[FFmpeg-devel] [PATCH 13/16] vmdaudio: simplify vmdaudio_decode_frame() by handling block_type first, then making a single call to vmdaudio_loadsound().

[Kostya]

On Tue, Feb 22, 2011 at 05:39:44PM -0500, Justin Ruggles wrote:
> On 02/22/2011 04:58 PM, Kostya wrote:
> 
> > On Tue, Feb 22, 2011 at 02:05:32PM -0500, Justin Ruggles wrote:
> >>
> >> This also adds output buffer size checks for AUDIO and SILENCE block types.
> >> ---
> >>  libavcodec/vmdav.c |   24 ++++++++++++------------
> >>  1 files changed, 12 insertions(+), 12 deletions(-)
> >>
> > 
> >> diff --git a/libavcodec/vmdav.c b/libavcodec/vmdav.c
> >> index 83173d0..92159f6 100644
> >> --- a/libavcodec/vmdav.c
> >> +++ b/libavcodec/vmdav.c
> >> @@ -506,7 +506,7 @@ static int vmdaudio_decode_frame(AVCodecContext *avctx,
> >>      const uint8_t *buf = avpkt->data;
> >>      int buf_size = avpkt->size;
> >>      VmdAudioContext *s = avctx->priv_data;
> >> -    int block_type;
> >> +    int block_type, silent_chunks;
> >>      unsigned char *output_samples = (unsigned char *)data;
> >>  
> >>      if (buf_size < 16) {
> >> @@ -523,23 +523,23 @@ static int vmdaudio_decode_frame(AVCodecContext *avctx,
> >>      buf      += 16;
> >>      buf_size -= 16;
> >>  
> >> -    if (block_type == BLOCK_TYPE_AUDIO) {
> >> -        /* the chunk contains audio */
> >> -        *data_size = vmdaudio_loadsound(s, output_samples, buf, 0, buf_size);
> >> -    } else if (block_type == BLOCK_TYPE_INITIAL) {
> >> -        /* initial chunk, may contain audio and silence */
> >> +    silent_chunks = 0;
> >> +    if (block_type == BLOCK_TYPE_INITIAL) {
> >>          uint32_t flags = AV_RB32(buf);
> >> -        int silent_chunks = av_popcount(flags);
> >> +        silent_chunks  = av_popcount(flags);
> >>          buf      += 4;
> >>          buf_size -= 4;
> >> -        if(*data_size < (s->block_align*silent_chunks + buf_size) * 2)
> >> -            return -1;
> >> -        *data_size = vmdaudio_loadsound(s, output_samples, buf, silent_chunks, buf_size);
> >>      } else if (block_type == BLOCK_TYPE_SILENCE) {
> >> -        /* silent chunk */
> >> -        *data_size = vmdaudio_loadsound(s, output_samples, buf, 1, 0);
> >> +        silent_chunks = 1;
> >> +        buf_size = 0; // should already be zero but set it just to be sure
> >>      }
> >>  
> >> +    /* ensure output buffer is large enough */
> >> +    if (*data_size < (s->block_align*silent_chunks + buf_size) * 2)
> >> +        return -1;
> >> +
> >> +    *data_size = vmdaudio_loadsound(s, output_samples, buf, silent_chunks, buf_size);
> >> +
> >>      return avpkt->size;
> >>  }
> >>  
> > 
> > will it work if initial block needs to interleave silent and non-silent chunks
> > in order given by flags?
> 
> 
> The structure of the patch will work. Just the silence handling would
> have to be changed and flags would have to be passed to
> vmdaudio_loadsound().  If you can figure it out, more power to you, but
> everything I tried involving interleaving the silence with audio ended
> up with silence in very obviously wrong places.
 
Yes, I remember my experiments resulted in something like that too.

Patch ok then.