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.
70915 Commits
32 Branches
913MB
Tree: 4ffbeddd35
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4ffbeddd35'
${ noResults }
FFmpeg/libavcodec/h264_direct.c

703 lines
29KB
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

  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 "avcodec.h"

  30. #include "h264.h"

  31. #include "mpegutils.h"

  32. #include "rectangle.h"

  33. #include "thread.h"

  34. 

  35. #include <assert.h>

  36. 

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

  38. {

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

  40.     int td = av_clip_int8(poc1 - poc0);

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

  42.         return 256;

  43.     } else {

  44.         int tb = av_clip_int8(poc - poc0);

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

  46.         return av_clip_intp2((tb * tx + 32) >> 6, 10);

  47.     }

  48. }

  49. 

  50. void ff_h264_direct_dist_scale_factor(H264Context *const h)

  51. {

  52.     const int poc  = FIELD_PICTURE(h) ? h->cur_pic_ptr->field_poc[h->picture_structure == PICT_BOTTOM_FIELD]

  53.                                       : h->cur_pic_ptr->poc;

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

  55.     int i, field;

  56. 

  57.     if (FRAME_MBAFF(h))

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

  59.             const int poc  = h->cur_pic_ptr->field_poc[field];

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

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

  62.                 h->dist_scale_factor_field[field][i ^ field] =

  63.                     get_scale_factor(h, poc, poc1, i + 16);

  64.         }

  65. 

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

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

  68. }

  69. 

  70. static void fill_colmap(H264Context *h, int map[2][16 + 32], int list,

  71.                         int field, int colfield, int mbafi)

  72. {

  73.     H264Picture *const ref1 = &h->ref_list[1][0];

  74.     int j, old_ref, rfield;

  75.     int start  = mbafi ? 16                       : 0;

  76.     int end    = mbafi ? 16 + 2 * h->ref_count[0] : h->ref_count[0];

  77.     int interl = mbafi || h->picture_structure != PICT_FRAME;

  78. 

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

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

  81. 

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

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

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

  85. 

  86.             if (!interl)

  87.                 poc |= 3;

  88.             // FIXME: store all MBAFF references so this is not needed

  89.             else if (interl && (poc & 3) == 3)

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

  91. 

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

  93.                 if (4 * h->ref_list[0][j].frame_num +

  94.                     (h->ref_list[0][j].reference & 3) == poc) {

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

  96.                     if (ref1->mbaff)

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

  98.                     if (rfield == field || !interl)

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

  100.                     break;

  101.                 }

  102.             }

  103.         }

  104.     }

  105. }

  106. 

  107. void ff_h264_direct_ref_list_init(H264Context *const h, H264SliceContext *sl)

  108. {

  109.     H264Picture *const ref1 = &h->ref_list[1][0];

  110.     H264Picture *const cur = h->cur_pic_ptr;

  111.     int list, j, field;

  112.     int sidx     = (h->picture_structure & 1) ^ 1;

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

  114. 

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

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

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

  118.             cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num +

  119.                                           (h->ref_list[list][j].reference & 3);

  120.     }

  121. 

  122.     if (h->picture_structure == PICT_FRAME) {

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

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

  125.     }

  126. 

  127.     cur->mbaff = FRAME_MBAFF(h);

  128. 

  129.     h->col_fieldoff = 0;

  130.     if (h->picture_structure == PICT_FRAME) {

  131.         int cur_poc  = h->cur_pic_ptr->poc;

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

  133.         h->col_parity = (FFABS(col_poc[0] - cur_poc) >=

  134.                          FFABS(col_poc[1] - cur_poc));

  135.         ref1sidx =

  136.         sidx     = h->col_parity;

  137.     // FL -> FL & differ parity

  138.     } else if (!(h->picture_structure & h->ref_list[1][0].reference) &&

  139.                !h->ref_list[1][0].mbaff) {

  140.         h->col_fieldoff = 2 * h->ref_list[1][0].reference - 3;

  141.     }

  142. 

  143.     if (sl->slice_type_nos != AV_PICTURE_TYPE_B || h->direct_spatial_mv_pred)

  144.         return;

  145. 

