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_array(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_array(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_array(pic_size_in_ctb,
 115                                       sizeof(*s->tab_slice_address));
 116     s->qp_y_tab           = av_malloc_array(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_array(2 * s->bs_width, (s->bs_height + 1));
 122     s->vertical_bs   = av_mallocz_array(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_array(sh->num_entry_point_offsets, sizeof(int));
 679             sh->offset = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
 680             sh->size = av_malloc_array(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     HEVCLocalContext *lc = s->HEVClc;
1052     GetBitContext gb;
1053     int cb_size   = 1 << log2_cb_size;
1054     int stride0   = s->frame->linesize[0];
1055     uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
1056     int   stride1 = s->frame->linesize[1];
1057     uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
1058     int   stride2 = s->frame->linesize[2];
1059     uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
1060
1061     int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
1062     const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
1063     int ret;
1064
1065     if (!s->sh.disable_deblocking_filter_flag)
1066         ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1067
1068     ret = init_get_bits(&gb, pcm, length);
1069     if (ret < 0)
1070         return ret;
1071
1072     s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
1073     s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1074     s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1075     return 0;
1076 }
1077
1078 /**
1079  * 8.5.3.2.2.1 Luma sample unidirectional interpolation process
1080  *
1081  * @param s HEVC decoding context
1082  * @param dst target buffer for block data at block position
1083  * @param dststride stride of the dst buffer
1084  * @param ref reference picture buffer at origin (0, 0)
1085  * @param mv motion vector (relative to block position) to get pixel data from
1086  * @param x_off horizontal position of block from origin (0, 0)
1087  * @param y_off vertical position of block from origin (0, 0)
1088  * @param block_w width of block
1089  * @param block_h height of block
1090  * @param luma_weight weighting factor applied to the luma prediction
1091  * @param luma_offset additive offset applied to the luma prediction value
1092  */
1093
1094 static void luma_mc_uni(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
1095                         AVFrame *ref, const Mv *mv, int x_off, int y_off,
1096                         int block_w, int block_h, int luma_weight, int luma_offset)
1097 {
1098     HEVCLocalContext *lc = s->HEVClc;
1099     uint8_t *src         = ref->data[0];
1100     ptrdiff_t srcstride  = ref->linesize[0];
1101     int pic_width        = s->sps->width;
1102     int pic_height       = s->sps->height;
1103     int mx               = mv->x & 3;
1104     int my               = mv->y & 3;
1105     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1106                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1107     int idx              = ff_hevc_pel_weight[block_w];
1108
1109     x_off += mv->x >> 2;
1110     y_off += mv->y >> 2;
1111     src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
1112
1113     if (x_off < QPEL_EXTRA_BEFORE || y_off < QPEL_EXTRA_AFTER ||
1114         x_off >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1115         y_off >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1116         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1117         int offset     = QPEL_EXTRA_BEFORE * srcstride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1118         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1119
1120         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1121                                  edge_emu_stride, srcstride,
1122                                  block_w + QPEL_EXTRA,
1123                                  block_h + QPEL_EXTRA,
1124                                  x_off - QPEL_EXTRA_BEFORE, y_off - QPEL_EXTRA_BEFORE,
1125                                  pic_width, pic_height);
1126         src = lc->edge_emu_buffer + buf_offset;
1127         srcstride = edge_emu_stride;
1128     }
1129
1130     if (!weight_flag)
1131         s->hevcdsp.put_hevc_qpel_uni[idx][!!my][!!mx](dst, dststride, src, srcstride,
1132                                                       block_h, mx, my, block_w);
1133     else
1134         s->hevcdsp.put_hevc_qpel_uni_w[idx][!!my][!!mx](dst, dststride, src, srcstride,
1135                                                         block_h, s->sh.luma_log2_weight_denom,
1136                                                         luma_weight, luma_offset, mx, my, block_w);
1137 }
1138
1139 /**
1140  * 8.5.3.2.2.1 Luma sample bidirectional interpolation process
1141  *
1142  * @param s HEVC decoding context
1143  * @param dst target buffer for block data at block position
1144  * @param dststride stride of the dst buffer
1145  * @param ref0 reference picture0 buffer at origin (0, 0)
1146  * @param mv0 motion vector0 (relative to block position) to get pixel data from
1147  * @param x_off horizontal position of block from origin (0, 0)
1148  * @param y_off vertical position of block from origin (0, 0)
1149  * @param block_w width of block
1150  * @param block_h height of block
1151  * @param ref1 reference picture1 buffer at origin (0, 0)
1152  * @param mv1 motion vector1 (relative to block position) to get pixel data from
1153  * @param current_mv current motion vector structure
1154  */
1155  static void luma_mc_bi(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
1156                        AVFrame *ref0, const Mv *mv0, int x_off, int y_off,
1157                        int block_w, int block_h, AVFrame *ref1, const Mv *mv1, struct MvField *current_mv)
1158 {
1159     HEVCLocalContext *lc = s->HEVClc;
1160     DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1161     ptrdiff_t src0stride  = ref0->linesize[0];
1162     ptrdiff_t src1stride  = ref1->linesize[0];
1163     int pic_width        = s->sps->width;
1164     int pic_height       = s->sps->height;
1165     int mx0              = mv0->x & 3;
1166     int my0              = mv0->y & 3;
1167     int mx1              = mv1->x & 3;
1168     int my1              = mv1->y & 3;
1169     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1170                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1171     int x_off0           = x_off + (mv0->x >> 2);
1172     int y_off0           = y_off + (mv0->y >> 2);
1173     int x_off1           = x_off + (mv1->x >> 2);
1174     int y_off1           = y_off + (mv1->y >> 2);
1175     int idx              = ff_hevc_pel_weight[block_w];
1176
1177     uint8_t *src0  = ref0->data[0] + y_off0 * src0stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
1178     uint8_t *src1  = ref1->data[0] + y_off1 * src1stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
1179
1180     if (x_off0 < QPEL_EXTRA_BEFORE || y_off0 < QPEL_EXTRA_AFTER ||
1181         x_off0 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1182         y_off0 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1183         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1184         int offset     = QPEL_EXTRA_BEFORE * src0stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1185         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1186
1187         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset,
1188                                  edge_emu_stride, src0stride,
1189                                  block_w + QPEL_EXTRA,
1190                                  block_h + QPEL_EXTRA,
1191                                  x_off0 - QPEL_EXTRA_BEFORE, y_off0 - QPEL_EXTRA_BEFORE,
1192                                  pic_width, pic_height);
1193         src0 = lc->edge_emu_buffer + buf_offset;
1194         src0stride = edge_emu_stride;
1195     }
1196
1197     if (x_off1 < QPEL_EXTRA_BEFORE || y_off1 < QPEL_EXTRA_AFTER ||
1198         x_off1 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
1199         y_off1 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
1200         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1201         int offset     = QPEL_EXTRA_BEFORE * src1stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1202         int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
1203
1204         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src1 - offset,
1205                                  edge_emu_stride, src1stride,
1206                                  block_w + QPEL_EXTRA,
1207                                  block_h + QPEL_EXTRA,
1208                                  x_off1 - QPEL_EXTRA_BEFORE, y_off1 - QPEL_EXTRA_BEFORE,
1209                                  pic_width, pic_height);
1210         src1 = lc->edge_emu_buffer2 + buf_offset;
1211         src1stride = edge_emu_stride;
1212     }
1213
1214     s->hevcdsp.put_hevc_qpel[idx][!!my0][!!mx0](tmp, MAX_PB_SIZE, src0, src0stride,
1215                                                 block_h, mx0, my0, block_w);
1216     if (!weight_flag)
1217         s->hevcdsp.put_hevc_qpel_bi[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
1218                                                        block_h, mx1, my1, block_w);
1219     else
1220         s->hevcdsp.put_hevc_qpel_bi_w[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
1221                                                          block_h, s->sh.luma_log2_weight_denom,
1222                                                          s->sh.luma_weight_l0[current_mv->ref_idx[0]],
1223                                                          s->sh.luma_weight_l1[current_mv->ref_idx[1]],
1224                                                          s->sh.luma_offset_l0[current_mv->ref_idx[0]],
1225                                                          s->sh.luma_offset_l1[current_mv->ref_idx[1]],
1226                                                          mx1, my1, block_w);
1227
1228 }
1229
1230 /**
1231  * 8.5.3.2.2.2 Chroma sample uniprediction interpolation process
1232  *
1233  * @param s HEVC decoding context
1234  * @param dst1 target buffer for block data at block position (U plane)
1235  * @param dst2 target buffer for block data at block position (V plane)
1236  * @param dststride stride of the dst1 and dst2 buffers
1237  * @param ref reference picture buffer at origin (0, 0)
1238  * @param mv motion vector (relative to block position) to get pixel data from
1239  * @param x_off horizontal position of block from origin (0, 0)
1240  * @param y_off vertical position of block from origin (0, 0)
1241  * @param block_w width of block
1242  * @param block_h height of block
1243  * @param chroma_weight weighting factor applied to the chroma prediction
1244  * @param chroma_offset additive offset applied to the chroma prediction value
1245  */
1246
1247 static void chroma_mc_uni(HEVCContext *s, uint8_t *dst0,
