yading@10: /*
yading@10:  * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
yading@10:  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: /**
yading@10:  * @file
yading@10:  * H.264 / AVC / MPEG4 part10 direct mb/block decoding.
yading@10:  * @author Michael Niedermayer <michaelni@gmx.at>
yading@10:  */
yading@10: 
yading@10: #include "internal.h"
yading@10: #include "avcodec.h"
yading@10: #include "mpegvideo.h"
yading@10: #include "h264.h"
yading@10: #include "rectangle.h"
yading@10: #include "thread.h"
yading@10: 
yading@10: //#undef NDEBUG
yading@10: #include <assert.h>
yading@10: 
yading@10: 
yading@10: static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
yading@10:     int poc0 = h->ref_list[0][i].poc;
yading@10:     int td = av_clip(poc1 - poc0, -128, 127);
yading@10:     if(td == 0 || h->ref_list[0][i].long_ref){
yading@10:         return 256;
yading@10:     }else{
yading@10:         int tb = av_clip(poc - poc0, -128, 127);
yading@10:         int tx = (16384 + (FFABS(td) >> 1)) / td;
yading@10:         return av_clip((tb*tx + 32) >> 6, -1024, 1023);
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_h264_direct_dist_scale_factor(H264Context * const h){
yading@10:     const int poc = h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD];
yading@10:     const int poc1 = h->ref_list[1][0].poc;
yading@10:     int i, field;
yading@10: 
yading@10:     if (FRAME_MBAFF(h))
yading@10:         for (field = 0; field < 2; field++){
yading@10:             const int poc  = h->cur_pic_ptr->field_poc[field];
yading@10:             const int poc1 = h->ref_list[1][0].field_poc[field];
yading@10:             for (i = 0; i < 2 * h->ref_count[0]; i++)
yading@10:                 h->dist_scale_factor_field[field][i^field] =
yading@10:                     get_scale_factor(h, poc, poc1, i+16);
yading@10:         }
yading@10: 
yading@10:     for (i = 0; i < h->ref_count[0]; i++){
yading@10:         h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
yading@10:     Picture * const ref1 = &h->ref_list[1][0];
yading@10:     int j, old_ref, rfield;
yading@10:     int start= mbafi ? 16                      : 0;
yading@10:     int end  = mbafi ? 16+2*h->ref_count[0]    : h->ref_count[0];
yading@10:     int interl= mbafi || h->picture_structure != PICT_FRAME;
yading@10: 
yading@10:     /* bogus; fills in for missing frames */
yading@10:     memset(map[list], 0, sizeof(map[list]));
yading@10: 
yading@10:     for(rfield=0; rfield<2; rfield++){
yading@10:         for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
yading@10:             int poc = ref1->ref_poc[colfield][list][old_ref];
yading@10: 
yading@10:             if     (!interl)
yading@10:                 poc |= 3;
yading@10:             else if( interl && (poc&3) == 3) // FIXME: store all MBAFF references so this is not needed
yading@10:                 poc= (poc&~3) + rfield + 1;
yading@10: 
yading@10:             for(j=start; j<end; j++){
yading@10:                 if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].reference & 3) == poc) {
yading@10:                     int cur_ref= mbafi ? (j-16)^field : j;
yading@10:                     if (ref1->mbaff)
yading@10:                         map[list][2 * old_ref + (rfield^field) + 16] = cur_ref;
yading@10:                     if(rfield == field || !interl)
yading@10:                         map[list][old_ref] = cur_ref;
yading@10:                     break;
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_h264_direct_ref_list_init(H264Context * const h){
yading@10:     Picture * const ref1 = &h->ref_list[1][0];
yading@10:     Picture * const cur = h->cur_pic_ptr;
yading@10:     int list, j, field;
yading@10:     int sidx= (h->picture_structure&1)^1;
yading@10:     int ref1sidx = (ref1->reference&1)^1;
yading@10: 
yading@10:     for(list=0; list<2; list++){
yading@10:         cur->ref_count[sidx][list] = h->ref_count[list];
yading@10:         for(j=0; j<h->ref_count[list]; j++)
yading@10:             cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference & 3);
yading@10:     }
yading@10: 
yading@10:     if(h->picture_structure == PICT_FRAME){
yading@10:         memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
yading@10:         memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));
yading@10:     }
yading@10: 
yading@10:     cur->mbaff = FRAME_MBAFF(h);
yading@10: 
yading@10:     h->col_fieldoff= 0;
yading@10:     if(h->picture_structure == PICT_FRAME){
yading@10:         int cur_poc = h->cur_pic_ptr->poc;
yading@10:         int *col_poc = h->ref_list[1]->field_poc;
yading@10:         h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
yading@10:         ref1sidx=sidx= h->col_parity;
yading@10:     } else if (!(h->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
yading@10:         h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;
yading@10:     }
yading@10: 
yading@10:     if (h->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)
yading@10:         return;