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

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

  148.         if (FRAME_MBAFF(h))

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

  150.                 fill_colmap(h, h->map_col_to_list0_field[field], list, field,

  151.                             field, 1);

  152.     }

  153. }

  154. 

  155. static void await_reference_mb_row(H264Context *const h, H264Picture *ref,

  156.                                    int mb_y)

  157. {

  158.     int ref_field         = ref->reference - 1;

  159.     int ref_field_picture = ref->field_picture;

  160.     int ref_height        = 16 * h->mb_height >> ref_field_picture;

  161. 

  162.     if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_FRAME))

  163.         return;

  164. 

  165.     /* FIXME: It can be safe to access mb stuff

  166.      * even if pixels aren't deblocked yet. */

  167. 

  168.     ff_thread_await_progress(&ref->tf,

  169.                              FFMIN(16 * mb_y >> ref_field_picture,

  170.                                    ref_height - 1),

  171.                              ref_field_picture && ref_field);

  172. }

  173. 

  174. static void pred_spatial_direct_motion(H264Context *const h, H264SliceContext *sl,

  175.                                        int *mb_type)

  176. {

  177.     int b8_stride = 2;

  178.     int b4_stride = h->b_stride;

  179.     int mb_xy = h->mb_xy, mb_y = h->mb_y;

  180.     int mb_type_col[2];

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

  182.     const int8_t *l1ref0, *l1ref1;

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

  184.     unsigned int sub_mb_type = MB_TYPE_L0L1;

  185.     int i8, i4;

  186.     int ref[2];

  187.     int mv[2];

  188.     int list;

  189. 

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

  191. 

  192.     await_reference_mb_row(h, &h->ref_list[1][0],

  193.                            h->mb_y + !!IS_INTERLACED(*mb_type));

  194. 

  195. #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16 | MB_TYPE_INTRA4x4 | \

  196.                                 MB_TYPE_INTRA16x16 | MB_TYPE_INTRA_PCM)

  197. 

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

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

  200.         int left_ref     = sl->ref_cache[list][scan8[0] - 1];

  201.         int top_ref      = sl->ref_cache[list][scan8[0] - 8];

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

  203.         const int16_t *C = sl->mv_cache[list][scan8[0]  - 8 + 4];

  204.         if (refc == PART_NOT_AVAILABLE) {

  205.             refc = sl->ref_cache[list][scan8[0] - 8 - 1];

  206.             C    = sl->mv_cache[list][scan8[0]  - 8 - 1];

  207.         }

  208.         ref[list] = FFMIN3((unsigned)left_ref,

  209.                            (unsigned)top_ref,

  210.                            (unsigned)refc);

  211.         if (ref[list] >= 0) {

  212.             /* This is just pred_motion() but with the cases removed that

  213.              * cannot happen for direct blocks. */

  214.             const int16_t *const A = sl->mv_cache[list][scan8[0] - 1];

  215.             const int16_t *const B = sl->mv_cache[list][scan8[0] - 8];

  216. 

  217.             int match_count = (left_ref == ref[list]) +

  218.                               (top_ref  == ref[list]) +

  219.                               (refc     == ref[list]);

  220. 

  221.             if (match_count > 1) { // most common

  222.                 mv[list] = pack16to32(mid_pred(A[0], B[0], C[0]),

  223.                                       mid_pred(A[1], B[1], C[1]));

  224.             } else {

  225.                 assert(match_count == 1);

  226.                 if (left_ref == ref[list])

  227.                     mv[list] = AV_RN32A(A);

  228.                 else if (top_ref == ref[list])

  229.                     mv[list] = AV_RN32A(B);

  230.                 else

  231.                     mv[list] = AV_RN32A(C);

  232.             }

  233.             av_assert2(ref[list] < (h->ref_count[list] << !!FRAME_MBAFF(h)));

  234.         } else {

  235.             int mask = ~(MB_TYPE_L0 << (2 * list));

  236.             mv[list]  = 0;

  237.             ref[list] = -1;

  238.             if (!is_b8x8)

  239.                 *mb_type &= mask;

  240.             sub_mb_type &= mask;

  241.         }

  242.     }

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

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

  245.         if (!is_b8x8)

  246.             *mb_type |= MB_TYPE_L0L1;

  247.         sub_mb_type |= MB_TYPE_L0L1;

  248.     }

  249. 

