00:58fdobridge: <butterflies> are you using the GSP crash info framework or does nouveau not have that yet?
00:59fdobridge: <airlied> not even using a firmware that has that
00:59fdobridge: <butterflies> oh, it has been quite useful here (different scenario however) so might be worth evaluating for your use case
01:00fdobridge: <airlied> what sort of info does it dump?
01:02fdobridge: <butterflies> full register state incl pc, CSRs
01:05fdobridge: <butterflies> https://github.com/NVIDIA/open-gpu-kernel-modules/blob/f59818b7517a913d93adf714f1f45ae69906e207/src/nvidia/src/libraries/crashcat/v1/crashcat_report_v1.c#L414
01:08fdobridge: <airlied> that's probably useful when a program running on the GPU crashes, not when GSP dies
01:08fdobridge: <butterflies> as in, meant GSP reg state
01:08fdobridge: <butterflies> incl CRs
01:09fdobridge: <butterflies> incl CSRs (edited)
01:50fdobridge: <gfxstrand> IDK what to make of that. It's clearly doing stuff but also it always sets `ALPHA_TO_COVERAGE = DISABLED` so... 🤷🏻♀️
01:51fdobridge: <gfxstrand> Maybe they're setting it for realz in a macro?
01:51fdobridge: <airlied> yeah it could be hiding in a macro, they also set the TIR use raster samples for alpha to coverage
02:08fdobridge: <gfxstrand> Yeah...
02:31fdobridge: <gfxstrand> I'm not really sure what that does, TBH.
02:32fdobridge: <gfxstrand> Got it!
02:35fdobridge: <airlied> nice
02:35fdobridge: <gfxstrand> Now for depth bias
03:53fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So that's ~50% of the final pass rate, right?
03:55fdobridge: <airlied> what do you mean "final" pass rate?
04:02fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> "We're at roughly 20 to 25% of what I would expect to be a "final" pass rate for uh... a for a competent driver these days" - Faith Ekstrand, 2022 (when looking at the ~193k passed tests in NVK)
04:09fdobridge: <airlied> yeah it's kinda a decision when to stop vs submit conformance for the personal buzz/publicity win
04:10fdobridge: <gfxstrand> IDK what our final rate will be. That estimate was before shader object massively bloated the CTS. 🙃
04:11fdobridge: <gfxstrand> This one still eludes me. I can get the test to pass by setting the depth bias clamp super close to zero but that seems wrong. 🤷🏻♀️
04:11fdobridge: <gfxstrand> @airlied If you wanted to dump that one, I wouldn't mind.
04:12fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> ESO is a EXT extension so it's definitely not required for conformance
04:12fdobridge: <gfxstrand> ```
04:12fdobridge: <gfxstrand> dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_posdEQP-VK.draw.dynamic_rendering.partial_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.complete_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> ```
04:13fdobridge: <gfxstrand> ```
04:13fdobridge: <gfxstrand> dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.partial_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.complete_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:13fdobridge: <gfxstrand> ``` (edited)
04:13fdobridge: <gfxstrand> CTS has coverage for lots of EXTs.
04:14fdobridge: <gfxstrand> And, yes, you're required to pass if you expose them.
04:15fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So we can simply not expose it for now?
04:15fdobridge: <gfxstrand> Yes, and that's what we do
04:15fdobridge: <gfxstrand> It's part of the 89% of tests returning `NotSupported`
04:16fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm not sure what's the point of ESO when GPL already exists (I remember them being somewhat related)
04:16fdobridge: <airlied> @gfxstrand https://people.freedesktop.org/~airlied/scratch/nv-traces/dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos
04:16fdobridge: <gfxstrand> Thanks!
04:16fdobridge: <gfxstrand> Oh, awesome, they're using depth bias factors of 0
04:17fdobridge: <gfxstrand> I wonder why they think that's reasonable
04:22fdobridge: <gfxstrand> I could use zero too but that's not what the test is requesting. 🤡
04:24fdobridge: <gfxstrand> Oh, well. I'll look more in the morning. For now, I'm going to let my computer do another run while I sleep and I'll make an NVK MR with a bunch of my fixes in the morning.
23:46benjaminl: is there a preferred way to handle methods that aren't included in the open-gpu-doc headers?
23:47benjaminl: I ended up just adding it to the nvidia-headers directory in this MR (https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25668/diffs#8f6029b3d9c79cf2844eea0932f1c8b490039924_1836_1836), but imagine that would get painful if anybody wanted to sync with the headers from the open-gpu-doc repo in the future
23:48airlied: yeah I'm not sure what is best there, we can open a request to nvidia to add any regs we feel are missing
23:48benjaminl: looks like people tried that before: https://github.com/NVIDIA/open-gpu-doc/pull/2
23:49benjaminl: and they don't accept PRs to that repo
23:52airlied: I've sent an email, will see if we get anything back
23:52airlied: it might that we need to add mesa specific ones
23:52airlied: like clc997_mesa.h that we run
23:52benjaminl: thanks! hope the email works :)