[FFmpeg-devel] [PATCH]lavc/mpegvideo_enc: Do not write an extra RST tag after final jpeg slice

[Carl Eugen Hoyos]

Am Mo., 20. Jan. 2020 um 22:51 Uhr schrieb Michael Niedermayer
<michael at niedermayer.cc>:
>
> On Mon, Jan 20, 2020 at 12:54:21AM +0100, Carl Eugen Hoyos wrote:
> > Hi!
> >
> > Attached patch fixes ticket #8412.
> >
> > Please comment, Carl Eugen
>
> >  mpegvideo_enc.c |    7 ++++---
> >  1 file changed, 4 insertions(+), 3 deletions(-)
> > 73d8636d0dba4f02cf1697b622c97945d738f06f  0001-lavc-mpegvideo_enc-Do-not-write-an-extra-RST-tag-aft.patch
> > From 7f62ffa46f9264c1fc7854def8467816e0173883 Mon Sep 17 00:00:00 2001
> > From: Carl Eugen Hoyos <ceffmpeg at gmail.com>
> > Date: Mon, 20 Jan 2020 00:51:33 +0100
> > Subject: [PATCH] lavc/mpegvideo_enc: Do not write an extra RST tag after final
> >  jpeg slice.
> >
> > Fixes ticket #8412.
> > ---
> >  libavcodec/mpegvideo_enc.c | 7 ++++---
> >  1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c
> > index 96f5b4a666..fc1c46fee7 100644
> > --- a/libavcodec/mpegvideo_enc.c
> > +++ b/libavcodec/mpegvideo_enc.c
> > @@ -3566,8 +3566,9 @@ static void merge_context_after_me(MpegEncContext *dst, MpegEncContext *src){
> >      MERGE(me.mb_var_sum_temp);
> >  }
> >
> > -static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src){
> > +static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src, int last){
> >      int i;
> > +    int put_bits = put_bits_count(&src->pb) - (src->out_format == FMT_MJPEG && src->avctx->active_thread_type & FF_THREAD_SLICE && last ? 32 : 0);
> >
> >      MERGE(dct_count[0]); //note, the other dct vars are not part of the context
> >      MERGE(dct_count[1]);
> > @@ -3594,7 +3595,7 @@ static void merge_context_after_encode(MpegEncContext *dst, MpegEncContext *src)
> >
> >      av_assert1(put_bits_count(&src->pb) % 8 ==0);
> >      av_assert1(put_bits_count(&dst->pb) % 8 ==0);
> > -    avpriv_copy_bits(&dst->pb, src->pb.buf, put_bits_count(&src->pb));
> > +    avpriv_copy_bits(&dst->pb, src->pb.buf, put_bits);
> >      flush_put_bits(&dst->pb);
> >  }
> >
> > @@ -3919,7 +3920,7 @@ static int encode_picture(MpegEncContext *s, int picture_number)
> >      for(i=1; i<context_count; i++){
> >          if (s->pb.buf_end == s->thread_context[i]->pb.buf)
> >              set_put_bits_buffer_size(&s->pb, FFMIN(s->thread_context[i]->pb.buf_end - s->pb.buf, INT_MAX/8-32));
> > -        merge_context_after_encode(s, s->thread_context[i]);
> > +        merge_context_after_encode(s, s->thread_context[i], i == context_count - 1);
> >      }
> >      emms_c();
> >      return 0;
> > --
> > 2.23.0
>
> this looks a bit unexpected
> is there a reason why the code that writes the RST is not skipped ?

Sorry if I misunderstand but how would one thread know that it is
the last one and no RST should be written?

> if it is doing one too much ?

A final RST is written before EOI, the ticket reporter claims that
it is one too much.

Carl Eugen