1248                           ptrdiff_t dststride, uint8_t *src0, ptrdiff_t srcstride, int reflist,
1249                           int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int chroma_weight, int chroma_offset)
1250 {
1251     HEVCLocalContext *lc = s->HEVClc;
1252     int pic_width        = s->sps->width >> s->sps->hshift[1];
1253     int pic_height       = s->sps->height >> s->sps->vshift[1];
1254     const Mv *mv         = &current_mv->mv[reflist];
1255     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1256                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1257     int idx              = ff_hevc_pel_weight[block_w];
1258     int hshift           = s->sps->hshift[1];
1259     int vshift           = s->sps->vshift[1];
1260     intptr_t mx          = mv->x & ((1 << (2 + hshift)) - 1);
1261     intptr_t my          = mv->y & ((1 << (2 + vshift)) - 1);
1262     intptr_t _mx         = mx << (1 - hshift);
1263     intptr_t _my         = my << (1 - vshift);
1264
1265     x_off += mv->x >> (2 + hshift);
1266     y_off += mv->y >> (2 + vshift);
1267     src0  += y_off * srcstride + (x_off << s->sps->pixel_shift);
1268
1269     if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1270         x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1271         y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1272         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1273         int offset0 = EPEL_EXTRA_BEFORE * (srcstride + (1 << s->sps->pixel_shift));
1274         int buf_offset0 = EPEL_EXTRA_BEFORE *
1275                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1276         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset0,
1277                                  edge_emu_stride, srcstride,
1278                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1279                                  x_off - EPEL_EXTRA_BEFORE,
1280                                  y_off - EPEL_EXTRA_BEFORE,
1281                                  pic_width, pic_height);
1282
1283         src0 = lc->edge_emu_buffer + buf_offset0;
1284         srcstride = edge_emu_stride;
1285     }
1286     if (!weight_flag)
1287         s->hevcdsp.put_hevc_epel_uni[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
1288                                                   block_h, _mx, _my, block_w);
1289     else
1290         s->hevcdsp.put_hevc_epel_uni_w[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
1291                                                         block_h, s->sh.chroma_log2_weight_denom,
1292                                                         chroma_weight, chroma_offset, _mx, _my, block_w);
1293 }
1294
1295 /**
1296  * 8.5.3.2.2.2 Chroma sample bidirectional interpolation process
1297  *
1298  * @param s HEVC decoding context
1299  * @param dst target buffer for block data at block position
1300  * @param dststride stride of the dst buffer
1301  * @param ref0 reference picture0 buffer at origin (0, 0)
1302  * @param mv0 motion vector0 (relative to block position) to get pixel data from
1303  * @param x_off horizontal position of block from origin (0, 0)
1304  * @param y_off vertical position of block from origin (0, 0)
1305  * @param block_w width of block
1306  * @param block_h height of block
1307  * @param ref1 reference picture1 buffer at origin (0, 0)
1308  * @param mv1 motion vector1 (relative to block position) to get pixel data from
1309  * @param current_mv current motion vector structure
1310  * @param cidx chroma component(cb, cr)
1311  */
1312 static void chroma_mc_bi(HEVCContext *s, uint8_t *dst0, ptrdiff_t dststride, AVFrame *ref0, AVFrame *ref1,
1313                          int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int cidx)
1314 {
1315     DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1316     int tmpstride = MAX_PB_SIZE;
1317     HEVCLocalContext *lc = s->HEVClc;
1318     uint8_t *src1        = ref0->data[cidx+1];
1319     uint8_t *src2        = ref1->data[cidx+1];
1320     ptrdiff_t src1stride = ref0->linesize[cidx+1];
1321     ptrdiff_t src2stride = ref1->linesize[cidx+1];
1322     int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1323                            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
1324     int pic_width        = s->sps->width >> s->sps->hshift[1];
1325     int pic_height       = s->sps->height >> s->sps->vshift[1];
1326     Mv *mv0              = &current_mv->mv[0];
1327     Mv *mv1              = &current_mv->mv[1];
1328     int hshift = s->sps->hshift[1];
1329     int vshift = s->sps->vshift[1];
1330
1331     intptr_t mx0 = mv0->x & ((1 << (2 + hshift)) - 1);
1332     intptr_t my0 = mv0->y & ((1 << (2 + vshift)) - 1);
1333     intptr_t mx1 = mv1->x & ((1 << (2 + hshift)) - 1);
1334     intptr_t my1 = mv1->y & ((1 << (2 + vshift)) - 1);
1335     intptr_t _mx0 = mx0 << (1 - hshift);
1336     intptr_t _my0 = my0 << (1 - vshift);
1337     intptr_t _mx1 = mx1 << (1 - hshift);
1338     intptr_t _my1 = my1 << (1 - vshift);
1339
1340     int x_off0 = x_off + (mv0->x >> (2 + hshift));
1341     int y_off0 = y_off + (mv0->y >> (2 + vshift));
1342     int x_off1 = x_off + (mv1->x >> (2 + hshift));
1343     int y_off1 = y_off + (mv1->y >> (2 + vshift));
1344     int idx = ff_hevc_pel_weight[block_w];
1345     src1  += y_off0 * src1stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
1346     src2  += y_off1 * src2stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
1347
1348     if (x_off0 < EPEL_EXTRA_BEFORE || y_off0 < EPEL_EXTRA_AFTER ||
1349         x_off0 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1350         y_off0 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1351         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1352         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1353         int buf_offset1 = EPEL_EXTRA_BEFORE *
1354                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1355
1356         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1357                                  edge_emu_stride, src1stride,
1358                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1359                                  x_off0 - EPEL_EXTRA_BEFORE,
1360                                  y_off0 - EPEL_EXTRA_BEFORE,
1361                                  pic_width, pic_height);
1362
1363         src1 = lc->edge_emu_buffer + buf_offset1;
1364         src1stride = edge_emu_stride;
1365     }
1366
1367     if (x_off1 < EPEL_EXTRA_BEFORE || y_off1 < EPEL_EXTRA_AFTER ||
1368         x_off1 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1369         y_off1 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1370         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1371         int offset1 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1372         int buf_offset1 = EPEL_EXTRA_BEFORE *
1373                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1374
1375         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src2 - offset1,
1376                                  edge_emu_stride, src2stride,
1377                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1378                                  x_off1 - EPEL_EXTRA_BEFORE,
1379                                  y_off1 - EPEL_EXTRA_BEFORE,
1380                                  pic_width, pic_height);
1381
1382         src2 = lc->edge_emu_buffer2 + buf_offset1;
1383         src2stride = edge_emu_stride;
1384     }
1385
1386     s->hevcdsp.put_hevc_epel[idx][!!my0][!!mx0](tmp, tmpstride, src1, src1stride,
1387                                                 block_h, _mx0, _my0, block_w);
1388     if (!weight_flag)
1389         s->hevcdsp.put_hevc_epel_bi[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
1390                                                        src2, src2stride, tmp, tmpstride,
1391                                                        block_h, _mx1, _my1, block_w);
1392     else
1393         s->hevcdsp.put_hevc_epel_bi_w[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
1394                                                          src2, src2stride, tmp, tmpstride,
1395                                                          block_h,
1396                                                          s->sh.chroma_log2_weight_denom,
1397                                                          s->sh.chroma_weight_l0[current_mv->ref_idx[0]][cidx],
1398                                                          s->sh.chroma_weight_l1[current_mv->ref_idx[1]][cidx],
1399                                                          s->sh.chroma_offset_l0[current_mv->ref_idx[0]][cidx],
1400                                                          s->sh.chroma_offset_l1[current_mv->ref_idx[1]][cidx],
1401                                                          _mx1, _my1, block_w);
1402 }
1403
1404 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1405                                 const Mv *mv, int y0, int height)
1406 {
1407     int y = (mv->y >> 2) + y0 + height + 9;
1408
1409     if (s->threads_type == FF_THREAD_FRAME )
1410         ff_thread_await_progress(&ref->tf, y, 0);
1411 }
1412
1413 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1414                                 int nPbW, int nPbH,
1415                                 int log2_cb_size, int partIdx)
1416 {
1417 #define POS(c_idx, x, y)                                                              \
1418     &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1419                            (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1420     HEVCLocalContext *lc = s->HEVClc;
1421     int merge_idx = 0;
1422     struct MvField current_mv = {{{ 0 }}};
1423
1424     int min_pu_width = s->sps->min_pu_width;
1425
1426     MvField *tab_mvf = s->ref->tab_mvf;
1427     RefPicList  *refPicList = s->ref->refPicList;
1428     HEVCFrame *ref0, *ref1;
1429     uint8_t *dst0 = POS(0, x0, y0);
1430     uint8_t *dst1 = POS(1, x0, y0);
1431     uint8_t *dst2 = POS(2, x0, y0);
1432     int log2_min_cb_size = s->sps->log2_min_cb_size;
1433     int min_cb_width     = s->sps->min_cb_width;
1434     int x_cb             = x0 >> log2_min_cb_size;
1435     int y_cb             = y0 >> log2_min_cb_size;
1436     int ref_idx[2];
1437     int mvp_flag[2];
1438     int x_pu, y_pu;
1439     int i, j;
1440
1441     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1442         if (s->sh.max_num_merge_cand > 1)
1443             merge_idx = ff_hevc_merge_idx_decode(s);
1444         else
1445             merge_idx = 0;
1446
1447         ff_hevc_luma_mv_merge_mode(s, x0, y0,
1448                                    1 << log2_cb_size,
1449                                    1 << log2_cb_size,
1450                                    log2_cb_size, partIdx,
1451                                    merge_idx, &current_mv);
1452         x_pu = x0 >> s->sps->log2_min_pu_size;
1453         y_pu = y0 >> s->sps->log2_min_pu_size;
1454
1455         for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1456             for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1457                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1458     } else { /* MODE_INTER */
