00:15 fdobridge: <g​fxstrand> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27927
00:16 fdobridge: <g​fxstrand> I think there's also a bit we can set to fix our scratch memory faults but I need to play around a bit with that
00:17 fdobridge: <m​arysaka> amazing 😄
00:19 fdobridge: <m​ohamexiety> really impressive debugging and investigating. nice!
00:24 fdobridge: <g​fxstrand> Next week, I'll see if I can find the magic scratch memory bit and we'll shut those faults off, too.
00:25 fdobridge: <g​fxstrand> Ugh... I'm seeing CTS fails
00:26 fdobridge: <g​fxstrand> But it breaks subgroup operations.
00:34 fdobridge: <g​fxstrand> Why? I have no clue!
01:13 fdobridge: <g​fxstrand> One theory is that we're using FALCON wrong and I'm accidentally stomping some other register I don't intend.
05:53 jcnrot: Well scala-lang generates arrays on the fly as it appears, cause when it really writes into the array, is where bitset gets iterated over division and modulo AND. SO access routine is multiply with wordlength i.e 64 after an average through a length of set. Next thing up is to learn howto make transition values through predicates. That is a larger subject, but i can describe how transition values work, but it appears the language is suitable for
05:53 jcnrot: compression of dictionary kind.
05:55 jcnrot: scala prime time is over, it was used when EU grant was won, but the code left to repos is still good.
08:38 fdobridge: <p​rop_energy_ball> SPH stands for spherical harmonics right? =D
08:38 fdobridge: <p​rop_energy_ball> (every acronym is so overloaded lol)
08:45 fdobridge: <!​DodoNVK (she) 🇱🇹> Shader Program Header
08:47 fdobridge: <!​DodoNVK (she) 🇱🇹> It even has a patent: https://patents.google.com/patent/US8786618
08:53 fdobridge: <m​arysaka> Better send the old documentation instead of the patent: https://nvidia.github.io/open-gpu-doc/Shader-Program-Header/Shader-Program-Header.html
08:53 fdobridge: <m​arysaka> (It's quite outdated but better than nothing)
09:42 fdobridge: <S​id> I forfeit the tournament I was participating in for reasons
09:42 fdobridge: <S​id> guess who's switching to full time nvk now
09:43 fdobridge: <t​om3026> i will when baldurs gate 3 runs :p
09:43 fdobridge: <S​id> Quake Live still doesn't run but
09:43 fdobridge: <S​id> I think that fix should be trivial
09:44 fdobridge: <r​edsheep> I wonder if the new kernel patches help
09:45 fdobridge: <t​om3026> sids or arlieds https://patchwork.freedesktop.org/patch/580696/ , applied both. still dies :p
09:45 fdobridge: <r​edsheep> I haven't had time to test at all this evening. Oh also @tiredchiku I probably won't get anywhere with calibrated timestamps all that soon, I still have git stuff to figure out that are in my way so go for it if you feel like it
09:45 fdobridge: <r​edsheep> Or I'll do it, like... next week
09:46 fdobridge: <S​id> I wonder if this fixes the Xid13 I saw on Sea of Thieves
09:49 fdobridge: <t​om3026> technically it might be worth testing BG3 on x11 , since the main menu is rather ohboy chunking slow even on blob. and nouveau seems to have some extra time when compiling shaders. wouldnt surprise me the least if its just compiling those shaders for to long it almost thinks its deadlocked and i begin missing and recieving wayland events in the right order and hitting protocol errors and death of seppuku in kwin
09:50 fdobridge: <r​edsheep> I have been testing on x11 this whole time, same result.
