00:55magic_rb: could someone quickly shoot me an invite to the discord again? i just got to it
00:55magic_rb: nevermind the one by sid still works
01:00fdobridge: <magic_rb.> i should be here now
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> MESA: error: ZINK: vkQueueSubmit failed (VK_ERROR_DEVICE_LOST)
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> again hit that, and
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> [Thu Mar 14 15:31:21 2024] nouveau 0000:01:00.0: java[933141]: job timeout, channel 32 killed!
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> and this, which is probably unrelated
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> [Thu Mar 14 14:00:32 2024] Asynchronous wait on fence 0000:00:02.0:picom[917222]:648c4 timed out (hint:intel_atomic_commit_ready [i915])
01:07fdobridge: <magic_rb.> [Thu Mar 14 14:00:32 2024] Asynchronous wait on fence drm_sched:nouveau_sched:48253c timed out (hint:submit_notify [i915]
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> please ignore the random times, something seems to be very wrong with how dmesg reports time :/
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> MESA: error: ZINK: vkQueueSubmit failed (VK_ERROR_DEVICE_LOST)
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> MESA: error: zink: DEVICE LOST!
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> again hit that, and
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> [Thu Mar 14 15:31:21 2024] nouveau 0000:01:00.0: java[933141]: job timeout, channel 32 killed!
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> and this, which is probably unrelated
01:07fdobridge: <magic_rb.>
01:07fdobridge: <magic_rb.> ```
01:07fdobridge: <magic_rb.> [Thu Mar 14 14:00:32 2024] Asynchronous wait on fence 0000:00:02.0:picom[917222]:648c4 timed out (hint:intel_atomic_commit_ready [i915])
01:07fdobridge: <magic_rb.> [Thu Mar 14 14:00:32 2024] Asynchronous wait on fence drm_sched:nouveau_sched:48253c timed out (hint:submit_notify [i915]
01:07fdobridge: <magic_rb.> ```
01:08fdobridge: <magic_rb.>
01:08fdobridge: <magic_rb.> please ignore the random times, something seems to be very wrong with how dmesg reports time :/ (edited)
01:47fdobridge: <airlied> okay 2080 Ti fix is on the list
03:32fdobridge: <gfxstrand> Run my NVK branch: https://gitlab.freedesktop.org/gfxstrand/linux/-/commits/nvk/?ref_type=heads
03:40fdobridge: <airlied> @gfxstrand would be good if you could drop the 2080 ti fix on there
04:01fdobridge: <gfxstrand> link?
04:02fdobridge: <airlied> https://patchwork.freedesktop.org/patch/582855/
04:03fdobridge: <redsheep> If you have an igpu and you don't need it for display output try disabling it. I had a very similar issue.
04:04fdobridge: <Joshie with Max-Q Design> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28164 Gamescope needs this now btw
04:04fdobridge: <Joshie with Max-Q Design> We should realllllllly do like... common feature enablement for WSI extensions at some point
04:04fdobridge: <gfxstrand> Yeah
04:05fdobridge: <Joshie with Max-Q Design> Will go add the other few drivers I missed I guess also
04:06fdobridge: <gfxstrand> Done
04:43fdobridge: <gfxstrand> @zmike.
04:43fdobridge: <gfxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1217694558121496637/0001-Fix-sparse_texture2-tests-for-GL_TEXTURE_2D_MULTISAM.patch?ex=6604f56e&is=65f2806e&hm=53410b58b7f57fe29768165e84fb6fd626c6d4f68344a0175f4ca77ee7fc7825&
04:43fdobridge: <gfxstrand> That fixes a bunch of sparse_texture2. I'm still fighting with KHR-GL46.sparse_texture2_tests.SparseTexture2Commitment
04:44fdobridge: <gfxstrand> I have no idea where the bug is at this point. Everything I'm seeing looks fine but it still fails. 😭
05:06StackDoubleFlow: hello
05:06StackDoubleFlow: is the discord invite public?
05:08fdobridge: <Sid> no, but here you go: https://discord.gg/vXqMgfUpGe
05:08StackDoubleFlow: Thank you
05:13fdobridge: <gfxstrand> I just realized while getting ready for bed that it may be due to the test doing "real" sampling without binding the miptail.
