Re: [Linphone-developers] patch for h263Oldenc

[Simon Morlat]

Hi Joakim,

Thank your for submitting this patch. The old H263 codec is fairly
untested and unused so I'm not suprised you could found a bug in it.
I think ffmpeg generates byte-aligned GOB and PSC.
The code has still one issue (not related to your patch): it is not able
to split at Macroblock level in case the distance between two GOBs is
greater that the requested mtu.
This is a problem for CIF streams and greater, but should work ok with
QCIF.

Simon

Le mercredi 30 décembre 2009 à 03:34 +0100, address@hidden a écrit :
> We have experienced issues with mediastreamers h263oldEnc and an ffmpeg
> based h263 decoder for Java Media Framework, JMF.
> 
> The problem seems to be that the JMF decoder is more stringent than
> other ffmpeg decoders and is sensitive to what we believe is an encoding
> error in videostreamer. When we wireshark h263 sessions we notice that the
> difference between working and non-working streams is that packets in
> the working stream begins at GOB or PSC boundaries, whereas packets in the non
> working mediastreamer encoded stream often doesn't begin at these
> boundaries. The visual result in the receiving end is something that
> looks like severe packet loss.
> 
> The attached patch solve these issues for us. It is a replacement for
> get_gbsc(). It works a little differently but has the same intent as
> get_gbsc(), finding the GBSC and PSC boundaries in a ffmpeg encoded
> stream. For simplicity is assumes the stream is byte aligned, which the
> original doesnt. This assumptions seems safe to us but if there are
> objections we can rework the patch. Furthermore it searches from the end
> of the packet size backwards for markers. This isn't really an
> optimization, we just found it easier to think this way.
> 
> To summarize, we believe this patch improves compatibility with old
> stringent h263 decoders. We are also happy to rework the patch to the
> standards needed for inclusion in the mediastreamer codebase.
> 
> 
> 
> 
> Index: videoenc.c
> ===================================================================
> --- videoenc.c        (revision 780)
> +++ videoenc.c        (working copy)
> @@ -371,12 +371,33 @@
>       return k;
>  }
>  
> +
> +static int get_gbsc_bytealigned(uint8_t *begin, uint8_t *end)
> +{
> +  int i;
> +  int len = end - begin;
> +  for (i = len - 2;  //len + length of scan window
> +       i > 2 + 2; //length of scan window + 2 avoidance of 1st gob or psc
> +       i--){
> +    if(*(begin + i) == 0 &&
> +       *(begin + i+1) == 0 &&
> +       (*(begin + i+2) & 0x80) == 0x80){
> +      //ms_message("JV psc/gob found! %2x %2x %2x", *(begin + i), *(begin + 
> i+1), *(begin + i + 2));
> +      return i;
> +    }
> +  }
> +  //ms_message("JV no psc or gob found!");
> +  return len;
> +  
> +}
> +
>  static void rfc2190_generate_packets(MSFilter *f, EncState *s, mblk_t 
> *frame, uint32_t timestamp){
>       mblk_t *packet=NULL;
>       
>       while (frame->b_rptr<frame->b_wptr){
>               packet=dupb(frame);
> -             
> frame->b_rptr=packet->b_wptr=packet->b_rptr+get_gbsc(packet->b_rptr, 
> MIN(packet->b_rptr+s->mtu,frame->b_wptr));
> +             frame->b_rptr = packet->b_wptr =
> +                  packet->b_rptr + get_gbsc_bytealigned(packet->b_rptr, 
> MIN(packet->b_rptr+s->mtu,frame->b_wptr));
>               add_rfc2190_header(&packet, &s->av_context);
>               mblk_set_timestamp_info(packet,timestamp);
>               ms_queue_put(f->outputs[0],packet);