Make the horizontal C scaler code clip only against INT16_MAX not 0,

this decreases the difference between C and MMX, its also faster.

Originally committed as revision 27593 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale

libswscale/swscale_template.c

@@ -2330,7 +2330,7 @@ static inline void RENAME(hScale)(int16_t *dst, int dstW, uint8_t *src, int srcW
             val += ((int)src[srcPos + j])*filter[filterSize*i + j];
         }
         //filter += hFilterSize;
-        dst[i] = av_clip(val>>7, 0, (1<<15)-1); // the cubic equation does overflow ...
+        dst[i] = FFMIN(val>>7, (1<<15)-1); // the cubic equation does overflow ...
         //dst[i] = val>>7;
     }
 #endif /* HAVE_ALTIVEC */