05:14fdobridge: <gfxstrand> However, I'm going to be a good girl and go to bed and leave debugging it the rest of the way for tomorrow.
07:26fdobridge: <nishii_ko> i assume you've done this via bios? i'm not at my pc currently so i can't check stuff, but i'm not seeing anything for disabling the igpu via anything else
07:27fdobridge: <nishii_ko> well actually, the issue with the ebuild was that it didn't have any nvk support at all yet, so i think i'll probably still have to wait anyways
07:45fdobridge: <valentineburley> Does this work for you guys on NVK? I built your branch and I'm getting segfaults on `dEQP-VK.wsi.*.maintenance1.*.` I think because of the reason I mentioned on https://gitlab.freedesktop.org/mesa/mesa/-/issues/10806
07:46fdobridge: <valentineburley> I also tried flipping them on earlier but maybe I'm missing something
07:47fdobridge: <valentineburley> `dEQP-VK.wsi.*.swapchain.*` also segfaults for me on `Test case 'dEQP-VK.wsi.xlib.swapchain.create.image_swapchain_create_info'.. Segmentation fault (core dumped)`
07:49fdobridge: <Joshie with Max-Q Design> Oh, I only tested Gamescope + VKD3D-Proton stuff on Turnip -- can you comment that on my MR that we need to solve that?
07:49fdobridge: <valentineburley> Sure!
07:49fdobridge: <Joshie with Max-Q Design> ty
11:38fdobridge: <zmike.> ugh I thought I'd fixed all of these
11:38fdobridge: <zmike.> how do they pass on blob?
13:25fdobridge: <gfxstrand> 🤷🏻♀️
13:26fdobridge: <gfxstrand> That shouldn't be too hard. You just have to detect the case in `CreateImage` and call a WSI helper. Look at how ANV does it.
13:27fdobridge: <gfxstrand> And then you have to add a special case to `BindImage`. Again, not hard.
13:31fdobridge: <zmike.> I'll wait a bit in case you find other things that need fixing before submitting this unless you were planning on gerriting yourself
13:31fdobridge: <gfxstrand> I'll Gerrit myself once I'm passing the tests.
13:31fdobridge: <zmike.> ok
13:31fdobridge: <!DodoNVK (she) 🇱🇹> Can common Vulkan runtime enable features/extensions without the driver needing to do that? 🍩
13:31fdobridge: <gfxstrand> These tests feel like nobody's run them.
13:31fdobridge: <zmike.> I think 4.6.2 branch is what I was using for my other sparse fixes
13:32fdobridge: <zmike.> haha
13:32fdobridge: <zmike.> imagine how bad they were before all my fixes
13:32fdobridge: <zmike.> they didn't even print errors
13:32fdobridge: <gfxstrand> No, not at present
14:45fdobridge: <gfxstrand> Damn. They're using TXL with lod=0. That *should* be fine.
14:58fdobridge: <marysaka> uurgh vulkaninfo doesn't query VK_KHR_cooperative_matrix properties :nya_sad:
14:59fdobridge: <marysaka> I think I might need an Ada card at some point
15:01karolherbst: Lyude: sooo.. it seems like I have a laptop with an Nvidia GPU capable of doing HDMI 2.1 and a TV capable of doing so as well... any tldr of things I'd need to do inside nouveau to make it all work? :D
15:03fdobridge: <gfxstrand> @zmike. How do I disable threaded context?
15:06fdobridge:<gfxstrand> inserts some `glFinish()` in the CTS
15:32fdobridge: <zmike.> you probably want `mesa_glthread=0 GALLIUM_THREAD=0`
15:39fdobridge: <gfxstrand> Uh oh... `_mesa_get_fallback_texture`...
15:39fdobridge: <zmike.> :stressheadache:
16:15fdobridge: <gfxstrand> Once again, NVK has no bugs.
16:17karolherbst: Lyude: mhh.. what's this scdc and scrambling business anyway and how does it relate to things like HDMI FRL?
16:21fdobridge: <zmike.> what was the problem this time
16:29fdobridge: <gfxstrand> Trying to use an integer texture format with GL_LINEAR. Mesa (correctly) declares your sampler incomplete and binds the fallback texture instead of the nice fancy sparse one you were trying to sample from.
