| 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 | lc->na.cand_up_right_sap |
| 56 | && (x0 + nPbW) < lc->end_of_tiles_x; |
| 57 | lc->na.cand_bottom_left = ((y0 + nPbH) >= lc->end_of_tiles_y) ? 0 : lc->na.cand_left; |
| 58 | } |
| 59 | |
| 60 | /* |
| 61 | * 6.4.1 Derivation process for z-scan order block availability |
| 62 | */ |
| 63 | static av_always_inline int z_scan_block_avail(HEVCContext *s, int xCurr, int yCurr, |
| 64 | int xN, int yN) |
| 65 | { |
| 66 | #define MIN_TB_ADDR_ZS(x, y) \ |
| 67 | s->pps->min_tb_addr_zs[(y) * (s->sps->tb_mask+2) + (x)] |
| 68 | |
| 69 | int xCurr_ctb = xCurr >> s->sps->log2_ctb_size; |
| 70 | int yCurr_ctb = yCurr >> s->sps->log2_ctb_size; |
| 71 | int xN_ctb = xN >> s->sps->log2_ctb_size; |
| 72 | int yN_ctb = yN >> s->sps->log2_ctb_size; |
| 73 | if( yN_ctb < yCurr_ctb || xN_ctb < xCurr_ctb ) |
| 74 | return 1; |
| 75 | else { |
| 76 | int Curr = MIN_TB_ADDR_ZS((xCurr >> s->sps->log2_min_tb_size) & s->sps->tb_mask, |
| 77 | (yCurr >> s->sps->log2_min_tb_size) & s->sps->tb_mask); |
| 78 | int N = MIN_TB_ADDR_ZS((xN >> s->sps->log2_min_tb_size) & s->sps->tb_mask, |
| 79 | (yN >> s->sps->log2_min_tb_size) & s->sps->tb_mask); |
| 80 | return N <= Curr; |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | //check if the two luma locations belong to the same mostion estimation region |
| 85 | static av_always_inline int is_diff_mer(HEVCContext *s, int xN, int yN, int xP, int yP) |
| 86 | { |
| 87 | uint8_t plevel = s->pps->log2_parallel_merge_level; |
| 88 | |
| 89 | return xN >> plevel == xP >> plevel && |
| 90 | yN >> plevel == yP >> plevel; |
| 91 | } |
| 92 | |
| 93 | #define MATCH_MV(x) (AV_RN32A(&A.x) == AV_RN32A(&B.x)) |
| 94 | #define MATCH(x) (A.x == B.x) |
| 95 | |
| 96 | // check if the mv's and refidx are the same between A and B |
| 97 | static av_always_inline int compare_mv_ref_idx(struct MvField A, struct MvField B) |
| 98 | { |
| 99 | int a_pf = A.pred_flag; |
| 100 | int b_pf = B.pred_flag; |
| 101 | if (a_pf == b_pf) { |
| 102 | if (a_pf == PF_BI) { |
| 103 | return MATCH(ref_idx[0]) && MATCH_MV(mv[0]) && |
| 104 | MATCH(ref_idx[1]) && MATCH_MV(mv[1]); |
| 105 | } else if (a_pf == PF_L0) { |
| 106 | return MATCH(ref_idx[0]) && MATCH_MV(mv[0]); |
| 107 | } else if (a_pf == PF_L1) { |
| 108 | return MATCH(ref_idx[1]) && MATCH_MV(mv[1]); |
| 109 | } |
| 110 | } |
| 111 | return 0; |
| 112 | } |
| 113 | |
| 114 | static av_always_inline void mv_scale(Mv *dst, Mv *src, int td, int tb) |
| 115 | { |
| 116 | int tx, scale_factor; |
| 117 | |
| 118 | td = av_clip_int8(td); |
| 119 | tb = av_clip_int8(tb); |
| 120 | tx = (0x4000 + abs(td / 2)) / td; |
| 121 | scale_factor = av_clip((tb * tx + 32) >> 6, -4096, 4095); |
| 122 | dst->x = av_clip_int16((scale_factor * src->x + 127 + |
| 123 | (scale_factor * src->x < 0)) >> 8); |
| 124 | dst->y = av_clip_int16((scale_factor * src->y + 127 + |
| 125 | (scale_factor * src->y < 0)) >> 8); |
| 126 | } |
| 127 | |
| 128 | static int check_mvset(Mv *mvLXCol, Mv *mvCol, |
| 129 | int colPic, int poc, |
| 130 | RefPicList *refPicList, int X, int refIdxLx, |
| 131 | RefPicList *refPicList_col, int listCol, int refidxCol) |
| 132 | { |
