git.sesse.net Git - ffmpeg/blob - libavcodec/hevc.c

   1 /*
   2  * HEVC video Decoder
   3  *
   4  * Copyright (C) 2012 - 2013 Guillaume Martres
   5  * Copyright (C) 2012 - 2013 Mickael Raulet
   6  * Copyright (C) 2012 - 2013 Gildas Cocherel
   7  * Copyright (C) 2012 - 2013 Wassim Hamidouche
   8  *
   9  * This file is part of FFmpeg.
  10  *
  11  * FFmpeg is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * FFmpeg is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with FFmpeg; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24  */
  25
  26 #include "libavutil/atomic.h"
  27 #include "libavutil/attributes.h"
  28 #include "libavutil/common.h"
  29 #include "libavutil/internal.h"
  30 #include "libavutil/md5.h"
  31 #include "libavutil/opt.h"
  32 #include "libavutil/pixdesc.h"
  33 #include "libavutil/stereo3d.h"
  34
  35 #include "bswapdsp.h"
  36 #include "bytestream.h"
  37 #include "cabac_functions.h"
  38 #include "golomb.h"
  39 #include "hevc.h"
  40
  41 const uint8_t ff_hevc_pel_weight[65] = { [2] = 0, [4] = 1, [6] = 2, [8] = 3, [12] = 4, [16] = 5, [24] = 6, [32] = 7, [48] = 8, [64] = 9 };
  42
  43 /**
  44  * NOTE: Each function hls_foo correspond to the function foo in the
  45  * specification (HLS stands for High Level Syntax).
  46  */
  47
  48 /**
  49  * Section 5.7
  50  */
  51
  52 /* free everything allocated  by pic_arrays_init() */
  53 static void pic_arrays_free(HEVCContext *s)
  54 {
  55     av_freep(&s->sao);
  56     av_freep(&s->deblock);
  57     av_freep(&s->split_cu_flag);
  58
  59     av_freep(&s->skip_flag);
  60     av_freep(&s->tab_ct_depth);
  61
  62     av_freep(&s->tab_ipm);
  63     av_freep(&s->cbf_luma);
  64     av_freep(&s->is_pcm);
  65
  66     av_freep(&s->qp_y_tab);
  67     av_freep(&s->tab_slice_address);
  68     av_freep(&s->filter_slice_edges);
  69
  70     av_freep(&s->horizontal_bs);
  71     av_freep(&s->vertical_bs);
  72
  73     av_freep(&s->sh.entry_point_offset);
  74     av_freep(&s->sh.size);
  75     av_freep(&s->sh.offset);
  76
  77     av_buffer_pool_uninit(&s->tab_mvf_pool);
  78     av_buffer_pool_uninit(&s->rpl_tab_pool);
  79 }
  80
  81 /* allocate arrays that depend on frame dimensions */
  82 static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
  83 {
  84     int log2_min_cb_size = sps->log2_min_cb_size;
  85     int width            = sps->width;
  86     int height           = sps->height;
  87     int pic_size         = width * height;
  88     int pic_size_in_ctb  = ((width  >> log2_min_cb_size) + 1) *
  89                            ((height >> log2_min_cb_size) + 1);
  90     int ctb_count        = sps->ctb_width * sps->ctb_height;
  91     int min_pu_size      = sps->min_pu_width * sps->min_pu_height;
  92
  93     s->bs_width  = width  >> 3;
  94     s->bs_height = height >> 3;
  95
  96     s->sao           = av_mallocz_array(ctb_count, sizeof(*s->sao));
  97     s->deblock       = av_mallocz_array(ctb_count, sizeof(*s->deblock));
  98     s->split_cu_flag = av_malloc(pic_size);
  99     if (!s->sao || !s->deblock || !s->split_cu_flag)
 100         goto fail;
 101
 102     s->skip_flag    = av_malloc(pic_size_in_ctb);
 103     s->tab_ct_depth = av_malloc(sps->min_cb_height * sps->min_cb_width);
 104     if (!s->skip_flag || !s->tab_ct_depth)
 105         goto fail;
 106
 107     s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
 108     s->tab_ipm  = av_mallocz(min_pu_size);
 109     s->is_pcm   = av_malloc(min_pu_size);
 110     if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
 111         goto fail;
 112
 113     s->filter_slice_edges = av_malloc(ctb_count);
 114     s->tab_slice_address  = av_malloc(pic_size_in_ctb *
 115                                       sizeof(*s->tab_slice_address));
 116     s->qp_y_tab           = av_malloc(pic_size_in_ctb *
 117                                       sizeof(*s->qp_y_tab));
 118     if (!s->qp_y_tab || !s->filter_slice_edges || !s->tab_slice_address)
 119         goto fail;
 120
 121     s->horizontal_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
 122     s->vertical_bs   = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
 123     if (!s->horizontal_bs || !s->vertical_bs)
 124         goto fail;
 125
 126     s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
 127                                           av_buffer_allocz);
 128     s->rpl_tab_pool = av_buffer_pool_init(ctb_count * sizeof(RefPicListTab),
 129                                           av_buffer_allocz);
 130     if (!s->tab_mvf_pool || !s->rpl_tab_pool)
 131         goto fail;
 132
 133     return 0;
 134
 135 fail:
 136     pic_arrays_free(s);
 137     return AVERROR(ENOMEM);
 138 }
 139
 140 static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
 141 {
 142     int i = 0;
 143     int j = 0;
 144     uint8_t luma_weight_l0_flag[16];
 145     uint8_t chroma_weight_l0_flag[16];
 146     uint8_t luma_weight_l1_flag[16];
 147     uint8_t chroma_weight_l1_flag[16];
 148
 149     s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
 150     if (s->sps->chroma_format_idc != 0) {
 151         int delta = get_se_golomb(gb);
 152         s->sh.chroma_log2_weight_denom = av_clip(s->sh.luma_log2_weight_denom + delta, 0, 7);
 153     }
 154
 155     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
 156         luma_weight_l0_flag[i] = get_bits1(gb);
 157         if (!luma_weight_l0_flag[i]) {
 158             s->sh.luma_weight_l0[i] = 1 << s->sh.luma_log2_weight_denom;
 159             s->sh.luma_offset_l0[i] = 0;
 160         }
 161     }
 162     if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
 163         for (i = 0; i < s->sh.nb_refs[L0]; i++)
 164             chroma_weight_l0_flag[i] = get_bits1(gb);
 165     } else {
 166         for (i = 0; i < s->sh.nb_refs[L0]; i++)
 167             chroma_weight_l0_flag[i] = 0;
 168     }
 169     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
 170         if (luma_weight_l0_flag[i]) {
 171             int delta_luma_weight_l0 = get_se_golomb(gb);
 172             s->sh.luma_weight_l0[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l0;
 173             s->sh.luma_offset_l0[i] = get_se_golomb(gb);
 174         }
 175         if (chroma_weight_l0_flag[i]) {
 176             for (j = 0; j < 2; j++) {
 177                 int delta_chroma_weight_l0 = get_se_golomb(gb);
 178                 int delta_chroma_offset_l0 = get_se_golomb(gb);
 179                 s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
 180                 s->sh.chroma_offset_l0[i][j] = av_clip((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
 181                                                                                     >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
 182             }
 183         } else {
 184             s->sh.chroma_weight_l0[i][0] = 1 << s->sh.chroma_log2_weight_denom;
 185             s->sh.chroma_offset_l0[i][0] = 0;
 186             s->sh.chroma_weight_l0[i][1] = 1 << s->sh.chroma_log2_weight_denom;
 187             s->sh.chroma_offset_l0[i][1] = 0;
 188         }
 189     }
 190     if (s->sh.slice_type == B_SLICE) {
 191         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
 192             luma_weight_l1_flag[i] = get_bits1(gb);
 193             if (!luma_weight_l1_flag[i]) {
 194                 s->sh.luma_weight_l1[i] = 1 << s->sh.luma_log2_weight_denom;
 195                 s->sh.luma_offset_l1[i] = 0;
 196             }
 197         }
 198         if (s->sps->chroma_format_idc != 0) {
 199             for (i = 0; i < s->sh.nb_refs[L1]; i++)
 200                 chroma_weight_l1_flag[i] = get_bits1(gb);
 201         } else {
 202             for (i = 0; i < s->sh.nb_refs[L1]; i++)
 203                 chroma_weight_l1_flag[i] = 0;
 204         }
 205         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
 206             if (luma_weight_l1_flag[i]) {
 207                 int delta_luma_weight_l1 = get_se_golomb(gb);
 208                 s->sh.luma_weight_l1[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l1;
 209                 s->sh.luma_offset_l1[i] = get_se_golomb(gb);
 210             }
 211             if (chroma_weight_l1_flag[i]) {
 212                 for (j = 0; j < 2; j++) {
 213                     int delta_chroma_weight_l1 = get_se_golomb(gb);
 214                     int delta_chroma_offset_l1 = get_se_golomb(gb);
 215                     s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
 216                     s->sh.chroma_offset_l1[i][j] = av_clip((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
 217                                                                                         >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
 218                 }
 219             } else {
 220                 s->sh.chroma_weight_l1[i][0] = 1 << s->sh.chroma_log2_weight_denom;
 221                 s->sh.chroma_offset_l1[i][0] = 0;
 222                 s->sh.chroma_weight_l1[i][1] = 1 << s->sh.chroma_log2_weight_denom;
 223                 s->sh.chroma_offset_l1[i][1] = 0;
 224             }
 225         }
 226     }
 227 }
 228
 229 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
 230 {
 231     const HEVCSPS *sps = s->sps;
 232     int max_poc_lsb    = 1 << sps->log2_max_poc_lsb;
 233     int prev_delta_msb = 0;
 234     unsigned int nb_sps = 0, nb_sh;
 235     int i;
 236
 237     rps->nb_refs = 0;
 238     if (!sps->long_term_ref_pics_present_flag)
 239         return 0;
 240
 241     if (sps->num_long_term_ref_pics_sps > 0)
 242         nb_sps = get_ue_golomb_long(gb);
 243     nb_sh = get_ue_golomb_long(gb);
 244
 245     if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
 246         return AVERROR_INVALIDDATA;
 247
 248     rps->nb_refs = nb_sh + nb_sps;
 249
 250     for (i = 0; i < rps->nb_refs; i++) {
 251         uint8_t delta_poc_msb_present;
 252
 253         if (i < nb_sps) {
 254             uint8_t lt_idx_sps = 0;
 255
 256             if (sps->num_long_term_ref_pics_sps > 1)
 257                 lt_idx_sps = get_bits(gb, av_ceil_log2(sps->num_long_term_ref_pics_sps));
 258
 259             rps->poc[i]  = sps->lt_ref_pic_poc_lsb_sps[lt_idx_sps];
 260             rps->used[i] = sps->used_by_curr_pic_lt_sps_flag[lt_idx_sps];
 261         } else {
 262             rps->poc[i]  = get_bits(gb, sps->log2_max_poc_lsb);
 263             rps->used[i] = get_bits1(gb);
 264         }
 265
 266         delta_poc_msb_present = get_bits1(gb);
 267         if (delta_poc_msb_present) {
 268             int delta = get_ue_golomb_long(gb);
 269
 270             if (i && i != nb_sps)
 271                 delta += prev_delta_msb;
 272
 273             rps->poc[i] += s->poc - delta * max_poc_lsb - s->sh.pic_order_cnt_lsb;
 274             prev_delta_msb = delta;
 275         }
 276     }
 277
 278     return 0;
 279 }
 280
 281 static int set_sps(HEVCContext *s, const HEVCSPS *sps)
 282 {
 283     int ret;
 284     unsigned int num = 0, den = 0;
 285
 286     pic_arrays_free(s);
 287     ret = pic_arrays_init(s, sps);
 288     if (ret < 0)
 289         goto fail;
 290
 291     s->avctx->coded_width         = sps->width;
 292     s->avctx->coded_height        = sps->height;
 293     s->avctx->width               = sps->output_width;
 294     s->avctx->height              = sps->output_height;
 295     s->avctx->pix_fmt             = sps->pix_fmt;
 296     s->avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
 297
 298     ff_set_sar(s->avctx, sps->vui.sar);
 299
 300     if (sps->vui.video_signal_type_present_flag)
 301         s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
 302                                                                : AVCOL_RANGE_MPEG;
 303     else
 304         s->avctx->color_range = AVCOL_RANGE_MPEG;
 305
 306     if (sps->vui.colour_description_present_flag) {
 307         s->avctx->color_primaries = sps->vui.colour_primaries;
 308         s->avctx->color_trc       = sps->vui.transfer_characteristic;
 309         s->avctx->colorspace      = sps->vui.matrix_coeffs;
 310     } else {
 311         s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
 312         s->avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
 313         s->avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
 314     }
 315
 316     ff_hevc_pred_init(&s->hpc,     sps->bit_depth);
 317     ff_hevc_dsp_init (&s->hevcdsp, sps->bit_depth);
 318     ff_videodsp_init (&s->vdsp,    sps->bit_depth);
 319
 320     if (sps->sao_enabled) {
 321         av_frame_unref(s->tmp_frame);
 322         ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
 323         if (ret < 0)
 324             goto fail;
 325         s->frame = s->tmp_frame;
 326     }
 327
 328     s->sps = sps;
 329     s->vps = (HEVCVPS*) s->vps_list[s->sps->vps_id]->data;
 330
 331     if (s->vps->vps_timing_info_present_flag) {
 332         num = s->vps->vps_num_units_in_tick;
 333         den = s->vps->vps_time_scale;
 334     } else if (sps->vui.vui_timing_info_present_flag) {
 335         num = sps->vui.vui_num_units_in_tick;
 336         den = sps->vui.vui_time_scale;
 337     }
 338
 339     if (num != 0 && den != 0)
 340         av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
 341                   num, den, 1 << 30);
 342
 343     return 0;
 344
 345 fail:
 346     pic_arrays_free(s);
 347     s->sps = NULL;
 348     return ret;
 349 }
 350
 351 static int is_sps_exist(HEVCContext *s, const HEVCSPS* last_sps)
 352 {
 353     int i;
 354
 355     for( i = 0; i < MAX_SPS_COUNT; i++)
 356         if(s->sps_list[i])
 357             if (last_sps == (HEVCSPS*)s->sps_list[i]->data)
 358                 return 1;
 359     return 0;
 360 }
 361
 362 static int hls_slice_header(HEVCContext *s)
 363 {
 364     GetBitContext *gb = &s->HEVClc->gb;
 365     SliceHeader *sh   = &s->sh;
 366     int i, j, ret;
 367
 368     // Coded parameters
 369     sh->first_slice_in_pic_flag = get_bits1(gb);
 370     if ((IS_IDR(s) || IS_BLA(s)) && sh->first_slice_in_pic_flag) {
 371         s->seq_decode = (s->seq_decode + 1) & 0xff;
 372         s->max_ra     = INT_MAX;
 373         if (IS_IDR(s))
 374             ff_hevc_clear_refs(s);
 375     }
 376     sh->no_output_of_prior_pics_flag = 0;
 377     if (IS_IRAP(s))
 378         sh->no_output_of_prior_pics_flag = get_bits1(gb);
 379     if (s->nal_unit_type == NAL_CRA_NUT && s->last_eos == 1)
 380         sh->no_output_of_prior_pics_flag = 1;
 381
 382     sh->pps_id = get_ue_golomb_long(gb);
 383     if (sh->pps_id >= MAX_PPS_COUNT || !s->pps_list[sh->pps_id]) {
 384         av_log(s->avctx, AV_LOG_ERROR, "PPS id out of range: %d\n", sh->pps_id);
 385         return AVERROR_INVALIDDATA;
 386     }
 387     if (!sh->first_slice_in_pic_flag &&
 388         s->pps != (HEVCPPS*)s->pps_list[sh->pps_id]->data) {
 389         av_log(s->avctx, AV_LOG_ERROR, "PPS changed between slices.\n");
 390         return AVERROR_INVALIDDATA;
 391     }
 392     s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
 393
 394     if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
 395         const HEVCSPS* last_sps = s->sps;
 396         s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
 397         if (last_sps) {
 398             if (is_sps_exist(s, last_sps)) {
 399                 if (s->sps->width !=  last_sps->width || s->sps->height != last_sps->height ||
 400                         s->sps->temporal_layer[s->sps->max_sub_layers - 1].max_dec_pic_buffering != last_sps->temporal_layer[last_sps->max_sub_layers - 1].max_dec_pic_buffering)
 401                     sh->no_output_of_prior_pics_flag = 0;
 402             } else
 403                 sh->no_output_of_prior_pics_flag = 0;
 404         }
 405         ff_hevc_clear_refs(s);
 406         ret = set_sps(s, s->sps);
 407         if (ret < 0)
 408             return ret;
 409
 410         s->seq_decode = (s->seq_decode + 1) & 0xff;
 411         s->max_ra     = INT_MAX;
 412     }
 413
 414     s->avctx->profile = s->sps->ptl.general_ptl.profile_idc;
 415     s->avctx->level   = s->sps->ptl.general_ptl.level_idc;
 416
 417     sh->dependent_slice_segment_flag = 0;
 418     if (!sh->first_slice_in_pic_flag) {
 419         int slice_address_length;
 420
 421         if (s->pps->dependent_slice_segments_enabled_flag)
 422             sh->dependent_slice_segment_flag = get_bits1(gb);
 423
 424         slice_address_length = av_ceil_log2(s->sps->ctb_width *
 425                                             s->sps->ctb_height);
 426         sh->slice_segment_addr = get_bits(gb, slice_address_length);
 427         if (sh->slice_segment_addr >= s->sps->ctb_width * s->sps->ctb_height) {
 428             av_log(s->avctx, AV_LOG_ERROR,
