bug was introduced in version 1.75 (2003-12-30)
this may have lead to a small drop in quality of the 4mv mode, but should have only affected the mbd=0 case
Originally committed as revision 2698 to svn://svn.ffmpeg.org/ffmpeg/trunk
uint8_t *dest_y = s->me.scratchpad + offset;
if(s->quarter_sample){
- uint8_t *ref= ref_data[0] + (mx4>>2) + (my4>>2)*stride + offset;
+ uint8_t *ref= ref_data[0] + (mx4>>2) + (my4>>2)*stride;
dxy = ((my4 & 3) << 2) | (mx4 & 3);
if(s->no_rounding)
else
s->dsp.put_qpel_pixels_tab [1][dxy](dest_y , ref , stride);
}else{
- uint8_t *ref= ref_data[0] + (mx4>>1) + (my4>>1)*stride + offset;
+ uint8_t *ref= ref_data[0] + (mx4>>1) + (my4>>1)*stride;
dxy = ((my4 & 1) << 1) | (mx4 & 1);
if(s->no_rounding)