Fix h264_loop_filter_strength_mmx2() so it works with PAFF.

fixed at least:
CVFI1_Sony_D.jsv
CVFI1_SVA_C.264
MR6_BT_B.h264

Originally committed as revision 14310 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/dsputil.h

@@ -347,7 +347,7 @@ typedef struct DSPContext {
     void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
     // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
     void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
-                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1);
+                                      int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
 
     void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
     void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);

libavcodec/h264.c

@@ -6495,7 +6495,7 @@ static void filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y,
             int step = IS_8x8DCT(mb_type) ? 2 : 1;
             edges = (mb_type & MB_TYPE_16x16) && !(h->cbp & 15) ? 1 : 4;
             s->dsp.h264_loop_filter_strength( bS, h->non_zero_count_cache, h->ref_cache, h->mv_cache,
-                                              (h->slice_type == FF_B_TYPE), edges, step, mask_edge0, mask_edge1 );
+                                              (h->slice_type == FF_B_TYPE), edges, step, mask_edge0, mask_edge1, FIELD_PICTURE);
         }
         if( IS_INTRA(s->current_picture.mb_type[mb_xy-1]) )
             bSv[0][0] = 0x0004000400040004ULL;

libavcodec/i386/h264dsp_mmx.c

@@ -20,6 +20,9 @@
 
 #include "dsputil_mmx.h"
 
+DECLARE_ALIGNED_8 (static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
+DECLARE_ALIGNED_8 (static const uint64_t, ff_pb_7_3  ) = 0x0307030703070307ULL;
+
 /***********************************/
 /* IDCT */
 
@@ -623,7 +626,7 @@ static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int a
 }
 
 static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
-                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1 ) {
+                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
     int dir;
     asm volatile(
         "pxor %%mm7, %%mm7 \n\t"
@@ -632,6 +635,13 @@ static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40]
         "movq %2, %%mm4 \n\t"
         ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
     );
+    if(field)
+        asm volatile(
+            "movq %0, %%mm5 \n\t"
+            "movq %1, %%mm4 \n\t"
+            ::"m"(ff_pb_3_1), "m"(ff_pb_7_3)
+        );
+
     // could do a special case for dir==0 && edges==1, but it only reduces the
     // average filter time by 1.2%
     for( dir=1; dir>=0; dir-- ) {