 429                    "Invalid slice segment address: %u.\n",
 430                    sh->slice_segment_addr);
 431             return AVERROR_INVALIDDATA;
 432         }
 433
 434         if (!sh->dependent_slice_segment_flag) {
 435             sh->slice_addr = sh->slice_segment_addr;
 436             s->slice_idx++;
 437         }
 438     } else {
 439         sh->slice_segment_addr = sh->slice_addr = 0;
 440         s->slice_idx           = 0;
 441         s->slice_initialized   = 0;
 442     }
 443
 444     if (!sh->dependent_slice_segment_flag) {
 445         s->slice_initialized = 0;
 446
 447         for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
 448             skip_bits(gb, 1);  // slice_reserved_undetermined_flag[]
 449
 450         sh->slice_type = get_ue_golomb_long(gb);
 451         if (!(sh->slice_type == I_SLICE ||
 452               sh->slice_type == P_SLICE ||
 453               sh->slice_type == B_SLICE)) {
 454             av_log(s->avctx, AV_LOG_ERROR, "Unknown slice type: %d.\n",
 455                    sh->slice_type);
 456             return AVERROR_INVALIDDATA;
 457         }
 458         if (IS_IRAP(s) && sh->slice_type != I_SLICE) {
 459             av_log(s->avctx, AV_LOG_ERROR, "Inter slices in an IRAP frame.\n");
 460             return AVERROR_INVALIDDATA;
 461         }
 462
 463         sh->pic_output_flag = 1;
 464         if (s->pps->output_flag_present_flag)
 465             sh->pic_output_flag = get_bits1(gb);
 466
 467         if (s->sps->separate_colour_plane_flag)
 468             sh->colour_plane_id = get_bits(gb, 2);
 469
 470         if (!IS_IDR(s)) {
 471             int short_term_ref_pic_set_sps_flag, poc;
 472
 473             sh->pic_order_cnt_lsb = get_bits(gb, s->sps->log2_max_poc_lsb);
 474             poc = ff_hevc_compute_poc(s, sh->pic_order_cnt_lsb);
 475             if (!sh->first_slice_in_pic_flag && poc != s->poc) {
 476                 av_log(s->avctx, AV_LOG_WARNING,
 477                        "Ignoring POC change between slices: %d -> %d\n", s->poc, poc);
 478                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
 479                     return AVERROR_INVALIDDATA;
 480                 poc = s->poc;
 481             }
 482             s->poc = poc;
 483
 484             short_term_ref_pic_set_sps_flag = get_bits1(gb);
 485             if (!short_term_ref_pic_set_sps_flag) {
 486                 ret = ff_hevc_decode_short_term_rps(s, &sh->slice_rps, s->sps, 1);
 487                 if (ret < 0)
 488                     return ret;
 489
 490                 sh->short_term_rps = &sh->slice_rps;
 491             } else {
 492                 int numbits, rps_idx;
 493
 494                 if (!s->sps->nb_st_rps) {
 495                     av_log(s->avctx, AV_LOG_ERROR, "No ref lists in the SPS.\n");
 496                     return AVERROR_INVALIDDATA;
 497                 }
 498
 499                 numbits = av_ceil_log2(s->sps->nb_st_rps);
 500                 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
 501                 sh->short_term_rps = &s->sps->st_rps[rps_idx];
 502             }
 503
 504             ret = decode_lt_rps(s, &sh->long_term_rps, gb);
 505             if (ret < 0) {
 506                 av_log(s->avctx, AV_LOG_WARNING, "Invalid long term RPS.\n");
 507                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
 508                     return AVERROR_INVALIDDATA;
 509             }
 510
 511             if (s->sps->sps_temporal_mvp_enabled_flag)
 512                 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
 513             else
 514                 sh->slice_temporal_mvp_enabled_flag = 0;
 515         } else {
 516             s->sh.short_term_rps = NULL;
 517             s->poc               = 0;
 518         }
 519
 520         /* 8.3.1 */
 521         if (s->temporal_id == 0 &&
 522             s->nal_unit_type != NAL_TRAIL_N &&
 523             s->nal_unit_type != NAL_TSA_N   &&
 524             s->nal_unit_type != NAL_STSA_N  &&
 525             s->nal_unit_type != NAL_RADL_N  &&
 526             s->nal_unit_type != NAL_RADL_R  &&
 527             s->nal_unit_type != NAL_RASL_N  &&
 528             s->nal_unit_type != NAL_RASL_R)
 529             s->pocTid0 = s->poc;
 530
 531         if (s->sps->sao_enabled) {
 532             sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
 533             sh->slice_sample_adaptive_offset_flag[1] =
 534             sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
 535         } else {
 536             sh->slice_sample_adaptive_offset_flag[0] = 0;
 537             sh->slice_sample_adaptive_offset_flag[1] = 0;
 538             sh->slice_sample_adaptive_offset_flag[2] = 0;
 539         }
 540
 541         sh->nb_refs[L0] = sh->nb_refs[L1] = 0;
 542         if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
 543             int nb_refs;
 544
 545             sh->nb_refs[L0] = s->pps->num_ref_idx_l0_default_active;
 546             if (sh->slice_type == B_SLICE)
 547                 sh->nb_refs[L1] = s->pps->num_ref_idx_l1_default_active;
 548
 549             if (get_bits1(gb)) { // num_ref_idx_active_override_flag
 550                 sh->nb_refs[L0] = get_ue_golomb_long(gb) + 1;
 551                 if (sh->slice_type == B_SLICE)
 552                     sh->nb_refs[L1] = get_ue_golomb_long(gb) + 1;
 553             }
 554             if (sh->nb_refs[L0] > MAX_REFS || sh->nb_refs[L1] > MAX_REFS) {
 555                 av_log(s->avctx, AV_LOG_ERROR, "Too many refs: %d/%d.\n",
 556                        sh->nb_refs[L0], sh->nb_refs[L1]);
 557                 return AVERROR_INVALIDDATA;
 558             }
 559
 560             sh->rpl_modification_flag[0] = 0;
 561             sh->rpl_modification_flag[1] = 0;
 562             nb_refs = ff_hevc_frame_nb_refs(s);
 563             if (!nb_refs) {
 564                 av_log(s->avctx, AV_LOG_ERROR, "Zero refs for a frame with P or B slices.\n");
 565                 return AVERROR_INVALIDDATA;
 566             }
 567
 568             if (s->pps->lists_modification_present_flag && nb_refs > 1) {
 569                 sh->rpl_modification_flag[0] = get_bits1(gb);
 570                 if (sh->rpl_modification_flag[0]) {
 571                     for (i = 0; i < sh->nb_refs[L0]; i++)
 572                         sh->list_entry_lx[0][i] = get_bits(gb, av_ceil_log2(nb_refs));
 573                 }
 574
 575                 if (sh->slice_type == B_SLICE) {
 576                     sh->rpl_modification_flag[1] = get_bits1(gb);
 577                     if (sh->rpl_modification_flag[1] == 1)
 578                         for (i = 0; i < sh->nb_refs[L1]; i++)
 579                             sh->list_entry_lx[1][i] = get_bits(gb, av_ceil_log2(nb_refs));
 580                 }
 581             }
 582
 583             if (sh->slice_type == B_SLICE)
 584                 sh->mvd_l1_zero_flag = get_bits1(gb);
 585
 586             if (s->pps->cabac_init_present_flag)
 587                 sh->cabac_init_flag = get_bits1(gb);
 588             else
 589                 sh->cabac_init_flag = 0;
 590
 591             sh->collocated_ref_idx = 0;
 592             if (sh->slice_temporal_mvp_enabled_flag) {
 593                 sh->collocated_list = L0;
 594                 if (sh->slice_type == B_SLICE)
 595                     sh->collocated_list = !get_bits1(gb);
 596
 597                 if (sh->nb_refs[sh->collocated_list] > 1) {
 598                     sh->collocated_ref_idx = get_ue_golomb_long(gb);
 599                     if (sh->collocated_ref_idx >= sh->nb_refs[sh->collocated_list]) {
 600                         av_log(s->avctx, AV_LOG_ERROR,
 601                                "Invalid collocated_ref_idx: %d.\n",
 602                                sh->collocated_ref_idx);
 603                         return AVERROR_INVALIDDATA;
 604                     }
 605                 }
 606             }
 607
 608             if ((s->pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
 609                 (s->pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
 610                 pred_weight_table(s, gb);
 611             }
 612
 613             sh->max_num_merge_cand = 5 - get_ue_golomb_long(gb);
 614             if (sh->max_num_merge_cand < 1 || sh->max_num_merge_cand > 5) {
 615                 av_log(s->avctx, AV_LOG_ERROR,
 616                        "Invalid number of merging MVP candidates: %d.\n",
 617                        sh->max_num_merge_cand);
 618                 return AVERROR_INVALIDDATA;
 619             }
 620         }
 621
 622         sh->slice_qp_delta = get_se_golomb(gb);
 623
 624         if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
 625             sh->slice_cb_qp_offset = get_se_golomb(gb);
 626             sh->slice_cr_qp_offset = get_se_golomb(gb);
 627         } else {
 628             sh->slice_cb_qp_offset = 0;
 629             sh->slice_cr_qp_offset = 0;
 630         }
 631
 632         if (s->pps->deblocking_filter_control_present_flag) {
 633             int deblocking_filter_override_flag = 0;
 634
 635             if (s->pps->deblocking_filter_override_enabled_flag)
 636                 deblocking_filter_override_flag = get_bits1(gb);
 637
 638             if (deblocking_filter_override_flag) {
 639                 sh->disable_deblocking_filter_flag = get_bits1(gb);
 640                 if (!sh->disable_deblocking_filter_flag) {
 641                     sh->beta_offset = get_se_golomb(gb) * 2;
 642                     sh->tc_offset   = get_se_golomb(gb) * 2;
 643                 }
 644             } else {
 645                 sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
 646                 sh->beta_offset                    = s->pps->beta_offset;
 647                 sh->tc_offset                      = s->pps->tc_offset;
 648             }
 649         } else {
 650             sh->disable_deblocking_filter_flag = 0;
 651             sh->beta_offset                    = 0;
 652             sh->tc_offset                      = 0;
 653         }
 654
 655         if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
 656             (sh->slice_sample_adaptive_offset_flag[0] ||
 657              sh->slice_sample_adaptive_offset_flag[1] ||
 658              !sh->disable_deblocking_filter_flag)) {
 659             sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
 660         } else {
 661             sh->slice_loop_filter_across_slices_enabled_flag = s->pps->seq_loop_filter_across_slices_enabled_flag;
 662         }
 663     } else if (!s->slice_initialized) {
 664         av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
 665         return AVERROR_INVALIDDATA;
 666     }
 667
 668     sh->num_entry_point_offsets = 0;
 669     if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
 670         sh->num_entry_point_offsets = get_ue_golomb_long(gb);
 671         if (sh->num_entry_point_offsets > 0) {
 672             int offset_len = get_ue_golomb_long(gb) + 1;
 673             int segments = offset_len >> 4;
 674             int rest = (offset_len & 15);
 675             av_freep(&sh->entry_point_offset);
 676             av_freep(&sh->offset);
 677             av_freep(&sh->size);
 678             sh->entry_point_offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
 679             sh->offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
 680             sh->size = av_malloc(sh->num_entry_point_offsets * sizeof(int));
 681             if (!sh->entry_point_offset || !sh->offset || !sh->size) {
 682                 sh->num_entry_point_offsets = 0;
 683                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate memory\n");
 684                 return AVERROR(ENOMEM);
 685             }
 686             for (i = 0; i < sh->num_entry_point_offsets; i++) {
 687                 int val = 0;
 688                 for (j = 0; j < segments; j++) {
 689                     val <<= 16;
 690                     val += get_bits(gb, 16);
 691                 }
 692                 if (rest) {
 693                     val <<= rest;
 694                     val += get_bits(gb, rest);
 695                 }
 696                 sh->entry_point_offset[i] = val + 1; // +1; // +1 to get the size
 697             }
 698             if (s->threads_number > 1 && (s->pps->num_tile_rows > 1 || s->pps->num_tile_columns > 1)) {
 699                 s->enable_parallel_tiles = 0; // TODO: you can enable tiles in parallel here
 700                 s->threads_number = 1;
 701             } else
 702                 s->enable_parallel_tiles = 0;
 703         } else
 704             s->enable_parallel_tiles = 0;
 705     }
 706
 707     if (s->pps->slice_header_extension_present_flag) {
 708         unsigned int length = get_ue_golomb_long(gb);
 709         for (i = 0; i < length; i++)
 710             skip_bits(gb, 8);  // slice_header_extension_data_byte
 711     }
 712
 713     // Inferred parameters
 714     sh->slice_qp = 26U + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
 715     if (sh->slice_qp > 51 ||
 716         sh->slice_qp < -s->sps->qp_bd_offset) {
 717         av_log(s->avctx, AV_LOG_ERROR,
 718                "The slice_qp %d is outside the valid range "
 719                "[%d, 51].\n",
 720                sh->slice_qp,
 721                -s->sps->qp_bd_offset);
 722         return AVERROR_INVALIDDATA;
 723     }
 724
 725     sh->slice_ctb_addr_rs = sh->slice_segment_addr;
 726
 727     if (!s->sh.slice_ctb_addr_rs && s->sh.dependent_slice_segment_flag) {
 728         av_log(s->avctx, AV_LOG_ERROR, "Impossible slice segment.\n");
 729         return AVERROR_INVALIDDATA;
 730     }
 731
 732     s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
 733
 734     if (!s->pps->cu_qp_delta_enabled_flag)
 735         s->HEVClc->qp_y = s->sh.slice_qp;
 736
 737     s->slice_initialized = 1;
 738
 739     return 0;
 740 }
 741
 742 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
 743
 744 #define SET_SAO(elem, value)                            \
 745 do {                                                    \
 746     if (!sao_merge_up_flag && !sao_merge_left_flag)     \
 747         sao->elem = value;                              \
 748     else if (sao_merge_left_flag)                       \
 749         sao->elem = CTB(s->sao, rx-1, ry).elem;         \
 750     else if (sao_merge_up_flag)                         \
 751         sao->elem = CTB(s->sao, rx, ry-1).elem;         \
 752     else                                                \
 753         sao->elem = 0;                                  \
 754 } while (0)
 755
 756 static void hls_sao_param(HEVCContext *s, int rx, int ry)
 757 {
 758     HEVCLocalContext *lc    = s->HEVClc;
 759     int sao_merge_left_flag = 0;
 760     int sao_merge_up_flag   = 0;
 761     int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
 762     SAOParams *sao          = &CTB(s->sao, rx, ry);
 763     int c_idx, i;
 764
 765     if (s->sh.slice_sample_adaptive_offset_flag[0] ||
 766         s->sh.slice_sample_adaptive_offset_flag[1]) {
 767         if (rx > 0) {
 768             if (lc->ctb_left_flag)
 769                 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
 770         }
 771         if (ry > 0 && !sao_merge_left_flag) {
 772             if (lc->ctb_up_flag)
 773                 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
 774         }
 775     }
 776
 777     for (c_idx = 0; c_idx < 3; c_idx++) {
 778         if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
 779             sao->type_idx[c_idx] = SAO_NOT_APPLIED;
 780             continue;
 781         }
 782
 783         if (c_idx == 2) {
 784             sao->type_idx[2] = sao->type_idx[1];
 785             sao->eo_class[2] = sao->eo_class[1];
 786         } else {
 787             SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
 788         }
 789
 790         if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
 791             continue;
 792
 793         for (i = 0; i < 4; i++)
 794             SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
 795
 796         if (sao->type_idx[c_idx] == SAO_BAND) {
 797             for (i = 0; i < 4; i++) {
 798                 if (sao->offset_abs[c_idx][i]) {
 799                     SET_SAO(offset_sign[c_idx][i],
 800                             ff_hevc_sao_offset_sign_decode(s));
 801                 } else {
 802                     sao->offset_sign[c_idx][i] = 0;
 803                 }
 804             }
 805             SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
 806         } else if (c_idx != 2) {
 807             SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
