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.
64885 Commits
32 Branches
1012MB
Tree: 560b0253ff
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '560b0253ff'
${ noResults }
FFmpeg/libavcodec/hevc_mvs.c

833 lines
28KB
Raw Blame History 

  1. /*

  2.  * HEVC video decoder

  3.  *

  4.  * Copyright (C) 2012 - 2013 Guillaume Martres

  5.  * Copyright (C) 2013 Anand Meher Kotra

  6.  *

  7.  * This file is part of FFmpeg.

  8.  *

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

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

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

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

  13.  *

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

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

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

  17.  * Lesser General Public License for more details.

  18.  *

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

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

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

  22.  */

  23. 

  24. #include "hevc.h"

  25. 

  26. static const uint8_t l0_l1_cand_idx[12][2] = {

  27.     { 0, 1, },

  28.     { 1, 0, },

  29.     { 0, 2, },

  30.     { 2, 0, },

  31.     { 1, 2, },

  32.     { 2, 1, },

  33.     { 0, 3, },

  34.     { 3, 0, },

  35.     { 1, 3, },

  36.     { 3, 1, },

  37.     { 2, 3, },

  38.     { 3, 2, },

  39. };

  40. 

  41. void ff_hevc_set_neighbour_available(HEVCContext *s, int x0, int y0,

  42.                                      int nPbW, int nPbH)

  43. {

  44.     HEVCLocalContext *lc = s->HEVClc;

  45.     int x0b = x0 & ((1 << s->sps->log2_ctb_size) - 1);

  46.     int y0b = y0 & ((1 << s->sps->log2_ctb_size) - 1);

  47. 

  48.     lc->na.cand_up       = (lc->ctb_up_flag   || y0b);

  49.     lc->na.cand_left     = (lc->ctb_left_flag || x0b);

  50.     lc->na.cand_up_left  = (!x0b && !y0b) ? lc->ctb_up_left_flag : lc->na.cand_left && lc->na.cand_up;

  51.     lc->na.cand_up_right_sap =

  52.             ((x0b + nPbW) == (1 << s->sps->log2_ctb_size)) ?

  53.                     lc->ctb_up_right_flag && !y0b : lc->na.cand_up;

  54.     lc->na.cand_up_right =

  55.             ((x0b + nPbW) == (1 << s->sps->log2_ctb_size) ?

  56.                     lc->ctb_up_right_flag && !y0b : lc->na.cand_up )

  57.                      && (x0 + nPbW) < lc->end_of_tiles_x;

  58.     lc->na.cand_bottom_left = ((y0 + nPbH) >= lc->end_of_tiles_y) ? 0 : lc->na.cand_left;

  59. }

  60. 

  61. /*

  62.  * 6.4.1 Derivation process for z-scan order block availability

  63.  */

  64. static int z_scan_block_avail(HEVCContext *s, int xCurr, int yCurr,

  65.                               int xN, int yN)

  66. {

  67. #define MIN_TB_ADDR_ZS(x, y)                                            \

  68.     s->pps->min_tb_addr_zs[(y) * (s->sps->tb_mask+2) + (x)]

  69. 

  70.     int xCurr_ctb = xCurr >> s->sps->log2_ctb_size;

  71.     int yCurr_ctb = yCurr >> s->sps->log2_ctb_size;

  72.     int xN_ctb    = xN    >> s->sps->log2_ctb_size;

  73.     int yN_ctb    = yN    >> s->sps->log2_ctb_size;

  74. 

  75.     if (xN < 0 || yN < 0 ||

  76.         xN >= s->sps->width ||

  77.         yN >= s->sps->height)

  78.         return 0;

  79. 

  80.     if( yN_ctb < yCurr_ctb || xN_ctb < xCurr_ctb )

  81.         return 1;

  82.     else {

  83.         int Curr = MIN_TB_ADDR_ZS((xCurr >> s->sps->log2_min_tb_size) & s->sps->tb_mask,

  84.                 (yCurr >> s->sps->log2_min_tb_size) & s->sps->tb_mask);

  85.         int N    = MIN_TB_ADDR_ZS((xN >> s->sps->log2_min_tb_size) & s->sps->tb_mask,

  86.                 (yN >> s->sps->log2_min_tb_size) & s->sps->tb_mask);

  87.         return N <= Curr;

  88.     }

  89. }

  90. 

  91. static int same_prediction_block(HEVCLocalContext *lc, int log2_cb_size,

  92.                                  int x0, int y0, int nPbW, int nPbH,

  93.                                  int xA1, int yA1, int partIdx)

  94. {

  95.     return !(nPbW << 1 == 1 << log2_cb_size &&

  96.              nPbH << 1 == 1 << log2_cb_size && partIdx == 1 &&

  97.              lc->cu.x + nPbW > xA1 &&

  98.              lc->cu.y + nPbH <= yA1);

  99. }

  100. 

  101. /*

  102.  * 6.4.2 Derivation process for prediction block availability

  103.  */

  104. static int check_prediction_block_available(HEVCContext *s, int log2_cb_size,

  105.                                             int x0, int y0, int nPbW, int nPbH,

  106.                                             int xA1, int yA1, int partIdx)

  107. {

  108.     HEVCLocalContext *lc = s->HEVClc;

  109. 