1459         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1460         if (lc->pu.merge_flag) {
1461             if (s->sh.max_num_merge_cand > 1)
1462                 merge_idx = ff_hevc_merge_idx_decode(s);
1463             else
1464                 merge_idx = 0;
1465
1466             ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1467                                        partIdx, merge_idx, &current_mv);
1468             x_pu = x0 >> s->sps->log2_min_pu_size;
1469             y_pu = y0 >> s->sps->log2_min_pu_size;
1470
1471             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1472                 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1473                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1474         } else {
1475             enum InterPredIdc inter_pred_idc = PRED_L0;
1476             ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1477             current_mv.pred_flag = 0;
1478             if (s->sh.slice_type == B_SLICE)
1479                 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1480
1481             if (inter_pred_idc != PRED_L1) {
1482                 if (s->sh.nb_refs[L0]) {
1483                     ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1484                     current_mv.ref_idx[0] = ref_idx[0];
1485                 }
1486                 current_mv.pred_flag = PF_L0;
1487                 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
1488                 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
1489                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1490                                          partIdx, merge_idx, &current_mv,
1491                                          mvp_flag[0], 0);
1492                 current_mv.mv[0].x += lc->pu.mvd.x;
1493                 current_mv.mv[0].y += lc->pu.mvd.y;
1494             }
1495
1496             if (inter_pred_idc != PRED_L0) {
1497                 if (s->sh.nb_refs[L1]) {
1498                     ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1499                     current_mv.ref_idx[1] = ref_idx[1];
1500                 }
1501
1502                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1503                     lc->pu.mvd.x = 0;
1504                     lc->pu.mvd.y = 0;
1505                 } else {
1506                     ff_hevc_hls_mvd_coding(s, x0, y0, 1);
1507                 }
1508
1509                 current_mv.pred_flag += PF_L1;
1510                 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
1511                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1512                                          partIdx, merge_idx, &current_mv,
1513                                          mvp_flag[1], 1);
1514                 current_mv.mv[1].x += lc->pu.mvd.x;
1515                 current_mv.mv[1].y += lc->pu.mvd.y;
1516             }
1517
1518             x_pu = x0 >> s->sps->log2_min_pu_size;
1519             y_pu = y0 >> s->sps->log2_min_pu_size;
1520
1521             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1522                 for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1523                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1524         }
1525     }
1526
1527     if (current_mv.pred_flag & PF_L0) {
1528         ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1529         if (!ref0)
1530             return;
1531         hevc_await_progress(s, ref0, &current_mv.mv[0], y0, nPbH);
1532     }
1533     if (current_mv.pred_flag & PF_L1) {
1534         ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1535         if (!ref1)
1536             return;
1537         hevc_await_progress(s, ref1, &current_mv.mv[1], y0, nPbH);
1538     }
1539
1540     if (current_mv.pred_flag == PF_L0) {
1541         int x0_c = x0 >> s->sps->hshift[1];
1542         int y0_c = y0 >> s->sps->vshift[1];
1543         int nPbW_c = nPbW >> s->sps->hshift[1];
1544         int nPbH_c = nPbH >> s->sps->vshift[1];
1545
1546         luma_mc_uni(s, dst0, s->frame->linesize[0], ref0->frame,
1547                     &current_mv.mv[0], x0, y0, nPbW, nPbH,
1548                     s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1549                     s->sh.luma_offset_l0[current_mv.ref_idx[0]]);
1550
1551         chroma_mc_uni(s, dst1, s->frame->linesize[1], ref0->frame->data[1], ref0->frame->linesize[1],
1552                       0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1553                       s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0]);
1554         chroma_mc_uni(s, dst2, s->frame->linesize[2], ref0->frame->data[2], ref0->frame->linesize[2],
1555                       0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1556                       s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1]);
1557     } else if (current_mv.pred_flag == PF_L1) {
1558         int x0_c = x0 >> s->sps->hshift[1];
1559         int y0_c = y0 >> s->sps->vshift[1];
1560         int nPbW_c = nPbW >> s->sps->hshift[1];
1561         int nPbH_c = nPbH >> s->sps->vshift[1];
1562
1563         luma_mc_uni(s, dst0, s->frame->linesize[0], ref1->frame,
1564                     &current_mv.mv[1], x0, y0, nPbW, nPbH,
1565                     s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1566                     s->sh.luma_offset_l1[current_mv.ref_idx[1]]);
1567
1568         chroma_mc_uni(s, dst1, s->frame->linesize[1], ref1->frame->data[1], ref1->frame->linesize[1],
1569                       1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1570                       s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0]);
1571
1572         chroma_mc_uni(s, dst2, s->frame->linesize[2], ref1->frame->data[2], ref1->frame->linesize[2],
1573                       1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
1574                       s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1]);
1575     } else if (current_mv.pred_flag == PF_BI) {
1576         int x0_c = x0 >> s->sps->hshift[1];
1577         int y0_c = y0 >> s->sps->vshift[1];
1578         int nPbW_c = nPbW >> s->sps->hshift[1];
1579         int nPbH_c = nPbH >> s->sps->vshift[1];
1580
1581         luma_mc_bi(s, dst0, s->frame->linesize[0], ref0->frame,
1582                    &current_mv.mv[0], x0, y0, nPbW, nPbH,
1583                    ref1->frame, &current_mv.mv[1], &current_mv);
1584
1585         chroma_mc_bi(s, dst1, s->frame->linesize[1], ref0->frame, ref1->frame,
1586                      x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 0);
1587
1588         chroma_mc_bi(s, dst2, s->frame->linesize[2], ref0->frame, ref1->frame,
1589                      x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 1);
1590     }
1591 }
1592
1593 /**
1594  * 8.4.1
1595  */
1596 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1597                                 int prev_intra_luma_pred_flag)
1598 {
1599     HEVCLocalContext *lc = s->HEVClc;
1600     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1601     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1602     int min_pu_width     = s->sps->min_pu_width;
1603     int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
1604     int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1605     int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1606
1607     int cand_up   = (lc->ctb_up_flag || y0b) ?
1608                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1609     int cand_left = (lc->ctb_left_flag || x0b) ?
1610                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
1611
1612     int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1613
1614     MvField *tab_mvf = s->ref->tab_mvf;
1615     int intra_pred_mode;
1616     int candidate[3];
1617     int i, j;
1618
1619     // intra_pred_mode prediction does not cross vertical CTB boundaries
1620     if ((y0 - 1) < y_ctb)
1621         cand_up = INTRA_DC;
1622
1623     if (cand_left == cand_up) {
1624         if (cand_left < 2) {
1625             candidate[0] = INTRA_PLANAR;
1626             candidate[1] = INTRA_DC;
1627             candidate[2] = INTRA_ANGULAR_26;
1628         } else {
1629             candidate[0] = cand_left;
1630             candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1631             candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1632         }
1633     } else {
1634         candidate[0] = cand_left;
1635         candidate[1] = cand_up;
1636         if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1637             candidate[2] = INTRA_PLANAR;
1638         } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1639             candidate[2] = INTRA_DC;
1640         } else {
1641             candidate[2] = INTRA_ANGULAR_26;
1642         }
1643     }
1644
1645     if (prev_intra_luma_pred_flag) {
1646         intra_pred_mode = candidate[lc->pu.mpm_idx];
1647     } else {
1648         if (candidate[0] > candidate[1])
1649             FFSWAP(uint8_t, candidate[0], candidate[1]);
1650         if (candidate[0] > candidate[2])
1651             FFSWAP(uint8_t, candidate[0], candidate[2]);
1652         if (candidate[1] > candidate[2])
1653             FFSWAP(uint8_t, candidate[1], candidate[2]);
1654
1655         intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1656         for (i = 0; i < 3; i++)
1657             if (intra_pred_mode >= candidate[i])
1658                 intra_pred_mode++;
1659     }
1660
1661     /* write the intra prediction units into the mv array */
1662     if (!size_in_pus)
1663         size_in_pus = 1;
1664     for (i = 0; i < size_in_pus; i++) {
1665         memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1666                intra_pred_mode, size_in_pus);
1667
1668         for (j = 0; j < size_in_pus; j++) {
1669             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag = PF_INTRA;
1670         }
1671     }
1672
1673     return intra_pred_mode;
1674 }
1675
1676 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1677                                           int log2_cb_size, int ct_depth)
1678 {
1679     int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1680     int x_cb   = x0 >> s->sps->log2_min_cb_size;
1681     int y_cb   = y0 >> s->sps->log2_min_cb_size;
1682     int y;
1683
1684     for (y = 0; y < length; y++)
1685         memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1686                ct_depth, length);
1687 }
1688
1689 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1690                                   int log2_cb_size)
1691 {
1692     HEVCLocalContext *lc = s->HEVClc;
1693     static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1694     uint8_t prev_intra_luma_pred_flag[4];
1695     int split   = lc->cu.part_mode == PART_NxN;
1696     int pb_size = (1 << log2_cb_size) >> split;
1697     int side    = split + 1;
1698     int chroma_mode;
1699     int i, j;
1700
1701     for (i = 0; i < side; i++)
1702         for (j = 0; j < side; j++)
1703             prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1704
1705     for (i = 0; i < side; i++) {
1706         for (j = 0; j < side; j++) {
1707             if (prev_intra_luma_pred_flag[2 * i + j])
