10:27 jglrxavpok: Hello, I was looking at Mesa/RADV's capabilities to output source locations after GPU hangs, and I saw that radv_dump_annotated_shader does not seem to use the generated debug info. I made a small (and naive) modification locally to use the debug info, the location of the hang does not exactly match with the line responsible for my hang, but is close enough, probably due to optimisations. Therefore I was wondering: is
10:27 jglrxavpok: there a reason not to use debug info when dumping the annotated shader, or is it just that nobody did it until now?
10:52 jannau: is there a reason why KMS drm devices do not have SYNCOBJ / SYNCOBJ_TIMELINE?
10:53 jannau: screen casting with explicit sync is currently broken on kmsro systems in kwin and mutter
10:56 jannau: debugged so far only in kwin but I'd surprised if if the issue in mutter has a different root cause
11:06 jannau: adding "| DRIVER_SYNCOBJ | DRIVER_SYNCOBJ_TIMELINE" to drm_driver.driver_features fixes screen casting with obs 31.1 in mutter and kwin
11:07 jannau: gnome issue is bad since it brings the whole session down
12:22 MrCooper: jannau: is there a mutter gitlab issue about that?
12:23 jannau: MrCooper: not yet, writing it atm
12:23 MrCooper: thanks
12:23 jannau: fix for kwin is in https://invent.kde.org/plasma/kwin/-/merge_requests/7941
12:42 jannau: MrCooper: https://gitlab.gnome.org/GNOME/mutter/-/issues/4224
12:47 MrCooper: thanks again
22:27 anonymix007[m]: Hello everyone! I have two DMA buffers: one allocated externally by a different process via EGL (src) and one allocated in my app using VA-API (dst, which later will be used for hw video encoding). What is the best (fastest and/or easiest) way to copy from one to another without copying to CPU given that sometimes the src buffer has an image upside down which needs to be flipped and also their format modifiers are different?