00:08fdobridge_: <gfxstrand> And I think this new form is a PLOP3 makes iadd_sat a lot easier.
00:09fdobridge_: <karolherbst🐧🦀> yeah.. looks like it 🙂
00:09fdobridge_: <gfxstrand> I need to think about it more
00:10fdobridge_: <butterflies> what are the odds of getting support for the nvgpu kernel module merged?
00:10fdobridge_: <butterflies> I care about headless only pretty much though
00:10fdobridge_: <karolherbst🐧🦀> with GSP it's kinda pointless, no?
00:11fdobridge_: <karolherbst🐧🦀> I'd rather we spend the time on debugging nouveau than dealing with unstable UAPI
00:14fdobridge_: <butterflies> Tegra
00:14fdobridge_: <butterflies> I meant nvgpu
00:14fdobridge_: <butterflies> not `nvidia`
00:14fdobridge_: <karolherbst🐧🦀> it's not in tree, so how would the UAPI be any more stable?
00:14fdobridge_: <butterflies> they ship cuda compat packages nowadays
00:14fdobridge_: <butterflies> for it
00:15fdobridge_: <butterflies> but yeah
00:15fdobridge_: <butterflies> you don't have to support a lot of UAPIs for Tegras at worst because of how few nvgpu major vers ship
00:17fdobridge_: <gfxstrand> I'd still rather we spend our time fixing nouveau.
00:18fdobridge_: <gfxstrand> I'd still rather we spend our time fixing nouveau.
00:18fdobridge_: <karolherbst🐧🦀> same
00:25fdobridge_: <gfxstrand> Like, if there's really solid use-cases for something and someone wants to write the code, sure. If it's an esoteric case with no code, it's probably not going to happen
00:29airlied: Lyude: is there that many callsites to need coccinelle? I didn't think we made that many calls yet, but yeah makes sense to use return codes instead of failing
00:30fdobridge_: <airlied> I've considered porting to nvidia just to see if there is any speed difference 😛
00:43fdobridge_: <gfxstrand> I've considered porting to Windows just to make a point. 😝
00:43fdobridge_: <karolherbst🐧🦀> 😄
00:43fdobridge_: <karolherbst🐧🦀> looking at nvidias compiler though I think it will take a while until we'd come even close
00:43fdobridge_: <karolherbst🐧🦀> like... there are just soo many of those little things
00:44fdobridge_: <karolherbst🐧🦀> like they use `SHF.HI` instead of `.LO` if they can move the source to the third slot and use `.reuse` on it
00:45fdobridge_: <gfxstrand> Yeah, maybe. They also generate some real garbage sometimes.
00:46fdobridge_: <karolherbst🐧🦀> I'm sure they have reasons for it all
00:46fdobridge_: <gfxstrand> I'm not
00:47fdobridge_: <gfxstrand> Don't get me wrong, they generate damn good code in a lot of cases. But they also have their stupid.
00:47fdobridge_: <gfxstrand> I doubt we'll be beating them any time soon but we can probably get close enough.
00:47fdobridge_: <karolherbst🐧🦀> yeah... maybe
00:48fdobridge_: <karolherbst🐧🦀> just all that instruction scheduling business will be interesting to figure out
00:48fdobridge_: <karolherbst🐧🦀> but we should at least come up with a loop merging pass in nir :ferrisUpsideDown:
00:49fdobridge_: <karolherbst🐧🦀> though that's more relevant for compute
00:49fdobridge_: <karolherbst🐧🦀> (and reducing barrier instruction usage)
01:02fdobridge_: <gfxstrand> Yeah, there's lots to do yet
01:15fdobridge_: <gfxstrand> Though I'm apparently already way faster. :triangle_nvk:
01:15fdobridge_: <gfxstrand>
01:15fdobridge_: <gfxstrand> https://gitlab.freedesktop.org/mesa/mesa/-/issues/10255
01:15fdobridge_: <gfxstrand> (IDK that I actually believe those numbers.)
01:37fdobridge_: <karolherbst🐧🦀> all those peak benchmarks are kinda .... uhm... funky
01:39fdobridge_: <butterflies> oh yeah
01:46fdobridge_: <airlied> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26513 for anyone who knows about transform feedback
06:36fdobridge_: <airlied> @gfxstrand did you mean to assign 26533 to label maker?
06:38fdobridge_: <gfxstrand> Yes
06:41fdobridge_: <gfxstrand> IDK if it's working, though. I've seen some unlabeled MRs stay the way for too long.
06:42fdobridge_: <airlied> does it work if the MR already has labels?
06:57fdobridge_: <airlied> Pass: 8896, Fail: 40, Crash: 14, Warn: 1, Skip: 90, Timeout: 8, Flake: 1, Duration: 16:24, Remaining: 0
06:57fdobridge_: <airlied> zink GL 4.5 CTS run just the main set of tests
06:58fdobridge_: <airlied> https://cdn.discordapp.com/attachments/1034184951790305330/1181852078235189338/message.txt?ex=65829085&is=65701b85&hm=b7f1ba5d096df3209faee8c462760bd3e8864e4cfae3e34f1857510233c68d2a&
07:57fdobridge_: <airlied> Pass: 8896, Fail: 52, Warn: 1, Skip: 90, Timeout: 11, Duration: 16:04, Remaining: 0 - less crashy version
08:10fdobridge_: <airlied> also a bunch of them appear to pass on their own
08:18fdobridge_: <phomes_> is this with 26211?
08:19fdobridge_: <airlied> yes
08:19fdobridge_: <airlied> the xfb and compute tests hit device loss badly
08:23fdobridge_: <airlied> turning off GALLIUM_THREAD in zink also fixes a few, so some work for zmike next year 😛
08:31fdobridge_: <dadschoorse> I thought the mr-label-maker::relabel label was supposed to work in this case, but apparently not
09:41fdobridge_: <airlied> Pass: 8900, Fail: 49, Warn: 1, Skip: 90, Timeout: 10, Duration: 13:17, Remaining: 0 gallium thread off
15:17fdobridge_: <gfxstrand> Yeah, I think it's down.
15:29fdobridge_: <karolherbst🐧🦀> it's fixed now I think
16:42fdobridge_: <!DodoNVK (she) 🇱🇹> Do these tests timeout with GSP?
16:50fdobridge_: <airlied> That is with gsp, some GL CTS tests take longer then the 3 minutes I set, some can take 5-10 mins but some might be bugs
23:23Liver_K: Hey, I've got some issues I haven't been able to fix easily, been a problem since I installed. When I move floating windows around (i3wm), the border trails a *ways* behind the window. I also get very bad tearing when watching videos or any moving pictures on screen. I cannot install any proprietary drivers (EVGA GeForce GTX 570 Superclocked) because they are EOL and will not compile on Debian Bookworm, so I am using Nouveau (1:1.0.17-2) on xorg
23:43karolherbst: Liver_K: it's generally advised to use the modesetting DDX as it has a few workaround in place to get rid of tearing
23:44karolherbst: alternatively, you could try to set "DRI" to "3" in the nouveau DDX via an xorg config
23:44karolherbst: but that's still not perfect
23:44Liver_K: What is DDX?
23:44karolherbst: the Xorg driver
23:45Liver_K: What does it stand for I mean
23:45karolherbst: decide dependent X I think
23:45karolherbst: *device