  110.     if (lc->cu.x < xA1 && lc->cu.y < yA1 &&

  111.         (lc->cu.x + (1 << log2_cb_size)) > xA1 &&

  112.         (lc->cu.y + (1 << log2_cb_size)) > yA1)

  113.         return same_prediction_block(lc, log2_cb_size, x0, y0,

  114.                                      nPbW, nPbH, xA1, yA1, partIdx);

  115.     else

  116.         return z_scan_block_avail(s, x0, y0, xA1, yA1);

  117. }

  118. 

  119. //check if the two luma locations belong to the same mostion estimation region

  120. static int isDiffMER(HEVCContext *s, int xN, int yN, int xP, int yP)

  121. {

  122.     uint8_t plevel = s->pps->log2_parallel_merge_level;

  123. 

  124.     return xN >> plevel == xP >> plevel &&

  125.            yN >> plevel == yP >> plevel;

  126. }

  127. 

  128. #define MATCH_MV(x) (AV_RN32A(&A.x) == AV_RN32A(&B.x))

  129. #define MATCH(x) (A.x == B.x)

  130. 

  131. // check if the mv's and refidx are the same between A and B

  132. static int compareMVrefidx(struct MvField A, struct MvField B)

  133. {

  134.     int a_pf = A.pred_flag;

  135.     int b_pf = B.pred_flag;

  136.     if (a_pf == b_pf) {

  137.         if (a_pf == PF_BI) {

  138.             return MATCH(ref_idx[0]) && MATCH_MV(mv[0]) &&

  139.                    MATCH(ref_idx[1]) && MATCH_MV(mv[1]);

  140.         } else if (a_pf == PF_L0) {

  141.             return MATCH(ref_idx[0]) && MATCH_MV(mv[0]);

  142.         } else if (a_pf == PF_L1) {

  143.             return MATCH(ref_idx[1]) && MATCH_MV(mv[1]);

  144.         }

  145.     }

  146.     return 0;

  147. }

  148. 

  149. static av_always_inline void mv_scale(Mv *dst, Mv *src, int td, int tb)

  150. {

  151.     int tx, scale_factor;

  152. 

  153.     td = av_clip_int8(td);

  154.     tb = av_clip_int8(tb);

  155.     tx = (0x4000 + abs(td / 2)) / td;

  156.     scale_factor = av_clip((tb * tx + 32) >> 6, -4096, 4095);

  157.     dst->x = av_clip_int16((scale_factor * src->x + 127 +

  158.                            (scale_factor * src->x < 0)) >> 8);

  159.     dst->y = av_clip_int16((scale_factor * src->y + 127 +

  160.                            (scale_factor * src->y < 0)) >> 8);

  161. }

  162. 

  163. static int check_mvset(Mv *mvLXCol, Mv *mvCol,

  164.                        int colPic, int poc,

  165.                        RefPicList *refPicList, int X, int refIdxLx,

  166.                        RefPicList *refPicList_col, int listCol, int refidxCol)

  167. {

  168.     int cur_lt = refPicList[X].isLongTerm[refIdxLx];

  169.     int col_lt = refPicList_col[listCol].isLongTerm[refidxCol];

  170.     int col_poc_diff, cur_poc_diff;

  171. 

  172.     if (cur_lt != col_lt) {

  173.         mvLXCol->x = 0;

  174.         mvLXCol->y = 0;

  175.         return 0;

  176.     }

  177. 

  178.     col_poc_diff = colPic - refPicList_col[listCol].list[refidxCol];

  179.     cur_poc_diff = poc    - refPicList[X].list[refIdxLx];

  180. 

  181.     if (cur_lt || col_poc_diff == cur_poc_diff || !col_poc_diff) {

  182.         mvLXCol->x = mvCol->x;

  183.         mvLXCol->y = mvCol->y;

  184.     } else {

  185.         mv_scale(mvLXCol, mvCol, col_poc_diff, cur_poc_diff);

  186.     }

  187.     return 1;

  188. }

  189. 

  190. #define CHECK_MVSET(l)                                          \

  191.     check_mvset(mvLXCol, temp_col.mv + l,                       \

  192.                 colPic, s->poc,                                 \

  193.                 refPicList, X, refIdxLx,                        \

  194.                 refPicList_col, L ## l, temp_col.ref_idx[l])

  195. 

  196. // derive the motion vectors section 8.5.3.1.8

  197. static int derive_temporal_colocated_mvs(HEVCContext *s, MvField temp_col,

  198.                                          int refIdxLx, Mv *mvLXCol, int X,

  199.                                          int colPic, RefPicList *refPicList_col)

  200. {

  201.     RefPicList *refPicList = s->ref->refPicList;

  202. 

  203.     if (temp_col.pred_flag == PF_INTRA)

  204.         return 0;

  205. 

