On 10/09/14 11:16, Gwenole Beauchesne wrote: > 2014-09-10 10:54 GMT+02:00 Gwenole Beauchesne <[email protected]>: >> Hi, >> >> 2014-09-10 10:38 GMT+02:00 Luca Barbato <[email protected]>: >>> On 31/08/14 21:24, Luca Barbato wrote: >>>> Bitstream-based hw acceleration does not fit well the per-slice model. >>>> Provide a decode() and a flush() callback to override completely the >>>> software decoder. >>>> --- >>>> libavcodec/avcodec.h | 12 ++++++++++++ >>>> libavcodec/utils.c | 9 +++++++-- >>>> 2 files changed, 19 insertions(+), 2 deletions(-) >>>> >>> >>> Are we ok with those two callbacks? >> >> That's ok with me, as this is overall needed improvement. > > On second thought, maybe disable bitstream-based hw acceleration if > CODEC_FLAG2_CHUNKS is enabled?
Might be safer to assume that even if we feed the bitstream serially anyway. > Or do we assume the hwaccel shall be supporting that in any case? Or, > what is the fallback plan if it comes to fail/not support in this > case? My current testcase should support it just fine and I wanted to keep that part simpler. A simple fallback-to-software could be implemented by just closing the hwaccell on failure and feeding the same data to the software decoder, assuming the failure happens at the first packet it would be quite seamless. lu _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
