diff mbox series

[FFmpeg-devel] avformat/wavdec: More accurate duration

Message ID tencent_815881C33F7BB353777238EB9BC130625908@qq.com
State New
Headers show
Series [FFmpeg-devel] avformat/wavdec: More accurate duration | expand

Checks

Context Check Description
andriy/x86_make_warn warning New warnings during build
andriy/x86_make success Make finished
andriy/x86_make_fate fail Make fate failed
andriy/PPC64_make success Make finished
andriy/PPC64_make_fate warning Make fate failed

Commit Message

Danivy July 28, 2021, 2:26 p.m. UTC 
Using data size and nAvgBytesPerSec to calculate the duration will be more accurate
---
 libavformat/wavdec.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

Comments

Andreas Rheinhardt July 28, 2021, 8:49 p.m. UTC | #1 
Danivy:
> Using data size and nAvgBytesPerSec to calculate the duration will be more accurate
> ---
>  libavformat/wavdec.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/libavformat/wavdec.c b/libavformat/wavdec.c
> index d2fb81ca7f..50b1f81e62 100644
> --- a/libavformat/wavdec.c
> +++ b/libavformat/wavdec.c
> @@ -185,7 +185,8 @@ static int wav_parse_fmt_tag(AVFormatContext *s, int64_t size, AVStream *st)
>  
>      st->internal->need_parsing = AVSTREAM_PARSE_FULL_RAW;
>  
> -    avpriv_set_pts_info(st, 64, 1, st->codecpar->sample_rate);
> +    uint64_t nAvgBytesPerSec = st->codecpar->bit_rate / 8;
> +    avpriv_set_pts_info(st, 64, 1, nAvgBytesPerSec);
>  
>      return 0;
>  }
> @@ -637,7 +638,9 @@ break_loop:
>                                    /
>                  (st->codecpar->channels * (uint64_t)av_get_bits_per_sample(st->codecpar->codec_id));
>  
> -    if (sample_count)
> +    if (data_size)
> +        st->duration = data_size;
> +    else if (sample_count)
>          st->duration = sample_count;
>  
>      if (st->codecpar->codec_id == AV_CODEC_ID_PCM_S32LE &&
> 
1. What is the point of using a finer timestamp scale than the sample rate?
2. There are scenarios where this leads to a coarser timestamp scale;
e.g. for mono adpcm yamaha files.
3. Parts of the code apparently still presume that the timebase is the
sample rate; therefore FATE (our regression testing system) fails with
your patch: https://patchwork.ffmpeg.org/check/36501/

- Andreas
diff mbox series

Patch

diff --git a/libavformat/wavdec.c b/libavformat/wavdec.c
index d2fb81ca7f..50b1f81e62 100644
--- a/libavformat/wavdec.c
+++ b/libavformat/wavdec.c
@@ -185,7 +185,8 @@  static int wav_parse_fmt_tag(AVFormatContext *s, int64_t size, AVStream *st)
 
     st->internal->need_parsing = AVSTREAM_PARSE_FULL_RAW;
 
-    avpriv_set_pts_info(st, 64, 1, st->codecpar->sample_rate);
+    uint64_t nAvgBytesPerSec = st->codecpar->bit_rate / 8;
+    avpriv_set_pts_info(st, 64, 1, nAvgBytesPerSec);
 
     return 0;
 }
@@ -637,7 +638,9 @@  break_loop:
                                   /
                 (st->codecpar->channels * (uint64_t)av_get_bits_per_sample(st->codecpar->codec_id));
 
-    if (sample_count)
+    if (data_size)
+        st->duration = data_size;
+    else if (sample_count)
         st->duration = sample_count;
 
     if (st->codecpar->codec_id == AV_CODEC_ID_PCM_S32LE &&