h264 does not depend on mpegvideo any more
[ffmpeg.git] / libavcodec / h264_direct.c
index a70e041..83c5958 100644 (file)
@@ -2,37 +2,36 @@
  * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
- * @file libavcodec/h264_direct.c
+ * @file
  * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
 #include "internal.h"
-#include "dsputil.h"
 #include "avcodec.h"
-#include "mpegvideo.h"
 #include "h264.h"
+#include "mpegutils.h"
 #include "rectangle.h"
+#include "thread.h"
 
-//#undef NDEBUG
 #include <assert.h>
 
 
@@ -49,29 +48,30 @@ static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
 }
 
 void ff_h264_direct_dist_scale_factor(H264Context * const h){
-    MpegEncContext * const s = &h->s;
-    const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];
+    const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD];
     const int poc1 = h->ref_list[1][0].poc;
     int i, field;
-    for(field=0; field<2; field++){
-        const int poc  = h->s.current_picture_ptr->field_poc[field];
-        const int poc1 = h->ref_list[1][0].field_poc[field];
-        for(i=0; i < 2*h->ref_count[0]; i++)
-            h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
-    }
 
-    for(i=0; i<h->ref_count[0]; i++){
+    if (FRAME_MBAFF(h))
+        for (field = 0; field < 2; field++){
+            const int poc  = h->cur_pic_ptr->field_poc[field];
+            const int poc1 = h->ref_list[1][0].field_poc[field];
+            for (i = 0; i < 2 * h->ref_count[0]; i++)
+                h->dist_scale_factor_field[field][i^field] =
+                    get_scale_factor(h, poc, poc1, i+16);
+        }
+
+    for (i = 0; i < h->ref_count[0]; i++){
         h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
     }
 }
 
 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
-    MpegEncContext * const s = &h->s;
-    Picture * const ref1 = &h->ref_list[1][0];
+    H264Picture * const ref1 = &h->ref_list[1][0];
     int j, old_ref, rfield;
     int start= mbafi ? 16                      : 0;
     int end  = mbafi ? 16+2*h->ref_count[0]    : h->ref_count[0];
-    int interl= mbafi || s->picture_structure != PICT_FRAME;
+    int interl= mbafi || h->picture_structure != PICT_FRAME;
 
     /* bogus; fills in for missing frames */
     memset(map[list], 0, sizeof(map[list]));
@@ -82,13 +82,14 @@ static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field,
 
             if     (!interl)
                 poc |= 3;
-            else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
+            else if( interl && (poc&3) == 3) // FIXME: store all MBAFF references so this is not needed
                 poc= (poc&~3) + rfield + 1;
 
             for(j=start; j<end; j++){
-                if(4*h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference&3) == poc){
+                if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference & 3) == poc) {
                     int cur_ref= mbafi ? (j-16)^field : j;
-                    map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
+                    if (ref1->mbaff)
+                        map[list][2 * old_ref + (rfield^field) + 16] = cur_ref;
                     if(rfield == field || !interl)
                         map[list][old_ref] = cur_ref;
                     break;
@@ -99,179 +100,209 @@ static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field,
 }
 
 void ff_h264_direct_ref_list_init(H264Context * const h){
-    MpegEncContext * const s = &h->s;
-    Picture * const ref1 = &h->ref_list[1][0];
-    Picture * const cur = s->current_picture_ptr;
+    H264Picture * const ref1 = &h->ref_list[1][0];
+    H264Picture * const cur = h->cur_pic_ptr;
     int list, j, field;
-    int sidx= (s->picture_structure&1)^1;
-    int ref1sidx= (ref1->reference&1)^1;
+    int sidx= (h->picture_structure&1)^1;
+    int ref1sidx = (ref1->reference&1)^1;
 
     for(list=0; list<2; list++){
         cur->ref_count[sidx][list] = h->ref_count[list];
         for(j=0; j<h->ref_count[list]; j++)
-            cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);
+            cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference & 3);
     }
 
-    if(s->picture_structure == PICT_FRAME){
+    if(h->picture_structure == PICT_FRAME){
         memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
         memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
     }
 
-    cur->mbaff= FRAME_MBAFF;
+    cur->mbaff = FRAME_MBAFF(h);
 
     h->col_fieldoff= 0;
-    if(s->picture_structure == PICT_FRAME){
-        int cur_poc = s->current_picture_ptr->poc;
+    if(h->picture_structure == PICT_FRAME){
+        int cur_poc = h->cur_pic_ptr->poc;
         int *col_poc = h->ref_list[1]->field_poc;
         h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
         ref1sidx=sidx= h->col_parity;
-    }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){ // FL -> FL & differ parity
-        h->col_fieldoff= s->mb_stride*(2*(h->ref_list[1][0].reference) - 3);
+    } else if (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
+        h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;
     }
 
