[FFmpeg-devel] [mov] Fix trampling of ctts during seeks when sidx support is enabled.
Dale Curtis
dalecurtis at chromium.org
Thu Jul 20 05:30:01 EEST 2017
Thanks will take a look. Is this test not part of fate? make fate passed
for me. The attached patch fixes this; the issue was that the index entries
are 1 to 1 with ctts values. When samples are added without ctts entries
we'd just initialize a single ctts entry with a count of 5. This left a gap
in the ctts table; the fix is to use only 1-count entries when this case is
hit.
Note: This made me realize the presence of a ctts box and a trun box with
ctts samples has always been broken. If the ctts box comes second it'll
wipe the trun's generated table, but if the trun box comes after the ctts
box it will try to insert samples at incorrect positions. Prior to my patch
they would be looked up at incorrect positions, so there shouldn't be any
new bad behavior here.
- dale
On Wed, Jul 19, 2017 at 3:28 PM, Michael Niedermayer <michael at niedermayer.cc
> wrote:
> On Tue, Jul 18, 2017 at 11:49:26AM -0700, Dale Curtis wrote:
> > Updated patch that fixes other ctts modification code to use the new
> > ctts_allocated_size value; I've verified this passes fate.
> >
> > - dale
> >
> > On Tue, Jul 18, 2017 at 9:53 AM, Dale Curtis <dalecurtis at chromium.org>
> > wrote:
> >
> > > Resending as it's own message per contribution rules. I've also
> attached
> > > the patch in case the formatting gets trashed.
> > >
> > > When sidx box support is enabled, the code will skip reading all
> > > trun boxes (each containing ctts entries for samples inthat box).
> > >
> > > If seeks are attempted before all ctts values are known, the old
> > > code would dump ctts entries into the wrong location. These are
> > > then used to compute pts values which leads to out of order and
> > > incorrectly timestamped packets.
> > >
> > > This patch fixes ctts processing by always using the index returned
> > > by av_add_index_entry() as the ctts_data index. When the index gains
> > > new entries old values are reshuffled as appropriate.
> > >
> > > This approach makes sense since the mov demuxer is already relying
> > > on the mapping of AVIndex entries to samples for correct demuxing.
> > >
> > > Notes for future improvement:
> > > Probably there are other boxes (stts, stsc, etc) that are impacted
> > > by this issue... this patch only attempts to fix ctts since it
> > > completely breaks packet timestamping.
> > >
> > > This patch continues using an array for the ctts data, which is not
> > > the most ideal given the rearrangement that needs to happen (via
> > > memmove as new entries are read in). Ideally AVIndex and the ctts
> > > data would be set-type structures so addition is always worst case
> > > O(lg(n)) instead of the O(n^2) that exists now; this slowdown is
> > > noticeable during seeks.
> > >
> > > Additionally since ctts samples from trun boxes always have a count
> > > of 1, there's probably an opportunity to avoid the post-seek fixup
> > > that iterates O(n) over all samples with an O(1) when no non-1 count
> > > samples are present.