  206.     if (!(temp_col.pred_flag & PF_L0))

  207.         return CHECK_MVSET(1);

  208.     else if (temp_col.pred_flag == PF_L0)

  209.         return CHECK_MVSET(0);

  210.     else if (temp_col.pred_flag == PF_BI) {

  211.         int check_diffpicount = 0;

  212.         int i, j;

  213.         for (j = 0; j < 2; j++) {

  214.             for (i = 0; i < refPicList[j].nb_refs; i++) {

  215.                 if (refPicList[j].list[i] > s->poc) {

  216.                     check_diffpicount++;

  217.                     break;

  218.                 }

  219.             }

  220.         }

  221.         if (!check_diffpicount) {

  222.             if (X==0)

  223.                 return CHECK_MVSET(0);

  224.             else

  225.                 return CHECK_MVSET(1);

  226.         } else {

  227.             if (s->sh.collocated_list == L1)

  228.                 return CHECK_MVSET(0);

  229.             else

  230.                 return CHECK_MVSET(1);

  231.         }

  232.     }

  233. 

  234.     return 0;

  235. }

  236. 

  237. #define TAB_MVF(x, y)                                                   \

  238.     tab_mvf[(y) * min_pu_width + x]

  239. 

  240. #define TAB_MVF_PU(v)                                                   \

  241.     TAB_MVF(x ## v ## _pu, y ## v ## _pu)

  242. 

  243. #define DERIVE_TEMPORAL_COLOCATED_MVS                                   \

  244.     derive_temporal_colocated_mvs(s, temp_col,                          \

  245.                                   refIdxLx, mvLXCol, X, colPic,         \

  246.                                   ff_hevc_get_ref_list(s, ref, x, y))

  247. 

  248. /*

  249.  * 8.5.3.1.7  temporal luma motion vector prediction

  250.  */

  251. static int temporal_luma_motion_vector(HEVCContext *s, int x0, int y0,

  252.                                        int nPbW, int nPbH, int refIdxLx,

  253.                                        Mv *mvLXCol, int X)

  254. {

  255.     MvField *tab_mvf;

  256.     MvField temp_col;

  257.     int x, y, x_pu, y_pu;

  258.     int min_pu_width = s->sps->min_pu_width;

  259.     int availableFlagLXCol = 0;

  260.     int colPic;

  261. 

  262.     HEVCFrame *ref = s->ref->collocated_ref;

  263. 

  264.     if (!ref)

  265.         return 0;

  266. 

  267.     tab_mvf = ref->tab_mvf;

  268.     colPic  = ref->poc;

  269. 

  270.     //bottom right collocated motion vector

  271.     x = x0 + nPbW;

  272.     y = y0 + nPbH;

  273. 

  274.     if (s->threads_type == FF_THREAD_FRAME )

  275.         ff_thread_await_progress(&ref->tf, y, 0);

  276.     if (tab_mvf &&

  277.         (y0 >> s->sps->log2_ctb_size) == (y >> s->sps->log2_ctb_size) &&

  278.         y < s->sps->height &&

  279.         x < s->sps->width) {

  280.         x                 &= -16;

  281.         y                 &= -16;

  282.         x_pu               = x >> s->sps->log2_min_pu_size;

  283.         y_pu               = y >> s->sps->log2_min_pu_size;

  284.         temp_col           = TAB_MVF(x_pu, y_pu);

  285.         availableFlagLXCol = DERIVE_TEMPORAL_COLOCATED_MVS;

  286.     }

  287. 

  288.     // derive center collocated motion vector

  289.     if (tab_mvf && !availableFlagLXCol) {

  290.         x                  = x0 + (nPbW >> 1);

  291.         y                  = y0 + (nPbH >> 1);

  292.         x                 &= -16;

  293.         y                 &= -16;

  294.         x_pu               = x >> s->sps->log2_min_pu_size;

  295.         y_pu               = y >> s->sps->log2_min_pu_size;

  296.         temp_col           = TAB_MVF(x_pu, y_pu);

  297.         availableFlagLXCol = DERIVE_TEMPORAL_COLOCATED_MVS;

  298.     }

  299.     return availableFlagLXCol;

  300. }

  301. 

  302. #define AVAILABLE(cand, v)                                      \

  303.     (cand && !(TAB_MVF_PU(v).pred_flag == PF_INTRA))

  304. 

  305. #define PRED_BLOCK_AVAILABLE(v)                                 \

  306.     check_prediction_block_available(s, log2_cb_size,           \

  307.                                      x0, y0, nPbW, nPbH,        \

  308.                                      x ## v, y ## v, part_idx)

  309. 

  310. #define COMPARE_MV_REFIDX(a, b)                                 \

  311.     compareMVrefidx(TAB_MVF_PU(a), TAB_MVF_PU(b))

  312. 

  313. /*

  314.  * 8.5.3.1.2  Derivation process for spatial merging candidates

  315.  */

  316. static void derive_spatial_merge_candidates(HEVCContext *s, int x0, int y0,

  317.                                             int nPbW, int nPbH,

  318.                                             int log2_cb_size,

  319.                                             int singleMCLFlag, int part_idx,

  320.                                             int merge_idx,

  321.                                             struct MvField mergecandlist[])

  322. {

  323.     HEVCLocalContext *lc   = s->HEVClc;

  324.     RefPicList *refPicList = s->ref->refPicList;

  325.     MvField *tab_mvf       = s->ref->tab_mvf;

  326. 

  327.     const int min_pu_width = s->sps->min_pu_width;

  328. 

