falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
22190 Commits
32 Branches
283MB
Tree: bdd19e29df
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'bdd19e29df'
${ noResults }
FFmpeg/libavcodec/h264_direct.c

470 lines
20KB
Raw Blame History 

  1. /*

  2.  * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding

  3.  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

  7.  * FFmpeg is free software; you can redistribute it and/or

  8.  * modify it under the terms of the GNU Lesser General Public

  9.  * License as published by the Free Software Foundation; either

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * FFmpeg is distributed in the hope that it will be useful,

  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  15.  * Lesser General Public License for more details.

  16.  *

  17.  * You should have received a copy of the GNU Lesser General Public

  18.  * License along with FFmpeg; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  */

  21. 

  22. /**

  23.  * @file libavcodec/h264_direct.c

  24.  * H.264 / AVC / MPEG4 part10 direct mb/block decoding.

  25.  * @author Michael Niedermayer <michaelni@gmx.at>

  26.  */

  27. 

  28. #include "internal.h"

  29. #include "dsputil.h"

  30. #include "avcodec.h"

  31. #include "mpegvideo.h"

  32. #include "h264.h"

  33. #include "h264_mvpred.h"

  34. #include "rectangle.h"

  35. 

  36. #if ARCH_X86

  37. #include "x86/h264_i386.h"

  38. #endif

  39. 

  40. //#undef NDEBUG

  41. #include <assert.h>

  42. 

  43. 

  44. static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){

  45.     int poc0 = h->ref_list[0][i].poc;

  46.     int td = av_clip(poc1 - poc0, -128, 127);

  47.     if(td == 0 || h->ref_list[0][i].long_ref){

  48.         return 256;

  49.     }else{

  50.         int tb = av_clip(poc - poc0, -128, 127);

  51.         int tx = (16384 + (FFABS(td) >> 1)) / td;

  52.         return av_clip((tb*tx + 32) >> 6, -1024, 1023);

  53.     }

  54. }

  55. 

  56. void ff_h264_direct_dist_scale_factor(H264Context * const h){

  57.     MpegEncContext * const s = &h->s;

  58.     const int poc = h->s.current_picture_ptr->field_poc[ s->picture_structure == PICT_BOTTOM_FIELD ];

  59.     const int poc1 = h->ref_list[1][0].poc;

  60.     int i, field;

  61.     for(field=0; field<2; field++){

  62.         const int poc  = h->s.current_picture_ptr->field_poc[field];

  63.         const int poc1 = h->ref_list[1][0].field_poc[field];

  64.         for(i=0; i < 2*h->ref_count[0]; i++)

  65.             h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);

  66.     }

  67. 

  68.     for(i=0; i<h->ref_count[0]; i++){

  69.         h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);

  70.     }

  71. }

  72. 

  73. static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){

  74.     MpegEncContext * const s = &h->s;

  75.     Picture * const ref1 = &h->ref_list[1][0];

  76.     int j, old_ref, rfield;

  77.     int start= mbafi ? 16                      : 0;

  78.     int end  = mbafi ? 16+2*h->ref_count[list] : h->ref_count[list];

  79.     int interl= mbafi || s->picture_structure != PICT_FRAME;

  80. 

  81.     /* bogus; fills in for missing frames */

  82.     memset(map[list], 0, sizeof(map[list]));

  83. 

  84.     for(rfield=0; rfield<2; rfield++){

  85.         for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){

  86.             int poc = ref1->ref_poc[colfield][list][old_ref];

  87. 

  88.             if     (!interl)

  89.                 poc |= 3;

  90.             else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed

  91.                 poc= (poc&~3) + rfield + 1;

  92. 

  93.             for(j=start; j<end; j++){

  94.                 if(4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3) == poc){

  95.                     int cur_ref= mbafi ? (j-16)^field : j;

  96.                     map[list][2*old_ref + (rfield^field) + 16] = cur_ref;

  97.                     if(rfield == field)

  98.                         map[list][old_ref] = cur_ref;

  99.                     break;

  100.                 }

  101.             }

  102.         }

  103.     }

  104. }

  105. 

  106. void ff_h264_direct_ref_list_init(H264Context * const h){

  107.     MpegEncContext * const s = &h->s;

  108.     Picture * const ref1 = &h->ref_list[1][0];

  109.     Picture * const cur = s->current_picture_ptr;

  110.     int list, j, field;

  111.     int sidx= (s->picture_structure&1)^1;

  112.     int ref1sidx= (ref1->reference&1)^1;

  113. 

  114.     for(list=0; list<2; list++){

  115.         cur->ref_count[sidx][list] = h->ref_count[list];

  116.         for(j=0; j<h->ref_count[list]; j++)

  117.             cur->ref_poc[sidx][list][j] = 4*h->ref_list[list][j].frame_num + (h->ref_list[list][j].reference&3);

  118.     }

  119. 