16:29fdobridge: <gfxstrand> Quality software, that CTS....
16:30fdobridge: <zmike.> 🤕
16:30fdobridge: <gfxstrand> https://cdn.discordapp.com/attachments/1034184951790305330/1217872583119208548/Z.png?ex=66059b3a&is=65f3263a&hm=e51ff55cb902061b4508d61ba13298c5f0ae0ccc20c4b620c64ca49db65e0d5c&
16:31fdobridge: <zmike.> because you wanna do your pal zmike a solid after he's fixed so many other tests for you, obviously
16:34fdobridge: <zmike.> I think all the buffer map test fixes have now landed?
16:35fdobridge: <zmike.> which means this should be the only one outstanding
16:38fdobridge: <gfxstrand> I fixed !28123. I just messed up the case for new style depth textures that return a scalar (vec2 with sparse info)
16:41fdobridge: <zmike.> and it now passes all the sparse test glob?
16:49fdobridge: <gfxstrand> ```
16:49fdobridge: <gfxstrand> Test run totals:
16:49fdobridge: <gfxstrand> Passed: 14/15 (93.3%)
16:49fdobridge: <gfxstrand> Failed: 0/15 (0.0%)
16:49fdobridge: <gfxstrand> Not supported: 1/15 (6.7%)
16:49fdobridge: <gfxstrand> Warnings: 0/15 (0.0%)
16:49fdobridge: <gfxstrand> Waived: 0/15 (0.0%)
16:49fdobridge: <gfxstrand> ```
16:49fdobridge: <gfxstrand> `KHR-GL46.sparse_texture\*`
16:49fdobridge: <gfxstrand> Let me make that `KHR-GL46.sparse\*` and run again just for grins
16:50fdobridge: <zmike.> huge
16:54fdobridge: <gfxstrand> ```
16:54fdobridge: <gfxstrand> Test run totals:
16:54fdobridge: <gfxstrand> Passed: 17/18 (94.4%)
16:54fdobridge: <gfxstrand> Failed: 0/18 (0.0%)
16:54fdobridge: <gfxstrand> Not supported: 1/18 (5.6%)
16:54fdobridge: <gfxstrand> Warnings: 0/18 (0.0%)
16:54fdobridge: <gfxstrand> Waived: 0/18 (0.0%)
16:54fdobridge: <gfxstrand> ```
16:54fdobridge: <redsheep> How close does that get you to full GL conformance?
16:57fdobridge: <redsheep> The fails list on main was still a few hundred lines last night but I know you have quite a few fixes cooking
17:01fdobridge: <zmike.> the fails list has been mostly piglit fails since last week
17:06fdobridge: <redsheep> Are the lines with @ piglit? If I'm reading that right sparse was the last 4.6 conformance blocker?
17:07fdobridge: <zmike.> yes
17:07fdobridge: <zmike.> and then there's some samplemask ones for ES
17:07fdobridge: <redsheep> That's awesome
17:19fdobridge: <gfxstrand> Okay, I need to go get some groceries and probably some lunch. I'll make CTS issues and gerrits once I'm back.
17:25fdobridge: <zmike.> 💪 :nvk: :muscleright:
17:54karolherbst: Lyude: ohh.. looks like there is a NV0073_CTRL_CMD_SPECIFIC_SET_HDMI_FRL_CONFIG thing
17:55karolherbst: no wonder I didn't get it to work...
17:57karolherbst: mhh.. `nvHdmiFrlSetConfig`
17:57karolherbst: guess there is a bit more to it
17:57karolherbst: but at least I know what I'm missing
17:58karolherbst: I'm wondering if we have to enable the FRL mode somewhere
18:16Lyude: karolherbst: yeah you probably know a bit more about it then I do - that ctrl cmd is the most I've really seen of it
18:16karolherbst: heh
18:16karolherbst: it's also the most _i've_ seen so far :D
18:16karolherbst: though there are also caps we have to set...
18:17karolherbst: but anyway.. it doesn't look to complicated overall
18:18karolherbst: mhh NV0073_CTRL_CMD_SPECIFIC_GET_HDMI_GPU_CAPS
18:20karolherbst: "This command is used to perform HDMI FRL link training and enable FRL mode for the specified displayId. The link configuration will be returned after link training success." 🙃
18:20karolherbst: awesome
18:21karolherbst: Lyude: soo.. I think my cable can't do HDMI 2.1 FRL so maybe that's enough to get me to fail link training.....