| 133 | int cur_lt = refPicList[X].isLongTerm[refIdxLx]; |
| 134 | int col_lt = refPicList_col[listCol].isLongTerm[refidxCol]; |
| 135 | int col_poc_diff, cur_poc_diff; |
| 136 | |
| 137 | if (cur_lt != col_lt) { |
| 138 | mvLXCol->x = 0; |
| 139 | mvLXCol->y = 0; |
| 140 | return 0; |
| 141 | } |
| 142 | |
| 143 | col_poc_diff = colPic - refPicList_col[listCol].list[refidxCol]; |
| 144 | cur_poc_diff = poc - refPicList[X].list[refIdxLx]; |
| 145 | |
| 146 | if (cur_lt || col_poc_diff == cur_poc_diff || !col_poc_diff) { |
| 147 | mvLXCol->x = mvCol->x; |
| 148 | mvLXCol->y = mvCol->y; |
| 149 | } else { |
| 150 | mv_scale(mvLXCol, mvCol, col_poc_diff, cur_poc_diff); |
| 151 | } |
| 152 | return 1; |
| 153 | } |
| 154 | |
| 155 | #define CHECK_MVSET(l) \ |
| 156 | check_mvset(mvLXCol, temp_col.mv + l, \ |
| 157 | colPic, s->poc, \ |
| 158 | refPicList, X, refIdxLx, \ |
| 159 | refPicList_col, L ## l, temp_col.ref_idx[l]) |
| 160 | |
| 161 | // derive the motion vectors section 8.5.3.1.8 |
| 162 | static int derive_temporal_colocated_mvs(HEVCContext *s, MvField temp_col, |
| 163 | int refIdxLx, Mv *mvLXCol, int X, |
| 164 | int colPic, RefPicList *refPicList_col) |
| 165 | { |
| 166 | RefPicList *refPicList = s->ref->refPicList; |
| 167 | |
| 168 | if (temp_col.pred_flag == PF_INTRA) |
| 169 | return 0; |
| 170 | |
| 171 | if (!(temp_col.pred_flag & PF_L0)) |
| 172 | return CHECK_MVSET(1); |
| 173 | else if (temp_col.pred_flag == PF_L0) |
| 174 | return CHECK_MVSET(0); |
| 175 | else if (temp_col.pred_flag == PF_BI) { |
| 176 | int check_diffpicount = 0; |
| 177 | int i, j; |
| 178 | for (j = 0; j < 2; j++) { |
| 179 | for (i = 0; i < refPicList[j].nb_refs; i++) { |
| 180 | if (refPicList[j].list[i] > s->poc) { |
| 181 | check_diffpicount++; |
| 182 | break; |
| 183 | } |
| 184 | } |
| 185 | } |
| 186 | if (!check_diffpicount) { |
| 187 | if (X==0) |
| 188 | return CHECK_MVSET(0); |
| 189 | else |
| 190 | return CHECK_MVSET(1); |
| 191 | } else { |
| 192 | if (s->sh.collocated_list == L1) |
| 193 | return CHECK_MVSET(0); |
| 194 | else |
| 195 | return CHECK_MVSET(1); |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | return 0; |
| 200 | } |
| 201 | |
| 202 | #define TAB_MVF(x, y) \ |
| 203 | tab_mvf[(y) * min_pu_width + x] |
| 204 | |
| 205 | #define TAB_MVF_PU(v) \ |
| 206 | TAB_MVF(((x ## v) >> s->sps->log2_min_pu_size), \ |
| 207 | ((y ## v) >> s->sps->log2_min_pu_size)) |
| 208 | |
| 209 | #define DERIVE_TEMPORAL_COLOCATED_MVS \ |
| 210 | derive_temporal_colocated_mvs(s, temp_col, \ |
| 211 | refIdxLx, mvLXCol, X, colPic, \ |
| 212 | ff_hevc_get_ref_list(s, ref, x, y)) |
| 213 | |
| 214 | /* |
| 215 | * 8.5.3.1.7 temporal luma motion vector prediction |
| 216 | */ |
| 217 | static int temporal_luma_motion_vector(HEVCContext *s, int x0, int y0, |
| 218 | int nPbW, int nPbH, int refIdxLx, |
| 219 | Mv *mvLXCol, int X) |
| 220 | { |
| 221 | MvField *tab_mvf; |
| 222 | MvField temp_col; |
| 223 | int x, y, x_pu, y_pu; |
| 224 | int min_pu_width = s->sps->min_pu_width; |
| 225 | int availableFlagLXCol = 0; |
| 226 | int colPic; |
| 227 | |
| 228 | HEVCFrame *ref = s->ref->collocated_ref; |
| 229 | |
| 230 | if (!ref) { |
