02:01 fdobridge: <g​fxstrand> NAK is now a net improvement over codegen for compute shaders.\
02:02 fdobridge: <g​fxstrand> Turns out being able to RA is important. 😅
02:02 fdobridge: <g​fxstrand> And, no, vkcube still doesn't work. 😛
03:49 fdobridge: <a​irlied> Just pull in my changes and it will 🙂
03:51 HdkR: Emulating Pixel Shader Interlock. That would be the nightmare feature
03:52 HdkR: Waiting for Nouveau to implement that feature for those emulators that want it :D
09:08 fdobridge: <k​arolherbst🐧🦀> when somebody from nvidia says "Anyway, thanks for all the great work on NVK." :3
09:09 fdobridge: <k​arolherbst🐧🦀> @gfxstrand ^^
09:47 fdobridge: <b​ylaws> HdkK: ryujinx does it for switch, but just pattern matches rather than emulating things at an instruction level
09:55 HdkR: That's what I would recommend anyway since trying to emulate the details of the implementation would be bonkers
13:33 fdobridge: <g​fxstrand> Oh? Do tell!
13:33 fdobridge: <k​arolherbst🐧🦀> not sure I can disclose that name, because it was in a private email 😄
13:33 fdobridge: <g​fxstrand> Heh
13:34 fdobridge: <k​arolherbst🐧🦀> but anyway, there are people at Nvidia being supportive and everything
13:35 fdobridge: <k​arolherbst🐧🦀> it's probably fine tho...
13:35 fdobridge: <k​arolherbst🐧🦀> but from my experience with Nvidia they are generally very reluctant of doing any public statements and I'd rather not want to put some people there in a weird spot
13:36 fdobridge: <k​arolherbst🐧🦀> hopefully some Nvidia folks will be at XDC tho
13:45 fdobridge: <g​fxstrand> That'd be cool
15:15 fdobridge: <g​fxstrand> `Pass: 404926, Fail: 287, Crash: 80, Skip: 3194648, Timeout: 2, Flake: 440, Duration: 1:33:03`
15:15 fdobridge: <g​fxstrand> That's with `NVK_USE_NAK=cs`
15:15 fdobridge: <g​fxstrand> Notably, all the SSBO tests now pass because NAK, unlike codegen, can register allocate. 🙃
15:40 fdobridge: <k​arolherbst🐧🦀> 😄
15:41 fdobridge: <k​arolherbst🐧🦀> I have this example where codegen fails to allocate 19 gprs...
15:54 fdobridge: <m​henning> some of the codegen RA stuff might be fixed by https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24541
15:55 fdobridge: <m​henning> although that still doesn't fix that we coalesce way too aggressively without any way to undo if it's a problem
15:58 fdobridge: <m​henning> (I haven't figured out how to survive a vulkan cts run locally yet, so I don't actually know how some of my changes affect it)
16:03 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> What's the result with `NVK_USE_NAK=all`?
19:20 fdobridge: <a​irlied> I think with my set of hacks most of fs was going and some of vs, gs and Tess need work
19:57 fdobridge: <g​fxstrand> Bad. We'll get there. Getting the bones good first is important.
19:58 fdobridge: <a​irlied> Have you gotten that possible intern onboard yet?
20:00 fdobridge: <k​arolherbst🐧🦀> wait.. another person? 😮
20:00 fdobridge: <k​arolherbst🐧🦀> or is that Mary?
20:01 fdobridge: <k​arolherbst🐧🦀> probably Mary...
20:16 fdobridge: <m​arysaka> I got my Turing GPU yesterday :nya_peek:
23:05 fdobridge: <g​fxstrand> Y'all are way too worried about 3D. 😂
23:40 fdobridge: <g​fxstrand> @marysaka I left you some notes here: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9777
23:41 fdobridge: <g​fxstrand> I'm not feeling great today. I'll add more later if something pops into my brain.