09:50 fdobridge: <t​om3026> ah okay
09:51 fdobridge: <r​edsheep> Maybe it needs x11 and kernel patches, I am just on 6.7.6 rn, but I kind of doubt it
09:51 fdobridge: <t​om3026> but yeah wayland has its flaws by design almost heh
09:52 fdobridge: <j​oobei> Care to elaborate please for a newbie? 🙂
09:53 fdobridge: <t​om3026> oh well im not an expert per se so take it with a grain of salt and research a bit on your own. but https://www.phoronix.com/news/Wayland-Proxy-Firefox to my mind it seems wayland is just a protoco
09:53 fdobridge: <S​id> yes, wayland is a set of protocols
09:53 fdobridge: <S​id> and a spec
09:53 fdobridge: <t​om3026> l, so clients and compositor <-> sends events back and forward. the protocol states which order and not these should come in
09:54 fdobridge: <t​om3026> if your compositor now chunks in bg3 and esentially your driver is waitin 10s on certain things it cant process events and your client starts sending more, and eventually it just goes in to the rabbit hole of errors
09:55 fdobridge: <S​id> ok nvm quake live launches
09:55 fdobridge: <S​id> my config was wrong
09:55 fdobridge: <t​om3026> so firefox seems to have made up some kind of man in the middle that caches events incase of such holdups and replays when it can
09:55 fdobridge: <S​id> oh
09:56 fdobridge: <S​id> wtf this is on iGPU
09:57 fdobridge: <r​edsheep> Yeah when I had my iGPU on accidentally I was having a heck of a time with it picking integrated without my knowledge
09:57 fdobridge: <r​edsheep> It confidently reported it was opengl 4.3 which I expected meant it was nvc0, but it was running on radeonsi
09:58 fdobridge: <S​id> I can't disable my iGPU :>
09:58 fdobridge: <r​edsheep> I learned my lesson, from here on I am specifying ICD for absolutely everything at all times
09:59 fdobridge: <S​id> how do you specify icd for an opengl game :>
09:59 fdobridge: <r​edsheep> That... I have noticed is not so consistent
09:59 fdobridge: <S​id> I've set the vulkan ICD, and I've set NOUVEAU_USE_ZINK=1 globally
09:59 fdobridge: <t​om3026> isnt it just DRI_PRIME flag?
09:59 fdobridge: <S​id> and I've set DRI_PRIME=pci-0000_01_00_0
09:59 fdobridge: <r​edsheep> Games seem to be able to ignore it somehow
09:59 fdobridge: <S​id> which points to my nv gpu
09:59 fdobridge: <S​id> yet it launches on igpu
10:01 fdobridge: <r​edsheep> You *should* be able to do LD_LIBRARY_PATH="$MESA_INSTALLDIR/lib64" as well as the ICD if you're also using zink, and maybe it would work to point it at a build that just entirely lacks your igpu drivers. Not really sure.
10:02 fdobridge: <r​edsheep> But games seem to be able to just ignore that, I tested like 8 different opengl apps and it was all over the place
10:03 fdobridge: <S​id> zinvk, yeah
10:03 fdobridge: <S​id> I refuse to call it zink + nvk
10:03 fdobridge: <r​edsheep> Maybe this was my issue? I didn't spend too much more time messing with it after I realized I should shut my igpu off
10:04 fdobridge: <r​edsheep> nink?
10:04 fdobridge: <r​edsheep> NVink maybe
10:05 fdobridge: <r​edsheep> If you say that out loud it sounds like you're talking about nvenc though
10:05 fdobridge: <S​id> wait
10:05 fdobridge: <S​id> hm
10:06 fdobridge: <S​id> ok, zinvk seems to be borked atm
10:06 fdobridge: <S​id> `glxinfo: ../mesa/src/nouveau/nil/nil_image.c:791: nil_sparse_standard_block_shape_px: Assertion dim != NIL_IMAGE_DIM_1D' failed.`
10:06 fdobridge: <S​id> wait
10:06 fdobridge: <S​id> sparse
10:06 fdobridge: <S​id> hm
10:08 fdobridge: <S​id> there we go
10:08 fdobridge: <S​id> OpenGL renderer string: zink Vulkan 1.3(TU116 (MESA_NVK))
10:08 fdobridge: <S​id> @mohamexiety sparse MR seems to break zinvk
10:09 fdobridge: <r​edsheep> I also had times where it claimed to be MESA_NVK where the performance plainly showed it was actually iGPU. Again though maybe just needed to set DRI_PRIME
10:09 fdobridge: <r​edsheep> (while using zink)
10:09 fdobridge: <S​id> hm
10:11 fdobridge: <S​id> ok yeah QL doesn't wanna work here
10:14 fdobridge: <S​id> wait wtf
10:14 fdobridge: <S​id> `Unhandled exception: assertion failed in wow64 32-bit code (0xed7e7579).`???
10:15 fdobridge: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1213429471990775808/image.png?ex=65f57142&is=65e2fc42&hm=0531930e90f27c3e69bf26c28bd36fe5eea15665e163254846b7f072aaafa464&
10:15 fdobridge: <S​id> 🤔
10:16 fdobridge: <S​id> ok rebuilt 32-bit nvk without sparse MR and
10:16 fdobridge: <S​id> here we go
10:16 fdobridge: <S​id> https://cdn.discordapp.com/attachments/1034184951790305330/1213429790997094460/image.png?ex=65f5718e&is=65e2fc8e&hm=ee9d0de24d68c8ec12e4fe7ccc3f9abb509abf78ab13435b57a4815337111f89&
10:17 fdobridge: <S​id> let's see perf...