| 231 | memset(mvLXCol, 0, sizeof(*mvLXCol)); |
| 232 | return 0; |
| 233 | } |
| 234 | |
| 235 | tab_mvf = ref->tab_mvf; |
| 236 | colPic = ref->poc; |
| 237 | |
| 238 | //bottom right collocated motion vector |
| 239 | x = x0 + nPbW; |
| 240 | y = y0 + nPbH; |
| 241 | |
| 242 | if (tab_mvf && |
| 243 | (y0 >> s->sps->log2_ctb_size) == (y >> s->sps->log2_ctb_size) && |
| 244 | y < s->sps->height && |
| 245 | x < s->sps->width) { |
| 246 | x &= ~15; |
| 247 | y &= ~15; |
| 248 | if (s->threads_type == FF_THREAD_FRAME) |
| 249 | ff_thread_await_progress(&ref->tf, y, 0); |
| 250 | x_pu = x >> s->sps->log2_min_pu_size; |
| 251 | y_pu = y >> s->sps->log2_min_pu_size; |
| 252 | temp_col = TAB_MVF(x_pu, y_pu); |
| 253 | availableFlagLXCol = DERIVE_TEMPORAL_COLOCATED_MVS; |
| 254 | } |
| 255 | |
| 256 | // derive center collocated motion vector |
| 257 | if (tab_mvf && !availableFlagLXCol) { |
| 258 | x = x0 + (nPbW >> 1); |
| 259 | y = y0 + (nPbH >> 1); |
| 260 | x &= ~15; |
| 261 | y &= ~15; |
| 262 | if (s->threads_type == FF_THREAD_FRAME) |
| 263 | ff_thread_await_progress(&ref->tf, y, 0); |
| 264 | x_pu = x >> s->sps->log2_min_pu_size; |
| 265 | y_pu = y >> s->sps->log2_min_pu_size; |
| 266 | temp_col = TAB_MVF(x_pu, y_pu); |
| 267 | availableFlagLXCol = DERIVE_TEMPORAL_COLOCATED_MVS; |
| 268 | } |
| 269 | return availableFlagLXCol; |
| 270 | } |
| 271 | |
| 272 | #define AVAILABLE(cand, v) \ |
| 273 | (cand && !(TAB_MVF_PU(v).pred_flag == PF_INTRA)) |
| 274 | |
| 275 | #define PRED_BLOCK_AVAILABLE(v) \ |
| 276 | z_scan_block_avail(s, x0, y0, x ## v, y ## v) |
| 277 | |
| 278 | #define COMPARE_MV_REFIDX(a, b) \ |
| 279 | compare_mv_ref_idx(TAB_MVF_PU(a), TAB_MVF_PU(b)) |
| 280 | |
| 281 | /* |
| 282 | * 8.5.3.1.2 Derivation process for spatial merging candidates |
| 283 | */ |
| 284 | static void derive_spatial_merge_candidates(HEVCContext *s, int x0, int y0, |
| 285 | int nPbW, int nPbH, |
| 286 | int log2_cb_size, |
| 287 | int singleMCLFlag, int part_idx, |
| 288 | int merge_idx, |
| 289 | struct MvField mergecandlist[]) |
| 290 | { |
| 291 | HEVCLocalContext *lc = s->HEVClc; |
| 292 | RefPicList *refPicList = s->ref->refPicList; |
| 293 | MvField *tab_mvf = s->ref->tab_mvf; |
| 294 | |
| 295 | const int min_pu_width = s->sps->min_pu_width; |
| 296 | |
| 297 | const int cand_bottom_left = lc->na.cand_bottom_left; |
| 298 | const int cand_left = lc->na.cand_left; |
| 299 | const int cand_up_left = lc->na.cand_up_left; |
| 300 | const int cand_up = lc->na.cand_up; |
| 301 | const int cand_up_right = lc->na.cand_up_right_sap; |
| 302 | |
| 303 | const int xA1 = x0 - 1; |
| 304 | const int yA1 = y0 + nPbH - 1; |
| 305 | |
| 306 | const int xB1 = x0 + nPbW - 1; |
| 307 | const int yB1 = y0 - 1; |
| 308 | |
| 309 | const int xB0 = x0 + nPbW; |
| 310 | const int yB0 = y0 - 1; |
| 311 | |
| 312 | const int xA0 = x0 - 1; |
| 313 | const int yA0 = y0 + nPbH; |
| 314 | |
| 315 | const int xB2 = x0 - 1; |
| 316 | const int yB2 = y0 - 1; |
| 317 | |
| 318 | const int nb_refs = (s->sh.slice_type == P_SLICE) ? |
| 319 | s->sh.nb_refs[0] : FFMIN(s->sh.nb_refs[0], s->sh.nb_refs[1]); |
| 320 | |
| 321 | int zero_idx = 0; |
| 322 | |
