03:47 fdobridge_: <a​irlied> just for reference adding vmbind to nouveau gallium driver is in the fuck it I'm starting the driver again, or at least import libdrm_nouveau into mesa arena
05:22 fdobridge_: <p​rop_energy_ball> zink hype
05:34 fdobridge_: <a​irlied> yeah trying to get zink to work as a fallback now, I'm not sure which horror show is worse
06:03 fdobridge_: <a​irlied> @gfxstrand should nvk expose .EXT_image_drm_format_modifier already?
06:06 fdobridge_: <a​irlied> oh I assume we have to hook up the modifiers first
06:59 fdobridge_: <a​irlied> oh I see you have a branch already, might be worth rebasing to see if we can get zink to work
08:30 fdobridge_: <m​ohamexiety> Iirc that branch is still incomplete but not sure what it's missing as I haven't looked yet. There's an annoying/obscure bug I am chasing in sparse right now and then I was supposed to finish modifiers
09:48 fdobridge_: <k​arolherbst🐧🦀> yep
09:48 fdobridge_: <k​arolherbst🐧🦀> I think we only have two options: rely on zink or write a new driver
12:38 fdobridge_: <z​mike.> surely job security demands you write a new driver
12:40 HdkR: Everybody, Zink...Two
12:44 fdobridge_: <z​mike.> as long as you keep in mind there will be no zink 3
12:45 HdkR: Can't count to three
13:41 dakr: Lyude: Cool! I think it depends, if we're confident the code is good to go upstream we can also keep some bits we don't need necessarily I think.
15:05 fdobridge_: <m​ohamexiety> Zink: The Prequel
15:08 fdobridge_: <g​fxstrand> As long as it's not Zink Episode I.
15:08 fdobridge_: <S​id> speaking of
15:09 fdobridge_: <S​id> Episode I is returning to theaters on May 4th
15:09 fdobridge_: <g​fxstrand> Yay?
15:09 fdobridge_: <g​fxstrand> I think I'll burn my theater spoons going to Dune 2
15:10 fdobridge_: <g​fxstrand> Though going to Episode 1 might be a fun flashback. I saw it in the theater with my dad when I was in Jr. High.
15:11 fdobridge_: <g​fxstrand> Or it could be a horrible flashback. Flashbacks can go either way for me. 🙃
15:11 fdobridge_: <S​id> I was -4 years old when phantom menace released...
15:12 fdobridge_: <S​id> also dunno if it'll be a global re-run, so
15:12 fdobridge_: <S​id> I don't have much hope
15:13 fdobridge_: <g​fxstrand> I'm at that age where I never know when I say something like that if people will think I'm old or young. 😂
16:15 fdobridge_: <t​om3026> 1999 jeez, yeah i saw it in cinema too. didnt realize it was that long ago
16:23 fdobridge_: <g​fxstrand> 1999? Really? Wow, I thought it was a year or two later than that.
16:24 fdobridge_: <r​inlovesyou> i was like -3 years old when the phantom menace released
16:24 fdobridge_: <S​id> high five
16:24 fdobridge_: <t​om3026> time flies faster then one might think, nvk is soon 2 years
16:24 fdobridge_: <t​om3026> 😄
18:39 fdobridge_: <g​fxstrand> https://gitlab.freedesktop.org/mesa/crucible/-/merge_requests/147
18:40 fdobridge_: <g​fxstrand> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27599
18:40 fdobridge_: <g​fxstrand> That's the one I meant to post
18:43 fdobridge_: <g​fxstrand> Is that enough for gamescope?
18:43 fdobridge_: <g​fxstrand> Because that I can implement without kernel work
18:44 fdobridge_: <!​DodoNVK (she) 🇱🇹> gamescope doesn't even use that
18:45 fdobridge_: <g​fxstrand> Yeah, but gamescope uses host pointer, doesn't it?
18:47 fdobridge_: <!​DodoNVK (she) 🇱🇹> It doesn't use the EXT_external_memory_host extension either and I think host pointer requires that extension to use it
18:48 fdobridge_: <g​fxstrand> Okay, then I got confused
18:49 fdobridge_: <g​fxstrand> In any case, this will make WoW64 happy
19:09 fdobridge_: <!​DodoNVK (she) 🇱🇹> Wine doesn't have the plumbing yet because of the extension being not merged (I found a branch for enabling that though so this should be easy to test)
19:46 fdobridge_: <a​irlied> @gfxstrand do we have users for host ptrs? like I could actually spend time figuring it out in the kernel
19:48 fdobridge_: <d​adschoorse> vkd3d-proton needs it for the d3d12 equivalent, which some games use
19:57 fdobridge_: <a​irlied> @gfxstrand @mohamexiety would be good to get the modifiers branch rebased at least
19:58 fdobridge_: <g​fxstrand> Yeah, the extension should land this week or next. I pulled the "Skip the usual Khronos process and ship it" card today.
20:13 fdobridge_: <!​DodoNVK (she) 🇱🇹> Wine uses it for hacky WoW64 Vulkan support (the proper solution is the extension mentioned above)
20:24 fdobridge_: <b​abblebones> Does the nvidia open gpu modules implement anything that can conceivably be used with userspace VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT vulkan calls for compute shaders?
20:24 fdobridge_: <b​abblebones> I've noticed the bin is only ever able to acquire a medium compute queue and *no higher* when asked
20:25 fdobridge_: <b​abblebones> Does the nvidia open gpu module implement anything that can conceivably be used with userspace VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT vulkan call for compute shaders? (edited)
20:25 fdobridge_: <!​DodoNVK (she) 🇱🇹> https://discord.com/channels/1033216351990456371/1034184951790305330/1205266596516466799
20:25 fdobridge_: <b​abblebones> Oh! So no priorities in their official open module at all, not just referring to nouveau here
20:27 fdobridge_: <!​DodoNVK (she) 🇱🇹> That's for nouveau
20:31 fdobridge_: <b​abblebones> Hmm, hoping someone here has peered at the open modules long enough they might know if there's shader priority bits in there that the prop userspace can engage with
20:33 fdobridge_: <b​abblebones> The binary libs appear to only return medium queues when requested highest possible, but there's article evidence suggesting nvidia implemented VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT in their 470 driver release
20:33 fdobridge_: <b​abblebones> Ah yes, changelog here
20:33 fdobridge_: <b​abblebones> https://forums.developer.nvidia.com/t/linux-solaris-and-freebsd-driver-470-42-01-beta/181536
20:41 fdobridge_: <a​irlied> you'd have to trace the open module calls with a non-pri and pri context and see if you can spot the difference
20:43 fdobridge_: <a​irlied> there are a bunch of NV0080 for fifo configuration, and NVCB97_SET_SCG_GRAPHICS_PRIORITY and NV2080_CTRL_PERF_BOOST_FLAGS_CUDA_PRIORITY just from glancing
20:54 fdobridge_: <r​hed0x> I hope they'll also do the rerun in Germany
21:37 fdobridge_: <g​fxstrand> Done. I think the next step is to play with it in the context of WSI and see how that goes.
21:39 fdobridge_: <b​abblebones> From those names I can make the assumption that ones for the graphics and compute queue, nice
21:39 fdobridge_: <b​abblebones>
21:39 fdobridge_: <b​abblebones> Well, there is something there, it just seems nvidia has disabled their binary userspace vulkan realtime bits