1708                 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1709             else
1710                 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1711
1712             lc->pu.intra_pred_mode[2 * i + j] =
1713                 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1714                                      prev_intra_luma_pred_flag[2 * i + j]);
1715         }
1716     }
1717
1718     chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1719     if (chroma_mode != 4) {
1720         if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1721             lc->pu.intra_pred_mode_c = 34;
1722         else
1723             lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1724     } else {
1725         lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1726     }
1727 }
1728
1729 static void intra_prediction_unit_default_value(HEVCContext *s,
1730                                                 int x0, int y0,
1731                                                 int log2_cb_size)
1732 {
1733     HEVCLocalContext *lc = s->HEVClc;
1734     int pb_size          = 1 << log2_cb_size;
1735     int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
1736     int min_pu_width     = s->sps->min_pu_width;
1737     MvField *tab_mvf     = s->ref->tab_mvf;
1738     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1739     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1740     int j, k;
1741
1742     if (size_in_pus == 0)
1743         size_in_pus = 1;
1744     for (j = 0; j < size_in_pus; j++)
1745         memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
1746     if (lc->cu.pred_mode == MODE_INTRA)
1747         for (j = 0; j < size_in_pus; j++)
1748             for (k = 0; k < size_in_pus; k++)
1749                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].pred_flag = PF_INTRA;
1750 }
1751
1752 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
1753 {
1754     int cb_size          = 1 << log2_cb_size;
1755     HEVCLocalContext *lc = s->HEVClc;
1756     int log2_min_cb_size = s->sps->log2_min_cb_size;
1757     int length           = cb_size >> log2_min_cb_size;
1758     int min_cb_width     = s->sps->min_cb_width;
1759     int x_cb             = x0 >> log2_min_cb_size;
1760     int y_cb             = y0 >> log2_min_cb_size;
1761     int x, y, ret;
1762     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1763
1764     lc->cu.x                = x0;
1765     lc->cu.y                = y0;
1766     lc->cu.rqt_root_cbf     = 1;
1767     lc->cu.pred_mode        = MODE_INTRA;
1768     lc->cu.part_mode        = PART_2Nx2N;
1769     lc->cu.intra_split_flag = 0;
1770     lc->cu.pcm_flag         = 0;
1771
1772     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
1773     for (x = 0; x < 4; x++)
1774         lc->pu.intra_pred_mode[x] = 1;
1775     if (s->pps->transquant_bypass_enable_flag) {
1776         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
1777         if (lc->cu.cu_transquant_bypass_flag)
1778             set_deblocking_bypass(s, x0, y0, log2_cb_size);
1779     } else
1780         lc->cu.cu_transquant_bypass_flag = 0;
1781
1782     if (s->sh.slice_type != I_SLICE) {
1783         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
1784
1785         x = y_cb * min_cb_width + x_cb;
1786         for (y = 0; y < length; y++) {
1787             memset(&s->skip_flag[x], skip_flag, length);
1788             x += min_cb_width;
1789         }
1790         lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
1791     }
1792
1793     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1794         hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1795         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1796
1797         if (!s->sh.disable_deblocking_filter_flag)
1798             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1799     } else {
1800         if (s->sh.slice_type != I_SLICE)
1801             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
1802         if (lc->cu.pred_mode != MODE_INTRA ||
1803             log2_cb_size == s->sps->log2_min_cb_size) {
1804             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
1805             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
1806                                       lc->cu.pred_mode == MODE_INTRA;
1807         }
1808
1809         if (lc->cu.pred_mode == MODE_INTRA) {
1810             if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
1811                 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
1812                 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
1813                 lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
1814             }
1815             if (lc->cu.pcm_flag) {
1816                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1817                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
1818                 if (s->sps->pcm.loop_filter_disable_flag)
1819                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
1820
1821                 if (ret < 0)
1822                     return ret;
1823             } else {
1824                 intra_prediction_unit(s, x0, y0, log2_cb_size);
1825             }
1826         } else {
1827             intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1828             switch (lc->cu.part_mode) {
1829             case PART_2Nx2N:
1830                 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1831                 break;
1832             case PART_2NxN:
1833                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
1834                 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
1835                 break;
1836             case PART_Nx2N:
1837                 hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
1838                 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
1839                 break;
1840             case PART_2NxnU:
1841                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
1842                 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
1843                 break;
1844             case PART_2NxnD:
1845                 hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
1846                 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
1847                 break;
1848             case PART_nLx2N:
1849                 hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
1850                 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
1851                 break;
1852             case PART_nRx2N:
1853                 hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
1854                 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
1855                 break;
1856             case PART_NxN:
1857                 hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
1858                 hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
1859                 hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
1860                 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
1861                 break;
1862             }
1863         }
1864
1865         if (!lc->cu.pcm_flag) {
1866             if (lc->cu.pred_mode != MODE_INTRA &&
1867                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
1868                 lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
1869             }
1870             if (lc->cu.rqt_root_cbf) {
1871                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
1872                                          s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
1873                                          s->sps->max_transform_hierarchy_depth_inter;
1874                 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
1875                                          log2_cb_size,
1876                                          log2_cb_size, 0, 0);
1877                 if (ret < 0)
1878                     return ret;
1879             } else {
1880                 if (!s->sh.disable_deblocking_filter_flag)
1881                     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1882             }
1883         }
1884     }
1885
1886     if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
1887         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
1888
1889     x = y_cb * min_cb_width + x_cb;
1890     for (y = 0; y < length; y++) {
1891         memset(&s->qp_y_tab[x], lc->qp_y, length);
1892         x += min_cb_width;
1893     }
1894
1895     if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1896        ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0) {
1897         lc->qPy_pred = lc->qp_y;
1898     }
1899
1900     set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
1901
1902     return 0;
1903 }
1904
1905 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
1906                                int log2_cb_size, int cb_depth)
1907 {
1908     HEVCLocalContext *lc = s->HEVClc;
1909     const int cb_size    = 1 << log2_cb_size;
1910     int ret;
1911     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1912
1913     lc->ct.depth = cb_depth;
1914     if (x0 + cb_size <= s->sps->width  &&
1915         y0 + cb_size <= s->sps->height &&
1916         log2_cb_size > s->sps->log2_min_cb_size) {
1917         SAMPLE(s->split_cu_flag, x0, y0) =
1918             ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
1919     } else {
1920         SAMPLE(s->split_cu_flag, x0, y0) =
1921             (log2_cb_size > s->sps->log2_min_cb_size);
1922     }
1923     if (s->pps->cu_qp_delta_enabled_flag &&
1924         log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
1925         lc->tu.is_cu_qp_delta_coded = 0;
1926         lc->tu.cu_qp_delta          = 0;
1927     }
1928
1929     if (SAMPLE(s->split_cu_flag, x0, y0)) {
1930         const int cb_size_split = cb_size >> 1;
1931         const int x1 = x0 + cb_size_split;
1932         const int y1 = y0 + cb_size_split;
1933
1934         int more_data = 0;
1935
1936         more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
1937         if (more_data < 0)
1938             return more_data;
1939
1940         if (more_data && x1 < s->sps->width) {
1941             more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
1942             if (more_data < 0)
1943                 return more_data;
1944         }
1945         if (more_data && y1 < s->sps->height) {
1946             more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
1947             if (more_data < 0)
1948                 return more_data;
1949         }
1950         if (more_data && x1 < s->sps->width &&
1951             y1 < s->sps->height) {
1952             more_data = hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
1953             if (more_data < 0)
1954                 return more_data;
1955         }
1956
1957         if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1958             ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0)
1959             lc->qPy_pred = lc->qp_y;
1960
1961         if (more_data)
1962             return ((x1 + cb_size_split) < s->sps->width ||
