13:27 fdobridge: <g​fxstrand> Bindless cbufs are working well enough that I can test them with The Witness now.
13:29 fdobridge: <g​fxstrand> 112 FPS -> 136 FPS (22%) on the 4060 in my laptop.
13:29 fdobridge: <g​fxstrand> And that's without my cbuf0 reworks
13:30 fdobridge: <g​fxstrand> That's just the difference between `ldg.constant` and `ldc`
13:30 fdobridge: <r​inlovesyou> Nice!
13:30 fdobridge: <r​inlovesyou> Can we test that bindless branch now? :3
13:30 fdobridge: <g​fxstrand> That's also a super hacky implementation
13:31 fdobridge: <g​fxstrand> You can test anything at any time. It's just a matter of how frustrating it's likely to be. 😂
13:31 fdobridge: <r​inlovesyou> Haha
13:31 fdobridge: <r​inlovesyou> Will have a lookie at how some of my worse performing games act with it
13:32 fdobridge: <g​fxstrand> Just pushed
13:32 fdobridge: <r​inlovesyou> Yakuza Kiwami 2 will probably be a good benchmark for a while to come as it even makes proprietary hit its limit sometimes
13:33 fdobridge: <r​inlovesyou> I think there's definitely some Linux shenanigans too as it's a 2017 game that almost maxes out my 2070 super lol
13:42 fdobridge: <!​DodoNVK (she) 🇱🇹> Found it: https://gitlab.freedesktop.org/gfxstrand/mesa/-/commits/nvk/bindless-ubo
13:42 fdobridge: <r​inlovesyou> Yee i know about it
13:43 fdobridge: <r​inlovesyou> I did try running it yesterday when i discovered it, but it didn't like running any games
13:43 fdobridge: <!​DodoNVK (she) 🇱🇹> It should run something with today's changes
13:43 fdobridge: <r​inlovesyou> Yep i will check in a bit
13:44 fdobridge: <r​inlovesyou> Still got two games that run pretty badly under nvk so I'm following any developments pretty closely :Hehe:
13:45 fdobridge: <g​fxstrand> It's failing a bunch of CTS tests but they seem to mostly be robustness2 tests and they're not faulting. I'll look into them later.
13:46 fdobridge: <g​fxstrand> My total hack is turning out to be surprisingly robust. 🤔
13:50 fdobridge: <!​DodoNVK (she) 🇱🇹> Do you want to get rid of that intrinsic entirely?
13:55 fdobridge: <r​inlovesyou> Just Cause 4 is still hitting something else,, although in situations where it ran better i can see better average frames
13:57 fdobridge: <g​fxstrand> I want to implement uniform ALU for real. Right now, it's doing some NIR lowering and betting that we don't use too many registers or anything like that.
13:57 fdobridge: <g​fxstrand> I don't think the NIR intrinsics will change.
13:57 fdobridge: <g​fxstrand> It'll just be more robust in the back-end.
13:58 fdobridge: <g​fxstrand> I also need to rework NVK so that it stores cbuf handles directly in the descriptor set rather than full vec4 SSBO addresses.
13:59 fdobridge: <g​fxstrand> Once I've got NAK handling full uniform integer ALU, it's all just a bunch of fiddly work from there.
13:59 fdobridge: <g​fxstrand> This is gonna be like 50 patches, all told... 😅
14:57 fdobridge: <d​jdeath3483> Not sure if you saw the work we did in Anv for uniform image/sampler handles
15:00 fdobridge: <g​fxstrand> No, I haven't been paying attention.
15:00 fdobridge: <d​jdeath3483> resource_intel, maybe there something similar would be useful for other backends
15:01 fdobridge: <d​jdeath3483> resource_intel, maybe something similar would be useful for other backends (edited)
15:02 fdobridge: <d​jdeath3483> Uniform ALUs is a more generic approach
15:03 fdobridge: <g​fxstrand> I added a `r2ur_nv` intrinsic which sounds pretty similar.
15:04 fdobridge: <g​fxstrand> But all that stuff lives in a different register file so I really need to implement a decent amount of ALU if I want it to be robust.
15:15 fdobridge: <d​jdeath3483> We reproduce the handle values in SIMD8 NoMask so no need to find the live channels
15:15 fdobridge: <d​jdeath3483> SIMD8 also save register file
15:16 fdobridge: <d​jdeath3483> The poor man's uniform register file 😁
23:25 killmlana: My gpu stopped being detected after an update, is this a known issue?
23:26 fdobridge: <c​langcat> Does the GPU show in dmesg at all?
23:26 fdobridge: <c​langcat> ```sh
23:26 fdobridge: <c​langcat> sudo dmesg | grep nouveau
23:26 fdobridge: <c​langcat> ```
23:47 killmlana: i just checked and yeah no output. I am also getting a simpledrm error
23:47 killmlana: i dont know if that's related