yading@10: 
yading@10:     for(list=0; list<2; list++){
yading@10:         fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
yading@10:         if (FRAME_MBAFF(h))
yading@10:         for(field=0; field<2; field++)
yading@10:             fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void await_reference_mb_row(H264Context * const h, Picture *ref, int mb_y)
yading@10: {
yading@10:     int ref_field = ref->reference - 1;
yading@10:     int ref_field_picture = ref->field_picture;
yading@10:     int ref_height = 16*h->mb_height >> ref_field_picture;
yading@10: 
yading@10:     if(!HAVE_THREADS || !(h->avctx->active_thread_type&FF_THREAD_FRAME))
yading@10:         return;
yading@10: 
yading@10:     //FIXME it can be safe to access mb stuff
yading@10:     //even if pixels aren't deblocked yet
yading@10: 
yading@10:     ff_thread_await_progress(&ref->tf,
yading@10:                              FFMIN(16 * mb_y >> ref_field_picture, ref_height - 1),
yading@10:                              ref_field_picture && ref_field);
yading@10: }
yading@10: 
yading@10: static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
yading@10:     int b8_stride = 2;
yading@10:     int b4_stride = h->b_stride;
yading@10:     int mb_xy = h->mb_xy, mb_y = h->mb_y;
yading@10:     int mb_type_col[2];
yading@10:     const int16_t (*l1mv0)[2], (*l1mv1)[2];
yading@10:     const int8_t *l1ref0, *l1ref1;
yading@10:     const int is_b8x8 = IS_8X8(*mb_type);
yading@10:     unsigned int sub_mb_type= MB_TYPE_L0L1;
yading@10:     int i8, i4;
yading@10:     int ref[2];
yading@10:     int mv[2];
yading@10:     int list;
yading@10: 
yading@10:     assert(h->ref_list[1][0].reference & 3);
yading@10: 
yading@10:     await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
yading@10: 
yading@10: #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
yading@10: 
yading@10: 
yading@10:     /* ref = min(neighbors) */
yading@10:     for(list=0; list<2; list++){
yading@10:         int left_ref = h->ref_cache[list][scan8[0] - 1];
yading@10:         int top_ref  = h->ref_cache[list][scan8[0] - 8];
yading@10:         int refc = h->ref_cache[list][scan8[0] - 8 + 4];
yading@10:         const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
yading@10:         if(refc == PART_NOT_AVAILABLE){
yading@10:             refc = h->ref_cache[list][scan8[0] - 8 - 1];
yading@10:             C    = h-> mv_cache[list][scan8[0] - 8 - 1];
yading@10:         }
yading@10:         ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
yading@10:         if(ref[list] >= 0){
yading@10:             //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
yading@10:             const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
yading@10:             const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
yading@10: 
yading@10:             int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
yading@10:             if(match_count > 1){ //most common
yading@10:                 mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
yading@10:                                      mid_pred(A[1], B[1], C[1]) );
yading@10:             }else {
yading@10:                 assert(match_count==1);
yading@10:                 if(left_ref==ref[list]){
yading@10:                     mv[list]= AV_RN32A(A);
yading@10:                 }else if(top_ref==ref[list]){
yading@10:                     mv[list]= AV_RN32A(B);
yading@10:                 }else{
yading@10:                     mv[list]= AV_RN32A(C);
yading@10:                 }
yading@10:             }
yading@10:             av_assert2(ref[list] < (h->ref_count[list] << !!FRAME_MBAFF(h)));
yading@10:         }else{
yading@10:             int mask= ~(MB_TYPE_L0 << (2*list));
yading@10:             mv[list] = 0;
yading@10:             ref[list] = -1;
yading@10:             if(!is_b8x8)
yading@10:                 *mb_type &= mask;
yading@10:             sub_mb_type &= mask;
yading@10:         }
yading@10:     }
yading@10:     if(ref[0] < 0 && ref[1] < 0){
yading@10:         ref[0] = ref[1] = 0;
yading@10:         if(!is_b8x8)
yading@10:             *mb_type |= MB_TYPE_L0L1;
yading@10:         sub_mb_type |= MB_TYPE_L0L1;
yading@10:     }
yading@10: 
yading@10:     if(!(is_b8x8|mv[0]|mv[1])){
yading@10:         fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
yading@10:         fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
yading@10:         fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
yading@10:         fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
yading@10:         *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;
yading@10:         return;
yading@10:     }
yading@10: 
yading@10:     if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
yading@10:         if (!IS_INTERLACED(*mb_type)) {                          //     AFR/FR    -> AFL/FL
yading@10:             mb_y = (h->mb_y&~1) + h->col_parity;
yading@10:             mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
yading@10:             b8_stride = 0;
