#asahi-dev · 2026-10-20 09:53
JamesCalligeros[m]has anyone with j293 noticed issues with 96 khz playback#asahi-dev · 2026-10-20 09:53
JamesCalligeros[m]seems to be random#asahi-dev · 2026-10-20 09:53
JamesCalligeros[m]hmmm at 96 khz some v/i samples are spiking at near full scale#asahi-dev · 2026-10-20 09:12
jannauthe main fix is the explicit vm_bo drop in KernelMappingInner's drop(). not sure if the deferred_cleanup() are necessary#asahi-dev · 2026-10-20 09:06
JamesCalligeros[m]still didnt fix it wth #asahi-dev · 2026-10-20 09:05
LarstiQjannau: was it that vm.vm.bo_deferred_cleanup()?#asahi-dev · 2026-10-20 09:04
JamesCalligeros[m]snd_soc_component_update_bits: "Return: 1 if the operation was successful"#asahi-dev · 2026-10-20 09:04
JamesCalligeros[m]"why are we getting EPERM when setting the codec idle props"#asahi-dev · 2026-10-20 08:57
jannaumemleak fix pushed. I'll look now at the import issue#asahi-dev · 2026-10-20 08:44
JamesCalligeros[m]I think i found the stupid visense data thing a few “people” have reported though so im looking at that right now#asahi-dev · 2026-10-20 08:44
JamesCalligeros[m]Perhaps#asahi-dev · 2026-10-20 08:43
sofusmaybe updating to expect multi-plane could also fix the other bug#asahi-dev · 2026-10-20 08:43
JamesCalligeros[m]Nice!#asahi-dev · 2026-10-20 08:42
jannaugpu memleak found and fixed. not yet pushed I want to run OpenGL and Vulkan CTS first#asahi-dev · 2026-10-20 08:42
JamesCalligeros[m]Would require changes to the layout struct and everything that references it either way#asahi-dev · 2026-10-20 08:41
JamesCalligeros[m]Interesting#asahi-dev · 2026-10-20 08:41
JamesCalligeros[m]That would be the best approach because the shader is um#asahi-dev · 2026-10-20 08:40
sofuscant it be called once per plane?#asahi-dev · 2026-10-20 07:34
chadmedidk if that is the only issue but i think it's wrong for planar formats#asahi-dev · 2026-10-20 07:33
chadmedcurrently it only takes one tile data and one metadata pointer#asahi-dev · 2026-10-20 07:33
chadmedsofus: i think we have to rework the libagx decompression and ail layout definiiton to deal with per-plane data and metadata#asahi-dev · 2026-10-19 14:07
sofusbtw, should the gpu export the buffer as interchange?#asahi-dev · 2026-10-19 14:05
sofusfair, i think mpv ignores height/width. The rest should be copied. Maybe the incorrect buffer size could come from a pitch*height (which would be wrong)#asahi-dev · 2026-10-19 14:02
chadmedits way past bedtime now though#asahi-dev · 2026-10-19 14:02
chadmedit may even be a gpu driver thing with yuv to rgb conversion#asahi-dev · 2026-10-19 14:02
chadmedim kinda stumped on the h264 one that looks wrong though. nothing in surface.c has any effect whatsoever#asahi-dev · 2026-10-19 14:02
chadmed(for the case of the hevc video with the incorrect buffer size) #asahi-dev · 2026-10-19 14:02
chadmedi think aligning the dimensions up might be causing issues but im not sure for certain#asahi-dev · 2026-10-19 14:00
sofuswhat do you think is wrong?#asahi-dev · 2026-10-19 13:57
chadmedi think the avd driver is wrong#asahi-dev · 2026-10-19 13:36
JamesCalligeros[m]Which diverges greatly from what one would expect#asahi-dev · 2026-10-19 13:36
JamesCalligeros[m]Interestingly ~nothing in surface.c seems to have any effect other than setting the modifier and the object size#asahi-dev · 2026-10-19 13:05
sofusno sorry i confused myself and looked at hwdec instead of vo#asahi-dev · 2026-10-19 13:02
sofusi think the format could come from ffmpeg#asahi-dev · 2026-10-19 13:01
sofusoh no :(#asahi-dev · 2026-10-19 12:45
sofusafbc is not even used, i have no clue why its there#asahi-dev · 2026-10-19 12:43
chadmedthe comment misled me#asahi-dev · 2026-10-19 12:43
chadmedoh uh nvm yeah this is for the fb descriptor going to drm#asahi-dev · 2026-10-19 12:42
jannauno, the queue leak already happens in 7.1, in 7.2 it results keeping GEM buffers pinned/unevictable. that could be change in the rust bindings or a change in the asahi driver#asahi-dev · 2026-10-19 12:41
sofuschadmed: are you sure? At least not for the dcp driver right?#asahi-dev · 2026-10-19 12:40
chadmedlinkmauve: yes i am aware#asahi-dev · 2026-10-19 12:40
chadmedinterchange is modelled after afbc which means we need to set twoplanes to false#asahi-dev · 2026-10-19 12:40
linkmauveIf you want mpv to present a NV12 buffer, use --vo=dmabuf-wayland or --vo=drm.#asahi-dev · 2026-10-19 12:40
sofussorry, brainfart. fullscreen and 'mpv --video-unscaled' can help#asahi-dev · 2026-10-19 12:40
chadmedanyway there are a couple of things in the vaapi wrapper ive found in surface.c #asahi-dev · 2026-10-19 12:39
chadmedit does not#asahi-dev · 2026-10-19 12:39
chadmedwdym? its clearly not lol, if it was doing direct scanout i would expect drm_info to show that there is an NV12 framebuffer on one of the planes#asahi-dev · 2026-10-19 12:38
sofusit works for me, are you sure its doing direct scanout?#asahi-dev · 2026-10-19 12:37
chadmedi think ive found something sussy though #asahi-dev · 2026-10-19 12:37
chadmedit doesnt, dwl never changes the format from xrgb8888