| 323 | int nb_merge_cand = 0; |
| 324 | int nb_orig_merge_cand = 0; |
| 325 | |
| 326 | int is_available_a0; |
| 327 | int is_available_a1; |
| 328 | int is_available_b0; |
| 329 | int is_available_b1; |
| 330 | int is_available_b2; |
| 331 | |
| 332 | |
| 333 | if (!singleMCLFlag && part_idx == 1 && |
| 334 | (lc->cu.part_mode == PART_Nx2N || |
| 335 | lc->cu.part_mode == PART_nLx2N || |
| 336 | lc->cu.part_mode == PART_nRx2N) || |
| 337 | is_diff_mer(s, xA1, yA1, x0, y0)) { |
| 338 | is_available_a1 = 0; |
| 339 | } else { |
| 340 | is_available_a1 = AVAILABLE(cand_left, A1); |
| 341 | if (is_available_a1) { |
| 342 | mergecandlist[nb_merge_cand] = TAB_MVF_PU(A1); |
| 343 | if (merge_idx == 0) |
| 344 | return; |
| 345 | nb_merge_cand++; |
| 346 | } |
| 347 | } |
| 348 | |
| 349 | if (!singleMCLFlag && part_idx == 1 && |
| 350 | (lc->cu.part_mode == PART_2NxN || |
| 351 | lc->cu.part_mode == PART_2NxnU || |
| 352 | lc->cu.part_mode == PART_2NxnD) || |
| 353 | is_diff_mer(s, xB1, yB1, x0, y0)) { |
| 354 | is_available_b1 = 0; |
| 355 | } else { |
| 356 | is_available_b1 = AVAILABLE(cand_up, B1); |
| 357 | if (is_available_b1 && |
| 358 | !(is_available_a1 && COMPARE_MV_REFIDX(B1, A1))) { |
| 359 | mergecandlist[nb_merge_cand] = TAB_MVF_PU(B1); |
| 360 | if (merge_idx == nb_merge_cand) |
| 361 | return; |
| 362 | nb_merge_cand++; |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | // above right spatial merge candidate |
| 367 | is_available_b0 = AVAILABLE(cand_up_right, B0) && |
| 368 | xB0 < s->sps->width && |
| 369 | PRED_BLOCK_AVAILABLE(B0) && |
| 370 | !is_diff_mer(s, xB0, yB0, x0, y0); |
| 371 | |
| 372 | if (is_available_b0 && |
| 373 | !(is_available_b1 && COMPARE_MV_REFIDX(B0, B1))) { |
| 374 | mergecandlist[nb_merge_cand] = TAB_MVF_PU(B0); |
| 375 | if (merge_idx == nb_merge_cand) |
| 376 | return; |
| 377 | nb_merge_cand++; |
| 378 | } |
| 379 | |
| 380 | // left bottom spatial merge candidate |
| 381 | is_available_a0 = AVAILABLE(cand_bottom_left, A0) && |
| 382 | yA0 < s->sps->height && |
| 383 | PRED_BLOCK_AVAILABLE(A0) && |
| 384 | !is_diff_mer(s, xA0, yA0, x0, y0); |
| 385 | |
| 386 | if (is_available_a0 && |
| 387 | !(is_available_a1 && COMPARE_MV_REFIDX(A0, A1))) { |
| 388 | mergecandlist[nb_merge_cand] = TAB_MVF_PU(A0); |
| 389 | if (merge_idx == nb_merge_cand) |
| 390 | return; |
| 391 | nb_merge_cand++; |
| 392 | } |
| 393 | |
| 394 | // above left spatial merge candidate |
| 395 | is_available_b2 = AVAILABLE(cand_up_left, B2) && |
| 396 | !is_diff_mer(s, xB2, yB2, x0, y0); |
| 397 | |
| 398 | if (is_available_b2 && |
| 399 | !(is_available_a1 && COMPARE_MV_REFIDX(B2, A1)) && |
| 400 | !(is_available_b1 && COMPARE_MV_REFIDX(B2, B1)) && |
| 401 | nb_merge_cand != 4) { |
| 402 | mergecandlist[nb_merge_cand] = TAB_MVF_PU(B2); |
| 403 | if (merge_idx == nb_merge_cand) |
| 404 | return; |
| 405 | nb_merge_cand++; |
| 406 | } |
| 407 | |
| 408 | // temporal motion vector candidate |
| 409 | if (s->sh.slice_temporal_mvp_enabled_flag && |
| 410 | nb_merge_cand < s->sh.max_num_merge_cand) { |
| 411 | Mv mv_l0_col, mv_l1_col; |
| 412 | int available_l0 = temporal_luma_motion_vector(s, x0, y0, nPbW, nPbH, |
| 413 | 0, &mv_l0_col, 0); |
