05:09chikuwad[d]: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/nouveau/compiler/nak/sm70_encode.rs?ref_type=heads#L3306-L3312
05:09chikuwad[d]: I imagine for fl16-vec I have to remove the second assert entirely and/or make it check only for the right atomtype, but is there any specific operation I should check for?
05:10chikuwad[d]: also, do shared atomics support 16-bit arithmetic at all?
05:36chikuwad[d]: chikuwad[d]: wait nvm I forgot the spec mentions `atomic add, min, max, and exchange`
06:57chikuwad[d]: okay progress
06:58chikuwad[d]: no more segfaults, just a plain failure on the CTS
06:58chikuwad[d]: but also my head hurts and so I will continue later
08:17mohamexiety[d]: getting there, good stuff :elysiaheart:
08:44chikuwad[d]: now I need to
08:44chikuwad[d]: uh
08:44chikuwad[d]: hm
08:45chikuwad[d]: right, lower the f16v2 into a 32bit atomic and run it that way?
08:45chikuwad[d]: and pack the 2xf16v2 that's lowered from a f16v4 into a 32bit atomic
15:49chikuwad[d]: yay :D
15:51blisto[d]: Go Sid
15:54chikuwad[d]: go where
15:54chikuwad[d]: https://tenor.com/view/cat-stare-blinking-loop-lopped-confused-ashley-gif-8279228928352683036
15:57blisto[d]: To a high paying engineer job
16:47blisto[d]: I be testing a bit of nvk yo
16:47blisto[d]: mmm
16:48chikuwad[d]: seems yummy, based on that reaction
16:49blisto[d]: Having some issues setting up a compilation
16:49blisto[d]: proc-macro2| ERROR: Subproject unicode-ident-1-rs is buildable: NO
16:49blisto[d]: subprojects/proc-macro2-1.0.86/meson.build:54:8: ERROR: Neither a subproject directory nor a unicode-ident-1-rs.wrap file was found.
16:50blisto[d]: https://discord.com/channels/1033216351990456371/1034796636834115584/1398476467855298624
16:50blisto[d]: mhm
16:51blisto[d]: Run-time dependency syn found: NO (tried pkgconfig and cmake)
16:51blisto[d]: Looking for a fallback subproject for the dependency syn
16:51blisto[d]: ERROR: Subproject syn is buildable: NO
16:51blisto[d]: src/compiler/rust/meson.build:125:10: ERROR: Neither a subproject directory nor a syn.wrap file was found.
16:51blisto[d]: now this :p
16:52blisto[d]: oh wait mmm
16:52blisto[d]: no nvm
16:53blisto[d]: Help Sid :BlobhajSadReach:
16:54mhenning[d]: Maybe try a `git clean -xdf` (warning: will delete any loose files) followed by a rebuild
16:55blisto[d]: Think i got it working
16:55blisto[d]: Not sure how lol
16:55mhenning[d]: (meson has some quirks around updating subprojects)
16:55blisto[d]: Checking out a potential issue in a game so wanted to compile older versions to see if it always looked like that
16:56blisto[d]: huzzah it is compiling
16:56chikuwad[d]: wahoo
16:56chikuwad[d]: sorry, gaming
16:57blisto[d]: How dare you
16:58chikuwad[d]: :c
16:58blisto[d]: :O
16:59blisto[d]: Is regression
16:59blisto[d]: Bisect time
16:59blisto[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1420454725937008742/image.png?ex=68d5750d&is=68d4238d&hm=8b36a7ecd3f8231c4b68bbec2bb98b10c13dad19c7ffff0d0d9af3c90647f027&
16:59blisto[d]: Icons being all weird in Endless Legends 2
17:00blisto[d]: At least on current master. 25.0.0 looked fine
17:03blisto[d]: between 25.1.0 and 25.2.0 is looks like
17:06blisto[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1420456410105511956/image.png?ex=68d5769e&is=68d4251e&hm=5dfd39031c096b82d2175a4dc6ac1ed05916a23b0f42434d182c193929013da6&
17:06blisto[d]: Look. It's sid when she does driver dev
17:18blisto[d]: I believe i have found the iconic killer. Just gonna double check to make sure
17:22blisto[d]: <https://gitlab.freedesktop.org/mesa/mesa/-/commit/7edb08668590e5040faf669e1c700a2b51475a8a>
17:23blisto[d]: Will make an issue
17:35blisto[d]: <https://gitlab.freedesktop.org/mesa/mesa/-/issues/13966>
17:42blisto[d]: Another issue in Stalker 2. Poking
17:47blisto[d]: 8932.058:0164:0320:err:seh:handle_syscall_fault Syscall stack overrun.
17:47blisto[d]: backtrace: --- Exception 0xc0000005 at 0x7f8942c717f1: /home/blisto/mesa-testing/nvk/target/lib/libvulkan_nouveau.so + 0x2717f1.
