00:00fdobridge: <esdrastarsis> Just another game doing cursed things, nothing new 🐸
00:13fdobridge: <gfxstrand> Damnit! My CTS runs are back up to an hour. I guess that's what I get for adding features. 😂
05:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I already hit that: https://discord.com/channels/1033216351990456371/1034184951790305330/1145387145255714846
05:32fdobridge: <airlied> Does NVIDIA claim to support it then just draw and copy to linear?
06:42fdobridge: <gfxstrand> 🤷🏻♀️ We're going to have to come up with some sort of plan if we're to make modifiers work. Something with a shadow copy. IDK if I want to copy at render pass boundaries or if I want to do image layout nonsense. I hate image layout nonsense. 🤮
06:43fdobridge: <gfxstrand> But image layout nonsense tends to be more efficient, even if it's harder to get right. 🤷🏻♀️
06:43fdobridge: <gfxstrand> The advantage of render pass copies is that I can skip the copy if there's no depth buffer bound in the render pass.
06:44fdobridge: <airlied> Yeah I suppose a tiled to linear on layout switches from color optimal might do it
06:44fdobridge: <gfxstrand> Depending on usage, that might be more efficient.
06:44fdobridge: <gfxstrand> The annoying part is that you have to make `GENERAL` work.
06:45fdobridge: <gfxstrand> So even if we do something fancy with layouts, we probably still have to do render pass load/store copies in `GENERAL`.
06:47fdobridge: <gfxstrand> I'm inclined to say "we have bandwidth" and do load/store copies and then we can optimize with layouts later if we run into perf issues.
07:00fdobridge: <airlied> Oh yeah general is a pain
08:44fdobridge: <marysaka> I just got the nv84_fence_context_new error on Ampere by just running all tests from api.txt after a while (5 min on 8 cores)
16:01fdobridge: <gfxstrand> Yeah...
16:02fdobridge: <gfxstrand> I've got an idea how to figure out a better size. I just need to type up some hacks and run it in my box for a while.
16:57fdobridge: <mhenning> Oh, that's interesting. I got it last night after a few minutes of running CTS, also on ampere. I wonder if a change on the userspace side made it more likely - I don't remember seeing it on my machine before last night.
16:58fdobridge: <mhenning> (that was on a run with 4 threads)
19:13fdobridge: <gfxstrand> Does someone want to file a bug? I can attach some notes to it but I'm not at my computer right now.
19:15fdobridge: <mhenning> Sure, I'll write one up
19:23fdobridge: <mhenning> Actually, now that I'm comparing my stack trace with one that was posted earlier, I'm not certain that the issue I saw last night is related
19:40fdobridge: <mhenning> Okay, yeah I was confused. The trace I got last night is the same as what I saw in https://gitlab.freedesktop.org/drm/nouveau/-/issues/265 which I suspect is different from the channel count thing despite both traces passing through nv84_fence_context_new
19:49fdobridge: <gfxstrand> Oh! That's the other one. Someone isn't handling `ERR_PTR()` somewhere and stuffing it into a data structure. I've seen 2-3 different iterations with different error codes. IDK that I've seen 8 before.
19:49fdobridge: <gfxstrand> `ERR_PTR()` considered harmful...
19:50fdobridge: <gfxstrand> Like, it's brilliant but also horrible. It desperately needs some type checking.
20:09fdobridge: <mhenning> Alright, I copy-pasted some info on the channel id thing so it's a little less scattered: https://gitlab.freedesktop.org/drm/nouveau/-/issues/277
20:57fdobridge: <gfxstrand> /me now has a Pascal
21:39fdobridge: <esdrastarsis> Shadow of Mordor on NVK (this is not my channel): https://video.hardlimit.com/w/j7hqZNAWS16HJpSPUk1Mya
21:41sravn: gfxstrand: Pascal, that was the programming language of my youth. Hmm, 25+ years ago - time flies. Ohh, and the most important GPU to support is Kepler (guess what I have :-) )
22:01karolherbst: pascal.... pain
22:02fdobridge: <gfxstrand> Oh, yeah, I mostly have it for completeness
22:02fdobridge: <gfxstrand> And I'll probably have to submit CTS on it at some point
22:02fdobridge: <gfxstrand> FYI: https://gitlab.freedesktop.org/mesa/mesa/-/issues/10172
22:02fdobridge: <gfxstrand> Feel freeo to provide feedback if you've spent time NAK hacking and been as FRUSTRATED WITH IT SHOUTING AT YOU as I have. 😂
22:03fdobridge: <gfxstrand> Feel free to provide feedback if you've spent time NAK hacking and been as FRUSTRATED WITH IT SHOUTING AT YOU as I have. 😂 (edited)
22:03fdobridge: <gfxstrand> I'll probably do the actual rework some time next week or the one after. It'll be tricky to get right if we want things aligned but I've got some ideas.