  329.     const int cand_bottom_left = lc->na.cand_bottom_left;

  330.     const int cand_left        = lc->na.cand_left;

  331.     const int cand_up_left     = lc->na.cand_up_left;

  332.     const int cand_up          = lc->na.cand_up;

  333.     const int cand_up_right    = lc->na.cand_up_right_sap;

  334. 

  335.     const int xA1    = x0 - 1;

  336.     const int yA1    = y0 + nPbH - 1;

  337.     const int xA1_pu = xA1 >> s->sps->log2_min_pu_size;

  338.     const int yA1_pu = yA1 >> s->sps->log2_min_pu_size;

  339. 

  340.     const int xB1    = x0 + nPbW - 1;

  341.     const int yB1    = y0 - 1;

  342.     const int xB1_pu = xB1 >> s->sps->log2_min_pu_size;

  343.     const int yB1_pu = yB1 >> s->sps->log2_min_pu_size;

  344. 

  345.     const int xB0    = x0 + nPbW;

  346.     const int yB0    = y0 - 1;

  347.     const int xB0_pu = xB0 >> s->sps->log2_min_pu_size;

  348.     const int yB0_pu = yB0 >> s->sps->log2_min_pu_size;

  349. 

  350.     const int xA0    = x0 - 1;

  351.     const int yA0    = y0 + nPbH;

  352.     const int xA0_pu = xA0 >> s->sps->log2_min_pu_size;

  353.     const int yA0_pu = yA0 >> s->sps->log2_min_pu_size;

  354. 

  355.     const int xB2    = x0 - 1;

  356.     const int yB2    = y0 - 1;

  357.     const int xB2_pu = xB2 >> s->sps->log2_min_pu_size;

  358.     const int yB2_pu = yB2 >> s->sps->log2_min_pu_size;

  359. 

  360.     const int nb_refs = (s->sh.slice_type == P_SLICE) ?

  361.                         s->sh.nb_refs[0] : FFMIN(s->sh.nb_refs[0], s->sh.nb_refs[1]);

  362.     int check_MER   = 1;

  363.     int check_MER_1 = 1;

  364. 

  365.     int zero_idx = 0;

  366. 

  367.     int nb_merge_cand = 0;

  368.     int nb_orig_merge_cand = 0;

  369. 

  370.     int is_available_a0;

  371.     int is_available_a1;

  372.     int is_available_b0;

  373.     int is_available_b1;

  374.     int is_available_b2;

  375.     int check_B0;

  376.     int check_A0;

  377. 

  378.     //first left spatial merge candidate

  379.     is_available_a1 = AVAILABLE(cand_left, A1);

  380. 

  381.     if (!singleMCLFlag && part_idx == 1 &&

  382.         (lc->cu.part_mode == PART_Nx2N ||

  383.          lc->cu.part_mode == PART_nLx2N ||

  384.          lc->cu.part_mode == PART_nRx2N) ||

  385.         isDiffMER(s, xA1, yA1, x0, y0)) {

  386.         is_available_a1 = 0;

  387.     }

  388. 

  389.     if (is_available_a1) {

  390.         mergecandlist[0] = TAB_MVF_PU(A1);

  391.         if (merge_idx == 0) return;

  392.         nb_merge_cand++;

  393.     }

  394. 

  395.     // above spatial merge candidate

  396.     is_available_b1 = AVAILABLE(cand_up, B1);

  397. 

  398.     if (!singleMCLFlag && part_idx == 1 &&

  399.         (lc->cu.part_mode == PART_2NxN ||

  400.          lc->cu.part_mode == PART_2NxnU ||

  401.          lc->cu.part_mode == PART_2NxnD) ||

  402.         isDiffMER(s, xB1, yB1, x0, y0)) {

  403.         is_available_b1 = 0;

  404.     }

  405. 

  406.     if (is_available_a1 && is_available_b1)

  407.         check_MER = !COMPARE_MV_REFIDX(B1, A1);

  408. 

  409.     if (is_available_b1 && check_MER)

  410.         mergecandlist[nb_merge_cand++] = TAB_MVF_PU(B1);

  411. 

  412.     // above right spatial merge candidate

  413.     check_MER = 1;

  414.     check_B0  = PRED_BLOCK_AVAILABLE(B0);

  415. 

  416.     is_available_b0 = check_B0 && AVAILABLE(cand_up_right, B0);

  417. 

  418.     if (isDiffMER(s, xB0, yB0, x0, y0))

  419.         is_available_b0 = 0;

  420. 

  421.     if (is_available_b1 && is_available_b0)

  422.         check_MER = !COMPARE_MV_REFIDX(B0, B1);

  423. 

  424.     if (is_available_b0 && check_MER) {

  425.         mergecandlist[nb_merge_cand] = TAB_MVF_PU(B0);

  426.         if (merge_idx == nb_merge_cand) return;

  427.         nb_merge_cand++;

  428.     }

  429. 

  430.     // left bottom spatial merge candidate

  431.     check_MER = 1;

  432.     check_A0  = PRED_BLOCK_AVAILABLE(A0);

  433. 

  434.     is_available_a0 = check_A0 && AVAILABLE(cand_bottom_left, A0);

  435. 

  436.     if (isDiffMER(s, xA0, yA0, x0, y0))

  437.         is_available_a0 = 0;

  438. 