-    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
+    if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
         return;
 
     for(list=0; list<2; list++){
         fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
-        if(FRAME_MBAFF)
+        if (FRAME_MBAFF(h))
         for(field=0; field<2; field++)
             fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
     }
 }
 
-void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
-    MpegEncContext * const s = &h->s;
-    int b8_stride = h->b8_stride;
+static void await_reference_mb_row(H264Context * const h, H264Picture *ref, int mb_y)
+{
+    int ref_field = ref->reference - 1;
+    int ref_field_picture = ref->field_picture;
+    int ref_height = 16*h->mb_height >> ref_field_picture;
+
+    if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME))
+        return;
+
+    //FIXME it can be safe to access mb stuff
+    //even if pixels aren't deblocked yet
+
+    ff_thread_await_progress(&ref->tf,
+                             FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1),
+                             ref_field_picture && ref_field);
+}
+
+static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
+    int b8_stride = 2;
     int b4_stride = h->b_stride;
-    int mb_xy = h->mb_xy;
+    int mb_xy = h->mb_xy, mb_y = h->mb_y;
     int mb_type_col[2];
     const int16_t (*l1mv0)[2], (*l1mv1)[2];
     const int8_t *l1ref0, *l1ref1;
     const int is_b8x8 = IS_8X8(*mb_type);
-    unsigned int sub_mb_type;
+    unsigned int sub_mb_type= MB_TYPE_L0L1;
     int i8, i4;
+    int ref[2];
+    int mv[2];
+    int list;
+
+    assert(h->ref_list[1][0].reference & 3);
 
-    assert(h->ref_list[1][0].reference&3);
+    await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
 
 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
 
-    if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL
-        if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL
-            mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
+
+    /* ref = min(neighbors) */
+    for(list=0; list<2; list++){
+        int left_ref = h->ref_cache[list][scan8[0] - 1];
+        int top_ref  = h->ref_cache[list][scan8[0] - 8];
+        int refc = h->ref_cache[list][scan8[0] - 8 + 4];
+        const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
+        if(refc == PART_NOT_AVAILABLE){
+            refc = h->ref_cache[list][scan8[0] - 8 - 1];
+            C    = h-> mv_cache[list][scan8[0] - 8 - 1];
+        }
+        ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
+        if(ref[list] >= 0){
+            //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
+            const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
+            const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
+
+            int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
+            if(match_count > 1){ //most common
+                mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
+                                     mid_pred(A[1], B[1], C[1]) );
+            }else {
+                assert(match_count==1);
+                if(left_ref==ref[list]){
+                    mv[list]= AV_RN32A(A);
+                }else if(top_ref==ref[list]){
+                    mv[list]= AV_RN32A(B);
+                }else{
+                    mv[list]= AV_RN32A(C);
+                }
+            }
+        }else{
+            int mask= ~(MB_TYPE_L0 << (2*list));
+            mv[list] = 0;
+            ref[list] = -1;
+            if(!is_b8x8)
+                *mb_type &= mask;
+            sub_mb_type &= mask;
+        }
+    }
+    if(ref[0] < 0 && ref[1] < 0){
+        ref[0] = ref[1] = 0;
+        if(!is_b8x8)
+            *mb_type |= MB_TYPE_L0L1;
+        sub_mb_type |= MB_TYPE_L0L1;
+    }
+
+    if(!(is_b8x8|mv[0]|mv[1])){
+        fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
+        fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
+        fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
+        fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
+        *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
+        return;
+    }
+
+    if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
+        if (!IS_INTERLACED(*mb_type)) {                          //     AFR/FR    -> AFL/FL
+            mb_y = (h->mb_y&~1) + h->col_parity;
+            mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
             b8_stride = 0;
         }else{
-            mb_xy += h->col_fieldoff; // non zero for FL -> FL & differ parity
+            mb_y  += h->col_fieldoff;
+            mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
         }
         goto single_col;
     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
-            mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
+            mb_y = h->mb_y&~1;
+            mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
-            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];
-            b8_stride *= 3;
+            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
+            b8_stride = 2+4*h->mb_stride;
             b4_stride *= 6;
+            if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
+                mb_type_col[0] &= ~MB_TYPE_INTERLACED;
+                mb_type_col[1] &= ~MB_TYPE_INTERLACED;
+            }
 
-            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
             if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
                 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
                 && !is_b8x8){
-                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
+                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
             }else{
-                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
+                *mb_type   |= MB_TYPE_8x8;
             }
         }else{                                           //     AFR/FR    -> AFR/FR
 single_col:
             mb_type_col[0] =
             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
-            if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){
-                /* FIXME save sub mb types from previous frames (or derive from MVs)
-                * so we know exactly what block size to use */
-                sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
-                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
-            }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
-                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
-                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
+
+            sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
+                *mb_type   |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
             }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
-                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
-                *mb_type   |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
+                *mb_type   |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
             }else{
-                sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
-                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
+                if(!h->sps.direct_8x8_inference_flag){
+                    /* FIXME save sub mb types from previous frames (or derive from MVs)
+                    * so we know exactly what block size to use */
+                    sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
+                }
+                *mb_type   |= MB_TYPE_8x8;
             }
         }
     }
 
