diff mbox

[FFmpeg-devel] lavf/mov.c: Make audio timestamps strictly monotonically increasing inside an edit list. Fixes gapless decoding.

Message ID 1474335091-17230-1-git-send-email-isasi@google.com
State Superseded
Headers show

Commit Message

Sasi Inguva Sept. 20, 2016, 1:31 a.m. UTC
Signed-off-by: Sasi Inguva <isasi@google.com>
---
 libavcodec/utils.c                           | 15 +++---
 libavformat/mov.c                            | 75 ++++++++++++++++++++++++----
 tests/ref/fate/gaplessenc-itunes-to-ipod-aac |  2 +-
 tests/ref/fate/gaplessenc-pcm-to-mov-aac     |  2 +-
 4 files changed, 74 insertions(+), 20 deletions(-)

Comments

Michael Niedermayer Sept. 20, 2016, 4:22 a.m. UTC | #1
On Mon, Sep 19, 2016 at 06:31:31PM -0700, Sasi Inguva wrote:
> Signed-off-by: Sasi Inguva <isasi@google.com>
> ---
>  libavcodec/utils.c                           | 15 +++---
>  libavformat/mov.c                            | 75 ++++++++++++++++++++++++----
>  tests/ref/fate/gaplessenc-itunes-to-ipod-aac |  2 +-
>  tests/ref/fate/gaplessenc-pcm-to-mov-aac     |  2 +-
>  4 files changed, 74 insertions(+), 20 deletions(-)

with this
./ffmpeg -i H263_NM_f.mp4
segfaults

also theres a problem with initial padding with libvorbis with this
change (similar to past acc/mp3 issues)

i can provide proper testcases if needed tomorrow ?
(too late ATM need to sleep)


[...]
Sasi Inguva Sept. 20, 2016, 9:29 p.m. UTC | #2
On Mon, Sep 19, 2016 at 9:22 PM, Michael Niedermayer <michael@niedermayer.cc
> wrote:

> On Mon, Sep 19, 2016 at 06:31:31PM -0700, Sasi Inguva wrote:
> > Signed-off-by: Sasi Inguva <isasi@google.com>
> > ---
> >  libavcodec/utils.c                           | 15 +++---
> >  libavformat/mov.c                            | 75
> ++++++++++++++++++++++++----
> >  tests/ref/fate/gaplessenc-itunes-to-ipod-aac |  2 +-
> >  tests/ref/fate/gaplessenc-pcm-to-mov-aac     |  2 +-
> >  4 files changed, 74 insertions(+), 20 deletions(-)
>
> with this
> ./ffmpeg -i H263_NM_f.mp4
> segfaults
>
fixed.


>
> also theres a problem with initial padding with libvorbis with this
> change (similar to past acc/mp3 issues)
>
> The problem is that libvorbis decoder is not producing any AVFrame, when
we decode the first packet. got_frame_ptr is zero, and frame->nb_samples is
zero when we decode the first packet.
Before this patch first packet is discarded , and skip_samples is set to
zero .
packet|codec_type=audio|stream_index=0|pts=-256|pts_time=-0.016000|dts=-256|dts_time=-0.016000|duration=256|duration_time=0.016000|convergence_duration=N/A|convergence_duration_time=N/A|size=1|pos=36|flags=KD
packet|codec_type=audio|stream_index=0|pts=0|pts_time=0.000000|dts=0|dts_time=0.000000|duration=384|duration_time=0.024000|convergence_duration=N/A|convergence_duration_time=N/A|size=1|pos=37|flags=K_

After this patch also, the first packet is discarded , but skip_samples is
also set to 256. However when we call decode on first packet  libvorbis
decoder , doesn't output anything in AVFrame and frame->nb_samples is zero.
So the skip_samples still remains to be consumed when the next packet is
decoded. Other decoders aac, mp3 do produce a decoded output frame on the
first packet and set the frame->nb_samples correctly, so it works for
others. I have just disabled setting skip_samples for libvorbis inside the
edit list code.


> i can provide proper testcases if needed tomorrow ?
> (too late ATM need to sleep)
>
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> Modern terrorism, a quick summary: Need oil, start war with country that
> has oil, kill hundread thousand in war. Let country fall into chaos,
> be surprised about raise of fundamantalists. Drop more bombs, kill more
> people, be surprised about them taking revenge and drop even more bombs
> and strip your own citizens of their rights and freedoms. to be continued
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
>
diff mbox

