100l, inv_zigzag_direct16 must be aligned to 16 bytes for dct_quantize_SSE2

(which uses movdqa to xmm1 on inv_zigzag_direct16).

Originally committed as revision 19711 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/dsputil.c

@@ -86,7 +86,7 @@ const uint8_t ff_zigzag248_direct[64] = {
 };
 
 /* not permutated inverse zigzag_direct + 1 for MMX quantizer */
-DECLARE_ALIGNED_8(uint16_t, inv_zigzag_direct16[64]);
+DECLARE_ALIGNED_16(uint16_t, inv_zigzag_direct16[64]);
 
 const uint8_t ff_alternate_horizontal_scan[64] = {
     0,  1,   2,  3,  8,  9, 16, 17,