+    await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
+
     l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
     l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
-    l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];
-    l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];
+    l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
+    l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
     if(!b8_stride){
-        if(s->mb_y&1){
-            l1ref0 += h->b8_stride;
-            l1ref1 += h->b8_stride;
+        if(h->mb_y&1){
+            l1ref0 += 2;
+            l1ref1 += 2;
             l1mv0  +=  2*b4_stride;
             l1mv1  +=  2*b4_stride;
         }
     }
 
-    if(h->direct_spatial_mv_pred){
-        int ref[2];
-        int mv[2][2];
-        int list;
-
-        /* FIXME interlacing + spatial direct uses wrong colocated block positions */
-
-        /* ref = min(neighbors) */
-        for(list=0; list<2; list++){
-            int left_ref = h->ref_cache[list][scan8[0] - 1];
-            int top_ref  = h->ref_cache[list][scan8[0] - 8];
-            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
-            const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
-            if(refc == PART_NOT_AVAILABLE){
-                refc = h->ref_cache[list][scan8[0] - 8 - 1];
-                C    = h-> mv_cache[list][scan8[0] - 8 - 1];
-            }
-            ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
-            if(ref[list] >= 0){
-                //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
-                const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
-                const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
-
-                int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
-                if(match_count > 1){ //most common
-                    mv[list][0]= mid_pred(A[0], B[0], C[0]);
-                    mv[list][1]= mid_pred(A[1], B[1], C[1]);
-                }else {
-                    assert(match_count==1);
-                    if(left_ref==ref[list]){
-                        mv[list][0]= A[0];
-                        mv[list][1]= A[1];
-                    }else if(top_ref==ref[list]){
-                        mv[list][0]= B[0];
-                        mv[list][1]= B[1];
-                    }else{
-                        mv[list][0]= C[0];
-                        mv[list][1]= C[1];
-                    }
-                }
-            }else{
-                int mask= ~(MB_TYPE_L0 << (2*list));
-                mv[list][0] = mv[list][1] = 0;
-                ref[list] = -1;
-                if(!is_b8x8)
-                    *mb_type &= mask;
-                sub_mb_type &= mask;
-            }
-        }
-        if(ref[0] < 0 && ref[1] < 0){
-            ref[0] = ref[1] = 0;
-            if(!is_b8x8)
-                *mb_type |= MB_TYPE_L0L1;
-            sub_mb_type |= MB_TYPE_L0L1;
-        }
 
         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
+            int n=0;
             for(i8=0; i8<4; i8++){
                 int x8 = i8&1;
                 int y8 = i8>>1;
@@ -290,16 +321,19 @@ single_col:
                        || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
                     a=b=0;
                     if(ref[0] > 0)
-                        a= pack16to32(mv[0][0],mv[0][1]);
+                        a= mv[0];
                     if(ref[1] > 0)
-                        b= pack16to32(mv[1][0],mv[1][1]);
+                        b= mv[1];
+                    n++;
                 }else{
-                    a= pack16to32(mv[0][0],mv[0][1]);
-                    b= pack16to32(mv[1][0],mv[1][1]);
+                    a= mv[0];
+                    b= mv[1];
                 }
                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
             }
+            if(!is_b8x8 && !(n&3))
+                *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
         }else if(IS_16X16(*mb_type)){
             int a,b;
 
@@ -311,16 +345,17 @@ single_col:
                        && h->x264_build>33U))){
                 a=b=0;
                 if(ref[0] > 0)
-                    a= pack16to32(mv[0][0],mv[0][1]);
+                    a= mv[0];
                 if(ref[1] > 0)
-                    b= pack16to32(mv[1][0],mv[1][1]);
+                    b= mv[1];
             }else{
-                a= pack16to32(mv[0][0],mv[0][1]);
-                b= pack16to32(mv[1][0],mv[1][1]);
+                a= mv[0];
+                b= mv[1];
             }
             fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
             fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
         }else{
+            int n=0;
             for(i8=0; i8<4; i8++){
                 const int x8 = i8&1;
                 const int y8 = i8>>1;
@@ -329,16 +364,17 @@ single_col:
                     continue;
                 h->sub_mb_type[i8] = sub_mb_type;
 
-                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
-                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
+                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
+                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
 