Patch

diff --git a/libavcodec/utils.c b/libavcodec/utils.c
index b0345b6..0c2d48c 100644
--- a/libavcodec/utils.c
+++ b/libavcodec/utils.c
@@ -2320,7 +2320,6 @@  int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
         uint32_t discard_padding = 0;
         uint8_t skip_reason = 0;
         uint8_t discard_reason = 0;
-        int demuxer_skip_samples = 0;
         // copy to ensure we do not change avpkt
         AVPacket tmp = *avpkt;
         int did_split = av_packet_split_side_data(&tmp);
@@ -2328,7 +2327,6 @@  int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
         if (ret < 0)
             goto fail;
 
-        demuxer_skip_samples = avctx->internal->skip_samples;
         avctx->internal->pkt = &tmp;
         if (HAVE_THREADS && avctx->active_thread_type & FF_THREAD_FRAME)
             ret = ff_thread_decode_frame(avctx, frame, got_frame_ptr, &tmp);
@@ -2353,13 +2351,6 @@  int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
                 frame->sample_rate = avctx->sample_rate;
         }
 
-
-        if (frame->flags & AV_FRAME_FLAG_DISCARD) {
-            // If using discard frame flag, ignore skip_samples set by the decoder.
-            avctx->internal->skip_samples = demuxer_skip_samples;
-            *got_frame_ptr = 0;
-        }
-
         side= av_packet_get_side_data(avctx->internal->pkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
         if(side && side_size>=10) {
             avctx->internal->skip_samples = AV_RL32(side);
@@ -2369,6 +2360,12 @@  int attribute_align_arg avcodec_decode_audio4(AVCodecContext *avctx,
             skip_reason = AV_RL8(side + 8);
             discard_reason = AV_RL8(side + 9);
         }
+
+        if (frame->flags & AV_FRAME_FLAG_DISCARD) {
+            avctx->internal->skip_samples -= frame->nb_samples;
+            *got_frame_ptr = 0;
+        }
+
         if (avctx->internal->skip_samples > 0 && *got_frame_ptr &&
             !(avctx->flags2 & AV_CODEC_FLAG2_SKIP_MANUAL)) {
             if(frame->nb_samples <= avctx->internal->skip_samples){
diff --git a/libavformat/mov.c b/libavformat/mov.c
index b84d9c0..0805139 100644
--- a/libavformat/mov.c
+++ b/libavformat/mov.c
@@ -2856,6 +2856,21 @@  static int64_t add_index_entry(AVStream *st, int64_t pos, int64_t timestamp,
 }
 
 /**
+ * Rewrite timestamps of index entries in the range [end_index - frame_duration_buffer_size, end_index)
+ * by subtracting end_ts successively by the amounts given in frame_duration_buffer.
+ */
+static void fix_index_entry_timestamps(AVStream* st, int end_index, int64_t end_ts,
+                                       int64_t* frame_duration_buffer,
+                                       int frame_duration_buffer_size) {
+    int i = 0;
+    av_assert0(end_index >= 0 && end_index <= st->nb_index_entries);
+    for (i = 0; i < frame_duration_buffer_size; i++) {
+        end_ts -= frame_duration_buffer[frame_duration_buffer_size - 1 - i];
+        st->index_entries[end_index - 1 - i].timestamp = end_ts;
+    }
+}
+
+/**
  * Append a new ctts entry to ctts_data.
  * Returns the new ctts_count if successful, else returns -1.
  */
@@ -2919,7 +2934,10 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
     int64_t edit_list_media_time_dts = 0;
     int64_t edit_list_start_encountered = 0;
     int64_t search_timestamp = 0;
-
+    int64_t* frame_duration_buffer = NULL;
+    int num_discarded_begin = 0;
+    int first_non_zero_audio_edit = -1;
+    int packet_skip_samples = 0;
 
     if (!msc->elst_data || msc->elst_count <= 0) {
         return;
@@ -2955,6 +2973,7 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
         edit_list_index++;
         edit_list_dts_counter = edit_list_dts_entry_end;
         edit_list_dts_entry_end += edit_list_duration;
+        num_discarded_begin = 0;
         if (edit_list_media_time == -1) {
             continue;
         }
@@ -2962,7 +2981,14 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
         // If we encounter a non-negative edit list reset the skip_samples/start_pad fields and set them
         // according to the edit list below.
         if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO) {
-            st->skip_samples = msc->start_pad = 0;
+            if (first_non_zero_audio_edit < 0) {
+                first_non_zero_audio_edit = 1;
+            } else {
+                first_non_zero_audio_edit = 0;
+            }
+
+            if (first_non_zero_audio_edit > 0)
+                st->skip_samples = msc->start_pad = 0;
         }
 
