09:47 Venemo: _ds_: why are you arguing with a spammer?
19:47 haasn: Is AMDVLK off-topic here?
19:49 airlied: if you want answers :-)
19:57 haasn: I don't suppose there were ever any plans of implementing VK_KHR_display in mesa? :)
19:58 emersion: it's supported, no?
19:58 airlied: pretty sure it's been implemented
19:58 haasn: Oh wait nvm, it is supported, and it shows the same bug I have in AMDVLK
19:58 haasn: So probably a bug in the open source stack :)
19:59 haasn: the tl;dr is I get regular flashing pixels, only ever a handful, every couple of seconds, when using VK_KHR_display (via mpv --vulkan-context=displayvk)
19:59 haasn: I tried forcing DSC on and off, respectively no change
20:00 haasn: Pixels are in all colors, very noticeable especially on dark content; seems to happen on every file I tested
20:00 haasn: Except very low res file so possibly to do with GPU load
20:00 haasn: Sound at all familiar to anybody? What could it even possibly be? Random debug flags I could try changing to see if they fix it?
20:01 haasn: Also sometimes after switching back to normal VT, bottom of my screen is replaced by colorful garbage pixels
20:18 haasn: Since a picture is worth a thousand words :) https://snow.haasn.xyz/~nand/photo_2023-05-23_22-12-11.jpg
20:19 sarnold: is that the big dipper?
20:21 airlied: haasn: that is strange, I wonder why direct to display would cause that
20:23 haasn: Seems to be much more frequent under AMDVLK than RADV but still not eliminated under RADV
20:25 haasn: https://snow.haasn.xyz/~nand/pixels1.jpg another instance
20:26 haasn: It's strange that they only appear to affect one color channel at a time
20:26 haasn: But it appears to be a cluster of pixels, each with the same color
20:26 haasn: Odd
20:30 airlied: is there some shader in that display pipeline or anything that isn't used under x11/wayland?
20:32 haasn: It’s the exact same shaders as on x11
20:33 haasn: Though on x11 it presumably doesn’t get directly sent to the display
20:33 haasn: Maybe something to do with the swapchain
20:34 haasn: Is there some layer I could use to dump the swapchain frames pixel data?
20:37 airlied: not that I know off, but those pixels are very random for display artifacts
20:37 airlied: does RADV_DEBUG=zerovram help
20:44 haasn: What's peculiar about these artefacts is how *regularly* they appear in particular with AMDVLK, with RADV seems more random
20:45 haasn: but on AMDVLK I get them as if by clock, whether the video is playing or not
20:45 haasn: So not related to GPU load, just one frame in every N is corrupted
20:46 haasn: Actually even on RADV it seems to happen at regular intervals, just less frequently
20:46 haasn: airlied: RADV_DEBUG=zerovram doesn't help unfortunately
20:48 haasn: Seems to happen about once every 4 seconds (at 120 fps), not sure if that's relevant
20:48 emersion: bad synchronization maybe?
20:49 emersion: does mesa set IN_FENCE_FD, for instance?