+                assert(b8_stride==2);
                 /* col_zero_flag */
-                if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[x8 + y8*b8_stride] == 0
-                                              || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0
+                if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[i8] == 0
+                                              || (l1ref0[i8] < 0 && l1ref1[i8] == 0
                                                   && h->x264_build>33U))){
-                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;
+                    const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
                     if(IS_SUB_8X8(sub_mb_type)){
                         const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
@@ -346,32 +382,125 @@ single_col:
                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                             if(ref[1] == 0)
                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
+                            n+=4;
                         }
-                    }else
+                    }else{
+                        int m=0;
                     for(i4=0; i4<4; i4++){
                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
                             if(ref[0] == 0)
-                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
+                                AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
                             if(ref[1] == 0)
-                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
+                                AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
+                            m++;
                         }
                     }
+                    if(!(m&3))
+                        h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
+                    n+=m;
+                    }
+                }
+            }
+            if(!is_b8x8 && !(n&15))
+                *mb_type= (*mb_type & ~(MB_TYPE_8x8|MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_P1L0|MB_TYPE_P1L1))|MB_TYPE_16x16|MB_TYPE_DIRECT2;
+        }
+}
+
+static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
+    int b8_stride = 2;
+    int b4_stride = h->b_stride;
+    int mb_xy = h->mb_xy, mb_y = h->mb_y;
+    int mb_type_col[2];
+    const int16_t (*l1mv0)[2], (*l1mv1)[2];
+    const int8_t *l1ref0, *l1ref1;
+    const int is_b8x8 = IS_8X8(*mb_type);
+    unsigned int sub_mb_type;
+    int i8, i4;
+
+    assert(h->ref_list[1][0].reference & 3);
+
+    await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
+
+    if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
+        if (!IS_INTERLACED(*mb_type)) {                          //     AFR/FR    -> AFL/FL
+            mb_y = (h->mb_y&~1) + h->col_parity;
+            mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
+            b8_stride = 0;
+        }else{
+            mb_y  += h->col_fieldoff;
+            mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
+        }
+        goto single_col;
+    }else{                                               // AFL/AFR/FR/FL -> AFR/FR
+        if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
+            mb_y = h->mb_y&~1;
+            mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
+            mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
+            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
+            b8_stride = 2+4*h->mb_stride;
+            b4_stride *= 6;
+            if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
+                mb_type_col[0] &= ~MB_TYPE_INTERLACED;
+                mb_type_col[1] &= ~MB_TYPE_INTERLACED;
+            }
+
+            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+
+            if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
+                && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
+                && !is_b8x8){
+                *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
+            }else{
+                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
+            }
+        }else{                                           //     AFR/FR    -> AFR/FR
+single_col:
+            mb_type_col[0] =
+            mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
+
+            sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
+            if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
+                *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
+            }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
+                *mb_type   |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
+            }else{
+                if(!h->sps.direct_8x8_inference_flag){
+                    /* FIXME save sub mb types from previous frames (or derive from MVs)
+                    * so we know exactly what block size to use */
+                    sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
                 }
+                *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
             }
         }
-    }else{ /* direct temporal mv pred */
+    }
+
+    await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
+
+    l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
+    l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
+    l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
+    l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
+    if(!b8_stride){
+        if(h->mb_y&1){
+            l1ref0 += 2;
+            l1ref1 += 2;
+            l1mv0  +=  2*b4_stride;
+            l1mv1  +=  2*b4_stride;
+        }
+    }
+
+    {
         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
         const int *dist_scale_factor = h->dist_scale_factor;
-        int ref_offset= 0;
+        int ref_offset;
 
-        if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
-            map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
-            map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
-            dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];
+        if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
+            map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0];
+            map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1];
+            dist_scale_factor   =h->dist_scale_factor_field[h->mb_y&1];
         }
-        if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))
-            ref_offset += 16;
+        ref_offset = (h->ref_list[1][0].mbaff<<4) & (mb_type_col[0]>>3); //if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0])) ref_offset=16 else 0
 
         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
             int y_shift  = 2*!IS_INTERLACED(*mb_type);
@@ -458,11 +587,12 @@ single_col:
                     continue;
                 }
 
-                ref0 = l1ref0[x8 + y8*b8_stride];
+                assert(b8_stride == 2);
+                ref0 = l1ref0[i8];
                 if(ref0 >= 0)
                     ref0 = map_col_to_list0[0][ref0 + ref_offset];
                 else{
-                    ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
+                    ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
                     l1mv= l1mv1;
                 }
                 scale = dist_scale_factor[ref0];
@@ -480,10 +610,18 @@ single_col:
                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
-                    *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
-                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
+                    AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
+                        pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
                 }
             }
         }
     }
 }
+
+void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
+    if(h->direct_spatial_mv_pred){
+        pred_spatial_direct_motion(h, mb_type);
+    }else{
+        pred_temp_direct_motion(h, mb_type);
+    }
+}