  120.     if(s->picture_structure == PICT_FRAME){

  121.         memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));

  122.         memcpy(cur->ref_poc  [1], cur->ref_poc  [0], sizeof(cur->ref_poc  [0]));

  123.     }

  124. 

  125.     cur->mbaff= FRAME_MBAFF;

  126. 

  127.     if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)

  128.         return;

  129. 

  130.     for(list=0; list<2; list++){

  131.         fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);

  132.         for(field=0; field<2; field++)

  133.             fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);

  134.     }

  135. }

  136. 

  137. void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){

  138.     MpegEncContext * const s = &h->s;

  139.     int b8_stride = h->b8_stride;

  140.     int b4_stride = h->b_stride;

  141.     int mb_xy = h->mb_xy;

  142.     int mb_type_col[2];

  143.     const int16_t (*l1mv0)[2], (*l1mv1)[2];

  144.     const int8_t *l1ref0, *l1ref1;

  145.     const int is_b8x8 = IS_8X8(*mb_type);

  146.     unsigned int sub_mb_type;

  147.     int i8, i4;

  148. 

  149.     assert(h->ref_list[1][0].reference&3);

  150. 

  151. #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)

  152. 

  153.     if(IS_INTERLACED(h->ref_list[1][0].mb_type[mb_xy])){ // AFL/AFR/FR/FL -> AFL/FL

  154.         if(!IS_INTERLACED(*mb_type)){                    //     AFR/FR    -> AFL/FL

  155.             int cur_poc = s->current_picture_ptr->poc;

  156.             int *col_poc = h->ref_list[1]->field_poc;

  157.             int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);

  158.             mb_xy= s->mb_x + ((s->mb_y&~1) + col_parity)*s->mb_stride;

  159.             b8_stride = 0;

  160.         }else if(!(s->picture_structure & h->ref_list[1][0].reference) && !h->ref_list[1][0].mbaff){// FL -> FL & differ parity

  161.             int fieldoff= 2*(h->ref_list[1][0].reference)-3;

  162.             mb_xy += s->mb_stride*fieldoff;

  163.         }

  164.         goto single_col;

  165.     }else{                                               // AFL/AFR/FR/FL -> AFR/FR

  166.         if(IS_INTERLACED(*mb_type)){                     // AFL       /FL -> AFR/FR

  167.             mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;

  168.             mb_type_col[0] = h->ref_list[1][0].mb_type[mb_xy];

  169.             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy + s->mb_stride];

  170.             b8_stride *= 3;

  171.             b4_stride *= 6;

  172.             //FIXME IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag

  173.             if(    (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)

  174.                 && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)

  175.                 && !is_b8x8){

  176.                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  177.                 *mb_type   |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */

  178.             }else{

  179.                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  180.                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;

  181.             }

  182.         }else{                                           //     AFR/FR    -> AFR/FR

  183. single_col:

  184.             mb_type_col[0] =

  185.             mb_type_col[1] = h->ref_list[1][0].mb_type[mb_xy];

  186.             if(IS_8X8(mb_type_col[0]) && !h->sps.direct_8x8_inference_flag){

  187.                 /* FIXME save sub mb types from previous frames (or derive from MVs)

  188.                 * so we know exactly what block size to use */

  189.                 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */

  190.                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;

  191.             }else if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){

  192.                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  193.                 *mb_type   |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */

  194.             }else{

  195.                 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  196.                 *mb_type   |= MB_TYPE_8x8|MB_TYPE_L0L1;

  197.             }

  198.         }

  199.     }

  200. 

  201.     l1mv0  = &h->ref_list[1][0].motion_val[0][h->mb2b_xy [mb_xy]];

  202.     l1mv1  = &h->ref_list[1][0].motion_val[1][h->mb2b_xy [mb_xy]];

  203.     l1ref0 = &h->ref_list[1][0].ref_index [0][h->mb2b8_xy[mb_xy]];

  204.     l1ref1 = &h->ref_list[1][0].ref_index [1][h->mb2b8_xy[mb_xy]];

  205.     if(!b8_stride){

  206.         if(s->mb_y&1){

  207.             l1ref0 += h->b8_stride;

  208.             l1ref1 += h->b8_stride;

  209.             l1mv0  +=  2*b4_stride;

  210.             l1mv1  +=  2*b4_stride;

  211.         }

  212.     }

  213. 

  214.     if(h->direct_spatial_mv_pred){

  215.         int ref[2];

  216.         int mv[2][2];

  217.         int list;

  218. 