 808         }
 809
 810         // Inferred parameters
 811         sao->offset_val[c_idx][0] = 0;
 812         for (i = 0; i < 4; i++) {
 813             sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
 814             if (sao->type_idx[c_idx] == SAO_EDGE) {
 815                 if (i > 1)
 816                     sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
 817             } else if (sao->offset_sign[c_idx][i]) {
 818                 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
 819             }
 820         }
 821     }
 822 }
 823
 824 #undef SET_SAO
 825 #undef CTB
 826
 827 static int hls_transform_unit(HEVCContext *s, int x0, int y0,
 828                               int xBase, int yBase, int cb_xBase, int cb_yBase,
 829                               int log2_cb_size, int log2_trafo_size,
 830                               int trafo_depth, int blk_idx)
 831 {
 832     HEVCLocalContext *lc = s->HEVClc;
 833
 834     if (lc->cu.pred_mode == MODE_INTRA) {
 835         int trafo_size = 1 << log2_trafo_size;
 836         ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
 837
 838         s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
 839         if (log2_trafo_size > 2) {
 840             trafo_size = trafo_size << (s->sps->hshift[1] - 1);
 841             ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
 842             s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 1);
 843             s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 2);
 844         } else if (blk_idx == 3) {
 845             trafo_size = trafo_size << s->sps->hshift[1];
 846             ff_hevc_set_neighbour_available(s, xBase, yBase,
 847                                             trafo_size, trafo_size);
 848             s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
 849             s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
 850         }
 851     }
 852
 853     if (lc->tt.cbf_luma ||
 854         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
 855         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
 856         int scan_idx   = SCAN_DIAG;
 857         int scan_idx_c = SCAN_DIAG;
 858
 859         if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
 860             lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
 861             if (lc->tu.cu_qp_delta != 0)
 862                 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
 863                     lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
 864             lc->tu.is_cu_qp_delta_coded = 1;
 865
 866             if (lc->tu.cu_qp_delta < -(26 + s->sps->qp_bd_offset / 2) ||
 867                 lc->tu.cu_qp_delta >  (25 + s->sps->qp_bd_offset / 2)) {
 868                 av_log(s->avctx, AV_LOG_ERROR,
 869                        "The cu_qp_delta %d is outside the valid range "
 870                        "[%d, %d].\n",
 871                        lc->tu.cu_qp_delta,
 872                        -(26 + s->sps->qp_bd_offset / 2),
 873                         (25 + s->sps->qp_bd_offset / 2));
 874                 return AVERROR_INVALIDDATA;
 875             }
 876
 877             ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
 878         }
 879
 880         if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
 881             if (lc->tu.cur_intra_pred_mode >= 6 &&
 882                 lc->tu.cur_intra_pred_mode <= 14) {
 883                 scan_idx = SCAN_VERT;
 884             } else if (lc->tu.cur_intra_pred_mode >= 22 &&
 885                        lc->tu.cur_intra_pred_mode <= 30) {
 886                 scan_idx = SCAN_HORIZ;
 887             }
 888
 889             if (lc->pu.intra_pred_mode_c >=  6 &&
 890                 lc->pu.intra_pred_mode_c <= 14) {
 891                 scan_idx_c = SCAN_VERT;
 892             } else if (lc->pu.intra_pred_mode_c >= 22 &&
 893                        lc->pu.intra_pred_mode_c <= 30) {
 894                 scan_idx_c = SCAN_HORIZ;
 895             }
 896         }
 897
 898         if (lc->tt.cbf_luma)
 899             ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
 900         if (log2_trafo_size > 2) {
 901             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
 902                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
 903             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
 904                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
 905         } else if (blk_idx == 3) {
 906             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
 907                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
 908             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
 909                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
 910         }
 911     }
 912     return 0;
 913 }
 914
 915 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
 916 {
 917     int cb_size          = 1 << log2_cb_size;
 918     int log2_min_pu_size = s->sps->log2_min_pu_size;
 919
 920     int min_pu_width     = s->sps->min_pu_width;
 921     int x_end = FFMIN(x0 + cb_size, s->sps->width);
 922     int y_end = FFMIN(y0 + cb_size, s->sps->height);
 923     int i, j;
 924
 925     for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
 926         for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
 927             s->is_pcm[i + j * min_pu_width] = 2;
 928 }
 929
 930 static int hls_transform_tree(HEVCContext *s, int x0, int y0,
 931                               int xBase, int yBase, int cb_xBase, int cb_yBase,
 932                               int log2_cb_size, int log2_trafo_size,
 933                               int trafo_depth, int blk_idx)
 934 {
 935     HEVCLocalContext *lc = s->HEVClc;
 936     uint8_t split_transform_flag;
 937     int ret;
 938
 939     if (trafo_depth > 0 && log2_trafo_size == 2) {
 940         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
 941             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase);
 942         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
 943             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase);
 944     } else {
 945         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
 946         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) = 0;
 947     }
 948
 949     if (lc->cu.intra_split_flag) {
 950         if (trafo_depth == 1)
 951             lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
 952     } else {
 953         lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
 954     }
 955
 956     lc->tt.cbf_luma = 1;
 957
 958     lc->tt.inter_split_flag = s->sps->max_transform_hierarchy_depth_inter == 0 &&
 959                               lc->cu.pred_mode == MODE_INTER &&
 960                               lc->cu.part_mode != PART_2Nx2N &&
 961                               trafo_depth == 0;
 962
 963     if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
 964         log2_trafo_size >  s->sps->log2_min_tb_size    &&
 965         trafo_depth     < lc->cu.max_trafo_depth       &&
 966         !(lc->cu.intra_split_flag && trafo_depth == 0)) {
 967         split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
 968     } else {
 969         split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
 970                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
 971                                lc->tt.inter_split_flag;
 972     }
 973
 974     if (log2_trafo_size > 2) {
 975         if (trafo_depth == 0 ||
 976             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase)) {
 977             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
 978                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 979         }
 980
 981         if (trafo_depth == 0 ||
 982             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase)) {
 983             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
 984                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 985         }
 986     }
 987
 988     if (split_transform_flag) {
 989         int x1 = x0 + ((1 << log2_trafo_size) >> 1);
 990         int y1 = y0 + ((1 << log2_trafo_size) >> 1);
 991
 992         ret = hls_transform_tree(s, x0, y0, x0, y0, cb_xBase, cb_yBase,
 993                                  log2_cb_size, log2_trafo_size - 1,
 994                                  trafo_depth + 1, 0);
 995         if (ret < 0)
 996             return ret;
 997         ret = hls_transform_tree(s, x1, y0, x0, y0, cb_xBase, cb_yBase,
 998                                  log2_cb_size, log2_trafo_size - 1,
 999                                  trafo_depth + 1, 1);
1000         if (ret < 0)
1001             return ret;
1002         ret = hls_transform_tree(s, x0, y1, x0, y0, cb_xBase, cb_yBase,
1003                                  log2_cb_size, log2_trafo_size - 1,
1004                                  trafo_depth + 1, 2);
1005         if (ret < 0)
1006             return ret;
1007         ret = hls_transform_tree(s, x1, y1, x0, y0, cb_xBase, cb_yBase,
1008                                  log2_cb_size, log2_trafo_size - 1,
1009                                  trafo_depth + 1, 3);
1010         if (ret < 0)
1011             return ret;
1012     } else {
1013         int min_tu_size      = 1 << s->sps->log2_min_tb_size;
1014         int log2_min_tu_size = s->sps->log2_min_tb_size;
1015         int min_tu_width     = s->sps->min_tb_width;
1016
1017         if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
1018             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
1019             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
1020             lc->tt.cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
1021         }
1022
1023         ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
1024                                  log2_cb_size, log2_trafo_size, trafo_depth,
1025                                  blk_idx);
1026         if (ret < 0)
1027             return ret;
1028         // TODO: store cbf_luma somewhere else
1029         if (lc->tt.cbf_luma) {
1030             int i, j;
1031             for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
1032                 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
1033                     int x_tu = (x0 + j) >> log2_min_tu_size;
1034                     int y_tu = (y0 + i) >> log2_min_tu_size;
1035                     s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
1036                 }
1037         }
1038         if (!s->sh.disable_deblocking_filter_flag) {
1039             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
1040             if (s->pps->transquant_bypass_enable_flag &&
1041                 lc->cu.cu_transquant_bypass_flag)
1042                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
1043         }
1044     }
1045     return 0;
1046 }
1047
1048 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
1049 {
1050     //TODO: non-4:2:0 support
1051     GetBitContext gb;
1052     int cb_size   = 1 << log2_cb_size;
1053     int stride0   = s->frame->linesize[0];
1054     uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
1055     int   stride1 = s->frame->linesize[1];
1056     uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
1057     int   stride2 = s->frame->linesize[2];
1058     uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
1059
1060     int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
1061     const uint8_t *pcm = skip_bytes(&s->HEVClc->cc, (length + 7) >> 3);
1062     int ret;
1063
1064     if (!s->sh.disable_deblocking_filter_flag)
1065         ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1066
1067     ret = init_get_bits(&gb, pcm, length);
1068     if (ret < 0)
1069         return ret;
1070
1071     s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
1072     s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1073     s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1074     return 0;
1075 }
1076
1077 /**
1078  * 8.5.3.2.2.1 Luma sample unidirectional interpolation process
1079  *
1080  * @param s HEVC decoding context
1081  * @param dst target buffer for block data at block position
1082  * @param dststride stride of the dst buffer
1083  * @param ref reference picture buffer at origin (0, 0)
1084  * @param mv motion vector (relative to block position) to get pixel data from
1085  * @param x_off horizontal position of block from origin (0, 0)
1086  * @param y_off vertical position of block from origin (0, 0)
1087  * @param block_w width of block
1088  * @param block_h height of block
1089  * @param luma_weight weighting factor applied to the luma prediction
1090  * @param luma_offset additive offset applied to the luma prediction value
1091  */
1092
1093 static void luma_mc_uni(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
1094                         AVFrame *ref, const Mv *mv, int x_off, int y_off,
1095                         int block_w, int block_h, int luma_weight, int luma_offset)
1096 {
1097     HEVCLocalContext *lc = s->HEVClc;
1098     uint8_t *src         = ref->data[0];
1099     ptrdiff_t srcstride  = ref->linesize[0];
1100     int pic_width        = s->sps->width;
1101     int pic_height       = s->sps->height;
1102     int mx               = mv->x & 3;
1103     int my               = mv->y & 3;
1104     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1105                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1106     int idx              = ff_hevc_pel_weight[block_w];
1107
1108     x_off += mv->x >> 2;
1109     y_off += mv->y >> 2;
1110     src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
1111
1112     if (x_off < QPEL_EXTRA_BEFORE || y_off < QPEL_EXTRA_AFTER ||
1113         x_off >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1114         y_off >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1115         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1116         int offset     = QPEL_EXTRA_BEFORE * srcstride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1117         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1118
1119         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1120                                  edge_emu_stride, srcstride,
1121                                  block_w + QPEL_EXTRA,
1122                                  block_h + QPEL_EXTRA,
1123                                  x_off - QPEL_EXTRA_BEFORE, y_off - QPEL_EXTRA_BEFORE,
1124                                  pic_width, pic_height);
1125         src = lc->edge_emu_buffer + buf_offset;
1126         srcstride = edge_emu_stride;
1127     }
1128
1129     if (!weight_flag)
1130         s->hevcdsp.put_hevc_qpel_uni[idx][!!my][!!mx](dst, dststride, src, srcstride,
1131                                                       block_h, mx, my, block_w);
1132     else
1133         s->hevcdsp.put_hevc_qpel_uni_w[idx][!!my][!!mx](dst, dststride, src, srcstride,
1134                                                         block_h, s->sh.luma_log2_weight_denom,
1135                                                         luma_weight, luma_offset, mx, my, block_w);
1136 }
1137
1138 /**
1139  * 8.5.3.2.2.1 Luma sample bidirectional interpolation process
1140  *
1141  * @param s HEVC decoding context
1142  * @param dst target buffer for block data at block position
1143  * @param dststride stride of the dst buffer
1144  * @param ref0 reference picture0 buffer at origin (0, 0)
1145  * @param mv0 motion vector0 (relative to block position) to get pixel data from
1146  * @param x_off horizontal position of block from origin (0, 0)
1147  * @param y_off vertical position of block from origin (0, 0)
1148  * @param block_w width of block
1149  * @param block_h height of block
1150  * @param ref1 reference picture1 buffer at origin (0, 0)
1151  * @param mv1 motion vector1 (relative to block position) to get pixel data from
1152  * @param current_mv current motion vector structure
1153  */
1154  static void luma_mc_bi(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
1155                        AVFrame *ref0, const Mv *mv0, int x_off, int y_off,
1156                        int block_w, int block_h, AVFrame *ref1, const Mv *mv1, struct MvField *current_mv)
1157 {
1158     HEVCLocalContext *lc = s->HEVClc;
1159     DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1160     ptrdiff_t src0stride  = ref0->linesize[0];
1161     ptrdiff_t src1stride  = ref1->linesize[0];
1162     int pic_width        = s->sps->width;
1163     int pic_height       = s->sps->height;
1164     int mx0              = mv0->x & 3;
1165     int my0              = mv0->y & 3;