yading@10:         }else{
yading@10:             mb_y  += h->col_fieldoff;
yading@10:             mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
yading@10:         }
yading@10:         goto single_col;
yading@10:     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
yading@10:         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
yading@10:             mb_y = h->mb_y&~1;
yading@10:             mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
yading@10:             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
yading@10:             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
yading@10:             b8_stride = 2+4*h->mb_stride;
yading@10:             b4_stride *= 6;
yading@10:             if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
yading@10:                 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
yading@10:                 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
yading@10:             }
yading@10: 
yading@10:             sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
yading@10:             if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
yading@10:                 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
yading@10:                 && !is_b8x8){
yading@10:                 *mb_type   |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
yading@10:             }else{
yading@10:                 *mb_type   |= MB_TYPE_8x8;
yading@10:             }
yading@10:         }else{                                           //     AFR/FR    -> AFR/FR
yading@10: single_col:
yading@10:             mb_type_col[0] =
yading@10:             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
yading@10: 
yading@10:             sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
yading@10:             if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
yading@10:                 *mb_type   |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
yading@10:             }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
yading@10:                 *mb_type   |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
yading@10:             }else{
yading@10:                 if(!h->sps.direct_8x8_inference_flag){
yading@10:                     /* FIXME save sub mb types from previous frames (or derive from MVs)
yading@10:                     * so we know exactly what block size to use */
yading@10:                     sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
yading@10:                 }
yading@10:                 *mb_type   |= MB_TYPE_8x8;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
yading@10: 
yading@10:     l1mv0  = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
yading@10:     l1mv1  = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
yading@10:     l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
yading@10:     l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
yading@10:     if(!b8_stride){
yading@10:         if(h->mb_y&1){
yading@10:             l1ref0 += 2;
yading@10:             l1ref1 += 2;
yading@10:             l1mv0  +=  2*b4_stride;
yading@10:             l1mv1  +=  2*b4_stride;
yading@10:         }
yading@10:     }
yading@10: 
yading@10: 
yading@10:         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
yading@10:             int n=0;
yading@10:             for(i8=0; i8<4; i8++){
yading@10:                 int x8 = i8&1;
yading@10:                 int y8 = i8>>1;
yading@10:                 int xy8 = x8+y8*b8_stride;
yading@10:                 int xy4 = 3*x8+y8*b4_stride;
yading@10:                 int a,b;
yading@10: 
yading@10:                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
yading@10:                     continue;
yading@10:                 h->sub_mb_type[i8] = sub_mb_type;
yading@10: 
yading@10:                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
yading@10:                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
yading@10:                 if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
yading@10:                    && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
yading@10:                        || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
yading@10:                     a=b=0;
yading@10:                     if(ref[0] > 0)
yading@10:                         a= mv[0];
yading@10:                     if(ref[1] > 0)
yading@10:                         b= mv[1];
yading@10:                     n++;
yading@10:                 }else{
yading@10:                     a= mv[0];
yading@10:                     b= mv[1];
yading@10:                 }
yading@10:                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
yading@10:                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
yading@10:             }
yading@10:             if(!is_b8x8 && !(n&3))
yading@10:                 *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;
yading@10:         }else if(IS_16X16(*mb_type)){
