Closed dwlsalmeida closed 1 year ago
Merged! Minus the last commit as I could not see how it improves clarity - add_to_ready_queue
is the only method where we add a frame to the queue, the other times we access it is to use extend. The new name was more confusing to me if anything. :)
One more test down, thanks!
@Gnurou it was a bit confusing when I was reading it again, because normal operation is to add to the dpb first until it fills up, then bump images from the DPB into the ready queue. add_to_ready_queue
is only executed on a corner case when the dpb is full and the frame is not a reference frame. It can cause a bit of confusion if one assumes that's the main function being used to fill the ready queue, that's my point.
But anyhow, a very minor commit, that's fine then
Fixes MR3_TANDBERG_B on Intel. Test remains broken on AMD both on GStreamer and cros-codecs.
Compared with JM, saw no difference in terms of the data sent at the affected frames (#283-#299)
ffmpeg reverts to software decoding.
My guess is that the AMD driver probably dislikes the long term reference introduced on frame #283 somehow.
ccdec.zip (on AMD)
Partially addresses #9