19:42fdobridge: <gfxstrand> I already fixed over a million Vulkan CTS tests. You're welcome. 😜
19:43fdobridge: <gfxstrand> (Why are there over a million Vulkan CTS tests? Well, that's a better question...)
19:43fdobridge: <gfxstrand> *(Why are there over a million Vulkan CTS tests? Well, that's a better question...)* (edited)
19:44fdobridge: <zmike.> smh you weren't there in the early days of shader object tests
19:44fdobridge: <gfxstrand> No but I've been there in the early days of most of the other extensions.
19:45fdobridge: <gfxstrand> I just had most of a decade of Vulkan spec to catch up on in less than 2 years. No biggie.
19:47fdobridge: <zmike.> then surely you can weigh in on whether GPL CTS times were worse than ESO CTS times
19:48fdobridge: <zmike.> oh no more failures
19:48fdobridge: <zmike.> :fullheadache:
19:53fdobridge: <gfxstrand> They both suck and we should only have done one of them.
19:53fdobridge: <gfxstrand> 🧠
19:54fdobridge: <zmike.> can't wait for KHR Pipeline Chunks
20:06fdobridge: <gfxstrand> @zmike. gerrit 14159 and 14160 if you're interested
20:12fdobridge: <zmike.> you have to mark them active
20:12fdobridge: <zmike.> and add the opengl reviewers
20:17fdobridge: <gfxstrand> Done.
20:17fdobridge: <gfxstrand> I always forget that step. 🙄
20:22fdobridge: <zmike.> gerrit is so intuitive
20:24fdobridge: <zmike.> I think that just leaves the samplemask ES fails and then you can avoid ever looking at glcts again until one of your optimization passes breaks something
20:42fdobridge: <valentineburley> So uh, why are 29 out of 30 of my sanity checks failing in Piglit?
20:42fdobridge: <valentineburley> And how would I only run the NV_compute_shader_derivatives tests?
20:55fdobridge: <gfxstrand> What samplemask ES fails?
21:04fdobridge: <airlied> dEQP-GLES31.functional.shaders.sample_variables.sample_mask_in.bit_count_per_two_samples.multisample_rbo_4,Fail
21:06fdobridge: <zmike.> yeah those
21:07fdobridge: <zmike.> you can do something like `./piglit run gpu -t '.*compute_shader_derivative.*' results`
21:08fdobridge: <gfxstrand> Okay, I was thinking it was the `interpolateAtSample()` thing I already triaged. I can look at that one.
21:08fdobridge: <zmike.> I was hoping it was the same issue, but sadly it isn't
21:14fdobridge: <gfxstrand> Ugh... Why am I getting EGL_NOT_INITIALIZED with the ES CTS but not the GL CTS
21:15fdobridge:<gfxstrand> hates dEQP
21:16fdobridge: <karolherbst🐧🦀> it could be worse
21:20fdobridge: <gfxstrand> probably
21:20fdobridge: <karolherbst🐧🦀> imagine people would have written CTS tests like for CL just insstead of 50 binaries you got 2000 😛
21:21fdobridge: <gfxstrand> I mean, have you looked at the dEQP copy_image test?
21:21fdobridge: <gfxstrand> Yes, I said "test" not "tests"
21:21fdobridge: <karolherbst🐧🦀> is this the one running for ~~hours~~ minutes?
21:23fdobridge: <gfxstrand> Yeah, it's like 10 minutes of copy image in a single test
21:23fdobridge: <dadschoorse> doesn't cl also literally try all inputs for some alu ops?
21:24fdobridge: <karolherbst🐧🦀> it's not that bad
21:25fdobridge: <karolherbst🐧🦀> it's actually configurable
21:36fdobridge: <zmike.> that's one of the ones I'm abusing my 🪑 powers to split
21:39HdkR: Mike coming in to the ring with the folding chair! Oh the humanity!
21:41fdobridge: <zmike.> as god as my witness that test is broken in half!
21:41fdobridge: <karolherbst🐧🦀> wasn't there also some `smoke` test which runs forever?