  439.     if (is_available_a1 && is_available_a0)

  440.         check_MER = !COMPARE_MV_REFIDX(A0, A1);

  441. 

  442.     if (is_available_a0 && check_MER) {

  443.         mergecandlist[nb_merge_cand] = TAB_MVF_PU(A0);

  444.         if (merge_idx == nb_merge_cand) return;

  445.         nb_merge_cand++;

  446.     }

  447. 

  448.     // above left spatial merge candidate

  449.     check_MER = 1;

  450. 

  451.     is_available_b2 = AVAILABLE(cand_up_left, B2);

  452. 

  453.     if (isDiffMER(s, xB2, yB2, x0, y0))

  454.         is_available_b2 = 0;

  455. 

  456.     if (is_available_a1 && is_available_b2)

  457.         check_MER = !COMPARE_MV_REFIDX(B2, A1);

  458. 

  459.     if (is_available_b1 && is_available_b2)

  460.         check_MER_1 = !COMPARE_MV_REFIDX(B2, B1);

  461. 

  462.     if (is_available_b2 && check_MER && check_MER_1 && nb_merge_cand != 4) {

  463.         mergecandlist[nb_merge_cand] = TAB_MVF_PU(B2);

  464.         if (merge_idx == nb_merge_cand) return;

  465.         nb_merge_cand++;

  466.     }

  467. 

  468.     // temporal motion vector candidate

  469.     if (s->sh.slice_temporal_mvp_enabled_flag &&

  470.         nb_merge_cand < s->sh.max_num_merge_cand) {

  471.         Mv mv_l0_col, mv_l1_col;

  472.         int available_l0 = temporal_luma_motion_vector(s, x0, y0, nPbW, nPbH,

  473.                                                        0, &mv_l0_col, 0);

  474.         int available_l1 = (s->sh.slice_type == B_SLICE) ?

  475.                            temporal_luma_motion_vector(s, x0, y0, nPbW, nPbH,

  476.                                                        0, &mv_l1_col, 1) : 0;

  477. 

  478.         if (available_l0 || available_l1) {

  479.             mergecandlist[nb_merge_cand].pred_flag = available_l0 + (available_l1 << 1);

  480.             if (available_l0) {

  481.                 mergecandlist[nb_merge_cand].mv[0]      = mv_l0_col;

  482.                 mergecandlist[nb_merge_cand].ref_idx[0] = 0;

  483.             }

  484.             if (available_l1) {

  485.                 mergecandlist[nb_merge_cand].mv[1]      = mv_l1_col;

  486.                 mergecandlist[nb_merge_cand].ref_idx[1] = 0;

  487.             }

  488.             if (merge_idx == nb_merge_cand) return;

  489.             nb_merge_cand++;

  490.         }

  491.     }

  492. 

  493.     nb_orig_merge_cand = nb_merge_cand;

  494. 

  495.     // combined bi-predictive merge candidates  (applies for B slices)

  496.     if (s->sh.slice_type == B_SLICE && nb_orig_merge_cand > 1 &&

  497.         nb_orig_merge_cand < s->sh.max_num_merge_cand) {

  498.         int comb_idx = 0;

  499. 

  500.         for (comb_idx = 0; nb_merge_cand < s->sh.max_num_merge_cand &&

  501.                            comb_idx < nb_orig_merge_cand * (nb_orig_merge_cand - 1); comb_idx++) {

  502.             int l0_cand_idx = l0_l1_cand_idx[comb_idx][0];

  503.             int l1_cand_idx = l0_l1_cand_idx[comb_idx][1];

  504.             MvField l0_cand = mergecandlist[l0_cand_idx];

  505.             MvField l1_cand = mergecandlist[l1_cand_idx];

  506. 

  507.             if ((l0_cand.pred_flag & PF_L0) && (l1_cand.pred_flag & PF_L1) &&

  508.                 (refPicList[0].list[l0_cand.ref_idx[0]] !=

  509.                  refPicList[1].list[l1_cand.ref_idx[1]] ||

  510.                  AV_RN32A(&l0_cand.mv[0]) != AV_RN32A(&l1_cand.mv[1]))) {

  511.                 mergecandlist[nb_merge_cand].ref_idx[0]   = l0_cand.ref_idx[0];

  512.                 mergecandlist[nb_merge_cand].ref_idx[1]   = l1_cand.ref_idx[1];

  513.                 mergecandlist[nb_merge_cand].pred_flag    = PF_BI;

  514.                 AV_COPY32(&mergecandlist[nb_merge_cand].mv[0], &l0_cand.mv[0]);

  515.                 AV_COPY32(&mergecandlist[nb_merge_cand].mv[1], &l1_cand.mv[1]);

  516.                 if (merge_idx == nb_merge_cand) return;

  517.                 nb_merge_cand++;

  518.             }

  519.         }

  520.     }

  521. 

  522.     // append Zero motion vector candidates

  523.     while (nb_merge_cand < s->sh.max_num_merge_cand) {

  524.         mergecandlist[nb_merge_cand].pred_flag    = PF_L0 + ((s->sh.slice_type == B_SLICE) << 1);

  525.         AV_ZERO32(mergecandlist[nb_merge_cand].mv+0);

  526.         AV_ZERO32(mergecandlist[nb_merge_cand].mv+1);

  527.         mergecandlist[nb_merge_cand].ref_idx[0]   = zero_idx < nb_refs ? zero_idx : 0;

  528.         mergecandlist[nb_merge_cand].ref_idx[1]   = zero_idx < nb_refs ? zero_idx : 0;

  529. 