  219.         /* FIXME interlacing + spatial direct uses wrong colocated block positions */

  220. 

  221.         /* ref = min(neighbors) */

  222.         for(list=0; list<2; list++){

  223.             int refa = h->ref_cache[list][scan8[0] - 1];

  224.             int refb = h->ref_cache[list][scan8[0] - 8];

  225.             int refc = h->ref_cache[list][scan8[0] - 8 + 4];

  226.             if(refc == PART_NOT_AVAILABLE)

  227.                 refc = h->ref_cache[list][scan8[0] - 8 - 1];

  228.             ref[list] = FFMIN3((unsigned)refa, (unsigned)refb, (unsigned)refc);

  229.             if(ref[list] < 0)

  230.                 ref[list] = -1;

  231.         }

  232. 

  233.         if(ref[0] < 0 && ref[1] < 0){

  234.             ref[0] = ref[1] = 0;

  235.             mv[0][0] = mv[0][1] =

  236.             mv[1][0] = mv[1][1] = 0;

  237.         }else{

  238.             for(list=0; list<2; list++){

  239.                 if(ref[list] >= 0)

  240.                     pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);

  241.                 else

  242.                     mv[list][0] = mv[list][1] = 0;

  243.             }

  244.         }

  245. 

  246.         if(ref[1] < 0){

  247.             if(!is_b8x8)

  248.                 *mb_type &= ~MB_TYPE_L1;

  249.             sub_mb_type &= ~MB_TYPE_L1;

  250.         }else if(ref[0] < 0){

  251.             if(!is_b8x8)

  252.                 *mb_type &= ~MB_TYPE_L0;

  253.             sub_mb_type &= ~MB_TYPE_L0;

  254.         }

  255. 

  256.         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){

  257.             for(i8=0; i8<4; i8++){

  258.                 int x8 = i8&1;

  259.                 int y8 = i8>>1;

  260.                 int xy8 = x8+y8*b8_stride;

  261.                 int xy4 = 3*x8+y8*b4_stride;

  262.                 int a=0, b=0;

  263. 

  264.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  265.                     continue;

  266.                 h->sub_mb_type[i8] = sub_mb_type;

  267. 

  268.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);

  269.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);

  270.                 if(!IS_INTRA(mb_type_col[y8])

  271.                    && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)

  272.                        || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){

  273.                     if(ref[0] > 0)

  274.                         a= pack16to32(mv[0][0],mv[0][1]);

  275.                     if(ref[1] > 0)

  276.                         b= pack16to32(mv[1][0],mv[1][1]);

  277.                 }else{

  278.                     a= pack16to32(mv[0][0],mv[0][1]);

  279.                     b= pack16to32(mv[1][0],mv[1][1]);

  280.                 }

  281.                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);

  282.                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);

  283.             }

  284.         }else if(IS_16X16(*mb_type)){

  285.             int a=0, b=0;

  286. 

  287.             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);

  288.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);

  289.             if(!IS_INTRA(mb_type_col[0])

  290.                && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)

  291.                    || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1

  292.                        && (h->x264_build>33 || !h->x264_build)))){

  293.                 if(ref[0] > 0)

  294.                     a= pack16to32(mv[0][0],mv[0][1]);

  295.                 if(ref[1] > 0)

  296.                     b= pack16to32(mv[1][0],mv[1][1]);

  297.             }else{

  298.                 a= pack16to32(mv[0][0],mv[0][1]);

  299.                 b= pack16to32(mv[1][0],mv[1][1]);

  300.             }

  301.             fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);

  302.             fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);

  303.         }else{

  304.             for(i8=0; i8<4; i8++){

  305.                 const int x8 = i8&1;

  306.                 const int y8 = i8>>1;

  307. 

  308.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  309.                     continue;

  310.                 h->sub_mb_type[i8] = sub_mb_type;

  311. 

  312.                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);

  313.                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);

  314.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);

  315.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);

  316. 

  317.                 /* col_zero_flag */

  318.                 if(!IS_INTRA(mb_type_col[0]) && (   l1ref0[x8 + y8*b8_stride] == 0

  319.                                               || (l1ref0[x8 + y8*b8_stride] < 0 && l1ref1[x8 + y8*b8_stride] == 0

  320.                                                   && (h->x264_build>33 || !h->x264_build)))){

  321.                     const int16_t (*l1mv)[2]= l1ref0[x8 + y8*b8_stride] == 0 ? l1mv0 : l1mv1;

  322.                     if(IS_SUB_8X8(sub_mb_type)){

  323.                         const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];

  324.                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){

  325.                             if(ref[0] == 0)

  326.                                 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  327.                             if(ref[1] == 0)

  328.                                 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  329.                         }

  330.                     }else

  331.                     for(i4=0; i4<4; i4++){

  332.                         const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];

  333.                         if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){

  334.                             if(ref[0] == 0)

  335.                                 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;

  336.                             if(ref[1] == 0)

  337.                                 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;

  338.                         }

  339.                     }

  340.                 }

  341.             }

  342.         }

  343.     }else{ /* direct temporal mv pred */

  344.         const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};

  345.         const int *dist_scale_factor = h->dist_scale_factor;

  346.         int ref_offset= 0;

  347. 

