[FFmpeg-devel] [PATCH] avformat/mp3dec: Make MP3 seek fast

wm4 nfxjfg at googlemail.com
Sat Sep 5 00:11:47 CEST 2015


On Fri, 4 Sep 2015 13:55:22 -0700
Tsung-Hung Wu <tsunghung at chromium.org> wrote:

> From 094c6efab6c5eec1fec274bf1bcace1987ae7d03 Mon Sep 17 00:00:00 2001
> From: Andy Wu <tsunghung at chromium.org>
> Date: Mon, 31 Aug 2015 17:08:30 -0700
> Subject: [PATCH] avformat/mp3dec: Make MP3 seek fast
> 
> When AVFMT_FLAG_FAST_SEEK is specified, make MP3 seek operation as
> fast as possible.
> 
> When no "-usetoc" is specified, the default operation is using TOC
> if available; otherwise, uses linear interpolation. This is useful
> when seeking a large MP3 file with no TOC available. One example is
> Podcast, many MP3 files are large, but no CBR/VBR tags. Most of
> them are actually CBR. Even in VBR cases, this option sacrifices the
> accuracy of playback time in exchange for responsiveness.
> ---
>  libavformat/mp3dec.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/libavformat/mp3dec.c b/libavformat/mp3dec.c
> index 007c6ea..d3080d7 100644
> --- a/libavformat/mp3dec.c
> +++ b/libavformat/mp3dec.c
> @@ -342,7 +342,7 @@ static int mp3_read_header(AVFormatContext *s)
>      int i;
> 
>      if (mp3->usetoc < 0)
> -        mp3->usetoc = (s->flags & AVFMT_FLAG_FAST_SEEK) ? 0 : 2;
> +        mp3->usetoc = (s->flags & AVFMT_FLAG_FAST_SEEK) ? 1 : 2;
> 
>      st = avformat_new_stream(s, NULL);
>      if (!st)
> @@ -489,19 +489,26 @@ static int mp3_seek(AVFormatContext *s, int
> stream_index, int64_t timestamp,
>      AVStream *st = s->streams[0];
>      int64_t ret  = av_index_search_timestamp(st, timestamp, flags);
>      int64_t best_pos;
> +    int fast_seek = (s->flags & AVFMT_FLAG_FAST_SEEK) ? 1 : 0;
> +    int64_t filesize = mp3->header_filesize;
> 
>      if (mp3->usetoc == 2)
>          return -1; // generic index code
> 
> -    if (   mp3->is_cbr
> +    if (filesize <= 0) {
> +        int64_t size = avio_size(s->pb);
> +        if (size > 0 && size > s->internal->data_offset)
> +            filesize = size - s->internal->data_offset;
> +    }
> +
> +    if (   (mp3->is_cbr || fast_seek)
>          && (mp3->usetoc == 0 || !mp3->xing_toc)
>          && st->duration > 0
> -        && mp3->header_filesize > s->internal->data_offset
> -        && mp3->frames) {
> +        && filesize > 0) {
>          ie = &ie1;
>          timestamp = av_clip64(timestamp, 0, st->duration);
>          ie->timestamp = timestamp;
> -        ie->pos       = av_rescale(timestamp, mp3->header_filesize,
> st->duration) + s->internal->data_offset;
> +        ie->pos       = av_rescale(timestamp, filesize, st->duration) +
> s->internal->data_offset;
>      } else if (mp3->xing_toc) {
>          if (ret < 0)
>              return ret;
> @@ -515,7 +522,7 @@ static int mp3_seek(AVFormatContext *s, int
> stream_index, int64_t timestamp,
>      if (best_pos < 0)
>          return best_pos;
> 
> -    if (mp3->is_cbr && ie == &ie1) {
> +    if (mp3->is_cbr && ie == &ie1 && mp3->frames) {
>          int frame_duration = av_rescale(st->duration, 1, mp3->frames);
>          ie1.timestamp = frame_duration * av_rescale(best_pos -
> s->internal->data_offset, mp3->frames, mp3->header_filesize);
>      }

Patch seems ok. I'll run FATE and apply it tomorrow, unless someone
else has comments. Thanks for the patch.

You've sent it twice; not sure if that was intended? I've looked at the
second one for the purpose of the review. (They're probably the same.)


More information about the ffmpeg-devel mailing list