Re: [RESEND PATCH v1 0/7] Performance improvement of decoder

[Nicolas Dufresne]

Jackson,

Le lundi 14 avril 2025 à 02:04 +0000, jackson.lee a écrit :
> > > Sequence Change test
> > > =====================
> > > gst-launch-1.0 filesrc location=./switch_1080p_720p_240frames.h264 !
> > > h264parse ! v4l2h264dec ! filesink location=./h264_output_420.yuv
> > > Setting pipeline to PAUSED ...
> > > Pipeline is PREROLLING ...
> > > Redistribute latency...
> > > Redistribute latency...
> > > Pipeline is PREROLLED ...
> > > Setting pipeline to PLAYING ...
> > > Redistribute latency...
> > > New clock: GstSystemClock
> > > Got EOS from element "pipeline0".
> > > Execution ended after 0:00:05.772414400 Setting pipeline to NULL ...
> > > Freeing pipeline ...
> > 
> > I did a test of my own here, and did get kernel splat. The warning
> > indicate that the state machine is no longer respected. This needs to be
> > address in v2, we added these check, since the locking is bound to legal
> > use of the state machine.
> 
> 
> Since applying the performance patch, device_run and wave5_vpu_dec_finish_decode is not synchronized any more.
> How about removing this warning message ?

I would like to understand how it makes sense the we've reached STOP
state while the HW is still actively decoding. It does not feel robust
to me, perhaps you want to introduce a new state ? Also, I'm concern
with overall thread-safety now.

regards,

-- 
Nicolas Dufresne
Principal Engineer at Collabora