  348.         if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){

  349.             map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];

  350.             map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];

  351.             dist_scale_factor   =h->dist_scale_factor_field[s->mb_y&1];

  352.         }

  353.         if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0]))

  354.             ref_offset += 16;

  355. 

  356.         if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){

  357.             /* FIXME assumes direct_8x8_inference == 1 */

  358.             int y_shift  = 2*!IS_INTERLACED(*mb_type);

  359. 

  360.             for(i8=0; i8<4; i8++){

  361.                 const int x8 = i8&1;

  362.                 const int y8 = i8>>1;

  363.                 int ref0, scale;

  364.                 const int16_t (*l1mv)[2]= l1mv0;

  365. 

  366.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  367.                     continue;

  368.                 h->sub_mb_type[i8] = sub_mb_type;

  369. 

  370.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  371.                 if(IS_INTRA(mb_type_col[y8])){

  372.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  373.                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  374.                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  375.                     continue;

  376.                 }

  377. 

  378.                 ref0 = l1ref0[x8 + y8*b8_stride];

  379.                 if(ref0 >= 0)

  380.                     ref0 = map_col_to_list0[0][ref0 + ref_offset];

  381.                 else{

  382.                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];

  383.                     l1mv= l1mv1;

  384.                 }

  385.                 scale = dist_scale_factor[ref0];

  386.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  387. 

  388.                 {

  389.                     const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];

  390.                     int my_col = (mv_col[1]<<y_shift)/2;

  391.                     int mx = (scale * mv_col[0] + 128) >> 8;

  392.                     int my = (scale * my_col + 128) >> 8;

  393.                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  394.                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);

  395.                 }

  396.             }

  397.             return;

  398.         }

  399. 

  400.         /* one-to-one mv scaling */

  401. 

  402.         if(IS_16X16(*mb_type)){

  403.             int ref, mv0, mv1;

  404. 

  405.             fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);

  406.             if(IS_INTRA(mb_type_col[0])){

  407.                 ref=mv0=mv1=0;

  408.             }else{

  409.                 const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]

  410.                                                 : map_col_to_list0[1][l1ref1[0] + ref_offset];

  411.                 const int scale = dist_scale_factor[ref0];

  412.                 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];

  413.                 int mv_l0[2];

  414.                 mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  415.                 mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  416.                 ref= ref0;

  417.                 mv0= pack16to32(mv_l0[0],mv_l0[1]);

  418.                 mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);

  419.             }

  420.             fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);

  421.             fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);

  422.             fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);

  423.         }else{

  424.             for(i8=0; i8<4; i8++){

  425.                 const int x8 = i8&1;

  426.                 const int y8 = i8>>1;

  427.                 int ref0, scale;

  428.                 const int16_t (*l1mv)[2]= l1mv0;

  429. 

  430.                 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))

  431.                     continue;

  432.                 h->sub_mb_type[i8] = sub_mb_type;

  433.                 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);

  434.                 if(IS_INTRA(mb_type_col[0])){

  435.                     fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);

  436.                     fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);

  437.                     fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);

  438.                     continue;

  439.                 }

  440. 

  441.                 ref0 = l1ref0[x8 + y8*b8_stride] + ref_offset;

  442.                 if(ref0 >= 0)

  443.                     ref0 = map_col_to_list0[0][ref0];

  444.                 else{

  445.                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];

  446.                     l1mv= l1mv1;

  447.                 }

  448.                 scale = dist_scale_factor[ref0];

  449. 

  450.                 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);

  451.                 if(IS_SUB_8X8(sub_mb_type)){

  452.                     const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];

  453.                     int mx = (scale * mv_col[0] + 128) >> 8;

  454.                     int my = (scale * mv_col[1] + 128) >> 8;

  455.                     fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);

  456.                     fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);

  457.                 }else

  458.                 for(i4=0; i4<4; i4++){

  459.                     const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];

  460.                     int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];

  461.                     mv_l0[0] = (scale * mv_col[0] + 128) >> 8;

  462.                     mv_l0[1] = (scale * mv_col[1] + 128) >> 8;

  463.                     *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =

  464.                         pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);

  465.                 }

  466.             }

  467.         }

  468.     }

  469. }