> > >
> > > Signed-off-by: Dale Curtis <dalecurtis at chromium.org>
> > > ---
> > > libavformat/isom.h | 1 +
> > > libavformat/mov.c | 46 ++++++++++++++++++++++++++++++
> +---------------
> > > 2 files changed, 32 insertions(+), 15 deletions(-)
> > >
> > > diff --git a/libavformat/isom.h b/libavformat/isom.h
> > > index ff009b0896..fdd98c28f5 100644
> > > --- a/libavformat/isom.h
> > > +++ b/libavformat/isom.h
> > > @@ -137,6 +137,7 @@ typedef struct MOVStreamContext {
> > > unsigned int stts_count;
> > > MOVStts *stts_data;
> > > unsigned int ctts_count;
> > > + unsigned int ctts_allocated_size;
> > > MOVStts *ctts_data;
> > > unsigned int stsc_count;
> > > MOVStsc *stsc_data;
> > > diff --git a/libavformat/mov.c b/libavformat/mov.c
> > > index 63f84be782..412290b435 100644
> > > --- a/libavformat/mov.c
> > > +++ b/libavformat/mov.c
> > > @@ -4297,11 +4297,6 @@ static int mov_read_trun(MOVContext *c,
> AVIOContext
> > > *pb, MOVAtom atom)
> > > }
> > > if ((uint64_t)entries+sc->ctts_count >=
> > > UINT_MAX/sizeof(*sc->ctts_data))
> > > return AVERROR_INVALIDDATA;
> > > - if ((err = av_reallocp_array(&sc->ctts_data, entries +
> > > sc->ctts_count,
> > > - sizeof(*sc->ctts_data))) < 0) {
> > > - sc->ctts_count = 0;
> > > - return err;
> > > - }
> > > if (flags & MOV_TRUN_DATA_OFFSET) data_offset =
> > > avio_rb32(pb);
> > > if (flags & MOV_TRUN_FIRST_SAMPLE_FLAGS) first_sample_flags =
> > > avio_rb32(pb);
> > > dts = sc->track_end - sc->time_offset;
> > > @@ -4312,26 +4307,28 @@ static int mov_read_trun(MOVContext *c,
> > > AVIOContext *pb, MOVAtom atom)
> > > unsigned sample_size = frag->size;
> > > int sample_flags = i ? frag->flags : first_sample_flags;
> > > unsigned sample_duration = frag->duration;
> > > + unsigned ctts_duration = 0;
> > > int keyframe = 0;
> > > + int ctts_index = 0;
> > > + int old_nb_index_entries = st->nb_index_entries;
> > >
> > > if (flags & MOV_TRUN_SAMPLE_DURATION) sample_duration =
> > > avio_rb32(pb);
> > > if (flags & MOV_TRUN_SAMPLE_SIZE) sample_size =
> > > avio_rb32(pb);
> > > if (flags & MOV_TRUN_SAMPLE_FLAGS) sample_flags =
> > > avio_rb32(pb);
> > > - sc->ctts_data[sc->ctts_count].count = 1;
> > > - sc->ctts_data[sc->ctts_count].duration = (flags &
> > > MOV_TRUN_SAMPLE_CTS) ?
> > > - avio_rb32(pb) : 0;
> > > - mov_update_dts_shift(sc, sc->ctts_data[sc->ctts_count].
> duration);
> > > + if (flags & MOV_TRUN_SAMPLE_CTS) ctts_duration =
> > > avio_rb32(pb);
> > > +
> > > + mov_update_dts_shift(sc, ctts_duration);
> > > if (frag->time != AV_NOPTS_VALUE) {
> > > if (c->use_mfra_for == FF_MOV_FLAG_MFRA_PTS) {
> > > int64_t pts = frag->time;
> > > av_log(c->fc, AV_LOG_DEBUG, "found frag time %"PRId64
> > > " sc->dts_shift %d ctts.duration %d"
> > > " sc->time_offset %"PRId64" flags &
> > > MOV_TRUN_SAMPLE_CTS %d\n", pts,
> > > - sc->dts_shift, sc->ctts_data[sc->ctts_count].
> > > duration,
> > > + sc->dts_shift, ctts_duration,
> > > sc->time_offset, flags & MOV_TRUN_SAMPLE_CTS);
> > > dts = pts - sc->dts_shift;
> > > if (flags & MOV_TRUN_SAMPLE_CTS) {
> > > - dts -= sc->ctts_data[sc->ctts_count].duration;
> > > + dts -= ctts_duration;
> > > } else {
> > > dts -= sc->time_offset;
> > > }
> > > @@ -4343,7 +4340,7 @@ static int mov_read_trun(MOVContext *c,
> AVIOContext
> > > *pb, MOVAtom atom)
> > > }
> > > frag->time = AV_NOPTS_VALUE;
> > > }
> > > - sc->ctts_count++;
> > > +
> > > if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO)
> > > keyframe = 1;
> > > else
> > > @@ -4352,11 +4349,30 @@ static int mov_read_trun(MOVContext *c,
> > > AVIOContext *pb, MOVAtom atom)
> > > MOV_FRAG_SAMPLE_FLAG_DEPENDS_YES));
> > > if (keyframe)
> > > distance = 0;
> > > - err = av_add_index_entry(st, offset, dts, sample_size,
> distance,
> > > - keyframe ? AVINDEX_KEYFRAME : 0);
> > > - if (err < 0) {
> > > + ctts_index = av_add_index_entry(st, offset, dts, sample_size,
> > > distance,
> > > + keyframe ? AVINDEX_KEYFRAME :
> 0);
> > > + if (ctts_index >= 0 && old_nb_index_entries <
> > > st->nb_index_entries) {
> > > + unsigned int size_needed = st->nb_index_entries *
> > > sizeof(*sc->ctts_data);
> > > + unsigned int request_size = size_needed >
> > > sc->ctts_allocated_size ?
