On 26/09/2023 10:19, Andreas Rheinhardt wrote:
Mark Thompson:
This is an option to modify the behaviour of the writer, not a syntax
field.
---
Tested by hacking av1_metadata.  For example, adding:

av_opt_set_int(ctx->common.output->priv_data, "fixed_obu_size_length",
7, 0);

gets you OBU headers that look like:

[trace_headers @ 0x55706fcb2880] OBU header
[trace_headers @ 0x55706fcb2880] 0
obu_forbidden_bit                                           0 = 0
[trace_headers @ 0x55706fcb2880] 1
obu_type                                                 0100 = 4
[trace_headers @ 0x55706fcb2880] 5
obu_extension_flag                                          0 = 0
[trace_headers @ 0x55706fcb2880] 6
obu_has_size_field                                          1 = 1
[trace_headers @ 0x55706fcb2880] 7
obu_reserved_1bit                                           0 = 0
[trace_headers @ 0x55706fcb2880] 8           obu_size
10101110100010101000000010000000100000001000000000000000 = 1326

It's not obvious that there is any value in exposing this option more
generally, though?  It could made a visible option of av1_metadata or
others if there is any use-case for it.

Thanks,

- Mark


  libavcodec/cbs_av1.c          | 31 +++++++++++++++++++++----------
  libavcodec/cbs_av1.h          |  5 ++++-
  libavcodec/vaapi_encode_av1.c |  4 +++-
  3 files changed, 28 insertions(+), 12 deletions(-)

What is the advantage of this? Why would people want to waste space?

Because they want to be able to write a different size into the field later 
without realigning all the following bytes.

VAAPI seems to require this, hence the previous code which added a fake syntax 
element field to the OBU structure containing the size to be written with, 
which this patch is replacing with an external option.

Thanks,

- Mark
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".

Reply via email to