21:43fdobridge: <gfxstrand> @zmike. You're setting `minSampleShading` in the pipeline but `rasterizationSamples = 0`
21:44fdobridge: <zmike.> 🤔
21:44fdobridge: <zmike.> well at least it's something simple
21:49fdobridge: <gfxstrand> How are `vkCmdSetRasterizationSamples()` and `minSampleShading` supposed to interact? This is GPL+EDS3 world. Who knows!
21:49fdobridge: <zmike.> if I had the spec in front of me I'd know, but unfortunately I'm on the couch
21:50fdobridge: <gfxstrand> I need to think a bit about how this should work on NV. I might be able to make it work.
21:51fdobridge: <zmike.> I'll 🙏
21:57fdobridge: <gfxstrand> Okay, I deleted 3 lines of code and now the test passes. If the CTS is happy with the patch, I'll write a massive comment and push it out.
21:57fdobridge: <gfxstrand> But ANV and others may have heartburn about that, too. I know in ANV I figured out how to solve it with way too much pain. IDK about anyone else.
21:58fdobridge: <illwieckz> patch: 1 line
21:58fdobridge: <illwieckz> comment: 3 books (4th one upcoming)
21:58fdobridge: <gfxstrand> Pretty much
21:59fdobridge: <illwieckz> Also sometime:
21:59fdobridge: <illwieckz> time spent on fixing the bug and understanding all implications required for a proper fix: 10 years
21:59fdobridge: <illwieckz> size of the patch fixing the bug: 1 line
22:00fdobridge: <gfxstrand> That's pretty much true of any bug fix regarding multisample interpolation (which this is). Also, all code touching multisample interpolation is probably wrong.
22:30fdobridge: <gfxstrand> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28192
22:33fdobridge: <zmike.> should this get cts coverage
22:36fdobridge: <gfxstrand> I'm not sure if it needs CTS coverage or a VU but it needs one of the two.
22:36fdobridge: <gfxstrand> I suspect this is a "write the CTS tests and see who blows up" kind of deal.
22:38fdobridge: <zmike.> mobile vendors will all explode
22:38fdobridge: <zmike.> everyone else is fine since nvk was the only driver failing this
22:45fdobridge: <gfxstrand> Want to start with a spec issue or a CTS issue?
22:46fdobridge: <zmike.> probably exploratory cts since it'll lead there anyway
22:48fdobridge: <gfxstrand> Lol. Fair.
22:49fdobridge: <gfxstrand> I don't want to run the risk of the CTS folks debating spec, though.
22:49fdobridge: <gfxstrand> How about I file both?
22:49fdobridge: <zmike.> just file a cts ticket and label it "exploratory"
22:49fdobridge: <gfxstrand> okay
22:49fdobridge: <zmike.> then they won't have to debate
22:50fdobridge: <zmike.> once we get results then we'll have something actionable for a ticket instead of arguing on a WG call for 20 minutes over nothing
22:51fdobridge: <zmike.> though I do enjoy arguing over nothing for extended periods on WG calls... 🤔
22:54fdobridge: <gfxstrand> https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/5019
22:55fdobridge: <gfxstrand> I'm gonna file a spec ticket, too. This feels like something people might have opinions on even though they haven't implemented EDS3 yet
22:58fdobridge: <gfxstrand> https://gitlab.khronos.org/vulkan/vulkan/-/issues/3826
22:59fdobridge:<zmike.> subscribes
23:00fdobridge: <gfxstrand> I've seen `if (sampleShadingEnable && minSampleShading * rasterizationSamples > 1.0)` enough times in Mesa that I'm a little afraid to make assumptions.
23:02fdobridge: <zmike.> will the results be as exciting as exploratory sparse testing?
23:02fdobridge: <zmike.> stay tuned to find out!
23:04fdobridge: <gfxstrand> It's multisample interpolation. That's like the habanero of specs. It does get spicier but not by much.
23:27fdobridge: <gfxstrand> Anyway, CTS is happy so I'm going to Marge it.
23:28fdobridge: <gfxstrand> Looks like some uniform control-flow tests are failing, though. Should have CTS'd that, I guess.
23:32fdobridge: <gfxstrand> Wow, those tests are crazypants.
23:32fdobridge: <gfxstrand> I'm not debugging that today