h264: mark some MC functions with av_always_inline instead of inline.

This actually causes them to be inlined, leading to a significant
speedup (1-1.5% in my measurements).

libavcodec/h264.c

@@ -438,7 +438,8 @@ static void chroma_dc_dct_c(DCTELEM *block){
 }
 #endif
 
-static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
+static av_always_inline void
+mc_dir_part(H264Context *h, Picture *pic, int n, int square,
             int height, int delta, int list,
             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
             int src_x_offset, int src_y_offset,
@@ -535,7 +536,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
               mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 }
 
-static inline void mc_part_std(H264Context *h, int n, int square, int height, int delta,
+static av_always_inline void
+mc_part_std(H264Context *h, int n, int square, int height, int delta,
             uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
             int x_offset, int y_offset,
             qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
@@ -578,7 +580,8 @@ static inline void mc_part_std(H264Context *h, int n, int square, int height, in
     }
 }
 
-static inline void mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
+static av_always_inline void
+mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
                  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
                  int x_offset, int y_offset,
                  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
@@ -664,7 +667,8 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int heigh
     }
 }
 
-static inline void mc_part(H264Context *h, int n, int square, int height, int delta,
+static av_always_inline void
+mc_part(H264Context *h, int n, int square, int height, int delta,
         uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
         int x_offset, int y_offset,
         qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
@@ -685,7 +689,9 @@ static inline void mc_part(H264Context *h, int n, int square, int height, int de
                     chroma_avg, list0, list1, pixel_shift, chroma_idc);
 }
 
-static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc){
+static av_always_inline void
+prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
+{
     /* fetch pixels for estimated mv 4 macroblocks ahead
      * optimized for 64byte cache lines */
     MpegEncContext * const s = &h->s;