1166     int mx1              = mv1->x & 3;
1167     int my1              = mv1->y & 3;
1168     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1169                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1170     int x_off0           = x_off + (mv0->x >> 2);
1171     int y_off0           = y_off + (mv0->y >> 2);
1172     int x_off1           = x_off + (mv1->x >> 2);
1173     int y_off1           = y_off + (mv1->y >> 2);
1174     int idx              = ff_hevc_pel_weight[block_w];
1175
1176     uint8_t *src0  = ref0->data[0] + y_off0 * src0stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
1177     uint8_t *src1  = ref1->data[0] + y_off1 * src1stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
1178
1179     if (x_off0 < QPEL_EXTRA_BEFORE || y_off0 < QPEL_EXTRA_AFTER ||
1180         x_off0 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1181         y_off0 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1182         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1183         int offset     = QPEL_EXTRA_BEFORE * src0stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1184         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1185
1186         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset,
1187                                  edge_emu_stride, src0stride,
1188                                  block_w + QPEL_EXTRA,
1189                                  block_h + QPEL_EXTRA,
1190                                  x_off0 - QPEL_EXTRA_BEFORE, y_off0 - QPEL_EXTRA_BEFORE,
1191                                  pic_width, pic_height);
1192         src0 = lc->edge_emu_buffer + buf_offset;
1193         src0stride = edge_emu_stride;
1194     }
1195
1196     if (x_off1 < QPEL_EXTRA_BEFORE || y_off1 < QPEL_EXTRA_AFTER ||
1197         x_off1 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1198         y_off1 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1199         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1200         int offset     = QPEL_EXTRA_BEFORE * src1stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1201         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1202
1203         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src1 - offset,
1204                                  edge_emu_stride, src1stride,
1205                                  block_w + QPEL_EXTRA,
1206                                  block_h + QPEL_EXTRA,
1207                                  x_off1 - QPEL_EXTRA_BEFORE, y_off1 - QPEL_EXTRA_BEFORE,
1208                                  pic_width, pic_height);
1209         src1 = lc->edge_emu_buffer2 + buf_offset;
1210         src1stride = edge_emu_stride;
1211     }
1212
1213     s->hevcdsp.put_hevc_qpel[idx][!!my0][!!mx0](tmp, MAX_PB_SIZE, src0, src0stride,
1214                                                 block_h, mx0, my0, block_w);
1215     if (!weight_flag)
1216         s->hevcdsp.put_hevc_qpel_bi[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
1217                                                        block_h, mx1, my1, block_w);
1218     else
1219         s->hevcdsp.put_hevc_qpel_bi_w[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
1220                                                          block_h, s->sh.luma_log2_weight_denom,
1221                                                          s->sh.luma_weight_l0[current_mv->ref_idx[0]],
1222                                                          s->sh.luma_weight_l1[current_mv->ref_idx[1]],
1223                                                          s->sh.luma_offset_l0[current_mv->ref_idx[0]],
1224                                                          s->sh.luma_offset_l1[current_mv->ref_idx[1]],
1225                                                          mx1, my1, block_w);
1226
1227 }
1228
1229 /**
1230  * 8.5.3.2.2.2 Chroma sample uniprediction interpolation process
1231  *
1232  * @param s HEVC decoding context
1233  * @param dst1 target buffer for block data at block position (U plane)
1234  * @param dst2 target buffer for block data at block position (V plane)
1235  * @param dststride stride of the dst1 and dst2 buffers
1236  * @param ref reference picture buffer at origin (0, 0)
1237  * @param mv motion vector (relative to block position) to get pixel data from
1238  * @param x_off horizontal position of block from origin (0, 0)
1239  * @param y_off vertical position of block from origin (0, 0)
1240  * @param block_w width of block
1241  * @param block_h height of block
1242  * @param chroma_weight weighting factor applied to the chroma prediction
1243  * @param chroma_offset additive offset applied to the chroma prediction value
1244  */
1245
1246 static void chroma_mc_uni(HEVCContext *s, uint8_t *dst0,
1247                           ptrdiff_t dststride, uint8_t *src0, ptrdiff_t srcstride, int reflist,
1248                           int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int chroma_weight, int chroma_offset)
1249 {
1250     HEVCLocalContext *lc = s->HEVClc;
1251     int pic_width        = s->sps->width >> s->sps->hshift[1];
1252     int pic_height       = s->sps->height >> s->sps->vshift[1];
1253     const Mv *mv         = &current_mv->mv[reflist];
1254     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1255                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1256     int idx              = ff_hevc_pel_weight[block_w];
1257     int hshift           = s->sps->hshift[1];
1258     int vshift           = s->sps->vshift[1];
1259     intptr_t mx          = mv->x & ((1 << (2 + hshift)) - 1);
1260     intptr_t my          = mv->y & ((1 << (2 + vshift)) - 1);
1261     intptr_t _mx         = mx << (1 - hshift);
1262     intptr_t _my         = my << (1 - vshift);
1263
1264     x_off += mv->x >> (2 + hshift);
1265     y_off += mv->y >> (2 + vshift);
1266     src0  += y_off * srcstride + (x_off << s->sps->pixel_shift);
1267
1268     if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1269         x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1270         y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1271         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1272         int offset0 = EPEL_EXTRA_BEFORE * (srcstride + (1 << s->sps->pixel_shift));
1273         int buf_offset0 = EPEL_EXTRA_BEFORE *
1274                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1275         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset0,
1276                                  edge_emu_stride, srcstride,
1277                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1278                                  x_off - EPEL_EXTRA_BEFORE,
1279                                  y_off - EPEL_EXTRA_BEFORE,
1280                                  pic_width, pic_height);
1281
1282         src0 = lc->edge_emu_buffer + buf_offset0;
1283         srcstride = edge_emu_stride;
1284     }
1285     if (!weight_flag)
1286         s->hevcdsp.put_hevc_epel_uni[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
1287                                                   block_h, _mx, _my, block_w);
1288     else
1289         s->hevcdsp.put_hevc_epel_uni_w[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
1290                                                         block_h, s->sh.chroma_log2_weight_denom,
1291                                                         chroma_weight, chroma_offset, _mx, _my, block_w);
1292 }
1293
1294 /**
1295  * 8.5.3.2.2.2 Chroma sample bidirectional interpolation process
1296  *
1297  * @param s HEVC decoding context
1298  * @param dst target buffer for block data at block position
1299  * @param dststride stride of the dst buffer
1300  * @param ref0 reference picture0 buffer at origin (0, 0)
1301  * @param mv0 motion vector0 (relative to block position) to get pixel data from
1302  * @param x_off horizontal position of block from origin (0, 0)
1303  * @param y_off vertical position of block from origin (0, 0)
1304  * @param block_w width of block
1305  * @param block_h height of block
1306  * @param ref1 reference picture1 buffer at origin (0, 0)
1307  * @param mv1 motion vector1 (relative to block position) to get pixel data from
1308  * @param current_mv current motion vector structure
1309  * @param cidx chroma component(cb, cr)
1310  */
1311 static void chroma_mc_bi(HEVCContext *s, uint8_t *dst0, ptrdiff_t dststride, AVFrame *ref0, AVFrame *ref1,
1312                          int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int cidx)
1313 {
1314     DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1315     int tmpstride = MAX_PB_SIZE;
1316     HEVCLocalContext *lc = s->HEVClc;
1317     uint8_t *src1        = ref0->data[cidx+1];
1318     uint8_t *src2        = ref1->data[cidx+1];
1319     ptrdiff_t src1stride = ref0->linesize[cidx+1];
1320     ptrdiff_t src2stride = ref1->linesize[cidx+1];
1321     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1322                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1323     int pic_width        = s->sps->width >> s->sps->hshift[1];
1324     int pic_height       = s->sps->height >> s->sps->vshift[1];
1325     Mv *mv0              = &current_mv->mv[0];
1326     Mv *mv1              = &current_mv->mv[1];
1327     int hshift = s->sps->hshift[1];
1328     int vshift = s->sps->vshift[1];
1329
1330     intptr_t mx0 = mv0->x & ((1 << (2 + hshift)) - 1);
1331     intptr_t my0 = mv0->y & ((1 << (2 + vshift)) - 1);
1332     intptr_t mx1 = mv1->x & ((1 << (2 + hshift)) - 1);
1333     intptr_t my1 = mv1->y & ((1 << (2 + vshift)) - 1);
1334     intptr_t _mx0 = mx0 << (1 - hshift);
1335     intptr_t _my0 = my0 << (1 - vshift);
1336     intptr_t _mx1 = mx1 << (1 - hshift);
1337     intptr_t _my1 = my1 << (1 - vshift);
1338
1339     int x_off0 = x_off + (mv0->x >> (2 + hshift));
1340     int y_off0 = y_off + (mv0->y >> (2 + vshift));
1341     int x_off1 = x_off + (mv1->x >> (2 + hshift));
1342     int y_off1 = y_off + (mv1->y >> (2 + vshift));
1343     int idx = ff_hevc_pel_weight[block_w];
1344     src1  += y_off0 * src1stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
1345     src2  += y_off1 * src2stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
1346
1347     if (x_off0 < EPEL_EXTRA_BEFORE || y_off0 < EPEL_EXTRA_AFTER ||
1348         x_off0 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1349         y_off0 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1350         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1351         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1352         int buf_offset1 = EPEL_EXTRA_BEFORE *
1353                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1354
1355         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1356                                  edge_emu_stride, src1stride,
1357                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1358                                  x_off0 - EPEL_EXTRA_BEFORE,
1359                                  y_off0 - EPEL_EXTRA_BEFORE,
1360                                  pic_width, pic_height);
1361
1362         src1 = lc->edge_emu_buffer + buf_offset1;
1363         src1stride = edge_emu_stride;
1364     }
1365
1366     if (x_off1 < EPEL_EXTRA_BEFORE || y_off1 < EPEL_EXTRA_AFTER ||
1367         x_off1 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1368         y_off1 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1369         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1370         int offset1 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1371         int buf_offset1 = EPEL_EXTRA_BEFORE *
1372                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1373
1374         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src2 - offset1,
1375                                  edge_emu_stride, src2stride,
1376                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1377                                  x_off1 - EPEL_EXTRA_BEFORE,
1378                                  y_off1 - EPEL_EXTRA_BEFORE,
1379                                  pic_width, pic_height);
1380
1381         src2 = lc->edge_emu_buffer2 + buf_offset1;
1382         src2stride = edge_emu_stride;
1383     }
1384
1385     s->hevcdsp.put_hevc_epel[idx][!!my0][!!mx0](tmp, tmpstride, src1, src1stride,
1386                                                 block_h, _mx0, _my0, block_w);
1387     if (!weight_flag)
1388         s->hevcdsp.put_hevc_epel_bi[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
1389                                                        src2, src2stride, tmp, tmpstride,
1390                                                        block_h, _mx1, _my1, block_w);
1391     else
1392         s->hevcdsp.put_hevc_epel_bi_w[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
1393                                                          src2, src2stride, tmp, tmpstride,
1394                                                          block_h,
1395                                                          s->sh.chroma_log2_weight_denom,
1396                                                          s->sh.chroma_weight_l0[current_mv->ref_idx[0]][cidx],
1397                                                          s->sh.chroma_weight_l1[current_mv->ref_idx[1]][cidx],
1398                                                          s->sh.chroma_offset_l0[current_mv->ref_idx[0]][cidx],
1399                                                          s->sh.chroma_offset_l1[current_mv->ref_idx[1]][cidx],
1400                                                          _mx1, _my1, block_w);
1401 }
1402
1403 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1404                                 const Mv *mv, int y0, int height)
1405 {
1406     int y = (mv->y >> 2) + y0 + height + 9;
1407
1408     if (s->threads_type == FF_THREAD_FRAME )
1409         ff_thread_await_progress(&ref->tf, y, 0);
1410 }
1411
1412 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1413                                 int nPbW, int nPbH,
1414                                 int log2_cb_size, int partIdx)
1415 {
1416 #define POS(c_idx, x, y)                                                              \
1417     &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1418                            (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1419     HEVCLocalContext *lc = s->HEVClc;
1420     int merge_idx = 0;
1421     struct MvField current_mv = {{{ 0 }}};
1422
1423     int min_pu_width = s->sps->min_pu_width;
1424
1425     MvField *tab_mvf = s->ref->tab_mvf;
1426     RefPicList  *refPicList = s->ref->refPicList;
1427     HEVCFrame *ref0, *ref1;
1428     uint8_t *dst0 = POS(0, x0, y0);
1429     uint8_t *dst1 = POS(1, x0, y0);
1430     uint8_t *dst2 = POS(2, x0, y0);
1431     int log2_min_cb_size = s->sps->log2_min_cb_size;
1432     int min_cb_width     = s->sps->min_cb_width;
1433     int x_cb             = x0 >> log2_min_cb_size;
1434     int y_cb             = y0 >> log2_min_cb_size;
1435     int ref_idx[2];
1436     int mvp_flag[2];
1437     int x_pu, y_pu;
1438     int i, j;
1439
1440     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1441         if (s->sh.max_num_merge_cand > 1)
1442             merge_idx = ff_hevc_merge_idx_decode(s);
1443         else
1444             merge_idx = 0;
1445
1446         ff_hevc_luma_mv_merge_mode(s, x0, y0,
1447                                    1 << log2_cb_size,
1448                                    1 << log2_cb_size,
1449                                    log2_cb_size, partIdx,
1450                                    merge_idx, &current_mv);
1451         x_pu = x0 >> s->sps->log2_min_pu_size;
1452         y_pu = y0 >> s->sps->log2_min_pu_size;
1453
1454         for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1455             for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1456                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1457     } else { /* MODE_INTER */
1458         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1459         if (lc->pu.merge_flag) {
1460             if (s->sh.max_num_merge_cand > 1)
1461                 merge_idx = ff_hevc_merge_idx_decode(s);
1462             else
1463                 merge_idx = 0;
1464
1465             ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1466                                        partIdx, merge_idx, &current_mv);
1467             x_pu = x0 >> s->sps->log2_min_pu_size;
1468             y_pu = y0 >> s->sps->log2_min_pu_size;
1469
1470             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1471                 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1472                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1473         } else {