  530.         if (merge_idx == nb_merge_cand) return;

  531.         nb_merge_cand++;

  532.         zero_idx++;

  533.     }

  534. }

  535. 

  536. /*

  537.  * 8.5.3.1.1 Derivation process of luma Mvs for merge mode

  538.  */

  539. void ff_hevc_luma_mv_merge_mode(HEVCContext *s, int x0, int y0, int nPbW,

  540.                                 int nPbH, int log2_cb_size, int part_idx,

  541.                                 int merge_idx, MvField *mv)

  542. {

  543.     int singleMCLFlag = 0;

  544.     int nCS = 1 << log2_cb_size;

  545.     LOCAL_ALIGNED(4, MvField, mergecand_list, [MRG_MAX_NUM_CANDS]);

  546.     int nPbW2 = nPbW;

  547.     int nPbH2 = nPbH;

  548.     HEVCLocalContext *lc = s->HEVClc;

  549. 

  550.     if (s->pps->log2_parallel_merge_level > 2 && nCS == 8) {

  551.         singleMCLFlag = 1;

  552.         x0            = lc->cu.x;

  553.         y0            = lc->cu.y;

  554.         nPbW          = nCS;

  555.         nPbH          = nCS;

  556.         part_idx      = 0;

  557.     }

  558. 

  559.     ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);

  560.     derive_spatial_merge_candidates(s, x0, y0, nPbW, nPbH, log2_cb_size,

  561.                                     singleMCLFlag, part_idx,

  562.                                     merge_idx, mergecand_list);

  563. 

  564.     if (mergecand_list[merge_idx].pred_flag == PF_BI &&

  565.         (nPbW2 + nPbH2) == 12) {

  566.         mergecand_list[merge_idx].pred_flag = PF_L0;

  567.     }

  568. 

  569.     *mv = mergecand_list[merge_idx];

  570. }

  571. 

  572. static av_always_inline void dist_scale(HEVCContext *s, Mv *mv,

  573.                                         int min_pu_width, int x, int y,

  574.                                         int elist, int ref_idx_curr, int ref_idx)

  575. {

  576.     RefPicList *refPicList = s->ref->refPicList;

  577.     MvField *tab_mvf       = s->ref->tab_mvf;

  578.     int ref_pic_elist      = refPicList[elist].list[TAB_MVF(x, y).ref_idx[elist]];

  579.     int ref_pic_curr       = refPicList[ref_idx_curr].list[ref_idx];

  580. 

  581.     if (ref_pic_elist != ref_pic_curr) {

  582.         int poc_diff = s->poc - ref_pic_elist;

  583.         if (!poc_diff)

  584.             poc_diff = 1;

  585.         mv_scale(mv, mv, poc_diff, s->poc - ref_pic_curr);

  586.     }

  587. }

  588. 

  589. static int mv_mp_mode_mx(HEVCContext *s, int x, int y, int pred_flag_index,

  590.                          Mv *mv, int ref_idx_curr, int ref_idx)

  591. {

  592.     MvField *tab_mvf = s->ref->tab_mvf;

  593.     int min_pu_width = s->sps->min_pu_width;

  594. 

  595.     RefPicList *refPicList = s->ref->refPicList;

  596. 

  597.     if (((TAB_MVF(x, y).pred_flag) & (1 << pred_flag_index)) &&

  598.         refPicList[pred_flag_index].list[TAB_MVF(x, y).ref_idx[pred_flag_index]] == refPicList[ref_idx_curr].list[ref_idx]) {

  599.         *mv = TAB_MVF(x, y).mv[pred_flag_index];

  600.         return 1;

  601.     }

  602.     return 0;

  603. }

  604. 

  605. static int mv_mp_mode_mx_lt(HEVCContext *s, int x, int y, int pred_flag_index,

  606.                             Mv *mv, int ref_idx_curr, int ref_idx)

  607. {

  608.     MvField *tab_mvf = s->ref->tab_mvf;

  609.     int min_pu_width = s->sps->min_pu_width;

  610. 

  611.     RefPicList *refPicList = s->ref->refPicList;

  612. 

  613.     if ((TAB_MVF(x, y).pred_flag) & (1 << pred_flag_index)) {

  614.         int currIsLongTerm     = refPicList[ref_idx_curr].isLongTerm[ref_idx];

  615. 

  616.         int colIsLongTerm =

  617.             refPicList[pred_flag_index].isLongTerm[(TAB_MVF(x, y).ref_idx[pred_flag_index])];

  618. 

  619.         if (colIsLongTerm == currIsLongTerm) {

  620.             *mv = TAB_MVF(x, y).mv[pred_flag_index];

  621.             if (!currIsLongTerm)

  622.                 dist_scale(s, mv, min_pu_width, x, y,

  623.                            pred_flag_index, ref_idx_curr, ref_idx);

  624.             return 1;

  625.         }

  626.     }

  627.     return 0;

  628. }

  629. 