10:20 JoshuaAshton: ninja
10:20 JoshuaAshton: this is not my terminal...
10:22 fdobridge: <S​id> 🥷
10:27 fdobridge: <S​id> Quake Live runs pretty well
10:28 fdobridge: <S​id> same framerate as proprietary driver
10:28 fdobridge: <S​id> 250fps, engine cap
10:39 fdobridge: <S​id> I do have one issue though
10:39 fdobridge: <S​id> with exclusive fullscreen, the game appears to be force vsync'd
10:39 fdobridge: <S​id> this wasn't an issue on nvidia
10:39 fdobridge: <S​id> and my config hasn't changed
10:40 fdobridge: <t​om3026> vblank_mode=0
10:41 fdobridge: <S​id> have that set already
10:41 fdobridge: <m​ohamexiety> Thanks! I suspected this could be problematic later on but wasn't sure how to deal with it. The spec doesn't really have a standard tile for 1D. I guess I could just let it go the normal path..
10:42 fdobridge: <S​id> you're most welcome :>
10:42 fdobridge: <t​om3026> ugh im running out of space i think its time to convert to btrfs
10:42 fdobridge: <t​om3026> 200gb of clones of repos lol
10:43 fdobridge: <t​om3026> suspect that should compress rather well
12:09 fdobridge: <S​id> it DOES fix the xid13's fuck yeah
12:53 fdobridge: <t​om3026> thats interesting, i wonder how many of the past Xid 13 that relates too been a few mentioned in here hm
12:54 fdobridge: <t​om3026> you had one on sea of thieves 01/21
12:54 fdobridge: <t​om3026> i had on cyberpunk
12:54 fdobridge: <t​om3026> 😄 time to commence testing once this btrfs defrag is done
12:56 fdobridge: <S​id> I've always had xid13 on sea of thieves
12:58 fdobridge: <S​id> on nvk
13:12 fdobridge: <S​id> sadly SoT needs more work :\
13:12 fdobridge: <S​id> ```
13:12 fdobridge: <S​id> [Sat Mar 2 18:41:15 2024] nouveau 0000:01:00.0: gsp: mmu fault queued
13:12 fdobridge: <S​id> [Sat Mar 2 18:41:15 2024] nouveau 0000:01:00.0: gsp: rc engn:00000001 chid:48 type:31 scope:1 part:233
13:12 fdobridge: <S​id> [Sat Mar 2 18:41:15 2024] nouveau 0000:01:00.0: fifo:001001:0006:0030:[SoTGame.exe[5167]] errored - disabling channel
13:12 fdobridge: <S​id> [Sat Mar 2 18:41:15 2024] nouveau 0000:01:00.0: SoTGame.exe[5167]: channel 48 killed!
13:12 fdobridge: <S​id> ```
13:16 fdobridge: <t​om3026> one error less tho!
13:17 fdobridge: <t​om3026> 01/29
13:17 fdobridge: <t​om3026> ```
13:17 fdobridge: <t​om3026> [Sun Jan 21 16:40:28 2024] nouveau 0000:01:00.0: SoTGame.exe[5699]: job timeout, channel 24 killed!
13:17 fdobridge: <t​om3026> [Sun Jan 21 16:40:28 2024] nouveau 0000:01:00.0: SoTGame.exe[5699]: error fencing pushbuf: -19
13:17 fdobridge: <t​om3026> [Sun Jan 21 16:40:32 2024] nouveau 0000:01:00.0: gsp: Xid:13 Graphics SM Warp Exception on (GPC 0, TPC 0, SM 0): Out Of Range Address
13:17 fdobridge: <t​om3026> [Sun Jan 21 16:40:32 2024] nouveau 0000:01:00.0: gsp: Xid:13 Graphics SM Global Exception on (GPC 0, TPC 0, SM 0): Multiple Warp Errors
13:17 fdobridge: <t​om3026> [Sun Jan 21 16:40:32 2024] nouveau 0000:01:00.0: gsp: Xid:13 Graphics Exception: ESR 0x504730=0xc00000e 0x504734=0x4 0x504728=0x4c1eb72 0x50472c=0x174
13:17 fdobridge: <t​om3026> ```
13:19 fdobridge: <S​id> yup
13:19 fdobridge: <S​id> no xid is good
13:19 fdobridge: <S​id> it means we're no longer doing something wrong with the driver
13:20 fdobridge: <S​id> and, well, hopefully this is the last missing piece
13:52 fdobridge: <g​fxstrand> Yeah, I think it'll fix a lot of our current crashes. I need to figure out why disabling OOR Address exceptions is breaking subgroups, though. That one really caught me off guard
13:52 fdobridge: <S​id> huh
13:52 fdobridge: <S​id> funny
13:54 fdobridge: <a​irlied> Do you have traces from NVIDIA using that flag?
