hopefully better for platforms not supporting misaligned read

Originally committed as revision 5647 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/cavs.c

@@ -299,7 +299,7 @@ static inline void load_intra_pred_luma(AVSContext *h, uint8_t *top,
 
 static void intra_pred_vert(uint8_t *d,uint8_t *top,uint8_t *left,int stride) {
     int y;
-    uint64_t a = *((uint64_t *)(&top[1]));
+    uint64_t a = unaligned64(&top[1]);
     for(y=0;y<8;y++) {
         *((uint64_t *)(d+y*stride)) = a;
     }