[FFmpeg-devel] [PATCH 1/2] MxPEG decoder

[Anatoly Nenashev]

On 01.11.2010 00:44, Michael Niedermayer wrote:
> On Sun, Oct 31, 2010 at 09:20:34PM +0300, Anatoly Nenashev wrote:
> [...]
>    
>> @@ -195,6 +198,37 @@ int ff_mjpeg_decode_dht(MJpegDecodeContext *s)
>>       return 0;
>>   }
>>
>> +static int mjpeg_allocate_picture(MJpegDecodeContext *s)
>> +{
>> +    if(s->picture.data[0])
>> +        s->avctx->release_buffer(s->avctx,&s->picture);
>> +
>> +    s->picture.reference= 0;
>> +    if(s->avctx->get_buffer(s->avctx,&s->picture)<  0){
>> +        av_log(s->avctx, AV_LOG_ERROR, "mjpeg get_buffer() failed\n");
>> +        return -1;
>> +    }
>> +
>> +    return 0;
>> +}
>> +
>> +static int mxpeg_allocate_picture(MJpegDecodeContext *s)
>> +{
>> +    if (s->mxctx.reference_picture.data[0])
>> +        s->avctx->release_buffer(s->avctx,&s->mxctx.reference_picture);
>> +    s->mxctx.reference_picture = s->picture;
>>      
>    
>> +    s->picture.data[0] = s->picture.data[1] =
>> +        s->picture.data[2] = s->picture.data[3] = 0;
>>      
> the indention is odd
>
>    
Just a feature of Emacs. Indention removed.

>    
>> +
>> +    s->picture.reference= 1;
>>      
>    
>> +    if(s->avctx->get_buffer(s->avctx,&s->picture)<  0){
>> +        av_log(s->avctx, AV_LOG_ERROR, "mxpeg get_buffer() failed\n");
>> +        return -1;
>> +    }
>>      
> duplicate code
>
>
>    
May be not because "mxpeg_allocate_picture" used in two places: 
ff_mjpeg_decode_sof and mxpeg_decode_mxm.

> [...]
>    
>> @@ -788,14 +823,29 @@ static int mjpeg_decode_scan(MJpegDecodeContext *s, int nb_components, int Ah, i
>>           }
>>       }
>>
>> +    if (use_mxm_bitmask) {
>> +        if (s->mb_width*s->mb_height>  (s->mxctx.mxm_bitmask_size<<  3)) {
>> +            av_log(s->avctx, AV_LOG_ERROR, "MXM bitmask is not complete\n");
>> +            return -1;
>> +        }
>> +
>> +        init_get_bits(&mxm_bitmask_gb, s->mxctx.mxm_bitmask,
>> +                      s->mxctx.mxm_bitmask_size<<  3);
>> +
>> +        s->picture.pict_type= FF_P_TYPE;
>> +        s->picture.key_frame= 0;
>> +    }
>>       for(mb_y = 0; mb_y<  s->mb_height; mb_y++) {
>>           for(mb_x = 0; mb_x<  s->mb_width; mb_x++) {
>> +            const int copy_mb = use_mxm_bitmask&&  !get_bits1(&mxm_bitmask_gb);
>> +            if (!reference_picture_avail&&  copy_mb) continue;
>>      
> cant use_mxm_bitmask be 0 when !reference_picture_avail
> (would avoid 1 check per mb)
>
>    
Fixed. Conditions are switched in "if()".

> [...]
>    
>> @@ -1505,6 +1631,15 @@ not_the_end:
>>   the_end:
>>       av_log(avctx, AV_LOG_DEBUG, "mjpeg decode frame unused %td bytes\n", buf_end - buf_ptr);
>>   //    return buf_end - buf_ptr;
>> +
>> +    if (s->avctx->codec_id == CODEC_ID_MXPEG&&
>> +        !s->mxctx.has_complete_frame) {
>> +        if (s->picture.key_frame)
>> +            s->mxctx.has_complete_frame = 1;
>>      
>    
>> +        else if (!(s->avctx->flags&  CODEC_FLAG_LOW_DELAY))
>> +            *data_size = 0;
>>      
> mxpeg is always low delay, unless i misunderstand something
>
>
>    

Not absolutely so. Generally there's no I-Frame concept for MxPEG known 
from other codecs. Every encoded frame contains some parts of all image.
All that we know is after fixed time delay we can collect a full puzzle. 
Actually some cameras send full JPEG  frame in connection and this 
feature is supported in attached patch but this is not that case on 
which it is necessary to rely. That's why if CODEC_FLAG_LOW_DELAY is 
disabled the decoder must collect a full image before return a first 
decoded frame. Otherwise, if CODEC_FLAG_LOW_DELAY is enabled we may see 
a frame full of holes at the start of played video.