yading@10:             int a,b;
yading@10: 
yading@10:             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
yading@10:             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
yading@10:             if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
yading@10:                && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
yading@10:                    || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
yading@10:                        && h->x264_build>33U))){
yading@10:                 a=b=0;
yading@10:                 if(ref[0] > 0)
yading@10:                     a= mv[0];
yading@10:                 if(ref[1] > 0)
yading@10:                     b= mv[1];
yading@10:             }else{
yading@10:                 a= mv[0];
yading@10:                 b= mv[1];
yading@10:             }
yading@10:             fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
yading@10:             fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
yading@10:         }else{
yading@10:             int n=0;
yading@10:             for(i8=0; i8<4; i8++){
yading@10:                 const int x8 = i8&1;
yading@10:                 const int y8 = i8>>1;
yading@10: 
yading@10:                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
yading@10:                     continue;
yading@10:                 h->sub_mb_type[i8] = sub_mb_type;
yading@10: 
yading@10:                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
yading@10:                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
yading@10:                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
yading@10:                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
yading@10: 
yading@10:                 assert(b8_stride==2);
yading@10:                 /* col_zero_flag */
yading@10:                 if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && (   l1ref0[i8] == 0
yading@10:                                               || (l1ref0[i8] < 0 && l1ref1[i8] == 0
yading@10:                                                   && h->x264_build>33U))){
yading@10:                     const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
yading@10:                     if(IS_SUB_8X8(sub_mb_type)){
yading@10:                         const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
yading@10:                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
yading@10:                             if(ref[0] == 0)
yading@10:                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                             if(ref[1] == 0)
yading@10:                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                             n+=4;
yading@10:                         }
yading@10:                     }else{
yading@10:                         int m=0;
yading@10:                     for(i4=0; i4<4; i4++){
yading@10:                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
yading@10:                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
yading@10:                             if(ref[0] == 0)
yading@10:                                 AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
yading@10:                             if(ref[1] == 0)
yading@10:                                 AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
yading@10:                             m++;
yading@10:                         }
yading@10:                     }
yading@10:                     if(!(m&3))
yading@10:                         h->sub_mb_type[i8]+= MB_TYPE_16x16 - MB_TYPE_8x8;
yading@10:                     n+=m;
yading@10:                     }
yading@10:                 }
yading@10:             }
yading@10:             if(!is_b8x8 && !(n&15))
yading@10:                 *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;
yading@10:         }
yading@10: }
yading@10: 
yading@10: static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
yading@10:     int b8_stride = 2;
yading@10:     int b4_stride = h->b_stride;
yading@10:     int mb_xy = h->mb_xy, mb_y = h->mb_y;
yading@10:     int mb_type_col[2];
yading@10:     const int16_t (*l1mv0)[2], (*l1mv1)[2];
yading@10:     const int8_t *l1ref0, *l1ref1;
yading@10:     const int is_b8x8 = IS_8X8(*mb_type);
yading@10:     unsigned int sub_mb_type;
yading@10:     int i8, i4;
yading@10: 
yading@10:     assert(h->ref_list[1][0].reference & 3);
yading@10: 
yading@10:     await_reference_mb_row(h, &h->ref_list[1][0], h->mb_y + !!IS_INTERLACED(*mb_type));
yading@10: 
yading@10:     if (IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
yading@10:         if (!IS_INTERLACED(*mb_type)) {                          //     AFR/FR    -> AFL/FL
yading@10:             mb_y = (h->mb_y&~1) + h->col_parity;
yading@10:             mb_xy= h->mb_x + ((h->mb_y&~1) + h->col_parity)*h->mb_stride;
yading@10:             b8_stride = 0;
yading@10:         }else{
yading@10:             mb_y  += h->col_fieldoff;
