00:23fdobridge: <airlied> should try the patches from my unrestricted depth mr on top to see if it fixes that
03:24fdobridge: <gfxstrand> `Pass: 405485, Fail: 66, Crash: 66, Warn: 4, Skip: 3194910, Timeout: 2, Flake: 350, Duration: 1:46:42`
03:24fdobridge: <gfxstrand> `Pass: 405250, Fail: 70, Crash: 76, Skip: 3195085, Timeout: 2, Flake: 400, Duration: 1:47:14` (edited)
04:20fdobridge: <gfxstrand> Depth clipping is such a giant PITA. We're going to need a macro pre-Volta. 🙄
04:21fdobridge: <gfxstrand> Also, points. Points suck.
04:22fdobridge: <gfxstrand> Right now I don't know how to disable depth clip for points. Like, at all.
04:39HdkR: I'm surprised it isn't just in the program header
04:41HdkR: There's so much random things shoved in there :D
04:49fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Fail == Crash 🤔
05:30fdobridge: <gfxstrand> The good news is that if I'm right about how clipping works, the macro is pretty easy.
05:30fdobridge: <gfxstrand> Just emulate the new feature on the old HW
05:37fdobridge: <gfxstrand> hrm... The annoying bit is that there are 8 viewports.
06:39fdobridge: <airlied> ah is this to replace the bit they added in the volta header? I have a patch to use that in my unrestricted depth
14:58fdobridge: <gfxstrand> Pass: 405328, Fail: 56, Crash: 65, Warn: 4, Skip: 3194680, Timeout: 2, Missing: 297, Flake: 451, Duration: 1:48:00
14:58fdobridge: <gfxstrand> `Pass: 405328, Fail: 56, Crash: 65, Warn: 4, Skip: 3194680, Timeout: 2, Missing: 297, Flake: 451, Duration: 1:48:00` (edited)
14:59fdobridge: <gfxstrand> Still got a few pesky depth clipping fails:
14:59fdobridge: <gfxstrand> ```
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.complete_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltaone_bias_clamp_neg,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.complete_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.complete_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltazero,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.partial_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltaone_bias_clamp_neg,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.partial_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.partial_secondary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltazero,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltaone_bias_clamp_neg,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.inverted_depth_ranges.nodepthclamp_deltazero,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltaone_bias_clamp_neg,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltasmall_bias_clamp_pos,Fail
14:59fdobridge: <gfxstrand> dEQP-VK.draw.renderpass.inverted_depth_ranges.nodepthclamp_deltazero,Fail
14:59fdobridge: <gfxstrand> ```
15:42fdobridge: <gfxstrand> I have a fix for 8 of those. I just don't know why it's needed.
15:43fdobridge: <gfxstrand> `bias_clamp_neg` eludes me, though.
15:59fdobridge: <gfxstrand> @airlied Feel free to ack the crap out of https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25357
15:59fdobridge: <gfxstrand> Or actually review. Review is good too
16:10fdobridge: <gfxstrand> I should probably think about switching to GSP, shouldn't i?
16:13fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> What does Missing mean?
16:14fdobridge: <gfxstrand> 🤷🏻♀️
16:14fdobridge: <gfxstrand> Typically it means your CTS source and build are out-of-sync but I know mine aren't
16:14fdobridge: <gfxstrand> Generally, it means that the CTS runner went "whoops! Where'd that test go?"
16:15fdobridge: <gfxstrand> If can also happen with crashes, I think.
16:15fdobridge: <gfxstrand> IDK why I suddenly have missing tests
16:15fdobridge: <gfxstrand> I guess I did re-build my debug build but not release. 🤔
16:18fdobridge: <gfxstrand> Unfortunately, for as good as those numbers look, the path to conformance still involves things like "Use NAK for tessellation shaders and figure out how barriers work" which aren't exactly 5m fixes. 😩
16:19fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So you're only using it for CS and FS, right?
16:19fdobridge: <gfxstrand> Yeah
16:20fdobridge: <gfxstrand> FS is basically done, modulo figuring out how to structure the SPH stuff. I just haven't had the right kind of spoons for that this week.
16:20fdobridge: <gfxstrand> @marysaka is trying to figure out GS right now.
16:20fdobridge: <gfxstrand> I think VS should mostly work once SPH gets sorted.
16:21fdobridge: <gfxstrand> VS is easy
16:22fdobridge: <marysaka> I think for VS there was stuffs for some intrinsic missing in between my kernel crashes but yeah it shouldn't be too hard hopefully
16:22fdobridge: <gfxstrand> (VS does need a patch from @marysaka to fix ALD encoding. I really should pull that patch in.)
16:22fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Hopefully NAK will be able to run D3D9-era games (or at least vkcube) fairly soon :ferris:
16:26fdobridge: <gfxstrand> I've got a run going right now for my latest depth fix. Once that's done, I should pull in Mary's ALD fix and give VS a try just for grins
18:07fdobridge: <gfxstrand> It's not looking great. 😂
18:08fdobridge: <gfxstrand> And by "not great" I mean only a 75-80% pass rate in the first few minutes.
18:08fdobridge: <gfxstrand> I'll let it run until my machine falls over and we'll see what's what
18:33fdobridge: <marysaka> For my patch? :blobcatnotlikethis:
19:20fdobridge: <gfxstrand> Your patch is fine, I think.
19:21fdobridge: <gfxstrand> IDK why stuff is falling.
19:21fdobridge: <gfxstrand> I've not looked at any of them yet
19:40fdobridge: <airlied> Don't go gsp yet for CTS runs
19:40fdobridge: <airlied> There is a fw crash taking out everything that occurs during CTS runs
20:56fdobridge: <gfxstrand> `Pass: 261145, Fail: 82383, Crash: 6408` with 30 min to go. Not horrible given that I've never run a VS since the very initial bring up with dEQP-VK.api.triangle.
21:00fdobridge: <gfxstrand> I'm going to guess 90% of those fails/crashes are probably like 2 bugs.
21:00fdobridge: <gfxstrand> But meh. Not gonna work on it today.
21:46fdobridge: <gfxstrand> Current VS baseline: `Pass: 302258, Fail: 95285, Crash: 7385, Warn: 4, Skip: 3194910, Timeout: 2, Flake: 1039, Duration: 3:15:54`