00:00fdobridge: <airlied> because it's not written yet
00:00fdobridge: <J., Echo (she) 🇱🇹> Is NAK necessary for NVK to work?
00:00fdobridge: <Esdras Tarsis> no
00:00fdobridge: <Esdras Tarsis> nvk uses codegen rn
00:00fdobridge: <Esdras Tarsis> which will be superseded by NAK iirc
00:02fdobridge: <J., Echo (she) 🇱🇹> Is NAK worse than codegen right now?
00:03fdobridge: <airlied> it's not written yet
00:03fdobridge: <airlied> so by default it would be worse
00:03fdobridge: <airlied> there's still a lot of typing before it can be used
00:08fdobridge: <airlied> though I'd hate comparing anything to codegen as worse 😛
00:15fdobridge: <karolherbst🐧🦀> NAK is perfectly fine code, everybody knows that
01:00fdobridge: <J., Echo (she) 🇱🇹> NVK runs PPSSPP 😈
01:01fdobridge: <karolherbst🐧🦀> noice
01:02fdobridge: <J., Echo (she) 🇱🇹> Though it sometimes loses the Vulkan device in vkQueueSubmit
01:09fdobridge: <J., Echo (she) 🇱🇹> I also managed to run an Unity game (Tonight We Riot) after forcing Vulkan ofc
01:13fdobridge: <J., Echo (she) 🇱🇹> PCSX2 requires vkGetBufferMemoryRequirements2, vkGetImageMemoryRequirements2, vkBindBufferMemory2 and vkBindImageMemory2
01:302:16fdobridge: <J., Echo (she) 🇱🇹> @Esdras Tarsis DXVK v1.5.1 returns `err: DxvkAdapter: Failed to create device` for some reason 🤔
03:33fdobridge: <Esdras Tarsis> enable geometryShaders
08:23fdobridge: <J., Echo (she) 🇱🇹> Why does the geometry shader PR require libnouveau_codegen_tgsi? 🤔
10:14fdobridge: <gouz> @ASDQueerFromEU the geometry shader work needs rebasing
10:15fdobridge: <gouz> But it still nvk works with codegen
10:15fdobridge: <gouz> But still nvk works with codegen (edited)
10:15fdobridge: <gouz> I do not think tgsi is used, only nir
10:41fdobridge: <karolherbst🐧🦀> some TGSI enums are used for the inter stage attributes
11:29fdobridge: <J., Echo (she) 🇱🇹> gouz: Your MR #137 still luckily applies without conflicts on latest nvk/main 🐸
11:55fdobridge: <J., Echo (she) 🇱🇹> @Esdras Tarsis Finally some progress 👀
11:55fdobridge: <J., Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1085894062751621171/Screenshot_20230316_134712.png
11:57fdobridge: <J., Echo (she) 🇱🇹> It crashes after some time though: `kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00800000 [DEVICE] ch 2 [00ffe46000 speed2.exe[149851]] subc 7 mthd 0000 data 0000fff9`
11:59fdobridge: <J., Echo (she) 🇱🇹> It freezes after some time though: `kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00800000 [DEVICE] ch 2 [00ffe46000 speed2.exe[149851]] subc 7 mthd 0000 data 0000fff9` (edited)
12:06fdobridge: <karolherbst🐧🦀> yeah.. guess there is a bug 🙃
12:06fdobridge: <karolherbst🐧🦀> subc7?
12:06fdobridge: <karolherbst🐧🦀> ehh
12:06fdobridge: <karolherbst🐧🦀> guess the command submission buffer gets corrupted
12:07fdobridge: <karolherbst🐧🦀> @ASDQueerFromEU do you run with assertions enabled or disabled?
12:08fdobridge: <karolherbst🐧🦀> though maybe we should make the code be available also in release builds...