| 414 | int available_l1 = (s->sh.slice_type == B_SLICE) ? |
| 415 | temporal_luma_motion_vector(s, x0, y0, nPbW, nPbH, |
| 416 | 0, &mv_l1_col, 1) : 0; |
| 417 | |
| 418 | if (available_l0 || available_l1) { |
| 419 | mergecandlist[nb_merge_cand].pred_flag = available_l0 + (available_l1 << 1); |
| 420 | AV_ZERO16(mergecandlist[nb_merge_cand].ref_idx); |
| 421 | mergecandlist[nb_merge_cand].mv[0] = mv_l0_col; |
| 422 | mergecandlist[nb_merge_cand].mv[1] = mv_l1_col; |
| 423 | |
| 424 | if (merge_idx == nb_merge_cand) |
| 425 | return; |
| 426 | nb_merge_cand++; |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | nb_orig_merge_cand = nb_merge_cand; |
| 431 | |
| 432 | // combined bi-predictive merge candidates (applies for B slices) |
| 433 | if (s->sh.slice_type == B_SLICE && nb_orig_merge_cand > 1 && |
| 434 | nb_orig_merge_cand < s->sh.max_num_merge_cand) { |
| 435 | int comb_idx = 0; |
| 436 | |
| 437 | for (comb_idx = 0; nb_merge_cand < s->sh.max_num_merge_cand && |
| 438 | comb_idx < nb_orig_merge_cand * (nb_orig_merge_cand - 1); comb_idx++) { |
| 439 | int l0_cand_idx = l0_l1_cand_idx[comb_idx][0]; |
| 440 | int l1_cand_idx = l0_l1_cand_idx[comb_idx][1]; |
| 441 | MvField l0_cand = mergecandlist[l0_cand_idx]; |
| 442 | MvField l1_cand = mergecandlist[l1_cand_idx]; |
| 443 | |
| 444 | if ((l0_cand.pred_flag & PF_L0) && (l1_cand.pred_flag & PF_L1) && |
| 445 | (refPicList[0].list[l0_cand.ref_idx[0]] != |
| 446 | refPicList[1].list[l1_cand.ref_idx[1]] || |
| 447 | AV_RN32A(&l0_cand.mv[0]) != AV_RN32A(&l1_cand.mv[1]))) { |
| 448 | mergecandlist[nb_merge_cand].ref_idx[0] = l0_cand.ref_idx[0]; |
| 449 | mergecandlist[nb_merge_cand].ref_idx[1] = l1_cand.ref_idx[1]; |
| 450 | mergecandlist[nb_merge_cand].pred_flag = PF_BI; |
| 451 | AV_COPY32(&mergecandlist[nb_merge_cand].mv[0], &l0_cand.mv[0]); |
| 452 | AV_COPY32(&mergecandlist[nb_merge_cand].mv[1], &l1_cand.mv[1]); |
| 453 | if (merge_idx == nb_merge_cand) |
| 454 | return; |
| 455 | nb_merge_cand++; |
| 456 | } |
| 457 | } |
| 458 | } |
| 459 | |
| 460 | // append Zero motion vector candidates |
| 461 | while (nb_merge_cand < s->sh.max_num_merge_cand) { |
| 462 | mergecandlist[nb_merge_cand].pred_flag = PF_L0 + ((s->sh.slice_type == B_SLICE) << 1); |
| 463 | AV_ZERO32(mergecandlist[nb_merge_cand].mv + 0); |
| 464 | AV_ZERO32(mergecandlist[nb_merge_cand].mv + 1); |
| 465 | mergecandlist[nb_merge_cand].ref_idx[0] = zero_idx < nb_refs ? zero_idx : 0; |
| 466 | mergecandlist[nb_merge_cand].ref_idx[1] = zero_idx < nb_refs ? zero_idx : 0; |
| 467 | |
| 468 | if (merge_idx == nb_merge_cand) |
| 469 | return; |
| 470 | nb_merge_cand++; |
| 471 | zero_idx++; |
| 472 | } |
| 473 | } |
| 474 | |
| 475 | /* |
| 476 | * 8.5.3.1.1 Derivation process of luma Mvs for merge mode |
| 477 | */ |
| 478 | void ff_hevc_luma_mv_merge_mode(HEVCContext *s, int x0, int y0, int nPbW, |
| 479 | int nPbH, int log2_cb_size, int part_idx, |
| 480 | int merge_idx, MvField *mv) |
| 481 | { |
| 482 | int singleMCLFlag = 0; |
| 483 | int nCS = 1 << log2_cb_size; |
| 484 | LOCAL_ALIGNED(4, MvField, mergecand_list, [MRG_MAX_NUM_CANDS]); |
| 485 | int nPbW2 = nPbW; |
| 486 | int nPbH2 = nPbH; |
