Created attachment 122250 [details]
With some movies gpu hangs are quite reproducable while playing with "mpv hwdec=vdpau vo=vdpau" in fullscreen (1920x1200), in the same time "mpv hwdec=vdpau vo=opengl" work fine there.
For me it is quite easy to reproduce this with such encoded video:
ffmpeg -i big_buck_bunny_1080p_surround.avi -vf scale=1024:576 -c:v libx264 -preset medium -x264-params keyint_min=23 -c:a copy output_1024_576.mkv
Mostly it requires less then one minutes of playing in full screen for GPU hangs.
I suppose that problem is rv730 specific, as I can not reproduce it with rv740 (but it PCIe and work with CPU which is mutch faster).
Created attachment 122251 [details]
Created attachment 122252 [details]
Most likely the name of the problem here is AGP.
We already tried a couple of different workarounds but it is quite hard to get fully stable with UVD.
Try to disable AGP with radeon.agpmode=-1 and see if that helps.
No it does not change it, I think it more likely size alignments related issue. If I change monitors video mode to 1366x768 (16/9 same as video 1024/576) it works in fullscreen too. More clearly would be a 1920x1080, but I was unable to set cvt 1920 1080 60 mode (screen corruption).
Ok, well that's a bit strange.
Does "mpv vo=vdpau" works? E.g. VDPAU output, but with software decoding?
Strange, but mpv --vo=vdpau works.
with 1920x1080 it also hangs (hwdec=vdpau vo=vdpau).
A complete shoot in the dark, but please try to disable semaphores by just returning false from uvd_v2_2_semaphore_emit() in drivers/gpu/drm/radeon/uvd_v2_2.c.
Otherwise I'm running out of ideas what that could be.
The only other explanation I have is that the added load of UVD+displaying at a certain resolution is crashing the GFX block. But I have no idea why that should happen.
bool uvd_v2_2_semaphore_emit(struct radeon_device *rdev,
struct radeon_ring *ring,
struct radeon_semaphore *semaphore,
-- GitLab Migration Automatic Message --
This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.
You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/581.