1474             enum InterPredIdc inter_pred_idc = PRED_L0;
1475             ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1476             current_mv.pred_flag = 0;
1477             if (s->sh.slice_type == B_SLICE)
1478                 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1479
1480             if (inter_pred_idc != PRED_L1) {
1481                 if (s->sh.nb_refs[L0]) {
1482                     ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1483                     current_mv.ref_idx[0] = ref_idx[0];
1484                 }
1485                 current_mv.pred_flag = PF_L0;
1486                 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
1487                 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
1488                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1489                                          partIdx, merge_idx, &current_mv,
1490                                          mvp_flag[0], 0);
1491                 current_mv.mv[0].x += lc->pu.mvd.x;
1492                 current_mv.mv[0].y += lc->pu.mvd.y;
1493             }
1494
1495             if (inter_pred_idc != PRED_L0) {
1496                 if (s->sh.nb_refs[L1]) {
1497                     ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1498                     current_mv.ref_idx[1] = ref_idx[1];
1499                 }
1500
1501                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1502                     lc->pu.mvd.x = 0;
1503                     lc->pu.mvd.y = 0;
1504                 } else {
1505                     ff_hevc_hls_mvd_coding(s, x0, y0, 1);
1506                 }
1507
1508                 current_mv.pred_flag += PF_L1;
1509                 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
1510                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1511                                          partIdx, merge_idx, &current_mv,
1512                                          mvp_flag[1], 1);
1513                 current_mv.mv[1].x += lc->pu.mvd.x;
1514                 current_mv.mv[1].y += lc->pu.mvd.y;
1515             }
1516
1517             x_pu = x0 >> s->sps->log2_min_pu_size;
1518             y_pu = y0 >> s->sps->log2_min_pu_size;
1519
1520             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1521                 for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1522                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1523         }
1524     }
1525
1526     if (current_mv.pred_flag & PF_L0) {
1527         ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1528         if (!ref0)
1529             return;
1530         hevc_await_progress(s, ref0, &current_mv.mv[0], y0, nPbH);
1531     }
1532     if (current_mv.pred_flag & PF_L1) {
1533         ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1534         if (!ref1)
1535             return;
1536         hevc_await_progress(s, ref1, &current_mv.mv[1], y0, nPbH);
1537     }
1538
1539     if (current_mv.pred_flag == PF_L0) {
1540         int x0_c = x0 >> s->sps->hshift[1];
1541         int y0_c = y0 >> s->sps->vshift[1];
1542         int nPbW_c = nPbW >> s->sps->hshift[1];
1543         int nPbH_c = nPbH >> s->sps->vshift[1];
1544
1545         luma_mc_uni(s, dst0, s->frame->linesize[0], ref0->frame,
1546                     &current_mv.mv[0], x0, y0, nPbW, nPbH,
1547                     s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1548                     s->sh.luma_offset_l0[current_mv.ref_idx[0]]);
1549
1550         chroma_mc_uni(s, dst1, s->frame->linesize[1], ref0->frame->data[1], ref0->frame->linesize[1],
1551                       0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1552                       s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0]);
1553         chroma_mc_uni(s, dst2, s->frame->linesize[2], ref0->frame->data[2], ref0->frame->linesize[2],
1554                       0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1555                       s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1]);
1556     } else if (current_mv.pred_flag == PF_L1) {
1557         int x0_c = x0 >> s->sps->hshift[1];
1558         int y0_c = y0 >> s->sps->vshift[1];
1559         int nPbW_c = nPbW >> s->sps->hshift[1];
1560         int nPbH_c = nPbH >> s->sps->vshift[1];
1561
1562         luma_mc_uni(s, dst0, s->frame->linesize[0], ref1->frame,
1563                     &current_mv.mv[1], x0, y0, nPbW, nPbH,
1564                     s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1565                     s->sh.luma_offset_l1[current_mv.ref_idx[1]]);
1566
1567         chroma_mc_uni(s, dst1, s->frame->linesize[1], ref1->frame->data[1], ref1->frame->linesize[1],
1568                       1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1569                       s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0]);
1570
1571         chroma_mc_uni(s, dst2, s->frame->linesize[2], ref1->frame->data[2], ref1->frame->linesize[2],
1572                       1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1573                       s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1]);
1574     } else if (current_mv.pred_flag == PF_BI) {
1575         int x0_c = x0 >> s->sps->hshift[1];
1576         int y0_c = y0 >> s->sps->vshift[1];
1577         int nPbW_c = nPbW >> s->sps->hshift[1];
1578         int nPbH_c = nPbH >> s->sps->vshift[1];
1579
1580         luma_mc_bi(s, dst0, s->frame->linesize[0], ref0->frame,
1581                    &current_mv.mv[0], x0, y0, nPbW, nPbH,
1582                    ref1->frame, &current_mv.mv[1], &current_mv);
1583
1584         chroma_mc_bi(s, dst1, s->frame->linesize[1], ref0->frame, ref1->frame,
1585                      x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 0);
1586
1587         chroma_mc_bi(s, dst2, s->frame->linesize[2], ref0->frame, ref1->frame,
1588                      x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 1);
1589     }
1590 }
1591
1592 /**
1593  * 8.4.1
1594  */
1595 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1596                                 int prev_intra_luma_pred_flag)
1597 {
1598     HEVCLocalContext *lc = s->HEVClc;
1599     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1600     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1601     int min_pu_width     = s->sps->min_pu_width;
1602     int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
1603     int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1604     int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1605
1606     int cand_up   = (lc->ctb_up_flag || y0b) ?
1607                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1608     int cand_left = (lc->ctb_left_flag || x0b) ?
1609                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
1610
1611     int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1612
1613     MvField *tab_mvf = s->ref->tab_mvf;
1614     int intra_pred_mode;
1615     int candidate[3];
1616     int i, j;
1617
1618     // intra_pred_mode prediction does not cross vertical CTB boundaries
1619     if ((y0 - 1) < y_ctb)
1620         cand_up = INTRA_DC;
1621
1622     if (cand_left == cand_up) {
1623         if (cand_left < 2) {
1624             candidate[0] = INTRA_PLANAR;
1625             candidate[1] = INTRA_DC;
1626             candidate[2] = INTRA_ANGULAR_26;
1627         } else {
1628             candidate[0] = cand_left;
1629             candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1630             candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1631         }
1632     } else {
1633         candidate[0] = cand_left;
1634         candidate[1] = cand_up;
1635         if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1636             candidate[2] = INTRA_PLANAR;
1637         } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1638             candidate[2] = INTRA_DC;
1639         } else {
1640             candidate[2] = INTRA_ANGULAR_26;
1641         }
1642     }
1643
1644     if (prev_intra_luma_pred_flag) {
1645         intra_pred_mode = candidate[lc->pu.mpm_idx];
1646     } else {
1647         if (candidate[0] > candidate[1])
1648             FFSWAP(uint8_t, candidate[0], candidate[1]);
1649         if (candidate[0] > candidate[2])
1650             FFSWAP(uint8_t, candidate[0], candidate[2]);
1651         if (candidate[1] > candidate[2])
1652             FFSWAP(uint8_t, candidate[1], candidate[2]);
1653
1654         intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1655         for (i = 0; i < 3; i++)
1656             if (intra_pred_mode >= candidate[i])
1657                 intra_pred_mode++;
1658     }
1659
1660     /* write the intra prediction units into the mv array */
1661     if (!size_in_pus)
1662         size_in_pus = 1;
1663     for (i = 0; i < size_in_pus; i++) {
1664         memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1665                intra_pred_mode, size_in_pus);
1666
1667         for (j = 0; j < size_in_pus; j++) {
1668             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag = PF_INTRA;
1669         }
1670     }
1671
1672     return intra_pred_mode;
1673 }
1674
1675 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1676                                           int log2_cb_size, int ct_depth)
1677 {
1678     int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1679     int x_cb   = x0 >> s->sps->log2_min_cb_size;
1680     int y_cb   = y0 >> s->sps->log2_min_cb_size;
1681     int y;
1682
1683     for (y = 0; y < length; y++)
1684         memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1685                ct_depth, length);
1686 }
1687
1688 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1689                                   int log2_cb_size)
1690 {
1691     HEVCLocalContext *lc = s->HEVClc;
1692     static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1693     uint8_t prev_intra_luma_pred_flag[4];
1694     int split   = lc->cu.part_mode == PART_NxN;
1695     int pb_size = (1 << log2_cb_size) >> split;
1696     int side    = split + 1;
1697     int chroma_mode;
1698     int i, j;
1699
1700     for (i = 0; i < side; i++)
1701         for (j = 0; j < side; j++)
1702             prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1703
1704     for (i = 0; i < side; i++) {
1705         for (j = 0; j < side; j++) {
1706             if (prev_intra_luma_pred_flag[2 * i + j])
1707                 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1708             else
1709                 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1710
1711             lc->pu.intra_pred_mode[2 * i + j] =
1712                 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1713                                      prev_intra_luma_pred_flag[2 * i + j]);
1714         }
1715     }
1716
1717     chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1718     if (chroma_mode != 4) {
1719         if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1720             lc->pu.intra_pred_mode_c = 34;
1721         else
1722             lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1723     } else {
1724         lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1725     }
1726 }
1727
1728 static void intra_prediction_unit_default_value(HEVCContext *s,
1729                                                 int x0, int y0,
1730                                                 int log2_cb_size)
1731 {
1732     HEVCLocalContext *lc = s->HEVClc;
1733     int pb_size          = 1 << log2_cb_size;
1734     int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
1735     int min_pu_width     = s->sps->min_pu_width;
1736     MvField *tab_mvf     = s->ref->tab_mvf;
1737     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1738     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1739     int j, k;
1740
1741     if (size_in_pus == 0)
1742         size_in_pus = 1;
1743     for (j = 0; j < size_in_pus; j++)
1744         memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
1745     if (lc->cu.pred_mode == MODE_INTRA)
1746         for (j = 0; j < size_in_pus; j++)
1747             for (k = 0; k < size_in_pus; k++)
1748                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].pred_flag = PF_INTRA;
1749 }
1750
1751 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
1752 {
1753     int cb_size          = 1 << log2_cb_size;
1754     HEVCLocalContext *lc = s->HEVClc;
1755     int log2_min_cb_size = s->sps->log2_min_cb_size;
1756     int length           = cb_size >> log2_min_cb_size;
1757     int min_cb_width     = s->sps->min_cb_width;
1758     int x_cb             = x0 >> log2_min_cb_size;
1759     int y_cb             = y0 >> log2_min_cb_size;
1760     int x, y, ret;
1761     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1762
1763     lc->cu.x                = x0;
1764     lc->cu.y                = y0;
1765     lc->cu.rqt_root_cbf     = 1;
1766     lc->cu.pred_mode        = MODE_INTRA;
1767     lc->cu.part_mode        = PART_2Nx2N;
1768     lc->cu.intra_split_flag = 0;
1769     lc->cu.pcm_flag         = 0;
1770
1771     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
1772     for (x = 0; x < 4; x++)
1773         lc->pu.intra_pred_mode[x] = 1;
1774     if (s->pps->transquant_bypass_enable_flag) {
1775         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
1776         if (lc->cu.cu_transquant_bypass_flag)
1777             set_deblocking_bypass(s, x0, y0, log2_cb_size);
1778     } else
1779         lc->cu.cu_transquant_bypass_flag = 0;
1780
1781     if (s->sh.slice_type != I_SLICE) {
1782         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
1783
1784         x = y_cb * min_cb_width + x_cb;
1785         for (y = 0; y < length; y++) {
1786             memset(&s->skip_flag[x], skip_flag, length);
1787             x += min_cb_width;
1788         }
1789         lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
1790     }
1791
1792     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1793         hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1794         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1795
1796         if (!s->sh.disable_deblocking_filter_flag)
1797             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1798     } else {
1799         if (s->sh.slice_type != I_SLICE)
1800             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
1801         if (lc->cu.pred_mode != MODE_INTRA ||
1802             log2_cb_size == s->sps->log2_min_cb_size) {
1803             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
1804             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
1805                                       lc->cu.pred_mode == MODE_INTRA;
1806         }
1807
1808         if (lc->cu.pred_mode == MODE_INTRA) {
1809             if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
1810                 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
1811                 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
1812                 lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
1813             }
1814             if (lc->cu.pcm_flag) {
1815                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1816                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
1817                 if (s->sps->pcm.loop_filter_disable_flag)
1818                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
1819
1820                 if (ret < 0)
1821                     return ret;
1822             } else {
1823                 intra_prediction_unit(s, x0, y0, log2_cb_size);
1824             }
1825         } else {
1826             intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1827             switch (lc->cu.part_mode) {
1828             case PART_2Nx2N:
1829                 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1830                 break;
1831             case PART_2NxN:
1832                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
1833                 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
1834                 break;
1835             case PART_Nx2N:
1836                 hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
1837                 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
1838                 break;
1839             case PART_2NxnU:
1840                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
1841                 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
1842                 break;
1843             case PART_2NxnD:
1844                 hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
