[FFmpeg-devel] [PATCH+RFC] AVFrame for audio

[Justin Ruggles]

Hi,

Justin Ruggles wrote:

> Michael Niedermayer wrote:
> 
>> On Thu, Oct 28, 2010 at 06:58:11PM -0400, Justin Ruggles wrote:
>>> Michael Niedermayer wrote:
>>>
>>>> On Wed, Oct 27, 2010 at 10:13:10PM -0400, Justin Ruggles wrote:
>>>>> Michael Niedermayer wrote:
>>>>>  doc/APIchanges       |    9 ++
>>>>>  libavcodec/avcodec.h |  100 ++++++++++++++++++++++++++++++--
>>>>>  libavcodec/pcm.c     |   41 +++++++++++--
>>>>>  libavcodec/utils.c   |  157 ++++++++++++++++++++++++++++++++++++++++++++-------
>>>>>  4 files changed, 275 insertions(+), 32 deletions(-)
>>>>> 39eb7fb791089c0822e3f87d3226b49131563a72  avcodec_decode_audio4.patch
>>>>> diff --git a/doc/APIchanges b/doc/APIchanges
>>>>> index 4155d32..a39d9fd 100644
>>>>> --- a/doc/APIchanges
>>>>> +++ b/doc/APIchanges
>>>>> @@ -13,6 +13,15 @@ libavutil:   2009-03-08
>>>>>  
>>>>>  API changes, most recent first:
>>>>>  
>>>>> +2010-XX-XX - rXXXXX - lavc 52.92.0 - AVFrame and avcodec_decode_audio
>>>>> +  Add nb_samples field to AVFrame.
>>>>> +  Add user_buffer, user_buffer_size, and user_buffer_in_use fields to AVCodecContext.
>>>>> +  Deprecate AVCODEC_MAX_AUDIO_FRAME_SIZE.
>>>>> +  Deprecate avcodec_decode_audio3() in favor of avcodec_decode_audio4().
>>>>> +  avcodec_decode_audio4() writes output samples to an AVFrame, which gives the
>>>>> +  audio decoders the ability to use get/release/reget_buffer() to get an
>>>>> +  output buffer.
>>>>> +
>>>>>  2010-10-10 - r25441 - lavfi 1.49.0 - AVFilterLink.time_base
>>>>>    Add time_base field to AVFilterLink.
>>>>>  
>>>>> diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
>>>>> index 4bddbaa..1aa1c8c 100644
>>>>> --- a/libavcodec/avcodec.h
>>>>> +++ b/libavcodec/avcodec.h
>>>>> @@ -31,7 +31,7 @@
>>>>>  #include "libavutil/cpu.h"
>>>>>  
>>>>>  #define LIBAVCODEC_VERSION_MAJOR 52
>>>>> -#define LIBAVCODEC_VERSION_MINOR 92
>>>>> +#define LIBAVCODEC_VERSION_MINOR 93
>>>>>  #define LIBAVCODEC_VERSION_MICRO  0
>>>>>  
>>>>>  #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
>>>>> @@ -467,8 +467,10 @@ enum SampleFormat {
>>>>>                                            CH_FRONT_LEFT_OF_CENTER|CH_FRONT_RIGHT_OF_CENTER)
>>>>>  #define CH_LAYOUT_STEREO_DOWNMIX    (CH_STEREO_LEFT|CH_STEREO_RIGHT)
>>>>>  
>>>>> +#if FF_API_AUDIO_OLD
>>>>>  /* in bytes */
>>>>>  #define AVCODEC_MAX_AUDIO_FRAME_SIZE 192000 // 1 second of 48khz 32bit audio
>>>>> +#endif
>>>>>  
>>>>>  /**
>>>>>   * Required number of additionally allocated bytes at the end of the input bitstream for decoding.
>>>>> @@ -988,7 +990,13 @@ typedef struct AVPanScan{
>>>>>       * - decoding: Set by libavcodec\
>>>>>       */\
>>>>>      void *hwaccel_picture_private;\
>>>>> -
>>>>> +\
>>>>> +    /**\
>>>>> +     * number of audio samples (per channel) described by this frame\
>>>>> +     * - encoding: Set by user.\
>>>>> +     * - decoding: Set by libavcodec.\
>>>>> +     */\
>>>>> +    int nb_samples;\
>>>>>  
>>>>>  #define FF_QSCALE_TYPE_MPEG1 0
>>>>>  #define FF_QSCALE_TYPE_MPEG2 1
>>>>> @@ -2744,6 +2752,33 @@ typedef struct AVCodecContext {
>>>>>       * - decoding: unused
>>>>>       */
>>>>>      int lpc_passes;
>>>>> +
>>>>> +    /**
>>>>> +     * User-allocated audio decoder output buffer & buffer size
>>>>> +     * If user_buffer is non-NULL and is large enough,
>>>>> +     * avcodec_default_get_buffer() may user it as an internal buffer instead
>>>>> +     * of allocating its own. This only works with decoders that support
>>>>> +     * CODEC_CAP_DR1. If the decoder uses this buffer, it will set the value
>>>>> +     * to NULL.
>>>>> +     *
>>>>> +     * @note The user may not use this field when using avcodec_decode_audio3()
>>>>> +     *       or avcodec_decode_audio2().
>>>> I dont see why we need such special casing. *samples could be passed in the new
>>>> api like it is in the old
>>> The old API already has the samples buffer passed directly.  Why should
>>> the old API be changed to accept user_buffer as an alternative to
>>> *samples?  Do they have to match?  If not, which takes priority?  This
>>> would require added documentation to an old API.  That seems more
>>> complexity than necessary when the ability to supply a direct buffer is
>>> already there.
>> you misunderstand
>> why is the new api having it passed over AVCodecContext and the old over an
>> function argument (in the sense why dont you change the new api to also take
>> a argument for that?) maybe i miss something but this seems to be simpler
> 
> Ah, thanks for clarifying.
> 
> The new API doesn't need it.  The old API needs it in order to avoid
> memcpy.  In the new API, if the user wants a direct buffer, she can
> override get/release_buffer().
> 
> So how about documenting user_buffer/_size for internal libavcodec use
> only?  We only really need it for backwards compatibility.  The fields
> and related code could also be deprecated along with
> avcodec_decode_audio3() so when the function goes away, so does the
> compatibility code.

In case this is ok, here is a patch+RFC for it.

-Justin


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: avcodec_decode_audio4.patch
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20101031/93865860/attachment.txt>