  250.     if (!(is_b8x8 | mv[0] | mv[1])) {

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

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

  253.         fill_rectangle(&sl->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);

  254.         fill_rectangle(&sl->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);

  255.         *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |

  256.                                  MB_TYPE_P1L0 | MB_TYPE_P1L1)) |

  257.                    MB_TYPE_16x16 | MB_TYPE_DIRECT2;

  258.         return;

  259.     }

  260. 

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

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

  263.             mb_y  = (h->mb_y & ~1) + h->col_parity;

  264.             mb_xy = h->mb_x +

  265.                     ((h->mb_y & ~1) + h->col_parity) * h->mb_stride;

  266.             b8_stride = 0;

  267.         } else {

  268.             mb_y  += h->col_fieldoff;

  269.             mb_xy += h->mb_stride * h->col_fieldoff; // non-zero for FL -> FL & differ parity

  270.         }

  271.         goto single_col;

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

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

  274.             mb_y           =  h->mb_y & ~1;

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

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

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

  278.             b8_stride      = 2 + 4 * h->mb_stride;

  279.             b4_stride     *= 6;

  280.             if (IS_INTERLACED(mb_type_col[0]) !=

  281.                 IS_INTERLACED(mb_type_col[1])) {

  282.                 mb_type_col[0] &= ~MB_TYPE_INTERLACED;

  283.                 mb_type_col[1] &= ~MB_TYPE_INTERLACED;

  284.             }

  285. 

  286.             sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */

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

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

  289.                 !is_b8x8) {

  290.                 *mb_type |= MB_TYPE_16x8 | MB_TYPE_DIRECT2;  /* B_16x8 */

  291.             } else {

  292.                 *mb_type |= MB_TYPE_8x8;

  293.             }

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

  295. single_col:

  296.             mb_type_col[0] =

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

  298. 

  299.             sub_mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_SUB_8x8 */

  300.             if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {

  301.                 *mb_type |= MB_TYPE_16x16 | MB_TYPE_DIRECT2; /* B_16x16 */

  302.             } else if (!is_b8x8 &&

  303.                        (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {

  304.                 *mb_type |= MB_TYPE_DIRECT2 |

  305.                             (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));

  306.             } else {

  307.                 if (!h->sps.direct_8x8_inference_flag) {

  308.                     /* FIXME: Save sub mb types from previous frames (or derive

  309.                      * from MVs) so we know exactly what block size to use. */

  310.                     sub_mb_type += (MB_TYPE_8x8 - MB_TYPE_16x16); /* B_SUB_4x4 */

  311.                 }

  312.                 *mb_type |= MB_TYPE_8x8;

  313.             }

  314.         }

  315.     }

  316. 

  317.     await_reference_mb_row(h, &h->ref_list[1][0], mb_y);

  318. 

  319.     l1mv0  = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy[mb_xy]];

  320.     l1mv1  = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy[mb_xy]];

  321.     l1ref0 = &h->ref_list[1][0].ref_index[0][4 * mb_xy];

  322.     l1ref1 = &h->ref_list[1][0].ref_index[1][4 * mb_xy];

  323.     if (!b8_stride) {

  324.         if (h->mb_y & 1) {

  325.             l1ref0 += 2;

  326.             l1ref1 += 2;

  327.             l1mv0  += 2 * b4_stride;

  328.             l1mv1  += 2 * b4_stride;

  329.         }

  330.     }

  331. 

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

  333.         int n = 0;

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

  335.             int x8  = i8 & 1;

  336.             int y8  = i8 >> 1;

  337.             int xy8 = x8     + y8 * b8_stride;

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

  339.             int a, b;

  340. 

  341.             if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))

  342.                 continue;

  343.             sl->sub_mb_type[i8] = sub_mb_type;

  344. 