yading@10:             mb_xy += h->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
yading@10:         }
yading@10:         goto single_col;
yading@10:     }else{                                               // AFL/AFR/FR/FL -> AFR/FR
yading@10:         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR
yading@10:             mb_y = h->mb_y&~1;
yading@10:             mb_xy= h->mb_x + (h->mb_y&~1)*h->mb_stride;
yading@10:             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];
yading@10:             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + h->mb_stride];
yading@10:             b8_stride = 2+4*h->mb_stride;
yading@10:             b4_stride *= 6;
yading@10:             if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
yading@10:                 mb_type_col[0] &= ~MB_TYPE_INTERLACED;
yading@10:                 mb_type_col[1] &= ~MB_TYPE_INTERLACED;
yading@10:             }
yading@10: 
yading@10:             sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
yading@10: 
yading@10:             if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
yading@10:                 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
yading@10:                 && !is_b8x8){
yading@10:                 *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
yading@10:             }else{
yading@10:                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
yading@10:             }
yading@10:         }else{                                           //     AFR/FR    -> AFR/FR
yading@10: single_col:
yading@10:             mb_type_col[0] =
yading@10:             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];
yading@10: 
yading@10:             sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
yading@10:             if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
yading@10:                 *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
yading@10:             }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
yading@10:                 *mb_type   |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
yading@10:             }else{
yading@10:                 if(!h->sps.direct_8x8_inference_flag){
yading@10:                     /* FIXME save sub mb types from previous frames (or derive from MVs)
yading@10:                     * so we know exactly what block size to use */
yading@10:                     sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
yading@10:                 }
yading@10:                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
yading@10: 
yading@10:     l1mv0  = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];
yading@10:     l1mv1  = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];
yading@10:     l1ref0 = &h->ref_list[1][0].ref_index [0][4 * mb_xy];
yading@10:     l1ref1 = &h->ref_list[1][0].ref_index [1][4 * mb_xy];
yading@10:     if(!b8_stride){
yading@10:         if(h->mb_y&1){
yading@10:             l1ref0 += 2;
yading@10:             l1ref1 += 2;
yading@10:             l1mv0  +=  2*b4_stride;
yading@10:             l1mv1  +=  2*b4_stride;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     {
yading@10:         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
yading@10:         const int *dist_scale_factor = h->dist_scale_factor;
yading@10:         int ref_offset;
yading@10: 
yading@10:         if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {
yading@10:             map_col_to_list0[0] = h->map_col_to_list0_field[h->mb_y&1][0];
yading@10:             map_col_to_list0[1] = h->map_col_to_list0_field[h->mb_y&1][1];
yading@10:             dist_scale_factor   =h->dist_scale_factor_field[h->mb_y&1];
yading@10:         }
yading@10:         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
yading@10: 
yading@10:         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
yading@10:             int y_shift  = 2*!IS_INTERLACED(*mb_type);
yading@10:             assert(h->sps.direct_8x8_inference_flag);
yading@10: 
yading@10:             for(i8=0; i8<4; i8++){
yading@10:                 const int x8 = i8&1;
yading@10:                 const int y8 = i8>>1;
yading@10:                 int ref0, scale;
yading@10:                 const int16_t (*l1mv)[2]= l1mv0;
yading@10: 
yading@10:                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
yading@10:                     continue;
yading@10:                 h->sub_mb_type[i8] = sub_mb_type;
yading@10: 
yading@10:                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
yading@10:                 if(IS_INTRA(mb_type_col[y8])){
yading@10:                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
yading@10:                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                     continue;
yading@10:                 }
yading@10: 
yading@10:                 ref0 = l1ref0[x8 + y8*b8_stride];
yading@10:                 if(ref0 >= 0)
yading@10:                     ref0 = map_col_to_list0[0][ref0 + ref_offset];