17:47blisto[d]: 8932.058:0164:0320:trace:seh:handle_syscall_fault returning to user mode ip=00006ffffbb8aa2f ret=c0000005
17:47blisto[d]: 8932.058:0164:0320:err:msvcrt:_wassert (L"!status && \"vkCreateComputePipelines\"",L"../src-wine/dlls/winevulkan/loader_thunks.c",3046)
17:47blisto[d]: happens during shader comp
17:48sonicadvance1[d]: Oh no, stacked too aggressively.
17:48blisto[d]: Let's see if it happens with older versions
17:49blisto[d]: (than master)
17:52mhenning[d]: There's a stack overflow fix in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37536 - it could be worth testing that
17:52blisto[d]: Will do!
17:52blisto[d]: happened the same with 25.0.0 at least
17:53snowycoder[d]: What are games doing with 1k+ CFG blocks 0_0
17:55mhenning[d]: ubershaders maybe?
17:56mhenning[d]: worth noting that that's after we inline and loop unroll, so it could possibly be our fault that the shaders are so big
17:58blisto[d]: mhenning[d]: Fixes it 👍
17:59mhenning[d]: okay, cool. That's another reason to get that one in then.
18:00blisto[d]: Now on to next issue heh. Getting a device lost when trying to load in to the actual game
18:00blisto[d]: [ons sep 24 19:59:14 2025] nouveau 0000:01:00.0: gsp: rc engn:00000001 chid:54 gfid:0 level:2 type:109 scope:1 part:233 fault_addr:0000000000000000 fault_type:00000000
18:00blisto[d]: [ons sep 24 19:59:14 2025] nouveau 0000:01:00.0: fifo:c00000:0036:0036:[GameThread[168721]] errored - disabling channel
18:00blisto[d]: [ons sep 24 19:59:14 2025] nouveau 0000:01:00.0: GameThread[168721]: channel 54 killed!
18:04blisto[d]: Does nvk have anything similar to `RADV_DEBUG=hang` to help debug such issues?
18:06mhenning[d]: There's NVK_DEBUG-push_sync which will print the failing command buffer
18:07blisto[d]: Cool
18:07blisto[d]: Or should i wait with making a issue until the MR has been merged?
18:07mhenning[d]: I think you can file an MR. It sounds like a separate issue in this case.
18:09blisto[d]: It's all this that gets printed in the Proton log right?
18:09blisto[d]: 0x00000000] HDR 20018040 subch 4 NINC
18:09blisto[d]: mthd 0100 NVC6B5_NOP
18:09blisto[d]: .PARAMETER = (0x0)
18:09blisto[d]: [0x00000002] HDR 800020a6 subch 1 IMMD
18:09blisto[d]: mthd 0298 NVC7C0_INVALIDATE_SKED_CACHES
18:09blisto[d]: .V = (0x0)
18:09blisto[d]: ...etc
18:11blisto[d]: mhenning[d]: Which renderdoc version do you use? Just so i make the capture for the other issue with the same
18:12mhenning[d]: I currently have 1.39 on linux/vulkan
18:13blisto[d]: Coolio
18:14chikuwad[d]: blisto[d]: https://tenor.com/view/coolio-gangsta%27s-pardise-glasses-gif-11659964042751685900
18:15blisto[d]: Bleh. Seems the game crash when trying to launch with renderdoc
18:28blisto[d]: Mkay made a issue for the stalker 2 hang
18:56blisto[d]: Dune Awakening also gives a GPU hang. But could be the same issue as it is also UE5
20:35superroman: Well i do not see under your hood or head, are you really that talented to say those things to me!? If you tested this formula, you would actually understand as to how powerful it is. I have seen collatz kekeci and quadratic before, but i developed my custom one. Maybe i am a joke but i try to fight back to some type of position and i give warning that what MaryTrampoline and teronimozuck
20:35superroman: documented on the logs is very powerful algorithm on it's own, the compression is unheard in literature which can be authored. Maybe you have not tested yourself i.e your skills on real stuff yet. Why should i be a joke, well i did those calculations for 2years which is pathetic, every day registering what are the outcomes, and found such model, i had long since suspected a relation to
20:35superroman: derivatives first and second degree derivatives in mathematics, i think that weird method relates to this field of science , i suspect it , i was never good in derivative calculations to tell for sure at 42years of age i have more important things to do, i suspect i am now already capable of solving derivatives however.
21:40blisto[d]: O!
21:40blisto[d]: Arch just got renderdoc 1.40 in the repos and now the game doesn't crash anymore
21:42blisto[d]: oh wait. is it because i switched to Nvidias prop driver? Forgot for a second
21:42blisto[d]: Something to poke at tomorrow
22:41mohamexiety[d]: cubanismo[d]: not sure if you can answer this but does tile mode factor in anything at all for programming PTEs on Turing+?
22:42mohamexiety[d]: looking to see if I can have YCbCr + compression or not, and the main deciding factor is this because depending on format, the different planes could have different dimensions which would lead to different tiling modes
22:42mohamexiety[d]: as far as I can tell nothing actually looks like it depends on it at all for Turing+ but I am not certain
22:46mohamexiety[d]: (in case of terminology mismatch, tiling mode is the log2 number of GOBs in a block)