  630. #define MP_MX(v, pred, mx)                                      \

  631.     mv_mp_mode_mx(s, x ## v ## _pu, y ## v ## _pu, pred,        \

  632.                   &mx, ref_idx_curr, ref_idx)

  633. 

  634. #define MP_MX_LT(v, pred, mx)                                   \

  635.     mv_mp_mode_mx_lt(s, x ## v ## _pu, y ## v ## _pu, pred,     \

  636.                      &mx, ref_idx_curr, ref_idx)

  637. 

  638. void ff_hevc_luma_mv_mvp_mode(HEVCContext *s, int x0, int y0, int nPbW,

  639.                               int nPbH, int log2_cb_size, int part_idx,

  640.                               int merge_idx, MvField *mv,

  641.                               int mvp_lx_flag, int LX)

  642. {

  643.     HEVCLocalContext *lc = s->HEVClc;

  644.     MvField *tab_mvf = s->ref->tab_mvf;

  645.     int isScaledFlag_L0 = 0;

  646.     int availableFlagLXA0 = 0;

  647.     int availableFlagLXB0 = 0;

  648.     int numMVPCandLX = 0;

  649.     int min_pu_width = s->sps->min_pu_width;

  650. 

  651.     int xA0, yA0;

  652.     int xA0_pu, yA0_pu;

  653.     int is_available_a0;

  654. 

  655.     int xA1, yA1;

  656.     int xA1_pu, yA1_pu;

  657.     int is_available_a1;

  658. 

  659.     int xB0, yB0;

  660.     int xB0_pu, yB0_pu;

  661.     int is_available_b0;

  662. 

  663.     int xB1, yB1;

  664.     int xB1_pu = 0, yB1_pu = 0;

  665.     int is_available_b1 = 0;

  666. 

  667.     int xB2, yB2;

  668.     int xB2_pu = 0, yB2_pu = 0;

  669.     int is_available_b2 = 0;

  670.     Mv mvpcand_list[2] = { { 0 } };

  671.     Mv mxA;

  672.     Mv mxB;

  673.     int ref_idx_curr = 0;

  674.     int ref_idx = 0;

  675.     int pred_flag_index_l0;

  676.     int pred_flag_index_l1;

  677.     int x0b = x0 & ((1 << s->sps->log2_ctb_size) - 1);

  678.     int y0b = y0 & ((1 << s->sps->log2_ctb_size) - 1);

  679. 

  680.     int cand_up = (lc->ctb_up_flag || y0b);

  681.     int cand_left = (lc->ctb_left_flag || x0b);

  682.     int cand_up_left =

  683.             (!x0b && !y0b) ? lc->ctb_up_left_flag : cand_left && cand_up;

  684.     int cand_up_right =

  685.             (x0b + nPbW == (1 << s->sps->log2_ctb_size) ||

  686.              x0  + nPbW >= lc->end_of_tiles_x) ? lc->ctb_up_right_flag && !y0b

  687.                                                : cand_up;

  688.     int cand_bottom_left = (y0 + nPbH >= lc->end_of_tiles_y) ? 0 : cand_left;

  689. 

  690.     ref_idx_curr       = LX;

  691.     ref_idx            = mv->ref_idx[LX];

  692.     pred_flag_index_l0 = LX;

  693.     pred_flag_index_l1 = !LX;

  694. 

  695.     // left bottom spatial candidate

  696.     xA0 = x0 - 1;

  697.     yA0 = y0 + nPbH;

  698.     xA0_pu = xA0 >> s->sps->log2_min_pu_size;

  699.     yA0_pu = yA0 >> s->sps->log2_min_pu_size;

  700. 

  701.     is_available_a0 = PRED_BLOCK_AVAILABLE(A0) && AVAILABLE(cand_bottom_left, A0);

  702. 

  703.     //left spatial merge candidate

  704.     xA1    = x0 - 1;

  705.     yA1    = y0 + nPbH - 1;

  706.     xA1_pu = xA1 >> s->sps->log2_min_pu_size;

  707.     yA1_pu = yA1 >> s->sps->log2_min_pu_size;

  708. 

  709.     is_available_a1 = AVAILABLE(cand_left, A1);

  710.     if (is_available_a0 || is_available_a1)

  711.         isScaledFlag_L0 = 1;

  712. 

  713.     if (is_available_a0) {

  714.         availableFlagLXA0 = MP_MX(A0, pred_flag_index_l0, mxA);

  715.         if (!availableFlagLXA0)

  716.             availableFlagLXA0 = MP_MX(A0, pred_flag_index_l1, mxA);

  717.     }

  718. 

  719.     if (is_available_a1 && !availableFlagLXA0) {

  720.         availableFlagLXA0 = MP_MX(A1, pred_flag_index_l0, mxA);

  721.         if (!availableFlagLXA0)

  722.             availableFlagLXA0 = MP_MX(A1, pred_flag_index_l1, mxA);

  723.     }

  724. 

  725.     if (is_available_a0 && !availableFlagLXA0) {

  726.         availableFlagLXA0 = MP_MX_LT(A0, pred_flag_index_l0, mxA);

  727.         if (!availableFlagLXA0)

  728.             availableFlagLXA0 = MP_MX_LT(A0, pred_flag_index_l1, mxA);

  729.     }

  730. 

