations that the GS can generate
20:11mareko: also all VS and GS sysvals are available
23:24Company: jenatali: do you have some location where I can download a pre-built DLL for GLon12?
23:25Company: the AMD Windows drivers are spamming my D3D12 debug output and also leaking my textures, so I'm looking for something better
23:25jenatali: https://github.com/pal1000/mesa-dist-win/releases ?
23:26Company: that works I guess - I was looking in freedesktop CI
23:26jenatali: That'd work too, any of the pipelines there that build Windows will have artifacts
23:27Company: I just looked and didn't find artifacts for https://gitlab.freedesktop.org/mesa/mesa/-/pipelines/1411218
23:28Company: which was the latest one
23:28jenatali: That didn't touch any code that builds for Windows so the Windows pipeline didn't run
23:29Company: ah, so I have to find commits that touched the right subdirs
23:29Company: that makes sense
23:38Company: yeah, I get transparency working, textures not leaking and not a single complaint by the d3d debug layers. So it's indeed the AMD GL driver being shite
23:39jenatali: That tracks with what I've heard, unfortunately
23:39jenatali: I wish we were able to convince them to drop their native GL driver and just use Mesa
23:40Company: I'm wondering if GTK should tell people to ship with the GLon12 DLL
23:41Company: and just use that
23:42zmike: smh not using zink
23:44Company: that would use Vulkan
23:45Company: and while AMD's Vulkan is better than AMD's GL, I don't think that goes for all drivers
23:45Company: on Windows with vendor drivers it's D3D or bust
23:47Company: Can DZN do multiplanar formats yet?
23:47jenatali: I honestly don't remember offhand
23:48jenatali: Been a while since I've touched it unfortunately
23:48Company: 6 months ago it didn't
23:49jenatali: Then no
23:49jenatali: GL does IIRC
23:49Company: if I use EGL I suppose
23:50Company: but WGL doesn't do that at all - I have to map the D3D subresources into individual GL textures I think
23:51Company: actually I don't even know - the EGL stuff is all dmabuf specific
23:52Company: I'd have to look if there's an extension for win32/d3d
23:52jenatali: I'm pretty sure I got NV12 stuff working through EGL at one point
23:55Company: lul
23:56Company: lavapipe is definitely cheating somehow
23:56jenatali: Hm?
23:57Company: it's getting 24fps playing a 4k video