yading@10:                 else{
yading@10:                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
yading@10:                     l1mv= l1mv1;
yading@10:                 }
yading@10:                 scale = dist_scale_factor[ref0];
yading@10:                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
yading@10: 
yading@10:                 {
yading@10:                     const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
yading@10:                     int my_col = (mv_col[1]<<y_shift)/2;
yading@10:                     int mx = (scale * mv_col[0] + 128) >> 8;
yading@10:                     int my = (scale * my_col + 128) >> 8;
yading@10:                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
yading@10:                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
yading@10:                 }
yading@10:             }
yading@10:             return;
yading@10:         }
yading@10: 
yading@10:         /* one-to-one mv scaling */
yading@10: 
yading@10:         if(IS_16X16(*mb_type)){
yading@10:             int ref, mv0, mv1;
yading@10: 
yading@10:             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
yading@10:             if(IS_INTRA(mb_type_col[0])){
yading@10:                 ref=mv0=mv1=0;
yading@10:             }else{
yading@10:                 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
yading@10:                                                 : map_col_to_list0[1][l1ref1[0] + ref_offset];
yading@10:                 const int scale = dist_scale_factor[ref0];
yading@10:                 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
yading@10:                 int mv_l0[2];
yading@10:                 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
yading@10:                 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
yading@10:                 ref= ref0;
yading@10:                 mv0= pack16to32(mv_l0[0],mv_l0[1]);
yading@10:                 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
yading@10:             }
yading@10:             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
yading@10:             fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
yading@10:             fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
yading@10:         }else{
yading@10:             for(i8=0; i8<4; i8++){
yading@10:                 const int x8 = i8&1;
yading@10:                 const int y8 = i8>>1;
yading@10:                 int ref0, scale;
yading@10:                 const int16_t (*l1mv)[2]= l1mv0;
yading@10: 
yading@10:                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
yading@10:                     continue;
yading@10:                 h->sub_mb_type[i8] = sub_mb_type;
yading@10:                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
yading@10:                 if(IS_INTRA(mb_type_col[0])){
yading@10:                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
yading@10:                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
yading@10:                     continue;
yading@10:                 }
yading@10: 
yading@10:                 assert(b8_stride == 2);
yading@10:                 ref0 = l1ref0[i8];
yading@10:                 if(ref0 >= 0)
yading@10:                     ref0 = map_col_to_list0[0][ref0 + ref_offset];
yading@10:                 else{
yading@10:                     ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
yading@10:                     l1mv= l1mv1;
yading@10:                 }
yading@10:                 scale = dist_scale_factor[ref0];
yading@10: 
yading@10:                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
yading@10:                 if(IS_SUB_8X8(sub_mb_type)){
yading@10:                     const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
yading@10:                     int mx = (scale * mv_col[0] + 128) >> 8;
yading@10:                     int my = (scale * mv_col[1] + 128) >> 8;
yading@10:                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
yading@10:                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
yading@10:                 }else
yading@10:                 for(i4=0; i4<4; i4++){
yading@10:                     const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
yading@10:                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
yading@10:                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
yading@10:                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
yading@10:                     AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
yading@10:                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
yading@10:                 }
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
yading@10:     if(h->direct_spatial_mv_pred){
yading@10:         pred_spatial_direct_motion(h, mb_type);
yading@10:     }else{
yading@10:         pred_temp_direct_motion(h, mb_type);
yading@10:     }
yading@10: }