| 487 | HEVCLocalContext *lc = s->HEVClc; |
| 488 | |
| 489 | if (s->pps->log2_parallel_merge_level > 2 && nCS == 8) { |
| 490 | singleMCLFlag = 1; |
| 491 | x0 = lc->cu.x; |
| 492 | y0 = lc->cu.y; |
| 493 | nPbW = nCS; |
| 494 | nPbH = nCS; |
| 495 | part_idx = 0; |
| 496 | } |
| 497 | |
| 498 | ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH); |
| 499 | derive_spatial_merge_candidates(s, x0, y0, nPbW, nPbH, log2_cb_size, |
| 500 | singleMCLFlag, part_idx, |
| 501 | merge_idx, mergecand_list); |
| 502 | |
| 503 | if (mergecand_list[merge_idx].pred_flag == PF_BI && |
| 504 | (nPbW2 + nPbH2) == 12) { |
| 505 | mergecand_list[merge_idx].pred_flag = PF_L0; |
| 506 | } |
| 507 | |
| 508 | *mv = mergecand_list[merge_idx]; |
| 509 | } |
| 510 | |
| 511 | static av_always_inline void dist_scale(HEVCContext *s, Mv *mv, |
| 512 | int min_pu_width, int x, int y, |
| 513 | int elist, int ref_idx_curr, int ref_idx) |
| 514 | { |
| 515 | RefPicList *refPicList = s->ref->refPicList; |
| 516 | MvField *tab_mvf = s->ref->tab_mvf; |
| 517 | int ref_pic_elist = refPicList[elist].list[TAB_MVF(x, y).ref_idx[elist]]; |
| 518 | int ref_pic_curr = refPicList[ref_idx_curr].list[ref_idx]; |
| 519 | |
| 520 | if (ref_pic_elist != ref_pic_curr) { |
| 521 | int poc_diff = s->poc - ref_pic_elist; |
| 522 | if (!poc_diff) |
| 523 | poc_diff = 1; |
| 524 | mv_scale(mv, mv, poc_diff, s->poc - ref_pic_curr); |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | static int mv_mp_mode_mx(HEVCContext *s, int x, int y, int pred_flag_index, |
| 529 | Mv *mv, int ref_idx_curr, int ref_idx) |
| 530 | { |
| 531 | MvField *tab_mvf = s->ref->tab_mvf; |
| 532 | int min_pu_width = s->sps->min_pu_width; |
| 533 | |
| 534 | RefPicList *refPicList = s->ref->refPicList; |
| 535 | |
| 536 | if (((TAB_MVF(x, y).pred_flag) & (1 << pred_flag_index)) && |
| 537 | refPicList[pred_flag_index].list[TAB_MVF(x, y).ref_idx[pred_flag_index]] == refPicList[ref_idx_curr].list[ref_idx]) { |
| 538 | *mv = TAB_MVF(x, y).mv[pred_flag_index]; |
| 539 | return 1; |
| 540 | } |
| 541 | return 0; |
| 542 | } |
| 543 | |
| 544 | static int mv_mp_mode_mx_lt(HEVCContext *s, int x, int y, int pred_flag_index, |
| 545 | Mv *mv, int ref_idx_curr, int ref_idx) |
| 546 | { |
| 547 | MvField *tab_mvf = s->ref->tab_mvf; |
| 548 | int min_pu_width = s->sps->min_pu_width; |
| 549 | |
| 550 | RefPicList *refPicList = s->ref->refPicList; |
| 551 | |
| 552 | if ((TAB_MVF(x, y).pred_flag) & (1 << pred_flag_index)) { |
| 553 | int currIsLongTerm = refPicList[ref_idx_curr].isLongTerm[ref_idx]; |
| 554 | |
| 555 | int colIsLongTerm = |
| 556 | refPicList[pred_flag_index].isLongTerm[(TAB_MVF(x, y).ref_idx[pred_flag_index])]; |
| 557 | |
| 558 | if (colIsLongTerm == currIsLongTerm) { |
| 559 | *mv = TAB_MVF(x, y).mv[pred_flag_index]; |
| 560 | if (!currIsLongTerm) |
| 561 | dist_scale(s, mv, min_pu_width, x, y, |
| 562 | pred_flag_index, ref_idx_curr, ref_idx); |
| 563 | return 1; |
| 564 | } |
| 565 | } |
| 566 | return 0; |
| 567 | } |
| 568 | |
| 569 | #define MP_MX(v, pred, mx) \ |
| 570 | mv_mp_mode_mx(s, \ |
| 571 | (x ## v) >> s->sps->log2_min_pu_size, \ |
| 572 | (y ## v) >> s->sps->log2_min_pu_size, \ |