> > > + FFMAX(size_needed, 2 * sc->ctts_allocated_size) :
> > > size_needed;
> > > + sc->ctts_data = av_fast_realloc(sc->ctts_data,
> > > &sc->ctts_allocated_size, request_size);
> > > + if (!sc->ctts_data) {
> > > + sc->ctts_count = 0;
> > > + return AVERROR(ENOMEM);
> > > + }
> > > +
> > > + if (ctts_index != old_nb_index_entries) {
> > > + memmove(sc->ctts_data + ctts_index + 1, sc->ctts_data
> +
> > > ctts_index,
> > > + sizeof(*sc->ctts_data) * (sc->ctts_count -
> > > ctts_index));
> > > + }
> > > +
> > > + sc->ctts_data[ctts_index].count = 1;
> > > + sc->ctts_data[ctts_index].duration = ctts_duration;
> > > + sc->ctts_count++;
> > > + } else {
> > > av_log(c->fc, AV_LOG_ERROR, "Failed to add index
> entry\n");
> > > }
> > > +
> > > av_log(c->fc, AV_LOG_TRACE, "AVIndex stream %d, sample %u,
> offset
> > > %"PRIx64", dts %"PRId64", "
> > > "size %u, distance %d, keyframe %d\n", st->index,
> > > sc->sample_count+i,
> > > offset, dts, sample_size, distance, keyframe);
> > > --
> > > 2.13.2.932.g7449e964c-goog
> > >
>
> > isom.h | 1 +
> > mov.c | 58 ++++++++++++++++++++++++++++++
> +++++++---------------------
> > 2 files changed, 38 insertions(+), 21 deletions(-)
> > f1e0078808ae40c006b68acfd075fb1cfe62acf2 0001-Fix-trampling-of-ctts-
> during-seeks-when-sidx-support.patch
> > From 0f94ed2f3cca57bd6dc164e750e92efbbf5b612f Mon Sep 17 00:00:00 2001
> > From: Dale Curtis <dalecurtis at chromium.org>
> > Date: Mon, 17 Jul 2017 17:38:09 -0700
> > Subject: [PATCH] Fix trampling of ctts during seeks when sidx support is
> > enabled.
> >
> > When sidx box support is enabled, the code will skip reading all
> > trun boxes (each containing ctts entries for samples inthat box).
> >
> > If seeks are attempted before all ctts values are known, the old
> > code would dump ctts entries into the wrong location. These are
> > then used to compute pts values which leads to out of order and
> > incorrectly timestamped packets.
> >
> > This patch fixes ctts processing by always using the index returned
> > by av_add_index_entry() as the ctts_data index. When the index gains
> > new entries old values are reshuffled as appropriate.
> >
> > This approach makes sense since the mov demuxer is already relying
> > on the mapping of AVIndex entries to samples for correct demuxing.
> >
> > Notes for future improvement:
> > Probably there are other boxes (stts, stsc, etc) that are impacted
> > by this issue... this patch only attempts to fix ctts since it
> > completely breaks packet timestamping.
> >
> > This patch continues using an array for the ctts data, which is not
> > the most ideal given the rearrangement that needs to happen (via
> > memmove as new entries are read in). Ideally AVIndex and the ctts
> > data would be set-type structures so addition is always worst case
> > O(lg(n)) instead of the O(n^2) that exists now; this slowdown is
> > noticeable during seeks.
> >
> > Additionally since ctts samples from trun boxes always have a count
> > of 1, there's probably an opportunity to avoid the post-seek fixup
> > that iterates O(n) over all samples with an O(1) when no non-1 count
> > samples are present.