1963                     (y1 + cb_size_split) < s->sps->height);
1964         else
1965             return 0;
1966     } else {
1967         ret = hls_coding_unit(s, x0, y0, log2_cb_size);
1968         if (ret < 0)
1969             return ret;
1970         if ((!((x0 + cb_size) %
1971                (1 << (s->sps->log2_ctb_size))) ||
1972              (x0 + cb_size >= s->sps->width)) &&
1973             (!((y0 + cb_size) %
1974                (1 << (s->sps->log2_ctb_size))) ||
1975              (y0 + cb_size >= s->sps->height))) {
1976             int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
1977             return !end_of_slice_flag;
1978         } else {
1979             return 1;
1980         }
1981     }
1982
1983     return 0;
1984 }
1985
1986 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
1987                                  int ctb_addr_ts)
1988 {
1989     HEVCLocalContext *lc  = s->HEVClc;
1990     int ctb_size          = 1 << s->sps->log2_ctb_size;
1991     int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1992     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
1993
1994     int tile_left_boundary, tile_up_boundary;
1995     int slice_left_boundary, slice_up_boundary;
1996
1997     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
1998
1999     if (s->pps->entropy_coding_sync_enabled_flag) {
2000         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
2001             lc->first_qp_group = 1;
2002         lc->end_of_tiles_x = s->sps->width;
2003     } else if (s->pps->tiles_enabled_flag) {
2004         if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
2005             int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
2006             lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
2007             lc->first_qp_group   = 1;
2008         }
2009     } else {
2010         lc->end_of_tiles_x = s->sps->width;
2011     }
2012
2013     lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
2014
2015     if (s->pps->tiles_enabled_flag) {
2016         tile_left_boundary = x_ctb > 0 &&
2017                              s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
2018         slice_left_boundary = x_ctb > 0 &&
2019                               s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
2020         tile_up_boundary  = y_ctb > 0 &&
2021                             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]];
2022         slice_up_boundary = y_ctb > 0 &&
2023                             s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
2024     } else {
2025         tile_left_boundary =
2026         tile_up_boundary   = 0;
2027         slice_left_boundary = ctb_addr_in_slice <= 0;
2028         slice_up_boundary   = ctb_addr_in_slice < s->sps->ctb_width;
2029     }
2030     lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
2031     lc->slice_or_tiles_up_boundary   = slice_up_boundary   + (tile_up_boundary   << 1);
2032     lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0)                  && !tile_left_boundary);
2033     lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
2034     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]]));
2035     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]]));
2036 }
2037
2038 static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
2039 {
2040     HEVCContext *s  = avctxt->priv_data;
2041     int ctb_size    = 1 << s->sps->log2_ctb_size;
2042     int more_data   = 1;
2043     int x_ctb       = 0;
2044     int y_ctb       = 0;
2045     int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
2046
2047     if (!ctb_addr_ts && s->sh.dependent_slice_segment_flag) {
2048         av_log(s->avctx, AV_LOG_ERROR, "Impossible initial tile.\n");
2049         return AVERROR_INVALIDDATA;
2050     }
2051
2052     if (s->sh.dependent_slice_segment_flag) {
2053         int prev_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts - 1];
2054         if (s->tab_slice_address[prev_rs] != s->sh.slice_addr) {
2055             av_log(s->avctx, AV_LOG_ERROR, "Previous slice segment missing\n");
2056             return AVERROR_INVALIDDATA;
2057         }
2058     }
2059
2060     while (more_data && ctb_addr_ts < s->sps->ctb_size) {
2061         int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2062
2063         x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2064         y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2065         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2066
2067         ff_hevc_cabac_init(s, ctb_addr_ts);
2068
2069         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2070
2071         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
2072         s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
2073         s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
2074
2075         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2076         if (more_data < 0) {
2077             s->tab_slice_address[ctb_addr_rs] = -1;
2078             return more_data;
2079         }
2080
2081
2082         ctb_addr_ts++;
2083         ff_hevc_save_states(s, ctb_addr_ts);
2084         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2085     }
2086
2087     if (x_ctb + ctb_size >= s->sps->width &&
2088         y_ctb + ctb_size >= s->sps->height)
2089         ff_hevc_hls_filter(s, x_ctb, y_ctb);
2090
2091     return ctb_addr_ts;
2092 }
2093
2094 static int hls_slice_data(HEVCContext *s)
2095 {
2096     int arg[2];
2097     int ret[2];
2098
2099     arg[0] = 0;
2100     arg[1] = 1;
2101
2102     s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
2103     return ret[0];
2104 }
2105 static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
2106 {
2107     HEVCContext *s1  = avctxt->priv_data, *s;
2108     HEVCLocalContext *lc;
2109     int ctb_size    = 1<< s1->sps->log2_ctb_size;
2110     int more_data   = 1;
2111     int *ctb_row_p    = input_ctb_row;
2112     int ctb_row = ctb_row_p[job];
2113     int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
2114     int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
2115     int thread = ctb_row % s1->threads_number;
2116     int ret;
2117
2118     s = s1->sList[self_id];
2119     lc = s->HEVClc;
2120
2121     if(ctb_row) {
2122         ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
2123
2124         if (ret < 0)
2125             return ret;
2126         ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
2127     }
2128
2129     while(more_data && ctb_addr_ts < s->sps->ctb_size) {
2130         int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
2131         int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
2132
2133         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2134
2135         ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
2136
2137         if (avpriv_atomic_int_get(&s1->wpp_err)){
2138             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2139             return 0;
2140         }
2141
2142         ff_hevc_cabac_init(s, ctb_addr_ts);
2143         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2144         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2145
2146         if (more_data < 0) {
2147             s->tab_slice_address[ctb_addr_rs] = -1;
2148             return more_data;
2149         }
2150
2151         ctb_addr_ts++;
2152
2153         ff_hevc_save_states(s, ctb_addr_ts);
2154         ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
2155         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2156
2157         if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
2158             avpriv_atomic_int_set(&s1->wpp_err,  1);
2159             ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2160             return 0;
2161         }
2162
2163         if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
2164             ff_hevc_hls_filter(s, x_ctb, y_ctb);
2165             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2166             return ctb_addr_ts;
2167         }
2168         ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2169         x_ctb+=ctb_size;
2170
2171         if(x_ctb >= s->sps->width) {
2172             break;
2173         }
2174     }
2175     ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2176
2177     return 0;
2178 }
2179
2180 static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
2181 {
2182     HEVCLocalContext *lc = s->HEVClc;
2183     int *ret = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
2184     int *arg = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
2185     int offset;
2186     int startheader, cmpt = 0;
2187     int i, j, res = 0;
2188
2189
2190     if (!s->sList[1]) {
2191         ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
2192
2193
2194         for (i = 1; i < s->threads_number; i++) {
2195             s->sList[i] = av_malloc(sizeof(HEVCContext));
2196             memcpy(s->sList[i], s, sizeof(HEVCContext));
2197             s->HEVClcList[i] = av_malloc(sizeof(HEVCLocalContext));
2198             s->sList[i]->HEVClc = s->HEVClcList[i];
2199         }
2200     }
2201
2202     offset = (lc->gb.index >> 3);
2203
2204     for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
2205         if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2206             startheader--;
2207             cmpt++;
2208         }
2209     }
2210
2211     for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
2212         offset += (s->sh.entry_point_offset[i - 1] - cmpt);
2213         for (j = 0, cmpt = 0, startheader = offset
2214              + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
2215             if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2216                 startheader--;
2217                 cmpt++;
2218             }
2219         }
2220         s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
2221         s->sh.offset[i - 1] = offset;
2222
2223     }
2224     if (s->sh.num_entry_point_offsets != 0) {
2225         offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
2226         s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
2227         s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
2228
2229     }
2230     s->data = nal;
2231
2232     for (i = 1; i < s->threads_number; i++) {
2233         s->sList[i]->HEVClc->first_qp_group = 1;
2234         s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
2235         memcpy(s->sList[i], s, sizeof(HEVCContext));
2236         s->sList[i]->HEVClc = s->HEVClcList[i];
2237     }
2238
2239     avpriv_atomic_int_set(&s->wpp_err, 0);
2240     ff_reset_entries(s->avctx);
2241
2242     for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
2243         arg[i] = i;
2244         ret[i] = 0;
2245     }
2246
2247     if (s->pps->entropy_coding_sync_enabled_flag)
2248         s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
2249
2250     for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
2251         res += ret[i];
2252     av_free(ret);
2253     av_free(arg);
2254     return res;
2255 }
2256
2257 /**
2258  * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2259  * 0 if the unit should be skipped, 1 otherwise
2260  */
2261 static int hls_nal_unit(HEVCContext *s)
2262 {
