00:14 fdobridge: <g​fxstrand> Better than nothing! I'm gonna call it there for tonight.
00:14 fdobridge: <g​fxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1157470377589604362/image.png?ex=6518b9ca&is=6517684a&hm=2c3cc6fc682110a7ec0ee83a760477684df9b3841dabfb99007e185c7bf7ea98&
07:23 fdobridge: <g​fxstrand> ```
07:23 fdobridge: <g​fxstrand> Test case 'dEQP-VK.binding_model.shader_access.primary_cmd_buf.sampler_mutable.tess_ctrl.single_descriptor.1d'..
07:23 fdobridge: <g​fxstrand> Pass (Pass)
07:23 fdobridge: <g​fxstrand>
07:23 fdobridge: <g​fxstrand> DONE!
07:23 fdobridge: <g​fxstrand>
07:23 fdobridge: <g​fxstrand> Test run totals:
07:23 fdobridge: <g​fxstrand> Passed: 1/1 (100.0%)
07:23 fdobridge: <g​fxstrand> Failed: 0/1 (0.0%)
07:23 fdobridge: <g​fxstrand> Not supported: 0/1 (0.0%)
07:23 fdobridge: <g​fxstrand> Warnings: 0/1 (0.0%)
07:23 fdobridge: <g​fxstrand> Waived: 0/1 (0.0%)
07:23 fdobridge: <g​fxstrand> ```
07:25 fdobridge: <g​fxstrand> @marysaka The per-vertex I/O patches in there may fix your GS woes.
07:25 fdobridge: <g​fxstrand> @marysaka The per-vertex I/O patches in in my nak/tess branch may fix your GS woes. (edited)
07:37 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> So you're already moving to D3D11-level shaders?
07:39 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> At this rate NAK is only missing RT shader support 🔦
08:19 HdkR: More common RT infrastructure in mesa sounds like a good thing to me. A mere Turnip user :)
15:39 fdobridge: <g​fxstrand> Don't get your hopes up. I'm not even going to think about RT until we have 1.3 conformance with most other things and decent perf.
15:41 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Aren't most missing things compiler-related though?
15:47 fdobridge: <g​fxstrand> For ray-tracing? No, there's a whole BVH format we don't understand, hardware bits for dispatch that aren't in the provided headers as far as I've seen, and a bunch of compiler stuff.
15:48 fdobridge: <g​fxstrand> And unlike the compiler stuff I've been doing which is a pretty straightforward translation of the higher-level shader language to instructions, RT tends to involve a bunch of meta stuff. Time will tell how bad that is on NVIDIA (Intel was a lot easier than AMD) but I know it's going to be more work than just plumbing an instruction or two through.
15:49 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> I mean for Vulkan conformance
15:50 fdobridge: <g​fxstrand> Oh. Yeah. TCS should get us most of the way there because a bunch of the TCS fails are because codegen doesn't know how to do barriers on Turing.
15:50 fdobridge: <g​fxstrand> Daniel is working on `imageAtomicCompSwap()` which should clen up most of the crashes.
15:50 fdobridge: <g​fxstrand> Then it's down to triage of a couple annoying bugs and a few API bits we need to polish to ensure we're advertising all the right things.
15:59 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Hopefully synchronization2 and vulkanMemoryModel aren't too far away (these are both needed by modern DXVK)
16:02 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> I'm advertising these but they will fail CTS (especially memory model on codegen)
16:40 fdobridge: <g​fxstrand> We'll get there. Need to get the core solid first.
19:23 fdobridge: <g​fxstrand> And... my machine is dead again. 🙄
19:24 fdobridge: <m​arysaka> :vReiAgony:
21:36 fdobridge: <g​fxstrand> `Pass: 405402, Fail: 84, Crash: 138, Warn: 4, Skip: 3194687, Timeout: 2, Missing: 280, Flake: 286, Duration: 1:43:28`