  345.             fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,

  346.                            (uint8_t)ref[0], 1);

  347.             fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,

  348.                            (uint8_t)ref[1], 1);

  349.             if (!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref &&

  350.                 ((l1ref0[xy8] == 0 &&

  351.                   FFABS(l1mv0[xy4][0]) <= 1 &&

  352.                   FFABS(l1mv0[xy4][1]) <= 1) ||

  353.                  (l1ref0[xy8] < 0 &&

  354.                   l1ref1[xy8] == 0 &&

  355.                   FFABS(l1mv1[xy4][0]) <= 1 &&

  356.                   FFABS(l1mv1[xy4][1]) <= 1))) {

  357.                 a =

  358.                 b = 0;

  359.                 if (ref[0] > 0)

  360.                     a = mv[0];

  361.                 if (ref[1] > 0)

  362.                     b = mv[1];

  363.                 n++;

  364.             } else {

  365.                 a = mv[0];

  366.                 b = mv[1];

  367.             }

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

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

  370.         }

  371.         if (!is_b8x8 && !(n & 3))

  372.             *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |

  373.                                      MB_TYPE_P1L0 | MB_TYPE_P1L1)) |

  374.                        MB_TYPE_16x16 | MB_TYPE_DIRECT2;

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

  376.         int a, b;

  377. 

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

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

  380.         if (!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref &&

  381.             ((l1ref0[0] == 0 &&

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

  383.               FFABS(l1mv0[0][1]) <= 1) ||

  384.              (l1ref0[0] < 0 && !l1ref1[0] &&

  385.               FFABS(l1mv1[0][0]) <= 1 &&

  386.               FFABS(l1mv1[0][1]) <= 1 &&

  387.               h->x264_build > 33U))) {

  388.             a = b = 0;

  389.             if (ref[0] > 0)

  390.                 a = mv[0];

  391.             if (ref[1] > 0)

  392.                 b = mv[1];

  393.         } else {

  394.             a = mv[0];

  395.             b = mv[1];

  396.         }

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

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

  399.     } else {

  400.         int n = 0;

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

  402.             const int x8 = i8 & 1;

  403.             const int y8 = i8 >> 1;

  404. 

  405.             if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))

  406.                 continue;

  407.             sl->sub_mb_type[i8] = sub_mb_type;

  408. 

  409.             fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8, mv[0], 4);

  410.             fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8, mv[1], 4);

  411.             fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,

  412.                            (uint8_t)ref[0], 1);

  413.             fill_rectangle(&sl->ref_cache[1][scan8[i8 * 4]], 2, 2, 8,

  414.                            (uint8_t)ref[1], 1);

  415. 

  416.             assert(b8_stride == 2);

  417.             /* col_zero_flag */

  418.             if (!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref &&

  419.                 (l1ref0[i8] == 0 ||

  420.                  (l1ref0[i8] < 0 &&

  421.                   l1ref1[i8] == 0 &&

  422.                   h->x264_build > 33U))) {

  423.                 const int16_t (*l1mv)[2] = l1ref0[i8] == 0 ? l1mv0 : l1mv1;

  424.                 if (IS_SUB_8X8(sub_mb_type)) {

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

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

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

  428.                             fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2,

  429.                                            8, 0, 4);

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

  431.                             fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2,

  432.                                            8, 0, 4);

  433.                         n += 4;

  434.                     }

  435.                 } else {

  436.                     int m = 0;

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

  438.                         const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +

  439.                                                      (y8 * 2 + (i4 >> 1)) * b4_stride];

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

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

  442.                                 AV_ZERO32(sl->mv_cache[0][scan8[i8 * 4 + i4]]);

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

  444.                                 AV_ZERO32(sl->mv_cache[1][scan8[i8 * 4 + i4]]);

  445.                             m++;

  446.                         }

  447.                     }

  448.                     if (!(m & 3))

  449.                         sl->sub_mb_type[i8] += MB_TYPE_16x16 - MB_TYPE_8x8;

  450.                     n += m;

  451.                 }

  452.             }

  453.         }

  454.         if (!is_b8x8 && !(n & 15))

  455.             *mb_type = (*mb_type & ~(MB_TYPE_8x8 | MB_TYPE_16x8 | MB_TYPE_8x16 |

  456.                                      MB_TYPE_P1L0 | MB_TYPE_P1L1)) |

  457.                        MB_TYPE_16x16 | MB_TYPE_DIRECT2;

  458.     }

  459. }

  460. 