12:08fdobridge: <J., Echo (she) 🇱🇹> I have b_ndebug enabled 🐸
12:09fdobridge: <karolherbst🐧🦀> yeah.. you might want to disable it 🙂
12:09fdobridge: <karolherbst🐧🦀> (but we should also make the validation be a runtime thing rather than compile time)
12:12fdobridge: <J., Echo (she) 🇱🇹> There's no difference when disabling it
12:13fdobridge: <J., Echo (she) 🇱🇹> Maybe NVK doesn't use LINEAR format modifier? I had an issue with this in OpenGL nouveau 🐸
12:17fdobridge: <J., Echo (she) 🇱🇹> Also vkcube-wayland segfaults (maybe that's because of the same modifier stuff?)
12:18fdobridge: <J., Echo (she) 🇱🇹> `kernel: nouveau 0000:01:00.0: fifo: PBDMA0: 00800000 [DEVICE] ch 2 [00ffe46000 vkcube-wayland[162716]] subc 7 mthd 0000 data 0000fff9` 🤔
12:22fdobridge: <karolherbst🐧🦀> mhh
12:22fdobridge: <karolherbst🐧🦀> it still smells like a corrupted command buffer
12:22fdobridge: <karolherbst🐧🦀> or maybe we submit total garbage and it lucks out
12:23fdobridge: <karolherbst🐧🦀> `NVK_DEBUG=push_dump`
12:23fdobridge: <karolherbst🐧🦀> and uhm.. try to find that part
12:23fdobridge: <karolherbst🐧🦀> ohh wait...
12:23fdobridge: <karolherbst🐧🦀> ahh, nvm
12:24fdobridge: <karolherbst🐧🦀> we bind 906e tu 7
12:24fdobridge: <karolherbst🐧🦀> which is... nothing
12:25fdobridge: <karolherbst🐧🦀> ehhhhhhhhhhh
12:25fdobridge: <karolherbst🐧🦀> I'm silly
12:25fdobridge: <karolherbst🐧🦀> that's the sw channel hack
12:25fdobridge: <karolherbst🐧🦀> I really should clean that up this week
12:31fdobridge: <karolherbst🐧🦀> @gfxstrand mind we drop the SW channel hack and just move to the new kernel patch?
12:32fdobridge: <J., Echo (she) 🇱🇹> Does that mean a custom kernel will be required?
12:34fdobridge: <karolherbst🐧🦀> yes
12:47fdobridge: <karolherbst🐧🦀> @ASDQueerFromEU I think that error on its own is harmless though
12:47fdobridge: <karolherbst🐧🦀> also, might want to apply this: https://gitlab.freedesktop.org/drm/nouveau/-/commit/99401da29004b777d6999bf78206f989423e3985
13:02fdobridge: <J., Echo (she) 🇱🇹> That's a kernel modification, so unless I can build a nouveau DKMS module I'll probably wait for GSP improvements
14:59fdobridge: <gfxstrand> Yeah, I just need to build a new kernel and test it all.
15:54fdobridge: <J., Echo (she) 🇱🇹> That patch didn't make the NFS Most Wanted look any better (still very corrupted) and I still get the 2 nouveau errors
15:55fdobridge: <J., Echo (she) 🇱🇹> However I know that NFS Most Wanted froze because of a vkUpdateDescriptorSetWithTemplateKHR failure 🐸
16:16fdobridge: <J., Echo (she) 🇱🇹> NFS Underground 1 and Most Wanted both encounter the same failure (so Underground 2 should be the same)
16:30fdobridge: <Esdras Tarsis> Nice
16:30fdobridge: <J., Echo (she) 🇱🇹> I got this with Underground 1 before it froze
16:30fdobridge: <J., Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1085963402972762192/Screenshot_20230316_181518.png
16:47fdobridge: <karolherbst🐧🦀> mhh
16:48fdobridge: <karolherbst🐧🦀> but got any other errors now?
17:12fdobridge: <J., Echo (she) 🇱🇹> Nothing new I think (but the game loads a bit further if you're quick)
17:17fdobridge: <J., Echo (she) 🇱🇹> Here are all the errors I got: https://pastebin.com/6J0Fxfpk
17:17fdobridge: <J., Echo (she) 🇱🇹> Are there any other debug tools I can use? Maybe the validation layers can help?
17:29fdobridge: <karolherbst🐧🦀> mhh, hard to tell. The only proper debugging tool we have atm is the semantically validation of the submitted commands, but they can still be wrong