14:27 fdobridge: <h​untercz122> 373 comments jeez
15:39 fdobridge: <p​avlo_it_115> I found it here (opennet)
15:39 fdobridge: <p​avlo_it_115> This is a troll from russia 🤣
15:39 fdobridge: <p​avlo_it_115> https://cdn.discordapp.com/attachments/1034184951790305330/1213511039769903155/image.png?ex=65f5bd39&is=65e34839&hm=8e5d0f902d6cfae0d84950e29831eebf691d0a2c3c807ed5908a6b8bfc73fddc&
15:41 fdobridge: <p​avlo_it_115> I found it here (opennet https://www.opennet.ru/opennews/art.shtml?num=60699#5)
15:41 fdobridge: <p​avlo_it_115> This is a troll from russia 🤣 (edited)
15:41 fdobridge: <p​avlo_it_115> https://cdn.discordapp.com/attachments/1034184951790305330/1213511039769903155/image.png?ex=65f5bd39&is=65e34839&hm=8e5d0f902d6cfae0d84950e29831eebf691d0a2c3c807ed5908a6b8bfc73fddc&
15:41 fdobridge: <p​avlo_it_115> I found it here (opennet https://www.opennet.ru/opennews/art.shtml?num=60699#5)
15:41 fdobridge: <p​avlo_it_115> This is a troll from russia 🤣 (edited)
15:41 fdobridge: <p​avlo_it_115> https://cdn.discordapp.com/attachments/1034184951790305330/1213511039769903155/image.png?ex=65f5bd39&is=65e34839&hm=8e5d0f902d6cfae0d84950e29831eebf691d0a2c3c807ed5908a6b8bfc73fddc&
15:41 fdobridge: <p​avlo_it_115> I found it here (opennet)
15:41 fdobridge: <p​avlo_it_115> This is a troll from russia 🤣 (edited)
15:41 fdobridge: <p​avlo_it_115> https://cdn.discordapp.com/attachments/1034184951790305330/1213511039769903155/image.png?ex=65f5bd39&is=65e34839&hm=8e5d0f902d6cfae0d84950e29831eebf691d0a2c3c807ed5908a6b8bfc73fddc&
16:13 fdobridge: <p​avlo_it_115> I found it here (opennet)
16:13 fdobridge: <p​avlo_it_115> This is a troll from russia (edited)
16:13 fdobridge: <p​avlo_it_115> https://cdn.discordapp.com/attachments/1034184951790305330/1213511039769903155/image.png?ex=65f5bd39&is=65e34839&hm=8e5d0f902d6cfae0d84950e29831eebf691d0a2c3c807ed5908a6b8bfc73fddc&
16:33 f_: ?
16:51 fdobridge: <p​avlo_it_115> https://discord.com/channels/1033216351990456371/1034184951790305330/1213162959606644777
16:56 fdobridge: <p​avlo_it_115> contact@collabora.cm received a letter from a troll to make an nvk driver for windows xp. Now I found this troll on a russian forum.
17:02 f_: That's doxxing, isn't it?
17:03 fdobridge: <t​om3026> well is it private or sensitive information?
17:53 sin_ack: hello! i am trying to get nouveau set up on my work laptop from hell. it has a "4090 laptop GPU" (nouveau says: NVIDIA AD103 (193000a1)) and a 4k120hz display. when i boot with nouveau, the screen works until nouveaudrmfb picks up the framebuffer, and then the screen becomes pitch black. the proprietary driver "works" but i'd rather use nouveau for several reasons. could you please
17:53 sin_ack: guide me on how to fix this? i'll stick in the channel in case you need some info from me. here's some initial stuff: dmesg https://paste.debian.net/1309255/ vbios.rom: https://files.catbox.moe/fg23zr.rom strap peek: 0x400080
17:56 sin_ack: i'm running linux 6.7.4 right now. CONFIG_FRAMEBUFFER_CONSOLE is enabled, and the nouveau driver shows up on lsmod. no errors that i can see in dmesg. mesa is 24.0.2 but that probably doesn't matter since i can't even get into the tty
17:59 sin_ack: oh i got something interesting when i tried to manually start xorg with startx. https://paste.debian.net/1309258/
18:03 fdobridge: <p​avlo_it_115> Doxxing? Maybe.. I came across it completely by accident
18:04 fdobridge: <p​avlo_it_115> I just read the comments and he exposed himself
18:05 fdobridge: <p​avlo_it_115> No. A fictitious nickname is not private information
18:20 fdobridge: <t​om3026> exactly
19:19 fdobridge: <t​om3026> does this completly disable OOR Address exceptions? because in my mind that feels a bit wrong then. the blob have a few Xid 13 out of range errors thats been going on since 535 up until 550 or im just misinterpreting this
19:25 fdobridge: <g​fxstrand> I think the exception still happens but it doesn't kill the context or something like that.