2263     GetBitContext *gb = &s->HEVClc->gb;
2264     int nuh_layer_id;
2265
2266     if (get_bits1(gb) != 0)
2267         return AVERROR_INVALIDDATA;
2268
2269     s->nal_unit_type = get_bits(gb, 6);
2270
2271     nuh_layer_id   = get_bits(gb, 6);
2272     s->temporal_id = get_bits(gb, 3) - 1;
2273     if (s->temporal_id < 0)
2274         return AVERROR_INVALIDDATA;
2275
2276     av_log(s->avctx, AV_LOG_DEBUG,
2277            "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2278            s->nal_unit_type, nuh_layer_id, s->temporal_id);
2279
2280     return nuh_layer_id == 0;
2281 }
2282
2283 static void restore_tqb_pixels(HEVCContext *s)
2284 {
2285     int min_pu_size = 1 << s->sps->log2_min_pu_size;
2286     int x, y, c_idx;
2287
2288     for (c_idx = 0; c_idx < 3; c_idx++) {
2289         ptrdiff_t stride = s->frame->linesize[c_idx];
2290         int hshift       = s->sps->hshift[c_idx];
2291         int vshift       = s->sps->vshift[c_idx];
2292         for (y = 0; y < s->sps->min_pu_height; y++) {
2293             for (x = 0; x < s->sps->min_pu_width; x++) {
2294                 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2295                     int n;
2296                     int len      = min_pu_size >> hshift;
2297                     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)];
2298                     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)];
2299                     for (n = 0; n < (min_pu_size >> vshift); n++) {
2300                         memcpy(dst, src, len);
2301                         src += stride;
2302                         dst += stride;
2303                     }
2304                 }
2305             }
2306         }
2307     }
2308 }
2309
2310 static int set_side_data(HEVCContext *s)
2311 {
2312     AVFrame *out = s->ref->frame;
2313
2314     if (s->sei_frame_packing_present &&
2315         s->frame_packing_arrangement_type >= 3 &&
2316         s->frame_packing_arrangement_type <= 5 &&
2317         s->content_interpretation_type > 0 &&
2318         s->content_interpretation_type < 3) {
2319         AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2320         if (!stereo)
2321             return AVERROR(ENOMEM);
2322
2323         switch (s->frame_packing_arrangement_type) {
2324         case 3:
2325             if (s->quincunx_subsampling)
2326                 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2327             else
2328                 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2329             break;
2330         case 4:
2331             stereo->type = AV_STEREO3D_TOPBOTTOM;
2332             break;
2333         case 5:
2334             stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2335             break;
2336         }
2337
2338         if (s->content_interpretation_type == 2)
2339             stereo->flags = AV_STEREO3D_FLAG_INVERT;
2340     }
2341
2342     return 0;
2343 }
2344
2345 static int hevc_frame_start(HEVCContext *s)
2346 {
2347     HEVCLocalContext *lc = s->HEVClc;
2348     int pic_size_in_ctb  = ((s->sps->width  >> s->sps->log2_min_cb_size) + 1) *
2349                            ((s->sps->height >> s->sps->log2_min_cb_size) + 1);
2350     int ret;
2351     AVFrame *cur_frame;
2352
2353     memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2354     memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
2355     memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
2356     memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
2357     memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
2358
2359     s->is_decoded        = 0;
2360     s->first_nal_type    = s->nal_unit_type;
2361
2362     if (s->pps->tiles_enabled_flag)
2363         lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2364
2365     ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2366                               s->poc);
2367     if (ret < 0)
2368         goto fail;
2369
2370     ret = ff_hevc_frame_rps(s);
2371     if (ret < 0) {
2372         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2373         goto fail;
2374     }
2375
2376     ret = set_side_data(s);
2377     if (ret < 0)
2378         goto fail;
2379
2380     cur_frame = s->sps->sao_enabled ? s->sao_frame : s->frame;
2381     cur_frame->pict_type = 3 - s->sh.slice_type;
2382
2383     av_frame_unref(s->output_frame);
2384     ret = ff_hevc_output_frame(s, s->output_frame, 0);
2385     if (ret < 0)
2386         goto fail;
2387
2388     ff_thread_finish_setup(s->avctx);
2389
2390     return 0;
2391
2392 fail:
2393     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2394         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2395     s->ref = NULL;
2396     return ret;
2397 }
2398
2399 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2400 {
2401     HEVCLocalContext *lc = s->HEVClc;
2402     GetBitContext *gb    = &lc->gb;
2403     int ctb_addr_ts, ret;
2404
2405     ret = init_get_bits8(gb, nal, length);
2406     if (ret < 0)
2407         return ret;
2408
2409     ret = hls_nal_unit(s);
2410     if (ret < 0) {
2411         av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2412                s->nal_unit_type);
2413         goto fail;
2414     } else if (!ret)
2415         return 0;
2416
2417     switch (s->nal_unit_type) {
2418     case NAL_VPS:
2419         ret = ff_hevc_decode_nal_vps(s);
2420         if (ret < 0)
2421             goto fail;
2422         break;
2423     case NAL_SPS:
2424         ret = ff_hevc_decode_nal_sps(s);
2425         if (ret < 0)
2426             goto fail;
2427         break;
2428     case NAL_PPS:
2429         ret = ff_hevc_decode_nal_pps(s);
2430         if (ret < 0)
2431             goto fail;
2432         break;
2433     case NAL_SEI_PREFIX:
2434     case NAL_SEI_SUFFIX:
2435         ret = ff_hevc_decode_nal_sei(s);
2436         if (ret < 0)
2437             goto fail;
2438         break;
2439     case NAL_TRAIL_R:
2440     case NAL_TRAIL_N:
2441     case NAL_TSA_N:
2442     case NAL_TSA_R:
2443     case NAL_STSA_N:
2444     case NAL_STSA_R:
2445     case NAL_BLA_W_LP:
2446     case NAL_BLA_W_RADL:
2447     case NAL_BLA_N_LP:
2448     case NAL_IDR_W_RADL:
2449     case NAL_IDR_N_LP:
2450     case NAL_CRA_NUT:
2451     case NAL_RADL_N:
2452     case NAL_RADL_R:
2453     case NAL_RASL_N:
2454     case NAL_RASL_R:
2455         ret = hls_slice_header(s);
2456         if (ret < 0)
2457             return ret;
2458
2459         if (s->max_ra == INT_MAX) {
2460             if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2461                 s->max_ra = s->poc;
2462             } else {
2463                 if (IS_IDR(s))
2464                     s->max_ra = INT_MIN;
2465             }
2466         }
2467
2468         if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2469             s->poc <= s->max_ra) {
2470             s->is_decoded = 0;
2471             break;
2472         } else {
2473             if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2474                 s->max_ra = INT_MIN;
2475         }
2476
2477         if (s->sh.first_slice_in_pic_flag) {
2478             ret = hevc_frame_start(s);
2479             if (ret < 0)
2480                 return ret;
2481         } else if (!s->ref) {
2482             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2483             goto fail;
2484         }
2485
2486         if (s->nal_unit_type != s->first_nal_type) {
2487             av_log(s->avctx, AV_LOG_ERROR,
2488                    "Non-matching NAL types of the VCL NALUs: %d %d\n",
2489                    s->first_nal_type, s->nal_unit_type);
2490             return AVERROR_INVALIDDATA;
2491         }
2492
2493         if (!s->sh.dependent_slice_segment_flag &&
2494             s->sh.slice_type != I_SLICE) {
2495             ret = ff_hevc_slice_rpl(s);
2496             if (ret < 0) {
2497                 av_log(s->avctx, AV_LOG_WARNING,
2498                        "Error constructing the reference lists for the current slice.\n");
2499                 goto fail;
2500             }
2501         }
2502
2503         if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
2504             ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
2505         else
2506             ctb_addr_ts = hls_slice_data(s);
2507         if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2508             s->is_decoded = 1;
2509             if ((s->pps->transquant_bypass_enable_flag ||
2510                  (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2511                 s->sps->sao_enabled)
2512                 restore_tqb_pixels(s);
2513         }
2514
2515         if (ctb_addr_ts < 0) {
2516             ret = ctb_addr_ts;
2517             goto fail;
2518         }
2519         break;
2520     case NAL_EOS_NUT:
2521     case NAL_EOB_NUT:
2522         s->seq_decode = (s->seq_decode + 1) & 0xff;
2523         s->max_ra     = INT_MAX;
2524         break;
2525     case NAL_AUD:
2526     case NAL_FD_NUT:
2527         break;
2528     default:
2529         av_log(s->avctx, AV_LOG_INFO,
2530                "Skipping NAL unit %d\n", s->nal_unit_type);
2531     }
2532
2533     return 0;
2534 fail:
2535     if (s->avctx->err_recognition & AV_EF_EXPLODE)
2536         return ret;
2537     return 0;
2538 }
2539
2540 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2541  * between these functions would be nice. */
2542 int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
2543                          HEVCNAL *nal)
2544 {
2545     int i, si, di;
2546     uint8_t *dst;
2547
2548     s->skipped_bytes = 0;
2549 #define STARTCODE_TEST                                                  \
2550         if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
2551             if (src[i + 2] != 3) {                                      \
2552                 /* startcode, so we must be past the end */             \
2553                 length = i;                                             \
2554             }                                                           \
2555             break;                                                      \
2556         }
2557 #if HAVE_FAST_UNALIGNED
2558 #define FIND_FIRST_ZERO                                                 \
2559         if (i > 0 && !src[i])                                           \
2560             i--;                                                        \
2561         while (src[i])                                                  \
2562             i++
2563 #if HAVE_FAST_64BIT
2564     for (i = 0; i + 1 < length; i += 9) {
2565         if (!((~AV_RN64A(src + i) &
2566                (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2567               0x8000800080008080ULL))
2568             continue;
2569         FIND_FIRST_ZERO;
2570         STARTCODE_TEST;
2571         i -= 7;
2572     }
2573 #else
2574     for (i = 0; i + 1 < length; i += 5) {
2575         if (!((~AV_RN32A(src + i) &
2576                (AV_RN32A(src + i) - 0x01000101U)) &
2577               0x80008080U))
2578             continue;
2579         FIND_FIRST_ZERO;
2580         STARTCODE_TEST;
2581         i -= 3;
2582     }
2583 #endif /* HAVE_FAST_64BIT */
2584 #else
2585     for (i = 0; i + 1 < length; i += 2) {