         //find closest previous key frame
@@ -3042,23 +3068,54 @@  static void mov_fix_index(MOVContext *mov, AVStream *st)
 
             if (curr_cts < edit_list_media_time || curr_cts >= (edit_list_duration + edit_list_media_time)) {
                 if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && curr_cts < edit_list_media_time &&
-                    curr_cts + frame_duration > edit_list_media_time &&
-                    st->skip_samples == 0 && msc->start_pad == 0) {
-                    st->skip_samples = msc->start_pad = edit_list_media_time - curr_cts;
+                    curr_cts + frame_duration > edit_list_media_time && first_non_zero_audio_edit > 0) {
+                     packet_skip_samples = edit_list_media_time - curr_cts;
+                     st->skip_samples += packet_skip_samples;
 
-                    // Shift the index entry timestamp by skip_samples to be correct.
-                    edit_list_dts_counter -= st->skip_samples;
+                    // Shift the index entry timestamp by packet_skip_samples to be correct.
+                    edit_list_dts_counter -= packet_skip_samples;
                     if (edit_list_start_encountered == 0)  {
-                      edit_list_start_encountered = 1;
+                        edit_list_start_encountered = 1;
+                        // Make timestamps strictly monotonically increasing for audio, by rewriting timestamps for
+                        // discarded packets.
+                        if (frame_duration_buffer) {
+                          fix_index_entry_timestamps(st, st->nb_index_entries, edit_list_dts_counter,
+                                                     frame_duration_buffer, num_discarded_begin);
+                          av_freep(&frame_duration_buffer);
+                        }
                     }
 
-                    av_log(mov->fc, AV_LOG_DEBUG, "skip %d audio samples from curr_cts: %"PRId64"\n", st->skip_samples, curr_cts);
+                    av_log(mov->fc, AV_LOG_DEBUG, "skip %d audio samples from curr_cts: %"PRId64"\n", packet_skip_samples, curr_cts);
                 } else {
                     flags |= AVINDEX_DISCARD_FRAME;
                     av_log(mov->fc, AV_LOG_DEBUG, "drop a frame at curr_cts: %"PRId64" @ %"PRId64"\n", curr_cts, index);
+
+                    if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && edit_list_start_encountered == 0) {
+                        num_discarded_begin++;
+                        frame_duration_buffer = av_realloc(frame_duration_buffer, num_discarded_begin);
+                        if (!frame_duration_buffer) {
+                            av_log(mov->fc, AV_LOG_ERROR, "Cannot reallocate frame duration buffer\n");
+                            break;
+                        }
+                        frame_duration_buffer[num_discarded_begin - 1] = frame_duration;
+
+                        // Increment skip_samples for the first non-zero audio edit list
+                        if (first_non_zero_audio_edit > 0) {
+                            st->skip_samples += frame_duration;
+                            msc->start_pad = st->skip_samples;
+                        }
+                    }
                 }
             } else if (edit_list_start_encountered == 0) {
                 edit_list_start_encountered = 1;
+                // Make timestamps strictly monotonically increasing for audio, by rewriting timestamps for
+                // discarded packets.
+                if (st->codecpar->codec_type == AVMEDIA_TYPE_AUDIO && frame_duration_buffer) {
+                    fix_index_entry_timestamps(st, st->nb_index_entries, edit_list_dts_counter,
+                                               frame_duration_buffer, num_discarded_begin);
+                    av_freep(&frame_duration_buffer);
+                }
+
             }
 
             if (add_index_entry(st, current->pos, edit_list_dts_counter, current->size,
diff --git a/tests/ref/fate/gaplessenc-itunes-to-ipod-aac b/tests/ref/fate/gaplessenc-itunes-to-ipod-aac
index 043c085..789681f 100644
--- a/tests/ref/fate/gaplessenc-itunes-to-ipod-aac
+++ b/tests/ref/fate/gaplessenc-itunes-to-ipod-aac
@@ -7,7 +7,7 @@  duration_ts=103326
 start_time=0.000000
 duration=2.367000
 [/FORMAT]
-packet|pts=0|dts=0|duration=N/A
+packet|pts=-1024|dts=-1024|duration=1024
 packet|pts=0|dts=0|duration=1024
 packet|pts=1024|dts=1024|duration=1024
 packet|pts=2048|dts=2048|duration=1024
diff --git a/tests/ref/fate/gaplessenc-pcm-to-mov-aac b/tests/ref/fate/gaplessenc-pcm-to-mov-aac
index 8b7e3f6..8702611 100644
--- a/tests/ref/fate/gaplessenc-pcm-to-mov-aac
+++ b/tests/ref/fate/gaplessenc-pcm-to-mov-aac
@@ -7,7 +7,7 @@  duration_ts=529200
 start_time=0.000000
 duration=12.024000
 [/FORMAT]
-packet|pts=0|dts=0|duration=N/A
+packet|pts=-1024|dts=-1024|duration=1024
 packet|pts=0|dts=0|duration=1024
 packet|pts=1024|dts=1024|duration=1024
 packet|pts=2048|dts=2048|duration=1024