  731.     if (is_available_a1 && !availableFlagLXA0) {

  732.         availableFlagLXA0 = MP_MX_LT(A1, pred_flag_index_l0, mxA);

  733.         if (!availableFlagLXA0)

  734.             availableFlagLXA0 = MP_MX_LT(A1, pred_flag_index_l1, mxA);

  735.     }

  736. 

  737.     if(availableFlagLXA0 && !mvp_lx_flag) {

  738.         mv->mv[LX] = mxA;

  739.         return;

  740.     }

  741. 

  742.     // B candidates

  743.     // above right spatial merge candidate

  744.     xB0    = x0 + nPbW;

  745.     yB0    = y0 - 1;

  746.     xB0_pu = xB0 >> s->sps->log2_min_pu_size;

  747.     yB0_pu = yB0 >> s->sps->log2_min_pu_size;

  748. 

  749.     is_available_b0 = PRED_BLOCK_AVAILABLE(B0) && AVAILABLE(cand_up_right, B0);

  750. 

  751.     if (is_available_b0) {

  752.         availableFlagLXB0 = MP_MX(B0, pred_flag_index_l0, mxB);

  753.         if (!availableFlagLXB0)

  754.             availableFlagLXB0 = MP_MX(B0, pred_flag_index_l1, mxB);

  755.     }

  756. 

  757.     if (!availableFlagLXB0) {

  758.         // above spatial merge candidate

  759.         xB1    = x0 + nPbW - 1;

  760.         yB1    = y0 - 1;

  761.         xB1_pu = xB1 >> s->sps->log2_min_pu_size;

  762.         yB1_pu = yB1 >> s->sps->log2_min_pu_size;

  763. 

  764.         is_available_b1 = AVAILABLE(cand_up, B1);

  765. 

  766.         if (is_available_b1) {

  767.             availableFlagLXB0 = MP_MX(B1, pred_flag_index_l0, mxB);

  768.             if (!availableFlagLXB0)

  769.                 availableFlagLXB0 = MP_MX(B1, pred_flag_index_l1, mxB);

  770.         }

  771.     }

  772. 

  773.     if (!availableFlagLXB0) {

  774.         // above left spatial merge candidate

  775.         xB2 = x0 - 1;

  776.         yB2 = y0 - 1;

  777.         xB2_pu = xB2 >> s->sps->log2_min_pu_size;

  778.         yB2_pu = yB2 >> s->sps->log2_min_pu_size;

  779.         is_available_b2 = AVAILABLE(cand_up_left, B2);

  780. 

  781.         if (is_available_b2) {

  782.             availableFlagLXB0 = MP_MX(B2, pred_flag_index_l0, mxB);

  783.             if (!availableFlagLXB0)

  784.                 availableFlagLXB0 = MP_MX(B2, pred_flag_index_l1, mxB);

  785.         }

  786.     }

  787. 

  788.     if (isScaledFlag_L0 == 0) {

  789.         if (availableFlagLXB0) {

  790.             availableFlagLXA0 = 1;

  791.             mxA = mxB;

  792.         }

  793.         availableFlagLXB0 = 0;

  794. 

  795.         // XB0 and L1

  796.         if (is_available_b0) {

  797.             availableFlagLXB0 = MP_MX_LT(B0, pred_flag_index_l0, mxB);

  798.             if (!availableFlagLXB0)

  799.                 availableFlagLXB0 = MP_MX_LT(B0, pred_flag_index_l1, mxB);

  800.         }

  801. 

  802.         if (is_available_b1 && !availableFlagLXB0) {

  803.             availableFlagLXB0 = MP_MX_LT(B1, pred_flag_index_l0, mxB);

  804.             if (!availableFlagLXB0)

  805.                 availableFlagLXB0 = MP_MX_LT(B1, pred_flag_index_l1, mxB);

  806.         }

  807. 

  808.         if (is_available_b2 && !availableFlagLXB0) {

  809.             availableFlagLXB0 = MP_MX_LT(B2, pred_flag_index_l0, mxB);

  810.             if (!availableFlagLXB0)

  811.                 availableFlagLXB0 = MP_MX_LT(B2, pred_flag_index_l1, mxB);

  812.         }

  813.     }

  814. 

  815.     if (availableFlagLXA0)

  816.         mvpcand_list[numMVPCandLX++] = mxA;

  817. 

  818.     if (availableFlagLXB0 && (!availableFlagLXA0 || mxA.x != mxB.x || mxA.y != mxB.y))

  819.         mvpcand_list[numMVPCandLX++] = mxB;

  820. 

  821.     //temporal motion vector prediction candidate

  822.     if (numMVPCandLX < 2 && s->sh.slice_temporal_mvp_enabled_flag) {

  823.         Mv mv_col;

  824.         int available_col = temporal_luma_motion_vector(s, x0, y0, nPbW,

  825.                                                         nPbH, ref_idx,

  826.                                                         &mv_col, LX);

  827.         if (available_col)

  828.             mvpcand_list[numMVPCandLX++] = mv_col;

  829.     }

  830. 

  831.     mv->mv[LX] = mvpcand_list[mvp_lx_flag];

  832. }