2586         if (src[i])
2587             continue;
2588         if (i > 0 && src[i - 1] == 0)
2589             i--;
2590         STARTCODE_TEST;
2591     }
2592 #endif /* HAVE_FAST_UNALIGNED */
2593
2594     if (i >= length - 1) { // no escaped 0
2595         nal->data = src;
2596         nal->size = length;
2597         return length;
2598     }
2599
2600     av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2601                    length + FF_INPUT_BUFFER_PADDING_SIZE);
2602     if (!nal->rbsp_buffer)
2603         return AVERROR(ENOMEM);
2604
2605     dst = nal->rbsp_buffer;
2606
2607     memcpy(dst, src, i);
2608     si = di = i;
2609     while (si + 2 < length) {
2610         // remove escapes (very rare 1:2^22)
2611         if (src[si + 2] > 3) {
2612             dst[di++] = src[si++];
2613             dst[di++] = src[si++];
2614         } else if (src[si] == 0 && src[si + 1] == 0) {
2615             if (src[si + 2] == 3) { // escape
2616                 dst[di++] = 0;
2617                 dst[di++] = 0;
2618                 si       += 3;
2619
2620                 s->skipped_bytes++;
2621                 if (s->skipped_bytes_pos_size < s->skipped_bytes) {
2622                     s->skipped_bytes_pos_size *= 2;
2623                     av_reallocp_array(&s->skipped_bytes_pos,
2624                             s->skipped_bytes_pos_size,
2625                             sizeof(*s->skipped_bytes_pos));
2626                     if (!s->skipped_bytes_pos)
2627                         return AVERROR(ENOMEM);
2628                 }
2629                 if (s->skipped_bytes_pos)
2630                     s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
2631                 continue;
2632             } else // next start code
2633                 goto nsc;
2634         }
2635
2636         dst[di++] = src[si++];
2637     }
2638     while (si < length)
2639         dst[di++] = src[si++];
2640
2641 nsc:
2642     memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2643
2644     nal->data = dst;
2645     nal->size = di;
2646     return si;
2647 }
2648
2649 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2650 {
2651     int i, consumed, ret = 0;
2652
2653     s->ref = NULL;
2654     s->last_eos = s->eos;
2655     s->eos = 0;
2656
2657     /* split the input packet into NAL units, so we know the upper bound on the
2658      * number of slices in the frame */
2659     s->nb_nals = 0;
2660     while (length >= 4) {
2661         HEVCNAL *nal;
2662         int extract_length = 0;
2663
2664         if (s->is_nalff) {
2665             int i;
2666             for (i = 0; i < s->nal_length_size; i++)
2667                 extract_length = (extract_length << 8) | buf[i];
2668             buf    += s->nal_length_size;
2669             length -= s->nal_length_size;
2670
2671             if (extract_length > length) {
2672                 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2673                 ret = AVERROR_INVALIDDATA;
2674                 goto fail;
2675             }
2676         } else {
2677             /* search start code */
2678             while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2679                 ++buf;
2680                 --length;
2681                 if (length < 4) {
2682                     av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
2683                     ret = AVERROR_INVALIDDATA;
2684                     goto fail;
2685                 }
2686             }
2687
2688             buf           += 3;
2689             length        -= 3;
2690         }
2691
2692         if (!s->is_nalff)
2693             extract_length = length;
2694
2695         if (s->nals_allocated < s->nb_nals + 1) {
2696             int new_size = s->nals_allocated + 1;
2697             HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2698             if (!tmp) {
2699                 ret = AVERROR(ENOMEM);
2700                 goto fail;
2701             }
2702             s->nals = tmp;
2703             memset(s->nals + s->nals_allocated, 0,
2704                    (new_size - s->nals_allocated) * sizeof(*tmp));
2705             av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
2706             av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
2707             av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
2708             s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
2709             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));
2710             s->nals_allocated = new_size;
2711         }
2712         s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
2713         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
2714         nal = &s->nals[s->nb_nals];
2715
2716         consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
2717
2718         s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
2719         s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
2720         s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
2721
2722
2723         if (consumed < 0) {
2724             ret = consumed;
2725             goto fail;
2726         }
2727
2728         ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
2729         if (ret < 0)
2730             goto fail;
2731         hls_nal_unit(s);
2732
2733         if (s->nal_unit_type == NAL_EOB_NUT ||
2734             s->nal_unit_type == NAL_EOS_NUT)
2735             s->eos = 1;
2736
2737         buf    += consumed;
2738         length -= consumed;
2739     }
2740
2741     /* parse the NAL units */
2742     for (i = 0; i < s->nb_nals; i++) {
2743         int ret;
2744         s->skipped_bytes = s->skipped_bytes_nal[i];
2745         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
2746
2747         ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2748         if (ret < 0) {
2749             av_log(s->avctx, AV_LOG_WARNING,
2750                    "Error parsing NAL unit #%d.\n", i);
2751             goto fail;
2752         }
2753     }
2754
2755 fail:
2756     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2757         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2758
2759     return ret;
2760 }
2761
2762 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2763 {
2764     int i;
2765     for (i = 0; i < 16; i++)
2766         av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2767 }
2768
2769 static int verify_md5(HEVCContext *s, AVFrame *frame)
2770 {
2771     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2772     int pixel_shift;
2773     int i, j;
2774
2775     if (!desc)
2776         return AVERROR(EINVAL);
2777
2778     pixel_shift = desc->comp[0].depth_minus1 > 7;
2779
2780     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2781            s->poc);
2782
2783     /* the checksums are LE, so we have to byteswap for >8bpp formats
2784      * on BE arches */
2785 #if HAVE_BIGENDIAN
2786     if (pixel_shift && !s->checksum_buf) {
2787         av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2788                        FFMAX3(frame->linesize[0], frame->linesize[1],
2789                               frame->linesize[2]));
2790         if (!s->checksum_buf)
2791             return AVERROR(ENOMEM);
2792     }
2793 #endif
2794
2795     for (i = 0; frame->data[i]; i++) {
2796         int width  = s->avctx->coded_width;
2797         int height = s->avctx->coded_height;
2798         int w = (i == 1 || i == 2) ? (width  >> desc->log2_chroma_w) : width;
2799         int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2800         uint8_t md5[16];
2801
2802         av_md5_init(s->md5_ctx);
2803         for (j = 0; j < h; j++) {
2804             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2805 #if HAVE_BIGENDIAN
2806             if (pixel_shift) {
2807                 s->bdsp.bswap16_buf((uint16_t *) s->checksum_buf,
2808                                     (const uint16_t *) src, w);
2809                 src = s->checksum_buf;
2810             }
2811 #endif
2812             av_md5_update(s->md5_ctx, src, w << pixel_shift);
2813         }
2814         av_md5_final(s->md5_ctx, md5);
2815
2816         if (!memcmp(md5, s->md5[i], 16)) {
2817             av_log   (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2818             print_md5(s->avctx, AV_LOG_DEBUG, md5);
2819             av_log   (s->avctx, AV_LOG_DEBUG, "; ");
2820         } else {
2821             av_log   (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2822             print_md5(s->avctx, AV_LOG_ERROR, md5);
2823             av_log   (s->avctx, AV_LOG_ERROR, " != ");
2824             print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2825             av_log   (s->avctx, AV_LOG_ERROR, "\n");
2826             return AVERROR_INVALIDDATA;
2827         }
2828     }
2829
2830     av_log(s->avctx, AV_LOG_DEBUG, "\n");
2831
2832     return 0;
2833 }
2834
2835 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2836                              AVPacket *avpkt)
2837 {
2838     int ret;
2839     HEVCContext *s = avctx->priv_data;
2840
2841     if (!avpkt->size) {
2842         ret = ff_hevc_output_frame(s, data, 1);
2843         if (ret < 0)
2844             return ret;
2845
2846         *got_output = ret;
2847         return 0;
2848     }
2849
2850     s->ref = NULL;
2851     ret    = decode_nal_units(s, avpkt->data, avpkt->size);
2852     if (ret < 0)
2853         return ret;
2854
2855     /* verify the SEI checksum */
2856     if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2857         s->is_md5) {
2858         ret = verify_md5(s, s->ref->frame);
2859         if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2860             ff_hevc_unref_frame(s, s->ref, ~0);
2861             return ret;
2862         }
2863     }
2864     s->is_md5 = 0;
2865
2866     if (s->is_decoded) {
2867         s->ref->frame->key_frame = IS_IRAP(s);
2868         av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2869         s->is_decoded = 0;
2870     }
2871
2872     if (s->output_frame->buf[0]) {
2873         av_frame_move_ref(data, s->output_frame);
2874         *got_output = 1;
2875     }
2876
2877     return avpkt->size;
2878 }
2879
2880 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2881 {
2882     int ret;
2883
2884     ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2885     if (ret < 0)
2886         return ret;
2887
2888     dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2889     if (!dst->tab_mvf_buf)
2890         goto fail;
2891     dst->tab_mvf = src->tab_mvf;
2892
2893     dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2894     if (!dst->rpl_tab_buf)
2895         goto fail;
2896     dst->rpl_tab = src->rpl_tab;
2897
2898     dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2899     if (!dst->rpl_buf)
2900         goto fail;
2901
2902     dst->poc        = src->poc;
2903     dst->ctb_count  = src->ctb_count;
2904     dst->window     = src->window;
2905     dst->flags      = src->flags;
2906     dst->sequence   = src->sequence;
2907
2908     return 0;
2909 fail:
2910     ff_hevc_unref_frame(s, dst, ~0);
2911     return AVERROR(ENOMEM);
2912 }
2913
2914 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2915 {
2916     HEVCContext       *s = avctx->priv_data;
