Open ids1024 opened 1 month ago
I'm also seeing this with the no-subsurfaces
feature. Not sure what made this start happening. That used to work.
https://github.com/iced-rs/iced/pull/2425 sounds related.
I noticed performance becomes really poor when wallpapers are set separately for each display.
Yeah, I think that's the same issue. When I saw this, one output had the right background. I guess I hadn't had external monitors connected when I last ran cosmic-bg
at the time.
Probably if we apply https://github.com/iced-rs/iced/pull/2425 and resolve conflicts (or update our Iced) it will fixed it.
If the same image is used for all monitors, it's fine. But otherwise, it's being evicted from the cache when the next monitor is rendered. So it has to uploaded to GPU every frame.
The fact I saw this running out of GPU memory, though, suggests wgpu or iced_wgpu is also leaking buffers somewhere...
I've noticed a weird slowness and crash, and it seems the issue was I didn't have cosmic-bg running...
Two issues here:
cosmic-bg
to be running to have~/.local/state/cosmic/com.system76.CosmicBackground
in a correct state. Not necessarily a problem to require cosmic-bg, but it should probably be handled more gracefully?iced_wgpu
and/orwgpu
issue, like https://github.com/pop-os/cosmic-workspaces-epoch/issues/40. But it's a weird thing to see...gdb
shows it's allocating wgpu buffers to upload image to the atlas. Not sure why caching isn't working, and only isn't working in this case?wgpu
issue