12:04fdobridge_: <!DodoNVK (she) 🇱🇹> What's the blocker for MR !27159? :frog_gears:
13:24fdobridge_: <!DodoNVK (she) 🇱🇹> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27654
13:24countrysergei: Mind cripples and piece of shits, i prefer to leave from your terrorist suicidal cripple environment until i come back with my pals to buttfuck you trashbags. You are incredible scum.
14:58fdobridge_: <clangcat> What?
14:59fdobridge_: <Sid> average OFTC #Nouveau spammer talk
15:28RSpliet: This happened in the freenode days too, and isn't exclusive to nouveau but rather anything that gets close to the Linux direct rendering manager
15:28RSpliet: ... just for context
15:35Ermine: back in freenode days this guy haunted #osdev as well, but now #osdev is on libera
15:57Sid127: I see
15:57Sid127: well, average linux hater spam then, I guess
16:08fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208082396885295144/image.png?ex=65e1fd68&is=65cf8868&hm=81ae51845ed7b620546742449a42ccfccd3caec702f85ae7f21318798c1f819f&
16:08fdobridge_: <rinlovesyou> how did you know
16:08fdobridge_: <mohamexiety> LMFAO
16:08Sid127:sweats
16:08fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208082575302721566/fa68c7260b492e66064842e1d6dab57cb33c7fbe.png?ex=65e1fd93&is=65cf8893&hm=e18bd2360231ad8f9a0419a3e6e87becb6006446942c5aa90fa5b29622f62421&
16:09Sid127:totally doesn't aspire to build a VR/simracing setup someday
16:53fdobridge_: <gfxstrand> I need to rework the optimization a bit and need someone to ack the NIR opcode.
16:53fdobridge_: <gfxstrand> It's not that high priority, though.
16:54fdobridge_: <gfxstrand> It's perf in that one benchmark but I doubt it'll show up anywhere else.
17:01fdobridge_: <!DodoNVK (she) 🇱🇹> I have applied 4 patches on top now
17:03fdobridge_: <!DodoNVK (she) 🇱🇹> Which are MR !27174, !27159, !27599 and !27654
17:16fdobridge_: <gfxstrand> Looks pretty good besides a few nits. I assume you've tested it with a real VR setup? (I don't have one.)
17:18fdobridge_: <!DodoNVK (she) 🇱🇹> I tricked my TV to be treated as a VR display (there's a person with a proper VR setup above)
17:22fdobridge_: <mohamexiety> @rinlovesyou ^
17:32fdobridge_: <gfxstrand> 27599 is on its way in.
17:37fdobridge_: <gfxstrand> Let me build myself some kernel patches and maybe we can do budget and REBAR
17:37fdobridge_: <redsheep> Speaking of perf, I've been trying to bottleneck hunt again and I realized that the witness is not as GPU bound as I thought it has been. From the heaviest spot I could find on the map 720p no msaa was 35 fps and that same spot was 30 fps with 4k 8x msaa
17:38fdobridge_: <redsheep> So those nasty shaders were probably never going to be that useful to address first, it's probably got a CPU bind
17:40fdobridge_: <redsheep> Not sure how I could test whether it's rendering too much geometry or not using LODs given the difficulty getting it to play nice with renderdoc
17:45fdobridge_: <gfxstrand> I'm starting to suspect it has nothing to do with shaders.
17:46fdobridge_: <gfxstrand> We're probably doing something with queries or similar that's causing unneeded stalls. The shaders looked mostly fine.
17:55fdobridge_: <rinlovesyou> i have
17:55fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1171505720349446276/1205237110152368240/nvk_vr.mp4?ex=65e0de07&is=65ce6907&hm=fbf5436bc3ec96e77d042e5f2c7979545e68c88682abea28791fb8f48db94b47&
17:56fdobridge_: <rinlovesyou> works pretty well
17:56fdobridge_: <rinlovesyou> steamvr still crashes and burns, but god knows why. Maybe because it can't do DP audio?
17:56fdobridge_: <rinlovesyou> steamvr still crashes and burns, but god knows why. Maybe because it can't do DP audio? Monado/OpenComposite work well though (edited)
18:03fdobridge_: <!DodoNVK (she) 🇱🇹> I wonder if Joshie works on SteamVR stuff
18:13fdobridge_: <babblebones> As soon as NVK can make realtime guarantees within the time window the compositor sets out for compute shaders under any ongoing GPU circumstances, its 100% set for high end VR, that is insane nvidia has been dragging their heels on just being able to do simple stuff direct mode for 2yrs now
18:18fdobridge_: <gfxstrand> Cool! Just wanted to double check.
18:19fdobridge_: <!DodoNVK (she) 🇱🇹> Can you get some SteamVR logs? 🐸
18:22fdobridge_: <rinlovesyou> This is all the vrcompositor says
18:22fdobridge_: <rinlovesyou> ```
18:22fdobridge_: <rinlovesyou> ASSERT: "unSize < 65536" at /data/src/common/vrcommon/vrrenderer/vulkanrenderer.cpp:5110.