> >
> > Signed-off-by: Dale Curtis <dalecurtis at chromium.org>
> > ---
> > libavformat/isom.h | 1 +
> > libavformat/mov.c | 58 ++++++++++++++++++++++++++++++
> ++++--------------------
> > 2 files changed, 38 insertions(+), 21 deletions(-)
>
> this seems to change timestamps for:
> ./ffmpeg -i matrixbench_mpeg2.mpg -t 1 -acodec aac -frag_duration 200k
> test.mov
> ./ffprobe -v 0 test.mov -show_packets -print_format compact
>
> http://samples.ffmpeg.org/benchmark/testsuite1/matrixbench_mpeg2.mpg
>
> The dts/pts pairs after the patch do not look good
>
> @@ -15,59 +15,59 @@
> packet|codec_type=video|stream_index=0|pts=5632|pts_
> time=0.440000|dts=3584|dts_time=0.280000|duration=512|
> duration_time=0.040000|convergence_duration=N/A|
> convergence_duration_time=N/A|size=3995|pos=29570|flags=__
> packet|codec_type=video|stream_index=0|pts=5120|pts_
> time=0.400000|dts=4096|dts_time=0.320000|duration=512|
> duration_time=0.040000|convergence_duration=N/A|
> convergence_duration_time=N/A|size=1526|pos=33565|flags=__
> packet|codec_type=video|stream_index=0|pts=4608|pts_
> time=0.360000|dts=4608|dts_time=0.360000|duration=512|
> duration_time=0.040000|convergence_duration=N/A|
> convergence_duration_time=N/A|size=1309|pos=35091|flags=__
> -packet|codec_type=audio|stream_index=1|pts=9984|pts_
> time=0.208000|dts=9984|dts_time=0.208000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=358|pos=36400|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=11008|pts_
> time=0.229333|dts=11008|dts_time=0.229333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=351|pos=36758|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=12032|pts_
> time=0.250667|dts=12032|dts_time=0.250667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=353|pos=37109|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=13056|pts_
> time=0.272000|dts=13056|dts_time=0.272000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=37462|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=14080|pts_
> time=0.293333|dts=14080|dts_time=0.293333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=351|pos=37799|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=15104|pts_
> time=0.314667|dts=15104|dts_time=0.314667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=38150|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=16128|pts_
> time=0.336000|dts=16128|dts_time=0.336000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=38487|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=17152|pts_
> time=0.357333|dts=17152|dts_time=0.357333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=353|pos=38824|flags=K_
> -packet|codec_type=audio|stream_index=1|pts=18176|pts_
> time=0.378667|dts=18176|dts_time=0.378667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=340|pos=39177|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=42509|pts_
> time=0.885604|dts=9984|dts_time=0.208000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=358|pos=36400|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=43533|pts_
> time=0.906937|dts=11008|dts_time=0.229333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=351|pos=36758|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=44557|pts_
> time=0.928271|dts=12032|dts_time=0.250667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=353|pos=37109|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=45581|pts_
> time=0.949604|dts=13056|dts_time=0.272000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=37462|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=46605|pts_
> time=0.970938|dts=14080|dts_time=0.293333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=351|pos=37799|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=47629|pts_
> time=0.992271|dts=15104|dts_time=0.314667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=38150|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=48653|pts_
> time=1.013604|dts=16128|dts_time=0.336000|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=337|pos=38487|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=49677|pts_
> time=1.034938|dts=17152|dts_time=0.357333|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=353|pos=38824|flags=K_
> +packet|codec_type=audio|stream_index=1|pts=50701|pts_
> time=1.056271|dts=18176|dts_time=0.378667|duration=1024|
> duration_time=0.021333|convergence_duration=N/A|
> convergence_duration_time=N/A|size=340|pos=39177|flags=K_
> packet|codec_type=video|stream_index=0|pts=7168|pts_
> time=0.560000|dts=5120|dts_time=0.400000|duration=512|
> duration_time=0.040000|convergence_duration=N/A|
> convergence_duration_time=N/A|size=4204|pos=39797|flags=__
> ...
>
> [...]
>
> --
> Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> I do not agree with what you have to say, but I'll defend to the death your
> right to say it. -- Voltaire
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Fix-trampling-of-ctts-during-seeks-when-sidx-support.patch
Type: application/octet-stream
Size: 10867 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20170719/f82a1dec/attachment.obj>
More information about the ffmpeg-devel
mailing list