1845                 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
1846                 break;
1847             case PART_nLx2N:
1848                 hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
1849                 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
1850                 break;
1851             case PART_nRx2N:
1852                 hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
1853                 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
1854                 break;
1855             case PART_NxN:
1856                 hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
1857                 hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
1858                 hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
1859                 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
1860                 break;
1861             }
1862         }
1863
1864         if (!lc->cu.pcm_flag) {
1865             if (lc->cu.pred_mode != MODE_INTRA &&
1866                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
1867                 lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
1868             }
1869             if (lc->cu.rqt_root_cbf) {
1870                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
1871                                          s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
1872                                          s->sps->max_transform_hierarchy_depth_inter;
1873                 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
1874                                          log2_cb_size,
1875                                          log2_cb_size, 0, 0);
1876                 if (ret < 0)
1877                     return ret;
1878             } else {
1879                 if (!s->sh.disable_deblocking_filter_flag)
1880                     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1881             }
1882         }
1883     }
1884
1885     if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
1886         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
1887
1888     x = y_cb * min_cb_width + x_cb;
1889     for (y = 0; y < length; y++) {
1890         memset(&s->qp_y_tab[x], lc->qp_y, length);
1891         x += min_cb_width;
1892     }
1893
1894     if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1895        ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0) {
1896         lc->qPy_pred = lc->qp_y;
1897     }
1898
1899     set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
1900
1901     return 0;
1902 }
1903
1904 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
1905                                int log2_cb_size, int cb_depth)
1906 {
1907     HEVCLocalContext *lc = s->HEVClc;
1908     const int cb_size    = 1 << log2_cb_size;
1909     int ret;
1910     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1911
1912     lc->ct.depth = cb_depth;
1913     if (x0 + cb_size <= s->sps->width  &&
1914         y0 + cb_size <= s->sps->height &&
1915         log2_cb_size > s->sps->log2_min_cb_size) {
1916         SAMPLE(s->split_cu_flag, x0, y0) =
1917             ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
1918     } else {
1919         SAMPLE(s->split_cu_flag, x0, y0) =
1920             (log2_cb_size > s->sps->log2_min_cb_size);
1921     }
1922     if (s->pps->cu_qp_delta_enabled_flag &&
1923         log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
1924         lc->tu.is_cu_qp_delta_coded = 0;
1925         lc->tu.cu_qp_delta          = 0;
1926     }
1927
1928     if (SAMPLE(s->split_cu_flag, x0, y0)) {
1929         const int cb_size_split = cb_size >> 1;
1930         const int x1 = x0 + cb_size_split;
1931         const int y1 = y0 + cb_size_split;
1932
1933         int more_data = 0;
1934
1935         more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
1936         if (more_data < 0)
1937             return more_data;
1938
1939         if (more_data && x1 < s->sps->width) {
1940             more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
1941             if (more_data < 0)
1942                 return more_data;
1943         }
1944         if (more_data && y1 < s->sps->height) {
1945             more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
1946             if (more_data < 0)
1947                 return more_data;
1948         }
1949         if (more_data && x1 < s->sps->width &&
1950             y1 < s->sps->height) {
1951             more_data = hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
1952             if (more_data < 0)
1953                 return more_data;
1954         }
1955
1956         if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1957             ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0)
1958             lc->qPy_pred = lc->qp_y;
1959
1960         if (more_data)
1961             return ((x1 + cb_size_split) < s->sps->width ||
1962                     (y1 + cb_size_split) < s->sps->height);
1963         else
1964             return 0;
1965     } else {
1966         ret = hls_coding_unit(s, x0, y0, log2_cb_size);
1967         if (ret < 0)
1968             return ret;
1969         if ((!((x0 + cb_size) %
1970                (1 << (s->sps->log2_ctb_size))) ||
1971              (x0 + cb_size >= s->sps->width)) &&
1972             (!((y0 + cb_size) %
1973                (1 << (s->sps->log2_ctb_size))) ||
1974              (y0 + cb_size >= s->sps->height))) {
1975             int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
1976             return !end_of_slice_flag;
1977         } else {
1978             return 1;
1979         }
1980     }
1981
1982     return 0;
1983 }
1984
1985 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
1986                                  int ctb_addr_ts)
1987 {
1988     HEVCLocalContext *lc  = s->HEVClc;
1989     int ctb_size          = 1 << s->sps->log2_ctb_size;
1990     int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1991     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
1992
1993     int tile_left_boundary, tile_up_boundary;
1994     int slice_left_boundary, slice_up_boundary;
1995
1996     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
1997
1998     if (s->pps->entropy_coding_sync_enabled_flag) {
1999         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
2000             lc->first_qp_group = 1;
2001         lc->end_of_tiles_x = s->sps->width;
2002     } else if (s->pps->tiles_enabled_flag) {
2003         if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
2004             int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
2005             lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
2006             lc->first_qp_group   = 1;
2007         }
2008     } else {
2009         lc->end_of_tiles_x = s->sps->width;
2010     }
2011
2012     lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
2013
2014     if (s->pps->tiles_enabled_flag) {
2015         tile_left_boundary = x_ctb > 0 &&
2016                              s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
2017         slice_left_boundary = x_ctb > 0 &&
2018                               s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
2019         tile_up_boundary  = y_ctb > 0 &&
2020                             s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
2021         slice_up_boundary = y_ctb > 0 &&
2022                             s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
2023     } else {
2024         tile_left_boundary =
2025         tile_up_boundary   = 0;
2026         slice_left_boundary = ctb_addr_in_slice <= 0;
2027         slice_up_boundary   = ctb_addr_in_slice < s->sps->ctb_width;
2028     }
2029     lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
2030     lc->slice_or_tiles_up_boundary   = slice_up_boundary   + (tile_up_boundary   << 1);
2031     lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0)                  && !tile_left_boundary);
2032     lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
2033     lc->ctb_up_right_flag = ((y_ctb > 0)                 && (ctb_addr_in_slice+1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->sps->ctb_width]]));
2034     lc->ctb_up_left_flag  = ((x_ctb > 0) && (y_ctb > 0)  && (ctb_addr_in_slice-1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1 - s->sps->ctb_width]]));
2035 }
2036
2037 static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
2038 {
2039     HEVCContext *s  = avctxt->priv_data;
2040     int ctb_size    = 1 << s->sps->log2_ctb_size;
2041     int more_data   = 1;
2042     int x_ctb       = 0;
2043     int y_ctb       = 0;
2044     int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
2045
2046     if (!ctb_addr_ts && s->sh.dependent_slice_segment_flag) {
2047         av_log(s->avctx, AV_LOG_ERROR, "Impossible initial tile.\n");
2048         return AVERROR_INVALIDDATA;
2049     }
2050
2051     if (s->sh.dependent_slice_segment_flag) {
2052         int prev_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts - 1];
2053         if (s->tab_slice_address[prev_rs] != s->sh.slice_addr) {
2054             av_log(s->avctx, AV_LOG_ERROR, "Previous slice segment missing\n");
2055             return AVERROR_INVALIDDATA;
2056         }
2057     }
2058
2059     while (more_data && ctb_addr_ts < s->sps->ctb_size) {
2060         int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2061
2062         x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2063         y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2064         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2065
2066         ff_hevc_cabac_init(s, ctb_addr_ts);
2067
2068         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2069
2070         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
2071         s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
2072         s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
2073
2074         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2075         if (more_data < 0) {
2076             s->tab_slice_address[ctb_addr_rs] = -1;
2077             return more_data;
2078         }
2079
2080
2081         ctb_addr_ts++;
2082         ff_hevc_save_states(s, ctb_addr_ts);
2083         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2084     }
2085
2086     if (x_ctb + ctb_size >= s->sps->width &&
2087         y_ctb + ctb_size >= s->sps->height)
2088         ff_hevc_hls_filter(s, x_ctb, y_ctb);
2089
2090     return ctb_addr_ts;
2091 }
2092
2093 static int hls_slice_data(HEVCContext *s)
2094 {
2095     int arg[2];
2096     int ret[2];
2097
2098     arg[0] = 0;
2099     arg[1] = 1;
2100
2101     s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
2102     return ret[0];
2103 }
2104 static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
2105 {
2106     HEVCContext *s1  = avctxt->priv_data, *s;
2107     HEVCLocalContext *lc;
2108     int ctb_size    = 1<< s1->sps->log2_ctb_size;
2109     int more_data   = 1;
2110     int *ctb_row_p    = input_ctb_row;
2111     int ctb_row = ctb_row_p[job];
2112     int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
2113     int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
2114     int thread = ctb_row % s1->threads_number;
2115     int ret;
2116
2117     s = s1->sList[self_id];
2118     lc = s->HEVClc;
2119
2120     if(ctb_row) {
2121         ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
2122
2123         if (ret < 0)
2124             return ret;
2125         ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
2126     }
2127
2128     while(more_data && ctb_addr_ts < s->sps->ctb_size) {
2129         int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
2130         int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
2131
2132         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2133
2134         ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
2135
2136         if (avpriv_atomic_int_get(&s1->wpp_err)){
2137             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2138             return 0;
2139         }
2140
2141         ff_hevc_cabac_init(s, ctb_addr_ts);
2142         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2143         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2144
2145         if (more_data < 0) {
2146             s->tab_slice_address[ctb_addr_rs] = -1;
2147             return more_data;
2148         }
2149
2150         ctb_addr_ts++;
2151
2152         ff_hevc_save_states(s, ctb_addr_ts);
2153         ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
2154         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2155
2156         if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
2157             avpriv_atomic_int_set(&s1->wpp_err,  1);
2158             ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2159             return 0;
2160         }
2161
2162         if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
2163             ff_hevc_hls_filter(s, x_ctb, y_ctb);
2164             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2165             return ctb_addr_ts;
2166         }
2167         ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2168         x_ctb+=ctb_size;
2169
2170         if(x_ctb >= s->sps->width) {
2171             break;
2172         }
2173     }
2174     ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2175
2176     return 0;
2177 }
2178
2179 static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
2180 {
2181     HEVCLocalContext *lc = s->HEVClc;
2182     int *ret = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2183     int *arg = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2184     int offset;
2185     int startheader, cmpt = 0;
2186     int i, j, res = 0;
2187
2188
2189     if (!s->sList[1]) {
2190         ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
2191
2192
2193         for (i = 1; i < s->threads_number; i++) {
2194             s->sList[i] = av_malloc(sizeof(HEVCContext));
2195             memcpy(s->sList[i], s, sizeof(HEVCContext));
2196             s->HEVClcList[i] = av_malloc(sizeof(HEVCLocalContext));
2197             s->sList[i]->HEVClc = s->HEVClcList[i];
2198         }
2199     }
2200
2201     offset = (lc->gb.index >> 3);
2202
2203     for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
2204         if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2205             startheader--;
2206             cmpt++;
2207         }
2208     }
2209
2210     for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
2211         offset += (s->sh.entry_point_offset[i - 1] - cmpt);
2212         for (j = 0, cmpt = 0, startheader = offset
2213              + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
2214             if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2215                 startheader--;
2216                 cmpt++;
2217             }
2218         }
2219         s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
2220         s->sh.offset[i - 1] = offset;
2221
2222     }
2223     if (s->sh.num_entry_point_offsets != 0) {
2224         offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
2225         s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
2226         s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
2227
2228     }
2229     s->data = nal;
2230
2231     for (i = 1; i < s->threads_number; i++) {
2232         s->sList[i]->HEVClc->first_qp_group = 1;
2233         s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
2234         memcpy(s->sList[i], s, sizeof(HEVCContext));
2235         s->sList[i]->HEVClc = s->HEVClcList[i];
2236     }
2237
2238     avpriv_atomic_int_set(&s->wpp_err, 0);
2239     ff_reset_entries(s->avctx);
2240
2241     for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
2242         arg[i] = i;
2243         ret[i] = 0;
2244     }
2245
2246     if (s->pps->entropy_coding_sync_enabled_flag)
2247         s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
2248
2249     for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
2250         res += ret[i];
2251     av_free(ret);
2252     av_free(arg);
2253     return res;
2254 }
2255
2256 /**
2257  * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2258  * 0 if the unit should be skipped, 1 otherwise
2259  */
2260 static int hls_nal_unit(HEVCContext *s)
2261 {
2262     GetBitContext *gb = &s->HEVClc->gb;
2263     int nuh_layer_id;
2264
2265     if (get_bits1(gb) != 0)
2266         return AVERROR_INVALIDDATA;
2267
2268     s->nal_unit_type = get_bits(gb, 6);
2269
2270     nuh_layer_id   = get_bits(gb, 6);
2271     s->temporal_id = get_bits(gb, 3) - 1;
2272     if (s->temporal_id < 0)
2273         return AVERROR_INVALIDDATA;