| 573 | pred, &mx, ref_idx_curr, ref_idx) |
| 574 | |
| 575 | #define MP_MX_LT(v, pred, mx) \ |
| 576 | mv_mp_mode_mx_lt(s, \ |
| 577 | (x ## v) >> s->sps->log2_min_pu_size, \ |
| 578 | (y ## v) >> s->sps->log2_min_pu_size, \ |
| 579 | pred, &mx, ref_idx_curr, ref_idx) |
| 580 | |
| 581 | void ff_hevc_luma_mv_mvp_mode(HEVCContext *s, int x0, int y0, int nPbW, |
| 582 | int nPbH, int log2_cb_size, int part_idx, |
| 583 | int merge_idx, MvField *mv, |
| 584 | int mvp_lx_flag, int LX) |
| 585 | { |
| 586 | HEVCLocalContext *lc = s->HEVClc; |
| 587 | MvField *tab_mvf = s->ref->tab_mvf; |
| 588 | int isScaledFlag_L0 = 0; |
| 589 | int availableFlagLXA0 = 1; |
| 590 | int availableFlagLXB0 = 1; |
| 591 | int numMVPCandLX = 0; |
| 592 | int min_pu_width = s->sps->min_pu_width; |
| 593 | |
| 594 | int xA0, yA0; |
| 595 | int is_available_a0; |
| 596 | int xA1, yA1; |
| 597 | int is_available_a1; |
| 598 | int xB0, yB0; |
| 599 | int is_available_b0; |
| 600 | int xB1, yB1; |
| 601 | int is_available_b1; |
| 602 | int xB2, yB2; |
| 603 | int is_available_b2; |
| 604 | |
| 605 | Mv mvpcand_list[2] = { { 0 } }; |
| 606 | Mv mxA; |
| 607 | Mv mxB; |
| 608 | int ref_idx_curr; |
| 609 | int ref_idx = 0; |
| 610 | int pred_flag_index_l0; |
| 611 | int pred_flag_index_l1; |
| 612 | |
| 613 | const int cand_bottom_left = lc->na.cand_bottom_left; |
| 614 | const int cand_left = lc->na.cand_left; |
| 615 | const int cand_up_left = lc->na.cand_up_left; |
| 616 | const int cand_up = lc->na.cand_up; |
| 617 | const int cand_up_right = lc->na.cand_up_right_sap; |
| 618 | ref_idx_curr = LX; |
| 619 | ref_idx = mv->ref_idx[LX]; |
| 620 | pred_flag_index_l0 = LX; |
| 621 | pred_flag_index_l1 = !LX; |
| 622 | |
| 623 | // left bottom spatial candidate |
| 624 | xA0 = x0 - 1; |
| 625 | yA0 = y0 + nPbH; |
| 626 | |
| 627 | is_available_a0 = AVAILABLE(cand_bottom_left, A0) && |
| 628 | yA0 < s->sps->height && |
| 629 | PRED_BLOCK_AVAILABLE(A0); |
| 630 | |
| 631 | //left spatial merge candidate |
| 632 | xA1 = x0 - 1; |
| 633 | yA1 = y0 + nPbH - 1; |
| 634 | |
| 635 | is_available_a1 = AVAILABLE(cand_left, A1); |
| 636 | if (is_available_a0 || is_available_a1) |
| 637 | isScaledFlag_L0 = 1; |
| 638 | |
| 639 | if (is_available_a0) { |
| 640 | if (MP_MX(A0, pred_flag_index_l0, mxA)) { |
| 641 | goto b_candidates; |
| 642 | } |
| 643 | if (MP_MX(A0, pred_flag_index_l1, mxA)) { |
| 644 | goto b_candidates; |
| 645 | } |
| 646 | } |
| 647 | |
| 648 | if (is_available_a1) { |
| 649 | if (MP_MX(A1, pred_flag_index_l0, mxA)) { |
| 650 | goto b_candidates; |
| 651 | } |
| 652 | if (MP_MX(A1, pred_flag_index_l1, mxA)) { |
| 653 | goto b_candidates; |
| 654 | } |
| 655 | } |
| 656 | |
| 657 | if (is_available_a0) { |
| 658 | if (MP_MX_LT(A0, pred_flag_index_l0, mxA)) { |
| 659 | goto b_candidates; |
| 660 | } |
| 661 | if (MP_MX_LT(A0, pred_flag_index_l1, mxA)) { |
| 662 | goto b_candidates; |
| 663 | } |
| 664 | } |
| 665 | |
| 666 | if (is_available_a1) { |
| 667 | if (MP_MX_LT(A1, pred_flag_index_l0, mxA)) { |
| 668 | goto b_candidates; |
| 669 | } |
| 670 | if (MP_MX_LT(A1, pred_flag_index_l1, mxA)) { |
| 671 | goto b_candidates; |
| 672 | } |
| 673 | } |
