[FFmpeg-devel] [PATCH] correctly handle MSRLE uncompressed

[Kostya]

On Tue, May 26, 2009 at 09:00:41PM -0700, Baptiste Coudurier wrote:
> Hi Kostya,
> 
> On 5/24/2009 3:17 AM, Kostya wrote:
> >> [...]
> >>
> >> @@ -88,8 +101,31 @@
> >>          }
> >>      }
> >>  
> >> -    ff_msrle_decode(avctx, (AVPicture*)&s->frame, avctx->bits_per_coded_sample, buf, buf_size);
> >> +    /* FIXME how to correctly detected RLE ??? */
> > 
> > how to correctly detect RLE
> > Hint: escape codes at the end
> 
> Humm, like 0 1 ?
> 
> >> +    if (avctx->height * istride == avpkt->size) {
> >> +        int linesize = avctx->width * avctx->bits_per_coded_sample / 8;
> >> +        uint8_t *ptr = s->frame.data[0];
> >> +        uint8_t *buf = avpkt->data + (avctx->height-1)*istride;
> >> +        int i, j;
> >>  
> >> +        for (i = 0; i < avctx->height; i++) {
> >> +            if (avctx->bits_per_coded_sample == 4) {
> >> +                for (j = 0; j < linesize; j++) {
> >> +                    ptr[j*2+0] = buf[j] >> 4;
> >> +                    ptr[j*2+1] = buf[j] & 0xf;
> >> +                }
> >> +                if (j*2 < avctx->width)
> >> +                    ptr[j*2+0] = buf[j] >> 4;
> > 
> > I'd rather write for(j = 0; j < avctx->width; j += 2){unpack byte}
> > without an additional condition
> 
> Wouldn't this write one sample too much ?

It will, but since stride >= width and stride is even, this is not an
out of bounds write.
Alternatively you can introduce PIX_FMT_PAL4 and do not unpack source at
all ;).
 
> [...]
> 
> -- 
> Baptiste COUDURIER                              GnuPG Key Id: 0x5C1ABAAA