[FFmpeg-devel] [PATCH][WIP]: multithreaded audio frame encoding

[Paul B Mahol]

On 1/25/16, Michael Niedermayer <michael at niedermayer.cc> wrote:
> On Mon, Jan 25, 2016 at 06:47:57PM +0100, Paul B Mahol wrote:
>> On 1/25/16, Paul B Mahol <onemda at gmail.com> wrote:
>> > Hi,
>> >
>> > nonworking patch atttached.
>> >
>>
>> Working patch attached.
>
>>  alacenc.c              |    2 +-
>>  frame_thread_encoder.c |   14 +++++++++++---
>>  frame_thread_encoder.h |    2 +-
>>  utils.c                |   14 ++++++++++----
>>  wavpackenc.c           |    8 ++++----
>
> is it intended that wavpack encoding differ with threads ?

It looks like somewhere something is cached by encoder.

> it seems they are smaller so nothing wrong, just unexpected

It is "random", can be bigger too.

>
> ./ffmpeg -i ~/videos/matrixbench_mpeg2.mpg  -flags +bitexact -fflags
> +bitexact -t 30 new.wv
> ./ffmpeg -i ~/videos/matrixbench_mpeg2.mpg  -flags +bitexact -fflags
> +bitexact -t 30 -threads 1 -thread_type 0 old.wv
> md5sum old.wv new.wv
> 4979304cc5b1149ebddc2a4c5c94aef7  old.wv
> f9a6cd192ace7ec69496096cd3bd1796  new.wv
> -rw-r----- 1 michael michael 2177102 Jan 25 19:16 new.wv
> -rw-r----- 1 michael michael 2199058 Jan 25 19:17 old.wv
>
> ./ffmpeg -i new.wv -f crc -
> ./ffmpeg -i old.wv -f crc -
> CRC=0x30bad4dd
> CRC=0x30bad4dd
>
>
> [...]
>
> --
> 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
>