| 674 | availableFlagLXA0 = 0; |
| 675 | |
| 676 | b_candidates: |
| 677 | // B candidates |
| 678 | // above right spatial merge candidate |
| 679 | xB0 = x0 + nPbW; |
| 680 | yB0 = y0 - 1; |
| 681 | |
| 682 | is_available_b0 = AVAILABLE(cand_up_right, B0) && |
| 683 | xB0 < s->sps->width && |
| 684 | PRED_BLOCK_AVAILABLE(B0); |
| 685 | |
| 686 | // above spatial merge candidate |
| 687 | xB1 = x0 + nPbW - 1; |
| 688 | yB1 = y0 - 1; |
| 689 | is_available_b1 = AVAILABLE(cand_up, B1); |
| 690 | |
| 691 | // above left spatial merge candidate |
| 692 | xB2 = x0 - 1; |
| 693 | yB2 = y0 - 1; |
| 694 | is_available_b2 = AVAILABLE(cand_up_left, B2); |
| 695 | |
| 696 | // above right spatial merge candidate |
| 697 | if (is_available_b0) { |
| 698 | if (MP_MX(B0, pred_flag_index_l0, mxB)) { |
| 699 | goto scalef; |
| 700 | } |
| 701 | if (MP_MX(B0, pred_flag_index_l1, mxB)) { |
| 702 | goto scalef; |
| 703 | } |
| 704 | } |
| 705 | |
| 706 | // above spatial merge candidate |
| 707 | if (is_available_b1) { |
| 708 | if (MP_MX(B1, pred_flag_index_l0, mxB)) { |
| 709 | goto scalef; |
| 710 | } |
| 711 | if (MP_MX(B1, pred_flag_index_l1, mxB)) { |
| 712 | goto scalef; |
| 713 | } |
| 714 | } |
| 715 | |
| 716 | // above left spatial merge candidate |
| 717 | if (is_available_b2) { |
| 718 | if (MP_MX(B2, pred_flag_index_l0, mxB)) { |
| 719 | goto scalef; |
| 720 | } |
| 721 | if (MP_MX(B2, pred_flag_index_l1, mxB)) { |
| 722 | goto scalef; |
| 723 | } |
| 724 | } |
| 725 | availableFlagLXB0 = 0; |
| 726 | |
| 727 | scalef: |
| 728 | if (!isScaledFlag_L0) { |
| 729 | if (availableFlagLXB0) { |
| 730 | availableFlagLXA0 = 1; |
| 731 | mxA = mxB; |
| 732 | } |
| 733 | availableFlagLXB0 = 0; |
| 734 | |
| 735 | // XB0 and L1 |
| 736 | if (is_available_b0) { |
| 737 | availableFlagLXB0 = MP_MX_LT(B0, pred_flag_index_l0, mxB); |
| 738 | if (!availableFlagLXB0) |
| 739 | availableFlagLXB0 = MP_MX_LT(B0, pred_flag_index_l1, mxB); |
| 740 | } |
| 741 | |
| 742 | if (is_available_b1 && !availableFlagLXB0) { |
| 743 | availableFlagLXB0 = MP_MX_LT(B1, pred_flag_index_l0, mxB); |
| 744 | if (!availableFlagLXB0) |
| 745 | availableFlagLXB0 = MP_MX_LT(B1, pred_flag_index_l1, mxB); |
| 746 | } |
| 747 | |
| 748 | if (is_available_b2 && !availableFlagLXB0) { |
| 749 | availableFlagLXB0 = MP_MX_LT(B2, pred_flag_index_l0, mxB); |
| 750 | if (!availableFlagLXB0) |
| 751 | availableFlagLXB0 = MP_MX_LT(B2, pred_flag_index_l1, mxB); |
| 752 | } |
| 753 | } |
| 754 | |
| 755 | if (availableFlagLXA0) |
| 756 | mvpcand_list[numMVPCandLX++] = mxA; |
| 757 | |
| 758 | if (availableFlagLXB0 && (!availableFlagLXA0 || mxA.x != mxB.x || mxA.y != mxB.y)) |
| 759 | mvpcand_list[numMVPCandLX++] = mxB; |
| 760 | |
| 761 | //temporal motion vector prediction candidate |
| 762 | if (numMVPCandLX < 2 && s->sh.slice_temporal_mvp_enabled_flag && |
| 763 | mvp_lx_flag == numMVPCandLX) { |
| 764 | Mv mv_col; |
| 765 | int available_col = temporal_luma_motion_vector(s, x0, y0, nPbW, |
| 766 | nPbH, ref_idx, |
| 767 | &mv_col, LX); |
| 768 | if (available_col) |
| 769 | mvpcand_list[numMVPCandLX++] = mv_col; |
| 770 | } |
| 771 | |
| 772 | mv->mv[LX] = mvpcand_list[mvp_lx_flag]; |
| 773 | } |