18:22fdobridge_: <rinlovesyou> ```
18:22fdobridge_: <rinlovesyou> the other logs don't seem to say anything of use
18:28fdobridge_: <!DodoNVK (she) 🇱🇹> Can you launch Steam with `SDL_VIDEODRIVER=x11`? Some people suggested that
18:28fdobridge_: <prop_energy_ball> I mean AMD still can't even do that
18:29fdobridge_: <prop_energy_ball> The best we have is CU reseevation on older gens and compute tunneling on RDNA2+
18:29fdobridge_: <prop_energy_ball> Apparently MES on RDNA3+ can do pre-emption but I have yet to see it
18:30fdobridge_: <babblebones> Actual preempt on 3+ is interesting
18:31fdobridge_: <babblebones> The compute tunnel carries enough of the experience to prevent it from becoming a problem even if its not airtight
18:31fdobridge_: <prop_energy_ball> Will not save you if the ring is full
18:31fdobridge_: <babblebones> Yep, I can still see a stutter or two when its full
18:32fdobridge_: <prop_energy_ball> I doubt its actually that
18:32fdobridge_: <prop_energy_ball> The async compute ring is separate
18:35fdobridge_: <babblebones> I've been going about it with the assumption that the rare stutter at especially loading screens where elements are being loaded to GPU is just the CUs loading up with a big dispatch which this does nothing to address
18:36fdobridge_: <babblebones> Have you gotten to try it yourself to see the difference it makes?
18:41fdobridge_: <babblebones> I'm probably mistaken but there appears to be a certain situation where the limited remaining stutters do still occur
18:41fdobridge_: <rinlovesyou> nope, same result
18:45fdobridge_: <prop_energy_ball> Such things are either done on the gfx queue or sdma queue, the ACE queue should be able to cut through that with tunnelling
18:45fdobridge_: <prop_energy_ball> it's much more likely that the amdgpu scheduling shit is broken as usual tho
18:46fdobridge_: <prop_energy_ball> the max hw submissions is like 2 at any time for a given ring, we bump this on SteamOS
18:46fdobridge_: <babblebones> Oh wow
18:46fdobridge_: <rinlovesyou> also i don't think your aur package builds anymore :(
18:46fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208122304593268746/image.png?ex=65e22293&is=65cfad93&hm=961bc9d4b949a51390001dcb427980684fa75fc826872cb757f712edb7699e47&
18:46fdobridge_: <babblebones> Do you happen to have to hw sub bump patch lying around?
18:47fdobridge_: <babblebones> I can try it on my own and see if the remaining blips are affected
18:48fdobridge_: <redsheep> Probably because 27599 finished merging
19:00fdobridge_: <prop_energy_ball> It's just a kernel command line argument
19:00fdobridge_: <prop_energy_ball> `amdgpu.sched_hw_submission=4`
19:00fdobridge_: <prop_energy_ball> That's what we ship on Steam Deck
19:10fdobridge_: <rinlovesyou> no even when i remove memory_placed from the pkgbuild it fails
19:10fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208128366251876392/image.png?ex=65e22838&is=65cfb338&hm=32e66640d5d3aa93575886673eb26452decaac273599cc291537b72cb2a81422&
19:14fdobridge_: <rinlovesyou> no even when i remove memory-placed from the pkgbuild it fails (edited)
19:14fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208128366251876392/image.png?ex=65e22838&is=65cfb338&hm=32e66640d5d3aa93575886673eb26452decaac273599cc291537b72cb2a81422&
19:18fdobridge_: <!DodoNVK (she) 🇱🇹> Faith was quick at merging one of the patches
19:21fdobridge_: <rinlovesyou> but like, i'm commenting out that patch and it's still failing
19:23fdobridge_: <babblebones> I'll give it a try and note any changes I find
19:30fdobridge_: <!DodoNVK (she) 🇱🇹> Try pulling the latest AUR package source
19:31fdobridge_: <!DodoNVK (she) 🇱🇹> I fixed the nits
19:35fdobridge_: <rinlovesyou> i see you've updated vulkan-nouveau-git, but my package manager is refusing to notice atm, also i think you haven't updated lib32-vulkan-nouveau-gi
19:35fdobridge_: <rinlovesyou> i see you've updated vulkan-nouveau-git, but my package manager is refusing to notice atm, also i think you haven't updated lib32-vulkan-nouveau-git (edited)
19:37fdobridge_: <Sid> delete the dir in .cache and tell package manager to install again
19:37fdobridge_: <Sid> what do yo use, yay? pikaur?
19:37fdobridge_: <rinlovesyou> paru
19:38fdobridge_: <rinlovesyou> but yeah yeeting cache seems to have done it
19:45fdobridge_: <rinlovesyou> https://cdn.discordapp.com/attachments/1034184951790305330/1208137022754267207/image.png?ex=65e23048&is=65cfbb48&hm=e9a97d60465b117f02d8554773880ccb8cd52d5db6ae4f2434dbcbe04f855f21&
19:45fdobridge_: <rinlovesyou> just a barrage of issues today :Hehe:
20:02fdobridge_: <!DodoNVK (she) 🇱🇹> This should be fixed too
20:09fdobridge_: <rinlovesyou> yep, looks like it's all good now
20:41fdobridge_: <rinlovesyou> is there any good way to track something like this down?
20:44fdobridge_: <rinlovesyou> one of my games (Just Cause 4) performs horribly even at 480p so we must be stalling somewhere hard
20:47fdobridge_: <!DodoNVK (she) 🇱🇹> Does it at least run at 30 FPS? If so then I guess you can make some PS2-like gameplay content
20:47fdobridge_: <rinlovesyou> *barely*
20:47fdobridge_: <rinlovesyou> many situations go down to 18-20
20:47fdobridge_: <rinlovesyou> smooth 60fps even at 1440p when i look at the sky though :clueless:
20:52fdobridge_: <rinlovesyou> many situations go down to 10-20 (edited)
22:29fdobridge_: <gfxstrand> It feels good to get back to writing code again... "real" VK_EXT_memory_budget and mappable VRAM are good to go as soon as @airlied reviews the mesa MRs and I ack the kernel patches.
22:30fdobridge_: <gfxstrand> Oh, and VK_EXT_map_memory_placed is landed
23:57fdobridge_: <gfxstrand> Looking at Kepler GL linear texturing now
23:57fdobridge_: <gfxstrand> IDK what's going wrong. Tracing through the GL driver, it all looks fine.