  461. static void pred_temp_direct_motion(H264Context *const h, H264SliceContext *sl,

  462.                                     int *mb_type)

  463. {

  464.     int b8_stride = 2;

  465.     int b4_stride = h->b_stride;

  466.     int mb_xy = h->mb_xy, mb_y = h->mb_y;

  467.     int mb_type_col[2];

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

  469.     const int8_t *l1ref0, *l1ref1;

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

  471.     unsigned int sub_mb_type;

  472.     int i8, i4;

  473. 

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

  475. 

  476.     await_reference_mb_row(h, &h->ref_list[1][0],

  477.                            h->mb_y + !!IS_INTERLACED(*mb_type));

  478. 

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

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

  481.             mb_y  = (h->mb_y & ~1) + h->col_parity;

  482.             mb_xy = h->mb_x +

  483.                     ((h->mb_y & ~1) + h->col_parity) * h->mb_stride;

  484.             b8_stride = 0;

  485.         } else {

  486.             mb_y  += h->col_fieldoff;

  487.             mb_xy += h->mb_stride * h->col_fieldoff; // non-zero for FL -> FL & differ parity

  488.         }

  489.         goto single_col;

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

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

  492.             mb_y           = h->mb_y & ~1;

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

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

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

  496.             b8_stride      = 2 + 4 * h->mb_stride;

  497.             b4_stride     *= 6;

  498.             if (IS_INTERLACED(mb_type_col[0]) !=

  499.                 IS_INTERLACED(mb_type_col[1])) {

  500.                 mb_type_col[0] &= ~MB_TYPE_INTERLACED;

  501.                 mb_type_col[1] &= ~MB_TYPE_INTERLACED;

  502.             }

  503. 

  504.             sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |

  505.                           MB_TYPE_DIRECT2;                  /* B_SUB_8x8 */

  506. 

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

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

  509.                 !is_b8x8) {

  510.                 *mb_type |= MB_TYPE_16x8 | MB_TYPE_L0L1 |

  511.                             MB_TYPE_DIRECT2;                /* B_16x8 */

  512.             } else {

  513.                 *mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;

  514.             }

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

  516. single_col:

  517.             mb_type_col[0]     =

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

  519. 

  520.             sub_mb_type = MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |

  521.                           MB_TYPE_DIRECT2;                  /* B_SUB_8x8 */

  522.             if (!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)) {

  523.                 *mb_type |= MB_TYPE_16x16 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |

  524.                             MB_TYPE_DIRECT2;                /* B_16x16 */

  525.             } else if (!is_b8x8 &&

  526.                        (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16))) {

  527.                 *mb_type |= MB_TYPE_L0L1 | MB_TYPE_DIRECT2 |

  528.                             (mb_type_col[0] & (MB_TYPE_16x8 | MB_TYPE_8x16));

  529.             } else {

  530.                 if (!h->sps.direct_8x8_inference_flag) {

  531.                     /* FIXME: save sub mb types from previous frames (or derive

  532.                      * from MVs) so we know exactly what block size to use */

  533.                     sub_mb_type = MB_TYPE_8x8 | MB_TYPE_P0L0 | MB_TYPE_P0L1 |

  534.                                   MB_TYPE_DIRECT2;          /* B_SUB_4x4 */

  535.                 }

  536.                 *mb_type |= MB_TYPE_8x8 | MB_TYPE_L0L1;

  537.             }

  538.         }

  539.     }

  540. 

  541.     await_reference_mb_row(h, &h->ref_list[1][0], mb_y);

  542. 

  543.     l1mv0  = (void*)&h->ref_list[1][0].motion_val[0][h->mb2b_xy[mb_xy]];

  544.     l1mv1  = (void*)&h->ref_list[1][0].motion_val[1][h->mb2b_xy[mb_xy]];

  545.     l1ref0 = &h->ref_list[1][0].ref_index[0][4 * mb_xy];

  546.     l1ref1 = &h->ref_list[1][0].ref_index[1][4 * mb_xy];

  547.     if (!b8_stride) {

  548.         if (h->mb_y & 1) {

  549.             l1ref0 += 2;

  550.             l1ref1 += 2;

  551.             l1mv0  += 2 * b4_stride;

  552.             l1mv1  += 2 * b4_stride;

  553.         }

  554.     }

  555. 

  556.     {

  557.         const int *map_col_to_list0[2] = { h->map_col_to_list0[0],

  558.                                            h->map_col_to_list0[1] };

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

  560.         int ref_offset;

  561. 