2274
2275     av_log(s->avctx, AV_LOG_DEBUG,
2276            "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2277            s->nal_unit_type, nuh_layer_id, s->temporal_id);
2278
2279     return nuh_layer_id == 0;
2280 }
2281
2282 static void restore_tqb_pixels(HEVCContext *s)
2283 {
2284     int min_pu_size = 1 << s->sps->log2_min_pu_size;
2285     int x, y, c_idx;
2286
2287     for (c_idx = 0; c_idx < 3; c_idx++) {
2288         ptrdiff_t stride = s->frame->linesize[c_idx];
2289         int hshift       = s->sps->hshift[c_idx];
2290         int vshift       = s->sps->vshift[c_idx];
2291         for (y = 0; y < s->sps->min_pu_height; y++) {
2292             for (x = 0; x < s->sps->min_pu_width; x++) {
2293                 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2294                     int n;
2295                     int len      = min_pu_size >> hshift;
2296                     uint8_t *src = &s->frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
2297                     uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
2298                     for (n = 0; n < (min_pu_size >> vshift); n++) {
2299                         memcpy(dst, src, len);
2300                         src += stride;
2301                         dst += stride;
2302                     }
2303                 }
2304             }
2305         }
2306     }
2307 }
2308
2309 static int set_side_data(HEVCContext *s)
2310 {
2311     AVFrame *out = s->ref->frame;
2312
2313     if (s->sei_frame_packing_present &&
2314         s->frame_packing_arrangement_type >= 3 &&
2315         s->frame_packing_arrangement_type <= 5 &&
2316         s->content_interpretation_type > 0 &&
2317         s->content_interpretation_type < 3) {
2318         AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2319         if (!stereo)
2320             return AVERROR(ENOMEM);
2321
2322         switch (s->frame_packing_arrangement_type) {
2323         case 3:
2324             if (s->quincunx_subsampling)
2325                 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2326             else
2327                 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2328             break;
2329         case 4:
2330             stereo->type = AV_STEREO3D_TOPBOTTOM;
2331             break;
2332         case 5:
2333             stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2334             break;
2335         }
2336
2337         if (s->content_interpretation_type == 2)
2338             stereo->flags = AV_STEREO3D_FLAG_INVERT;
2339     }
2340
2341     return 0;
2342 }
2343
2344 static int hevc_frame_start(HEVCContext *s)
2345 {
2346     HEVCLocalContext *lc = s->HEVClc;
2347     int pic_size_in_ctb  = ((s->sps->width  >> s->sps->log2_min_cb_size) + 1) *
2348                            ((s->sps->height >> s->sps->log2_min_cb_size) + 1);
2349     int ret;
2350     AVFrame *cur_frame;
2351
2352     memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2353     memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
2354     memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
2355     memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
2356     memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
2357
2358     s->is_decoded        = 0;
2359     s->first_nal_type    = s->nal_unit_type;
2360
2361     if (s->pps->tiles_enabled_flag)
2362         lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2363
2364     ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2365                               s->poc);
2366     if (ret < 0)
2367         goto fail;
2368
2369     ret = ff_hevc_frame_rps(s);
2370     if (ret < 0) {
2371         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2372         goto fail;
2373     }
2374
2375     ret = set_side_data(s);
2376     if (ret < 0)
2377         goto fail;
2378
2379     cur_frame = s->sps->sao_enabled ? s->sao_frame : s->frame;
2380     cur_frame->pict_type = 3 - s->sh.slice_type;
2381
2382     av_frame_unref(s->output_frame);
2383     ret = ff_hevc_output_frame(s, s->output_frame, 0);
2384     if (ret < 0)
2385         goto fail;
2386
2387     ff_thread_finish_setup(s->avctx);
2388
2389     return 0;
2390
2391 fail:
2392     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2393         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2394     s->ref = NULL;
2395     return ret;
2396 }
2397
2398 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2399 {
2400     HEVCLocalContext *lc = s->HEVClc;
2401     GetBitContext *gb    = &lc->gb;
2402     int ctb_addr_ts, ret;
2403
2404     ret = init_get_bits8(gb, nal, length);
2405     if (ret < 0)
2406         return ret;
2407
2408     ret = hls_nal_unit(s);
2409     if (ret < 0) {
2410         av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2411                s->nal_unit_type);
2412         goto fail;
2413     } else if (!ret)
2414         return 0;
2415
2416     switch (s->nal_unit_type) {
2417     case NAL_VPS:
2418         ret = ff_hevc_decode_nal_vps(s);
2419         if (ret < 0)
2420             goto fail;
2421         break;
2422     case NAL_SPS:
2423         ret = ff_hevc_decode_nal_sps(s);
2424         if (ret < 0)
2425             goto fail;
2426         break;
2427     case NAL_PPS:
2428         ret = ff_hevc_decode_nal_pps(s);
2429         if (ret < 0)
2430             goto fail;
2431         break;
2432     case NAL_SEI_PREFIX:
2433     case NAL_SEI_SUFFIX:
2434         ret = ff_hevc_decode_nal_sei(s);
2435         if (ret < 0)
2436             goto fail;
2437         break;
2438     case NAL_TRAIL_R:
2439     case NAL_TRAIL_N:
2440     case NAL_TSA_N:
2441     case NAL_TSA_R:
2442     case NAL_STSA_N:
2443     case NAL_STSA_R:
2444     case NAL_BLA_W_LP:
2445     case NAL_BLA_W_RADL:
2446     case NAL_BLA_N_LP:
2447     case NAL_IDR_W_RADL:
2448     case NAL_IDR_N_LP:
2449     case NAL_CRA_NUT:
2450     case NAL_RADL_N:
2451     case NAL_RADL_R:
2452     case NAL_RASL_N:
2453     case NAL_RASL_R:
2454         ret = hls_slice_header(s);
2455         if (ret < 0)
2456             return ret;
2457
2458         if (s->max_ra == INT_MAX) {
2459             if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2460                 s->max_ra = s->poc;
2461             } else {
2462                 if (IS_IDR(s))
2463                     s->max_ra = INT_MIN;
2464             }
2465         }
2466
2467         if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2468             s->poc <= s->max_ra) {
2469             s->is_decoded = 0;
2470             break;
2471         } else {
2472             if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2473                 s->max_ra = INT_MIN;
2474         }
2475
2476         if (s->sh.first_slice_in_pic_flag) {
2477             ret = hevc_frame_start(s);
2478             if (ret < 0)
2479                 return ret;
2480         } else if (!s->ref) {
2481             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2482             goto fail;
2483         }
2484
2485         if (s->nal_unit_type != s->first_nal_type) {
2486             av_log(s->avctx, AV_LOG_ERROR,
2487                    "Non-matching NAL types of the VCL NALUs: %d %d\n",
2488                    s->first_nal_type, s->nal_unit_type);
2489             return AVERROR_INVALIDDATA;
2490         }
2491
2492         if (!s->sh.dependent_slice_segment_flag &&
2493             s->sh.slice_type != I_SLICE) {
2494             ret = ff_hevc_slice_rpl(s);
2495             if (ret < 0) {
2496                 av_log(s->avctx, AV_LOG_WARNING,
2497                        "Error constructing the reference lists for the current slice.\n");
2498                 goto fail;
2499             }
2500         }
2501
2502         if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
2503             ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
2504         else
2505             ctb_addr_ts = hls_slice_data(s);
2506         if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2507             s->is_decoded = 1;
2508             if ((s->pps->transquant_bypass_enable_flag ||
2509                  (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2510                 s->sps->sao_enabled)
2511                 restore_tqb_pixels(s);
2512         }
2513
2514         if (ctb_addr_ts < 0) {
2515             ret = ctb_addr_ts;
2516             goto fail;
2517         }
2518         break;
2519     case NAL_EOS_NUT:
2520     case NAL_EOB_NUT:
2521         s->seq_decode = (s->seq_decode + 1) & 0xff;
2522         s->max_ra     = INT_MAX;
2523         break;
2524     case NAL_AUD:
2525     case NAL_FD_NUT:
2526         break;
2527     default:
2528         av_log(s->avctx, AV_LOG_INFO,
2529                "Skipping NAL unit %d\n", s->nal_unit_type);
2530     }
2531
2532     return 0;
2533 fail:
2534     if (s->avctx->err_recognition & AV_EF_EXPLODE)
2535         return ret;
2536     return 0;
2537 }
2538
2539 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2540  * between these functions would be nice. */
2541 int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
2542                          HEVCNAL *nal)
2543 {
2544     int i, si, di;
2545     uint8_t *dst;
2546
2547     s->skipped_bytes = 0;
2548 #define STARTCODE_TEST                                                  \
2549         if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
2550             if (src[i + 2] != 3) {                                      \
2551                 /* startcode, so we must be past the end */             \
2552                 length = i;                                             \
2553             }                                                           \
2554             break;                                                      \
2555         }
2556 #if HAVE_FAST_UNALIGNED
2557 #define FIND_FIRST_ZERO                                                 \
2558         if (i > 0 && !src[i])                                           \
2559             i--;                                                        \
2560         while (src[i])                                                  \
2561             i++
2562 #if HAVE_FAST_64BIT
2563     for (i = 0; i + 1 < length; i += 9) {
2564         if (!((~AV_RN64A(src + i) &
2565                (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2566               0x8000800080008080ULL))
2567             continue;
2568         FIND_FIRST_ZERO;
2569         STARTCODE_TEST;
2570         i -= 7;
2571     }
2572 #else
2573     for (i = 0; i + 1 < length; i += 5) {
2574         if (!((~AV_RN32A(src + i) &
2575                (AV_RN32A(src + i) - 0x01000101U)) &
2576               0x80008080U))
2577             continue;
2578         FIND_FIRST_ZERO;
2579         STARTCODE_TEST;
2580         i -= 3;
2581     }
2582 #endif /* HAVE_FAST_64BIT */
2583 #else
2584     for (i = 0; i + 1 < length; i += 2) {
2585         if (src[i])
2586             continue;
2587         if (i > 0 && src[i - 1] == 0)
2588             i--;
2589         STARTCODE_TEST;
2590     }
2591 #endif /* HAVE_FAST_UNALIGNED */
2592
2593     if (i >= length - 1) { // no escaped 0
2594         nal->data = src;
2595         nal->size = length;
2596         return length;
2597     }
2598
2599     av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2600                    length + FF_INPUT_BUFFER_PADDING_SIZE);
2601     if (!nal->rbsp_buffer)
2602         return AVERROR(ENOMEM);
2603
2604     dst = nal->rbsp_buffer;
2605
2606     memcpy(dst, src, i);
2607     si = di = i;
2608     while (si + 2 < length) {
2609         // remove escapes (very rare 1:2^22)
2610         if (src[si + 2] > 3) {
2611             dst[di++] = src[si++];
2612             dst[di++] = src[si++];
2613         } else if (src[si] == 0 && src[si + 1] == 0) {
2614             if (src[si + 2] == 3) { // escape
2615                 dst[di++] = 0;
2616                 dst[di++] = 0;
2617                 si       += 3;
2618
2619                 s->skipped_bytes++;
2620                 if (s->skipped_bytes_pos_size < s->skipped_bytes) {
2621                     s->skipped_bytes_pos_size *= 2;
2622                     av_reallocp_array(&s->skipped_bytes_pos,
2623                             s->skipped_bytes_pos_size,
2624                             sizeof(*s->skipped_bytes_pos));
2625                     if (!s->skipped_bytes_pos)
2626                         return AVERROR(ENOMEM);
2627                 }
2628                 if (s->skipped_bytes_pos)
2629                     s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
2630                 continue;
2631             } else // next start code
2632                 goto nsc;
2633         }
2634
2635         dst[di++] = src[si++];
2636     }
2637     while (si < length)
2638         dst[di++] = src[si++];
2639
2640 nsc:
2641     memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2642
2643     nal->data = dst;
2644     nal->size = di;
2645     return si;
2646 }
2647
2648 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2649 {
2650     int i, consumed, ret = 0;
2651
2652     s->ref = NULL;
2653     s->last_eos = s->eos;
2654     s->eos = 0;
2655
2656     /* split the input packet into NAL units, so we know the upper bound on the
2657      * number of slices in the frame */
2658     s->nb_nals = 0;
2659     while (length >= 4) {
2660         HEVCNAL *nal;
2661         int extract_length = 0;
2662
2663         if (s->is_nalff) {
2664             int i;
2665             for (i = 0; i < s->nal_length_size; i++)
2666                 extract_length = (extract_length << 8) | buf[i];
2667             buf    += s->nal_length_size;
2668             length -= s->nal_length_size;
2669
2670             if (extract_length > length) {
2671                 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2672                 ret = AVERROR_INVALIDDATA;
2673                 goto fail;
2674             }
2675         } else {
2676             /* search start code */
2677             while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2678                 ++buf;
2679                 --length;
2680                 if (length < 4) {
2681                     av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
2682                     ret = AVERROR_INVALIDDATA;
2683                     goto fail;
2684                 }
2685             }
2686
2687             buf           += 3;
2688             length        -= 3;
2689         }
2690
2691         if (!s->is_nalff)
2692             extract_length = length;
2693
2694         if (s->nals_allocated < s->nb_nals + 1) {
2695             int new_size = s->nals_allocated + 1;
2696             HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2697             if (!tmp) {
2698                 ret = AVERROR(ENOMEM);
2699                 goto fail;
2700             }
2701             s->nals = tmp;
2702             memset(s->nals + s->nals_allocated, 0,
2703                    (new_size - s->nals_allocated) * sizeof(*tmp));
2704             av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
2705             av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
2706             av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
2707             s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
2708             s->skipped_bytes_pos_nal[s->nals_allocated] = av_malloc_array(s->skipped_bytes_pos_size_nal[s->nals_allocated], sizeof(*s->skipped_bytes_pos));
2709             s->nals_allocated = new_size;
2710         }
2711         s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
2712         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
2713         nal = &s->nals[s->nb_nals];
2714
2715         consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
2716
2717         s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
2718         s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
2719         s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
2720
2721
2722         if (consumed < 0) {
2723             ret = consumed;
2724             goto fail;
2725         }
2726
2727         ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
2728         if (ret < 0)
2729             goto fail;
2730         hls_nal_unit(s);
2731
2732         if (s->nal_unit_type == NAL_EOB_NUT ||
2733             s->nal_unit_type == NAL_EOS_NUT)
2734             s->eos = 1;
2735
2736         buf    += consumed;
2737         length -= consumed;
2738     }
2739
2740     /* parse the NAL units */
2741     for (i = 0; i < s->nb_nals; i++) {
2742         int ret;
2743         s->skipped_bytes = s->skipped_bytes_nal[i];
2744         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
2745