2917     HEVCLocalContext *lc = s->HEVClc;
2918     int i;
2919
2920     pic_arrays_free(s);
2921
2922     av_freep(&s->md5_ctx);
2923
2924     for(i=0; i < s->nals_allocated; i++) {
2925         av_freep(&s->skipped_bytes_pos_nal[i]);
2926     }
2927     av_freep(&s->skipped_bytes_pos_size_nal);
2928     av_freep(&s->skipped_bytes_nal);
2929     av_freep(&s->skipped_bytes_pos_nal);
2930
2931     av_freep(&s->cabac_state);
2932
2933     av_frame_free(&s->tmp_frame);
2934     av_frame_free(&s->output_frame);
2935
2936     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2937         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2938         av_frame_free(&s->DPB[i].frame);
2939     }
2940
2941     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2942         av_buffer_unref(&s->vps_list[i]);
2943     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2944         av_buffer_unref(&s->sps_list[i]);
2945     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2946         av_buffer_unref(&s->pps_list[i]);
2947
2948     av_freep(&s->sh.entry_point_offset);
2949     av_freep(&s->sh.offset);
2950     av_freep(&s->sh.size);
2951
2952     for (i = 1; i < s->threads_number; i++) {
2953         lc = s->HEVClcList[i];
2954         if (lc) {
2955             av_freep(&s->HEVClcList[i]);
2956             av_freep(&s->sList[i]);
2957         }
2958     }
2959     if (s->HEVClc == s->HEVClcList[0])
2960         s->HEVClc = NULL;
2961     av_freep(&s->HEVClcList[0]);
2962
2963     for (i = 0; i < s->nals_allocated; i++)
2964         av_freep(&s->nals[i].rbsp_buffer);
2965     av_freep(&s->nals);
2966     s->nals_allocated = 0;
2967
2968     return 0;
2969 }
2970
2971 static av_cold int hevc_init_context(AVCodecContext *avctx)
2972 {
2973     HEVCContext *s = avctx->priv_data;
2974     int i;
2975
2976     s->avctx = avctx;
2977
2978     s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
2979     if (!s->HEVClc)
2980         goto fail;
2981     s->HEVClcList[0] = s->HEVClc;
2982     s->sList[0] = s;
2983
2984     s->cabac_state = av_malloc(HEVC_CONTEXTS);
2985     if (!s->cabac_state)
2986         goto fail;
2987
2988     s->tmp_frame = av_frame_alloc();
2989     if (!s->tmp_frame)
2990         goto fail;
2991
2992     s->output_frame = av_frame_alloc();
2993     if (!s->output_frame)
2994         goto fail;
2995
2996     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2997         s->DPB[i].frame = av_frame_alloc();
2998         if (!s->DPB[i].frame)
2999             goto fail;
3000         s->DPB[i].tf.f = s->DPB[i].frame;
3001     }
3002
3003     s->max_ra = INT_MAX;
3004
3005     s->md5_ctx = av_md5_alloc();
3006     if (!s->md5_ctx)
3007         goto fail;
3008
3009     ff_bswapdsp_init(&s->bdsp);
3010
3011     s->context_initialized = 1;
3012     s->eos = 0;
3013
3014     return 0;
3015
3016 fail:
3017     hevc_decode_free(avctx);
3018     return AVERROR(ENOMEM);
3019 }
3020
3021 static int hevc_update_thread_context(AVCodecContext *dst,
3022                                       const AVCodecContext *src)
3023 {
3024     HEVCContext *s  = dst->priv_data;
3025     HEVCContext *s0 = src->priv_data;
3026     int i, ret;
3027
3028     if (!s->context_initialized) {
3029         ret = hevc_init_context(dst);
3030         if (ret < 0)
3031             return ret;
3032     }
3033
3034     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
3035         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
3036         if (s0->DPB[i].frame->buf[0]) {
3037             ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
3038             if (ret < 0)
3039                 return ret;
3040         }
3041     }
3042
3043     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
3044         av_buffer_unref(&s->vps_list[i]);
3045         if (s0->vps_list[i]) {
3046             s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
3047             if (!s->vps_list[i])
3048                 return AVERROR(ENOMEM);
3049         }
3050     }
3051
3052     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
3053         av_buffer_unref(&s->sps_list[i]);
3054         if (s0->sps_list[i]) {
3055             s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
3056             if (!s->sps_list[i])
3057                 return AVERROR(ENOMEM);
3058         }
3059     }
3060
3061     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
3062         av_buffer_unref(&s->pps_list[i]);
3063         if (s0->pps_list[i]) {
3064             s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
3065             if (!s->pps_list[i])
3066                 return AVERROR(ENOMEM);
3067         }
3068     }
3069
3070     if (s->sps != s0->sps)
3071         ret = set_sps(s, s0->sps);
3072
3073     s->seq_decode = s0->seq_decode;
3074     s->seq_output = s0->seq_output;
3075     s->pocTid0    = s0->pocTid0;
3076     s->max_ra     = s0->max_ra;
3077     s->eos        = s0->eos;
3078
3079     s->is_nalff        = s0->is_nalff;
3080     s->nal_length_size = s0->nal_length_size;
3081
3082     s->threads_number      = s0->threads_number;
3083     s->threads_type        = s0->threads_type;
3084
3085     if (s0->eos) {
3086         s->seq_decode = (s->seq_decode + 1) & 0xff;
3087         s->max_ra = INT_MAX;
3088     }
3089
3090     return 0;
3091 }
3092
3093 static int hevc_decode_extradata(HEVCContext *s)
3094 {
3095     AVCodecContext *avctx = s->avctx;
3096     GetByteContext gb;
3097     int ret;
3098
3099     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
3100
3101     if (avctx->extradata_size > 3 &&
3102         (avctx->extradata[0] || avctx->extradata[1] ||
3103          avctx->extradata[2] > 1)) {
3104         /* It seems the extradata is encoded as hvcC format.
3105          * Temporarily, we support configurationVersion==0 until 14496-15 3rd
3106          * is finalized. When finalized, configurationVersion will be 1 and we
3107          * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
3108         int i, j, num_arrays, nal_len_size;
3109
3110         s->is_nalff = 1;
3111
3112         bytestream2_skip(&gb, 21);
3113         nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
3114         num_arrays   = bytestream2_get_byte(&gb);
3115
3116         /* nal units in the hvcC always have length coded with 2 bytes,
3117          * so put a fake nal_length_size = 2 while parsing them */
3118         s->nal_length_size = 2;
3119
3120         /* Decode nal units from hvcC. */
3121         for (i = 0; i < num_arrays; i++) {
3122             int type = bytestream2_get_byte(&gb) & 0x3f;
3123             int cnt  = bytestream2_get_be16(&gb);
3124
3125             for (j = 0; j < cnt; j++) {
3126                 // +2 for the nal size field
3127                 int nalsize = bytestream2_peek_be16(&gb) + 2;
3128                 if (bytestream2_get_bytes_left(&gb) < nalsize) {
3129                     av_log(s->avctx, AV_LOG_ERROR,
3130                            "Invalid NAL unit size in extradata.\n");
3131                     return AVERROR_INVALIDDATA;
3132                 }
3133
3134                 ret = decode_nal_units(s, gb.buffer, nalsize);
3135                 if (ret < 0) {
3136                     av_log(avctx, AV_LOG_ERROR,
3137                            "Decoding nal unit %d %d from hvcC failed\n",
3138                            type, i);
3139                     return ret;
3140                 }
3141                 bytestream2_skip(&gb, nalsize);
3142             }
3143         }
3144
3145         /* Now store right nal length size, that will be used to parse
3146          * all other nals */
3147         s->nal_length_size = nal_len_size;
3148     } else {
3149         s->is_nalff = 0;
3150         ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
3151         if (ret < 0)
3152             return ret;
3153     }
3154     return 0;
3155 }
3156
3157 static av_cold int hevc_decode_init(AVCodecContext *avctx)
3158 {
3159     HEVCContext *s = avctx->priv_data;
3160     int ret;
3161
3162     ff_init_cabac_states();
3163
3164     avctx->internal->allocate_progress = 1;
3165
3166     ret = hevc_init_context(avctx);
3167     if (ret < 0)
3168         return ret;
3169
3170     s->enable_parallel_tiles = 0;
3171     s->picture_struct = 0;
3172
3173     if(avctx->active_thread_type & FF_THREAD_SLICE)
3174         s->threads_number = avctx->thread_count;
3175     else
3176         s->threads_number = 1;
3177
3178     if (avctx->extradata_size > 0 && avctx->extradata) {
3179         ret = hevc_decode_extradata(s);
3180         if (ret < 0) {
3181             hevc_decode_free(avctx);
3182             return ret;
3183         }
3184     }
3185
3186     if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
3187             s->threads_type = FF_THREAD_FRAME;
3188         else
3189             s->threads_type = FF_THREAD_SLICE;
3190
3191     return 0;
3192 }
3193
3194 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
3195 {
3196     HEVCContext *s = avctx->priv_data;
3197     int ret;
3198
3199     memset(s, 0, sizeof(*s));
3200
3201     ret = hevc_init_context(avctx);
3202     if (ret < 0)
3203         return ret;
3204
3205     return 0;
3206 }
3207
3208 static void hevc_decode_flush(AVCodecContext *avctx)
3209 {
3210     HEVCContext *s = avctx->priv_data;
3211     ff_hevc_flush_dpb(s);
3212     s->max_ra = INT_MAX;
3213 }
3214
3215 #define OFFSET(x) offsetof(HEVCContext, x)
3216 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
3217
3218 static const AVProfile profiles[] = {
3219     { FF_PROFILE_HEVC_MAIN,                 "Main"                },
3220     { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
3221     { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
3222     { FF_PROFILE_UNKNOWN },
3223 };
3224
3225 static const AVOption options[] = {
3226     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
3227         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3228     { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
3229         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3230     { NULL },
3231 };
3232
3233 static const AVClass hevc_decoder_class = {
3234     .class_name = "HEVC decoder",
3235     .item_name  = av_default_item_name,
3236     .option     = options,
3237     .version    = LIBAVUTIL_VERSION_INT,
3238 };
3239
3240 AVCodec ff_hevc_decoder = {
3241     .name                  = "hevc",
3242     .long_name             = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3243     .type                  = AVMEDIA_TYPE_VIDEO,
3244     .id                    = AV_CODEC_ID_HEVC,
3245     .priv_data_size        = sizeof(HEVCContext),
3246     .priv_class            = &hevc_decoder_class,
3247     .init                  = hevc_decode_init,
3248     .close                 = hevc_decode_free,
3249     .decode                = hevc_decode_frame,
3250     .flush                 = hevc_decode_flush,
3251     .update_thread_context = hevc_update_thread_context,
3252     .init_thread_copy      = hevc_init_thread_copy,
3253     .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3254                              CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
3255     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
3256 };