00:33 Sachiel: any clues what happened here? https://gitlab.freedesktop.org/mesa/mesa/-/jobs/58014490
00:34 Sachiel: oh, it revived after marge aborted
00:58 Kayden: yeah...had two MRs fail to marge because CI was taking too long :/
00:58 Kayden: just one of those days I suppose :)
03:03 DemiMarie: dj-death: is there any way to validate a DMA-BUF?
03:15 fastrabbit: Derivative itself is easy concept, first derivative is linear algebra where the formula is derived from relation term for the focal term of interest, and second is substitution with values where the phenomenon of interest gets exact value and the dimension is at the time of interest, that's in mathematics as high maths and part of the integral computation as sub class or subject of math it's not complex to grown up man.
03:32 dwfreed: .
06:03 Kayden: I think something is wrong with the iris-kbl runners
06:04 Kayden: been trying to merge a patch series that doesn't affect kbl for 7 hours now
06:04 Kayden: and on one run they ran for 54 minutes and timed out, and on another they just sat queued for ages
06:06 Kayden: guess I can write a patch to disable them
06:08 Kayden: DavidHeidelberg: that's your farm, right? does disabling iris-kbl for now sound like the right thing to do?
06:11 Kayden: (also, no worries, just wondering what to do)
08:04 daniels: Kayden: he’s in the US atm, but indeed please disable them if they’re problematic
09:34 colinmarc: why is using VkMemoryDedicatedAllocateInfo necessary/relevant when importing a dmabuf? is that documented somewhere? I just fixed this nvidia bug I've been struggling with for months by chaining that into the memory creation call, and I have no idea why. (Credit to nowrep, who suggested it, but I guess he's not in this channel)
09:34 colinmarc: I would like to submit docs upstream somewhere so the next person doesn't get stuck...
12:01 jani: mlankhorst: when do you plan to send the next (last?) drm-misc-next pull request? I'd like to get https://lore.kernel.org/r/20240422121011.4133236-1-jani.nikula@intel.com included
12:01 jani: mlankhorst: it touches so many places, that would be good to get them in sync for the merge window
12:44 kusma: tsk tsk: https://gitlab.freedesktop.org/mesa/mesa/-/tree/elima/radv-video-encode-caps-maxbitrate?ref_type=heads
12:53 demarchi: for these emails from lkp about build regressions... do you know where the logs for recent errors are stored?
13:46 JoshuaAshton: colinmarc: We also use VkMemoryDedicatedAllocateInfo in Gamescope for dmabuf imports fwiw
13:47 JoshuaAshton: colinmarc: https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VkMemoryDedicatedAllocateInfo.html#VUID-VkMemoryDedicatedAllocateInfo-image-01876
13:49 colinmarc: JoshuaAshton: Right, I'm mostly wondering why that's important
13:49 colinmarc: it's not intuitive
13:50 colinmarc: I wonder if that validation error should cover dmabuf too?
13:53 mlankhorst: jani: looks sane, planning on later today. :)
13:53 mlankhorst: touches many places though
13:53 mlankhorst: Not sure what the policy is in that case
13:54 jani: mlankhorst: I pretty much said I'll take acks but merge anyway :)
13:55 mlankhorst: Ok, go for it so I can do a pull request
14:17 jani: mlankhorst: done. thanks!
14:22 mlankhorst: Ok, I' ll start writing. :)
15:08 zmike: jenatali: I assume you do shader rewrites for legacy (4 component) shadow samplers?
15:08 jenatali: I think so?
15:08 zmike: I'm gonna move some of my handling into common
15:09 zmike: so maybe you can delete some code too
15:09 jenatali: I'm pretty sure our lowering is already in common
15:09 zmike: 🤔
15:10 zmike: I don't see any tracking for this in shader_info?
15:10 jenatali: We barely use shader_info, most everything is variables and derefs
15:10 jenatali: But no, what I'm thinking of is nir_lower_tex_shadow which we only run on shadow textures with txl, txb, or txd
15:11 zmike: ah
15:11 jenatali: As of D3D shader model 6.7 we can support txl, and 6.8 finally added txb and txd so we can opt out of that lowering on new D3D drivers
15:11 jenatali: I'm trying to remember what we do for legacy shadow sampling instrs
15:12 zmike: does your shadow sampling have 4 components or 1?
15:12 jenatali: 1
15:12 zmike: that's what I thought
15:12 zmike: so this will help you in some way
15:12 zmike: maybe
15:13 jenatali: Oh hm, the HLSL returns a float but I think the DXIL actually still returns a 4-component result
15:13 zmike: hm
15:13 jenatali: Yep, returns a "%dx.types.ResRet.f32" which is a 4-component result
15:14 zmike: I see
15:14 zmike: so then maybe this isn't useful after all...
15:14 jenatali: So we only do the 4 -> 1 thing when removing the shadow aspect of it
17:46 Lyude: In a low memory situation (say, you have 16 GB of ram but only like 300MB free) a 77kb cache-coherent dma allocation shouldn't be that likely to fail should it? or am I underestimating the power of memory fragmentation?
19:48 robmur01: Lyude: out of 16GB, I'd expect that 300MB-worth to be scattered all over the place, and finding a contiguous 128KB for an order 5 allocation quite possibly tricky indeed
19:54 robmur01: especially if it's also constrained by trying to match a less-than-64-bit DMA mask