2746         ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2747         if (ret < 0) {
2748             av_log(s->avctx, AV_LOG_WARNING,
2749                    "Error parsing NAL unit #%d.\n", i);
2750             goto fail;
2751         }
2752     }
2753
2754 fail:
2755     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2756         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2757
2758     return ret;
2759 }
2760
2761 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2762 {
2763     int i;
2764     for (i = 0; i < 16; i++)
2765         av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2766 }
2767
2768 static int verify_md5(HEVCContext *s, AVFrame *frame)
2769 {
2770     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2771     int pixel_shift;
2772     int i, j;
2773
2774     if (!desc)
2775         return AVERROR(EINVAL);
2776
2777     pixel_shift = desc->comp[0].depth_minus1 > 7;
2778
2779     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2780            s->poc);
2781
2782     /* the checksums are LE, so we have to byteswap for >8bpp formats
2783      * on BE arches */
2784 #if HAVE_BIGENDIAN
2785     if (pixel_shift && !s->checksum_buf) {
2786         av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2787                        FFMAX3(frame->linesize[0], frame->linesize[1],
2788                               frame->linesize[2]));
2789         if (!s->checksum_buf)
2790             return AVERROR(ENOMEM);
2791     }
2792 #endif
2793
2794     for (i = 0; frame->data[i]; i++) {
2795         int width  = s->avctx->coded_width;
2796         int height = s->avctx->coded_height;
2797         int w = (i == 1 || i == 2) ? (width  >> desc->log2_chroma_w) : width;
2798         int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2799         uint8_t md5[16];
2800
2801         av_md5_init(s->md5_ctx);
2802         for (j = 0; j < h; j++) {
2803             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2804 #if HAVE_BIGENDIAN
2805             if (pixel_shift) {
2806                 s->bdsp.bswap16_buf((uint16_t *) s->checksum_buf,
2807                                     (const uint16_t *) src, w);
2808                 src = s->checksum_buf;
2809             }
2810 #endif
2811             av_md5_update(s->md5_ctx, src, w << pixel_shift);
2812         }
2813         av_md5_final(s->md5_ctx, md5);
2814
2815         if (!memcmp(md5, s->md5[i], 16)) {
2816             av_log   (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2817             print_md5(s->avctx, AV_LOG_DEBUG, md5);
2818             av_log   (s->avctx, AV_LOG_DEBUG, "; ");
2819         } else {
2820             av_log   (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2821             print_md5(s->avctx, AV_LOG_ERROR, md5);
2822             av_log   (s->avctx, AV_LOG_ERROR, " != ");
2823             print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2824             av_log   (s->avctx, AV_LOG_ERROR, "\n");
2825             return AVERROR_INVALIDDATA;
2826         }
2827     }
2828
2829     av_log(s->avctx, AV_LOG_DEBUG, "\n");
2830
2831     return 0;
2832 }
2833
2834 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2835                              AVPacket *avpkt)
2836 {
2837     int ret;
2838     HEVCContext *s = avctx->priv_data;
2839
2840     if (!avpkt->size) {
2841         ret = ff_hevc_output_frame(s, data, 1);
2842         if (ret < 0)
2843             return ret;
2844
2845         *got_output = ret;
2846         return 0;
2847     }
2848
2849     s->ref = NULL;
2850     ret    = decode_nal_units(s, avpkt->data, avpkt->size);
2851     if (ret < 0)
2852         return ret;
2853
2854     /* verify the SEI checksum */
2855     if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2856         s->is_md5) {
2857         ret = verify_md5(s, s->ref->frame);
2858         if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2859             ff_hevc_unref_frame(s, s->ref, ~0);
2860             return ret;
2861         }
2862     }
2863     s->is_md5 = 0;
2864
2865     if (s->is_decoded) {
2866         s->ref->frame->key_frame = IS_IRAP(s);
2867         av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2868         s->is_decoded = 0;
2869     }
2870
2871     if (s->output_frame->buf[0]) {
2872         av_frame_move_ref(data, s->output_frame);
2873         *got_output = 1;
2874     }
2875
2876     return avpkt->size;
2877 }
2878
2879 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2880 {
2881     int ret;
2882
2883     ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2884     if (ret < 0)
2885         return ret;
2886
2887     dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2888     if (!dst->tab_mvf_buf)
2889         goto fail;
2890     dst->tab_mvf = src->tab_mvf;
2891
2892     dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2893     if (!dst->rpl_tab_buf)
2894         goto fail;
2895     dst->rpl_tab = src->rpl_tab;
2896
2897     dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2898     if (!dst->rpl_buf)
2899         goto fail;
2900
2901     dst->poc        = src->poc;
2902     dst->ctb_count  = src->ctb_count;
2903     dst->window     = src->window;
2904     dst->flags      = src->flags;
2905     dst->sequence   = src->sequence;
2906
2907     return 0;
2908 fail:
2909     ff_hevc_unref_frame(s, dst, ~0);
2910     return AVERROR(ENOMEM);
2911 }
2912
2913 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2914 {
2915     HEVCContext       *s = avctx->priv_data;
2916     HEVCLocalContext *lc = s->HEVClc;
2917     int i;
2918
2919     pic_arrays_free(s);
2920
2921     av_freep(&s->md5_ctx);
2922
2923     for(i=0; i < s->nals_allocated; i++) {
2924         av_freep(&s->skipped_bytes_pos_nal[i]);
2925     }
2926     av_freep(&s->skipped_bytes_pos_size_nal);
2927     av_freep(&s->skipped_bytes_nal);
2928     av_freep(&s->skipped_bytes_pos_nal);
2929
2930     av_freep(&s->cabac_state);
2931
2932     av_frame_free(&s->tmp_frame);
2933     av_frame_free(&s->output_frame);
2934
2935     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2936         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2937         av_frame_free(&s->DPB[i].frame);
2938     }
2939
2940     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2941         av_buffer_unref(&s->vps_list[i]);
2942     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2943         av_buffer_unref(&s->sps_list[i]);
2944     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2945         av_buffer_unref(&s->pps_list[i]);
2946
2947     av_freep(&s->sh.entry_point_offset);
2948     av_freep(&s->sh.offset);
2949     av_freep(&s->sh.size);
2950
2951     for (i = 1; i < s->threads_number; i++) {
2952         lc = s->HEVClcList[i];
2953         if (lc) {
2954             av_freep(&s->HEVClcList[i]);
2955             av_freep(&s->sList[i]);
2956         }
2957     }
2958     if (s->HEVClc == s->HEVClcList[0])
2959         s->HEVClc = NULL;
2960     av_freep(&s->HEVClcList[0]);
2961
2962     for (i = 0; i < s->nals_allocated; i++)
2963         av_freep(&s->nals[i].rbsp_buffer);
2964     av_freep(&s->nals);
2965     s->nals_allocated = 0;
2966
2967     return 0;
2968 }
2969
2970 static av_cold int hevc_init_context(AVCodecContext *avctx)
2971 {
2972     HEVCContext *s = avctx->priv_data;
2973     int i;
2974
2975     s->avctx = avctx;
2976
2977     s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
2978     if (!s->HEVClc)
2979         goto fail;
2980     s->HEVClcList[0] = s->HEVClc;
2981     s->sList[0] = s;
2982
2983     s->cabac_state = av_malloc(HEVC_CONTEXTS);
2984     if (!s->cabac_state)
2985         goto fail;
2986
2987     s->tmp_frame = av_frame_alloc();
2988     if (!s->tmp_frame)
2989         goto fail;
2990
2991     s->output_frame = av_frame_alloc();
2992     if (!s->output_frame)
2993         goto fail;
2994
2995     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2996         s->DPB[i].frame = av_frame_alloc();
2997         if (!s->DPB[i].frame)
2998             goto fail;
2999         s->DPB[i].tf.f = s->DPB[i].frame;
3000     }
3001
3002     s->max_ra = INT_MAX;
3003
3004     s->md5_ctx = av_md5_alloc();
3005     if (!s->md5_ctx)
3006         goto fail;
3007
3008     ff_bswapdsp_init(&s->bdsp);
3009
3010     s->context_initialized = 1;
3011     s->eos = 0;
3012
3013     return 0;
3014
3015 fail:
3016     hevc_decode_free(avctx);
3017     return AVERROR(ENOMEM);
3018 }
3019
3020 static int hevc_update_thread_context(AVCodecContext *dst,
3021                                       const AVCodecContext *src)
3022 {
3023     HEVCContext *s  = dst->priv_data;
3024     HEVCContext *s0 = src->priv_data;
3025     int i, ret;
3026
3027     if (!s->context_initialized) {
3028         ret = hevc_init_context(dst);
3029         if (ret < 0)
3030             return ret;
3031     }
3032
3033     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
3034         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
3035         if (s0->DPB[i].frame->buf[0]) {
3036             ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
3037             if (ret < 0)
3038                 return ret;
3039         }
3040     }
3041
3042     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
3043         av_buffer_unref(&s->vps_list[i]);
3044         if (s0->vps_list[i]) {
3045             s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
3046             if (!s->vps_list[i])
3047                 return AVERROR(ENOMEM);
3048         }
3049     }
3050
3051     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
3052         av_buffer_unref(&s->sps_list[i]);
3053         if (s0->sps_list[i]) {
3054             s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
3055             if (!s->sps_list[i])
3056                 return AVERROR(ENOMEM);
3057         }
3058     }
3059
3060     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
3061         av_buffer_unref(&s->pps_list[i]);
3062         if (s0->pps_list[i]) {
3063             s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
3064             if (!s->pps_list[i])
3065                 return AVERROR(ENOMEM);
3066         }
3067     }
3068
3069     if (s->sps != s0->sps)
3070         ret = set_sps(s, s0->sps);
3071
3072     s->seq_decode = s0->seq_decode;
3073     s->seq_output = s0->seq_output;
3074     s->pocTid0    = s0->pocTid0;
3075     s->max_ra     = s0->max_ra;
3076     s->eos        = s0->eos;
3077
3078     s->is_nalff        = s0->is_nalff;
3079     s->nal_length_size = s0->nal_length_size;
3080
3081     s->threads_number      = s0->threads_number;
3082     s->threads_type        = s0->threads_type;
3083
3084     if (s0->eos) {
3085         s->seq_decode = (s->seq_decode + 1) & 0xff;
3086         s->max_ra = INT_MAX;
3087     }
3088
3089     return 0;
3090 }
3091
3092 static int hevc_decode_extradata(HEVCContext *s)
3093 {
3094     AVCodecContext *avctx = s->avctx;
3095     GetByteContext gb;
3096     int ret;
3097
3098     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
3099
3100     if (avctx->extradata_size > 3 &&
3101         (avctx->extradata[0] || avctx->extradata[1] ||
3102          avctx->extradata[2] > 1)) {
3103         /* It seems the extradata is encoded as hvcC format.
3104          * Temporarily, we support configurationVersion==0 until 14496-15 3rd
3105          * is finalized. When finalized, configurationVersion will be 1 and we
3106          * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
3107         int i, j, num_arrays, nal_len_size;
3108
3109         s->is_nalff = 1;
3110
3111         bytestream2_skip(&gb, 21);
3112         nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
3113         num_arrays   = bytestream2_get_byte(&gb);
3114
3115         /* nal units in the hvcC always have length coded with 2 bytes,
3116          * so put a fake nal_length_size = 2 while parsing them */
3117         s->nal_length_size = 2;
3118
3119         /* Decode nal units from hvcC. */
3120         for (i = 0; i < num_arrays; i++) {
3121             int type = bytestream2_get_byte(&gb) & 0x3f;
3122             int cnt  = bytestream2_get_be16(&gb);
3123
3124             for (j = 0; j < cnt; j++) {
3125                 // +2 for the nal size field
3126                 int nalsize = bytestream2_peek_be16(&gb) + 2;
3127                 if (bytestream2_get_bytes_left(&gb) < nalsize) {
3128                     av_log(s->avctx, AV_LOG_ERROR,
3129                            "Invalid NAL unit size in extradata.\n");
3130                     return AVERROR_INVALIDDATA;
3131                 }
3132
3133                 ret = decode_nal_units(s, gb.buffer, nalsize);
3134                 if (ret < 0) {
3135                     av_log(avctx, AV_LOG_ERROR,
3136                            "Decoding nal unit %d %d from hvcC failed\n",
3137                            type, i);
3138                     return ret;
3139                 }
3140                 bytestream2_skip(&gb, nalsize);
3141             }
3142         }
3143
3144         /* Now store right nal length size, that will be used to parse
3145          * all other nals */
3146         s->nal_length_size = nal_len_size;
3147     } else {
3148         s->is_nalff = 0;
3149         ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
3150         if (ret < 0)
3151             return ret;
3152     }
3153     return 0;
3154 }
3155
3156 static av_cold int hevc_decode_init(AVCodecContext *avctx)
3157 {
3158     HEVCContext *s = avctx->priv_data;
3159     int ret;
3160
3161     ff_init_cabac_states();
3162
3163     avctx->internal->allocate_progress = 1;
3164
3165     ret = hevc_init_context(avctx);
3166     if (ret < 0)
3167         return ret;
3168
3169     s->enable_parallel_tiles = 0;
3170     s->picture_struct = 0;
3171
3172     if(avctx->active_thread_type & FF_THREAD_SLICE)
3173         s->threads_number = avctx->thread_count;
3174     else
3175         s->threads_number = 1;
3176
3177     if (avctx->extradata_size > 0 && avctx->extradata) {
3178         ret = hevc_decode_extradata(s);
3179         if (ret < 0) {
3180             hevc_decode_free(avctx);
3181             return ret;
3182         }
3183     }
3184
3185     if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
3186             s->threads_type = FF_THREAD_FRAME;
3187         else
3188             s->threads_type = FF_THREAD_SLICE;
3189
3190     return 0;
3191 }
3192
3193 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
3194 {
3195     HEVCContext *s = avctx->priv_data;
3196     int ret;
3197
3198     memset(s, 0, sizeof(*s));
3199
3200     ret = hevc_init_context(avctx);
3201     if (ret < 0)
3202         return ret;
3203
3204     return 0;
3205 }
3206
3207 static void hevc_decode_flush(AVCodecContext *avctx)
3208 {
3209     HEVCContext *s = avctx->priv_data;
3210     ff_hevc_flush_dpb(s);
3211     s->max_ra = INT_MAX;
3212 }
3213
3214 #define OFFSET(x) offsetof(HEVCContext, x)
3215 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
3216
3217 static const AVProfile profiles[] = {
3218     { FF_PROFILE_HEVC_MAIN,                 "Main"                },
3219     { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
3220     { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
3221     { FF_PROFILE_UNKNOWN },
3222 };
3223
3224 static const AVOption options[] = {
3225     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
3226         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3227     { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
3228         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3229     { NULL },
3230 };
3231
3232 static const AVClass hevc_decoder_class = {
3233     .class_name = "HEVC decoder",
3234     .item_name  = av_default_item_name,
3235     .option     = options,
3236     .version    = LIBAVUTIL_VERSION_INT,
3237 };
3238
3239 AVCodec ff_hevc_decoder = {
3240     .name                  = "hevc",
3241     .long_name             = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3242     .type                  = AVMEDIA_TYPE_VIDEO,
3243     .id                    = AV_CODEC_ID_HEVC,
3244     .priv_data_size        = sizeof(HEVCContext),
3245     .priv_class            = &hevc_decoder_class,
3246     .init                  = hevc_decode_init,
3247     .close                 = hevc_decode_free,
3248     .decode                = hevc_decode_frame,
3249     .flush                 = hevc_decode_flush,
3250     .update_thread_context = hevc_update_thread_context,
3251     .init_thread_copy      = hevc_init_thread_copy,
3252     .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3253                              CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
3254     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
3255 };