  562.         if (FRAME_MBAFF(h) && IS_INTERLACED(*mb_type)) {

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

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

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

  566.         }

  567.         ref_offset = (h->ref_list[1][0].mbaff << 4) & (mb_type_col[0] >> 3);

  568. 

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

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

  571.             assert(h->sps.direct_8x8_inference_flag);

  572. 

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

  574.                 const int x8 = i8 & 1;

  575.                 const int y8 = i8 >> 1;

  576.                 int ref0, scale;

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

  578. 

  579.                 if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))

  580.                     continue;

  581.                 sl->sub_mb_type[i8] = sub_mb_type;

  582. 

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

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

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

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

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

  588.                     continue;

  589.                 }

  590. 

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

  592.                 if (ref0 >= 0)

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

  594.                 else {

  595.                     ref0 = map_col_to_list0[1][l1ref1[x8 + y8 * b8_stride] +

  596.                                                ref_offset];

  597.                     l1mv = l1mv1;

  598.                 }

  599.                 scale = dist_scale_factor[ref0];

  600.                 fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,

  601.                                ref0, 1);

  602. 

  603.                 {

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

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

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

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

  608.                     fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,

  609.                                    pack16to32(mx, my), 4);

  610.                     fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,

  611.                                    pack16to32(mx - mv_col[0], my - my_col), 4);

  612.                 }

  613.             }

  614.             return;

  615.         }

  616. 

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

  618. 

  619.         if (IS_16X16(*mb_type)) {

  620.             int ref, mv0, mv1;

  621. 

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

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

  624.                 ref = mv0 = mv1 = 0;

  625.             } else {

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

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

  628.                 const int scale = dist_scale_factor[ref0];

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

  630.                 int mv_l0[2];

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

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

  633.                 ref      = ref0;

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

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

  636.             }

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

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

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

  640.         } else {

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

  642.                 const int x8 = i8 & 1;

  643.                 const int y8 = i8 >> 1;

  644.                 int ref0, scale;

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

  646. 

  647.                 if (is_b8x8 && !IS_DIRECT(sl->sub_mb_type[i8]))

  648.                     continue;

  649.                 sl->sub_mb_type[i8] = sub_mb_type;

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

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

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

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

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

  655.                     continue;

  656.                 }

  657. 

  658.                 assert(b8_stride == 2);

  659.                 ref0 = l1ref0[i8];

  660.                 if (ref0 >= 0)

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

  662.                 else {

  663.                     ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];

  664.                     l1mv = l1mv1;

  665.                 }

  666.                 scale = dist_scale_factor[ref0];

  667. 

  668.                 fill_rectangle(&sl->ref_cache[0][scan8[i8 * 4]], 2, 2, 8,

  669.                                ref0, 1);

  670.                 if (IS_SUB_8X8(sub_mb_type)) {

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

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

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

  674.                     fill_rectangle(&sl->mv_cache[0][scan8[i8 * 4]], 2, 2, 8,

  675.                                    pack16to32(mx, my), 4);

  676.                     fill_rectangle(&sl->mv_cache[1][scan8[i8 * 4]], 2, 2, 8,

  677.                                    pack16to32(mx - mv_col[0], my - mv_col[1]), 4);

  678.                 } else {

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

  680.                         const int16_t *mv_col = l1mv[x8 * 2 + (i4 & 1) +

  681.                                                      (y8 * 2 + (i4 >> 1)) * b4_stride];

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

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

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

  685.                         AV_WN32A(sl->mv_cache[1][scan8[i8 * 4 + i4]],

  686.                                  pack16to32(mv_l0[0] - mv_col[0],

  687.                                             mv_l0[1] - mv_col[1]));

  688.                     }

  689.                 }

  690.             }

  691.         }

  692.     }

  693. }

  694. 

  695. void ff_h264_pred_direct_motion(H264Context *const h, H264SliceContext *sl,

  696.                                 int *mb_type)

  697. {

  698.     if (h->direct_spatial_mv_pred)

  699.         pred_spatial_direct_motion(h, sl, mb_type);

  700.     else

  701.         pred_temp_direct_motion(h, sl, mb_type);

  702. }