00:27 alyssa: (apparently Apple`s compiler does do that lowering but the documentation is very very confusing about it)
04:16 DemiMarie:wonders if any proprietary vendor has ever used Mesa
04:24 airlied: would they be a proprietary vendor then? :-)
04:42 DemiMarie: airlied: Mesa is MIT licensed so they could make a proprietary downstream fork.
04:43 DemiMarie:dreams that someday Mesa will be the only user-mode graphics driver, used on Windows, macOS, Fuchsia, and Android as well as desktop Linux
04:49 airlied: it would be a bit of a dead end, since we move so fast, at one point in AMD's history they used to reuse some mesa code for dri bits in their driver
04:53 HdkR: IMGTec also pulled some bits for their Linux driver at some point as well?
06:46 dj-death: I didn't realize lower_io wasn't able to limit structure access to only the data accessed
06:46 dj-death: like for struct { uint a; uint b; }
06:47 dj-death: any access to a or b will have the base=0 range=8
08:34 dschuermann: can someone from Intel confirm that https://gitlab.freedesktop.org/ShriramShastry is a real person and working for Intel?
08:46 dj-death: dschuermann: the email used on this MR https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30840 is in the employee db
09:44 dschuermann: dj-death: thanks. kind of odd contributions, though
09:45 dj-death: dschuermann: what's the one you're looking at?
09:46 dschuermann: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30915
09:47 ccr: eh
09:47 ccr: his replies smell like AI to me.
09:48 ity: Why does it read like ChatGPT
09:49 ity: Ah, so we weren't the only one that were thinking as such when reading the MR comment
09:49 dschuermann: yeah, it's really wild
09:51 ity: Yea...
09:54 dj-death: AI is a bloody plague right? :)
09:56 ity: We'd say misused GenAI is :D The technology behind GenAI is interesting. Of course as long as nobody makes MRs written by an LLM.
09:58 dj-death: it's all great unless you have to deal with the result ;)
10:00 ity: True enough...
10:01 pq: Ermine, I think yes, although usually it's the vert.refresh that's inaccurate and computed with the help of dotclock.
15:25 jenatali: Conan Kudo: Got a game that did this as an example? Since Mesa on Windows used to be software-only I'm not aware of this being common. I *am* aware of apps bundling ANGLE though
15:26 jenatali: Huh. I assume it bundles llvmpipe then?
16:14 Company: emersion: Do you know any place where pixel formats are matched with https://learn.microsoft.com/en-us/windows/win32/api/dxgiformat/ne-dxgiformat-dxgi_format ? Because your and afrantzis' page don't have them
16:15 Company: because I want to map them to GTK formats
16:19 emersion: Company: i have no idea about these (but could accept a patch to add these to pixfmtdb)
16:23 Company: I'll leave that to someone who knows all those formats
16:24 Company: I guess dzn might know?
16:45 landownerjobs: I think most here know how PCI interconnect works. I know some little too (inspected some cpu soc cores which had it). So the encoder variants are designed for simplicity under DMA device, at one end you toggle in the next compressed power representer, by adding one to last power transferred (rewriting byte count by last address which had 1), and another operand counts bytes transferred
16:45 landownerjobs: through writing 1s to enable address register to dest, and zeros where channel or byte-enable goes off so it skips zeroes. Plausibly enough there is two counters for bytes and words. Do not know when they do not match what happens though at hand. So two pipelined bursts aka one bursts latency makes the encoder functional. Technically those operations are easiest. I hope to do yet some
16:45 landownerjobs: research but i am fairly sure all hw has those functions, byte and word count and channel or byte enable regs.
17:22 cheako: Hi, I asked about this yesterday but I do hate to bump. On intel(a770) it feels like making too many image views causes both kernel drivers to error. I wonder if there is anyone who would dig into this more, or if I should try and guess?
17:24 jenatali: Company: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/microsoft/vulkan/dzn_util.c?ref_type=heads
17:25 Sachiel: cheako: got an issue with details?
17:27 jenatali: Company: Combine that with https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/vulkan/util/vk_format.c?ref_type=heads for the full mapping
17:28 Company: thx, will have a look
17:54 rattledjaws: Along with jews and even before them estonians are biggest freaks ever lived on this planet, entirely deluded fradulent activities. And through their terror to learn psychology is the biggest mistake possible to make, this science is not a real science. Where as you only get confirmation to the suspected fact that there amoral terrorists are entirely ill, and that opinion and force as to
17:54 rattledjaws: how abortion leftover leech and sick terrorist is also still a human comes from western world, not from Russia -- which is a country whiches scientists docked first their modules to the international space station. So they are pretty much a real people, not as toxic as Kalev Kurg bragging with his fuckshows with Gloria Terreur and Laura Keskinen, extorting money killing more powerful
17:54 rattledjaws: people, running around hotels with a knife extorting for money and resorting to scam how others disgust towards such people are mentally ill, they get shot by russians in masses the same way in front of their kids like Kalev Kurg got. Jewish and Estonians are the most toxic people on the planet with totally absurd and terror actions, so Estonians are ranked before them cause jewish had
17:54 rattledjaws: some real scientists too next to their medicines blood business.
17:58 olv: Hi, does anyone have any suggestion on whether syncobj query should guarantee progress? https://lore.kernel.org/lkml/20241022161825.228278-1-olvaffe@gmail.com/
18:01 olv: panvk needs that for timeline support in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31720
18:03 olv: another options is for vk_drm_syncobj to do a 0-timeout wait before a query. It can probably have some smartness to skip the wait when possible as long as the syncobj is not external
19:01 cheako: Sachiel: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31384#note_2628568
19:03 cheako: I don't think for details I really only have"it broken," so I wonder what more I should look into. I guess I should look at create image view and follow it to the ioctl or whatever it's calling.
19:19 cheako: From my little bit of looking into it, create image view is userspace only... so can't be the cause of the angry dmesg message.
19:21 Sachiel: ah, dgc... yeah, I'd assume "it broken" would be right for it at the moment...