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