19:29 Sid127: that sounds about right
19:32 fdobridge: <t​om3026> okay
19:33 fdobridge: <j​oobei> I'm not sure if this is the right channel to ask: I'm trying to build mesa and I'm missing some dependencies. if I "sudo apt install ..." these dependencies, I will just get the latest version available for my system. Is that ok or should I open the meson.build file and try to see what versions are needed and install those?
19:33 fdobridge: <j​oobei> https://cdn.discordapp.com/attachments/1034184951790305330/1213569813322932265/image.png?ex=65f5f3f6&is=65e37ef6&hm=c31d43de34b2f09b539cd5bb8a6c648b780a101d1a8c5f7bfa0a50f40c46c713&
19:34 fdobridge: <j​oobei> or better yet, is there some guide for building?
19:35 tiredchiku: test!
19:35 fdobridge: <S​id> aw
19:35 fdobridge: <S​id> that should be alright
19:36 fdobridge: <S​id> afaik there's no guide for building atm
19:36 fdobridge: <S​id> apart from the usual mesa build instructions
20:53 fdobridge: <t​riang3l> Am I going to hell for this? :happy_gears: :evil_gears: https://gitlab.freedesktop.org/Triang3l/mesa/-/commit/ac1a66aeef11bfc6346d6a186063d0e289b6a9b9
20:53 fdobridge: <t​riang3l> But at least I'm going from the greed circle to the gluttony one 😋
20:55 fdobridge: <t​riang3l> By the way @gfxstrand, I think I found a smol leak in error handling in the shader object code, but I need to verify if that's actually a leak before opening an MR/issue
20:56 fdobridge: <t​riang3l> I fixed it in that dynamic allocation commit
21:01 fdobridge: <t​riang3l> While there's a `for (uint32_t l = 0; l < linked_count; l++) { ralloc_free(infos[l].nir); }` loop on the non-VK_SUCCESS path for linked shaders, for standalone ones, if `ops->compile` fails, it looks like the `break` happens, and `nir` gets leaked, and there doesn't seem to be any cleanup for standalone shaders after that `switch`
21:02 fdobridge: <t​riang3l> and this is the `vkCreateShadersEXT` entry point, so there doesn't seem to be a wider context where it would be deleted alongside other object allocations (aside from the entire process 🙃)
21:03 fdobridge: <t​riang3l> so I guess it actually ends up being leaked in that case
21:05 fdobridge: <t​riang3l> By the way @gfxstrand, I think I found a smol leak in error handling in the shader object code, but I need to verify if that's actually a leak before opening an MR/issue :ferris: (edited)
21:06 fdobridge: <t​riang3l> and this is the `vkCreateShadersEXT` entry point, so there doesn't seem to be a wider context where it would be deleted alongside other allocations for that object (aside from the entire process 🙃) (edited)
21:06 fdobridge: <t​riang3l> and this is the `vkCreateShadersEXT` entry point, so there doesn't seem to be a wider context where it would be deleted alongside other allocations for the relevant object (aside from the entire process 🙃) (edited)
21:07 fdobridge: <t​riang3l> and `ralloc_free(nir);` needs to be done if `ops->compile` fails
21:09 fdobridge: <t​riang3l> I fixed it in that dynamic allocation commit on my Terakan branch (edited)
22:02 fdobridge: <g​fxstrand> That's plausible
22:11 sin_ack: could anyone help me out with what i posted earlier?
22:12 fdobridge: <S​id> ..what did you post earlier
22:15 fdobridge: <S​id> oh the 4090 laptop thing
22:15 fdobridge: <S​id> hm
23:44 sin_ack: oh sorry, i was cleaning my keyboard and didn't notice your message. yes, it's about the 4090. i'm going to bed now, but if you need me to run some commands, just let me know and i'll get the output for you in the morning. i have ssh access to the laptop so i can run stuff with both nouveau and nvidia loaded.
23:45 sin_ack: s/both/either/