Merge commit 'a7b365ae191f45a0d7ed7b34033d5d0cbdd47139'
authorMichael Niedermayer <michaelni@gmx.at>
Mon, 17 Nov 2014 10:54:41 +0000 (11:54 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Mon, 17 Nov 2014 10:54:41 +0000 (11:54 +0100)
* commit 'a7b365ae191f45a0d7ed7b34033d5d0cbdd47139':
  hevc: reduce code duplication in hls_prediction_unit()

Conflicts:
libavcodec/hevc.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
1  2 
libavcodec/hevc.c

diff --combined libavcodec/hevc.c
@@@ -1,29 -1,28 +1,29 @@@
  /*
 - * HEVC video decoder
 + * HEVC video Decoder
   *
   * Copyright (C) 2012 - 2013 Guillaume Martres
   * Copyright (C) 2012 - 2013 Mickael Raulet
   * Copyright (C) 2012 - 2013 Gildas Cocherel
   * Copyright (C) 2012 - 2013 Wassim Hamidouche
   *
 - * This file is part of Libav.
 + * This file is part of FFmpeg.
   *
 - * Libav is free software; you can redistribute it and/or
 + * FFmpeg is free software; you can redistribute it and/or
   * modify it under the terms of the GNU Lesser General Public
   * License as published by the Free Software Foundation; either
   * version 2.1 of the License, or (at your option) any later version.
   *
 - * Libav is distributed in the hope that it will be useful,
 + * FFmpeg is distributed in the hope that it will be useful,
   * but WITHOUT ANY WARRANTY; without even the implied warranty of
   * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   * Lesser General Public License for more details.
   *
   * You should have received a copy of the GNU Lesser General Public
 - * License along with Libav; if not, write to the Free Software
 + * License along with FFmpeg; if not, write to the Free Software
   * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
   */
  
 +#include "libavutil/atomic.h"
  #include "libavutil/attributes.h"
  #include "libavutil/common.h"
  #include "libavutil/display.h"
  #include "golomb.h"
  #include "hevc.h"
  
 -const uint8_t ff_hevc_qpel_extra_before[4] = { 0, 3, 3, 2 };
 -const uint8_t ff_hevc_qpel_extra_after[4]  = { 0, 3, 4, 4 };
 -const uint8_t ff_hevc_qpel_extra[4]        = { 0, 6, 7, 6 };
 -
 -static const uint8_t scan_1x1[1] = { 0 };
 -
 -static const uint8_t horiz_scan2x2_x[4] = { 0, 1, 0, 1 };
 -
 -static const uint8_t horiz_scan2x2_y[4] = { 0, 0, 1, 1 };
 -
 -static const uint8_t horiz_scan4x4_x[16] = {
 -    0, 1, 2, 3,
 -    0, 1, 2, 3,
 -    0, 1, 2, 3,
 -    0, 1, 2, 3,
 -};
 -
 -static const uint8_t horiz_scan4x4_y[16] = {
 -    0, 0, 0, 0,
 -    1, 1, 1, 1,
 -    2, 2, 2, 2,
 -    3, 3, 3, 3,
 -};
 -
 -static const uint8_t horiz_scan8x8_inv[8][8] = {
 -    {  0,  1,  2,  3, 16, 17, 18, 19, },
 -    {  4,  5,  6,  7, 20, 21, 22, 23, },
 -    {  8,  9, 10, 11, 24, 25, 26, 27, },
 -    { 12, 13, 14, 15, 28, 29, 30, 31, },
 -    { 32, 33, 34, 35, 48, 49, 50, 51, },
 -    { 36, 37, 38, 39, 52, 53, 54, 55, },
 -    { 40, 41, 42, 43, 56, 57, 58, 59, },
 -    { 44, 45, 46, 47, 60, 61, 62, 63, },
 -};
 -
 -static const uint8_t diag_scan2x2_x[4] = { 0, 0, 1, 1 };
 -
 -static const uint8_t diag_scan2x2_y[4] = { 0, 1, 0, 1 };
 -
 -static const uint8_t diag_scan2x2_inv[2][2] = {
 -    { 0, 2, },
 -    { 1, 3, },
 -};
 -
 -const uint8_t ff_hevc_diag_scan4x4_x[16] = {
 -    0, 0, 1, 0,
 -    1, 2, 0, 1,
 -    2, 3, 1, 2,
 -    3, 2, 3, 3,
 -};
 -
 -const uint8_t ff_hevc_diag_scan4x4_y[16] = {
 -    0, 1, 0, 2,
 -    1, 0, 3, 2,
 -    1, 0, 3, 2,
 -    1, 3, 2, 3,
 -};
 -
 -static const uint8_t diag_scan4x4_inv[4][4] = {
 -    { 0,  2,  5,  9, },
 -    { 1,  4,  8, 12, },
 -    { 3,  7, 11, 14, },
 -    { 6, 10, 13, 15, },
 -};
 -
 -const uint8_t ff_hevc_diag_scan8x8_x[64] = {
 -    0, 0, 1, 0,
 -    1, 2, 0, 1,
 -    2, 3, 0, 1,
 -    2, 3, 4, 0,
 -    1, 2, 3, 4,
 -    5, 0, 1, 2,
 -    3, 4, 5, 6,
 -    0, 1, 2, 3,
 -    4, 5, 6, 7,
 -    1, 2, 3, 4,
 -    5, 6, 7, 2,
 -    3, 4, 5, 6,
 -    7, 3, 4, 5,
 -    6, 7, 4, 5,
 -    6, 7, 5, 6,
 -    7, 6, 7, 7,
 -};
 -
 -const uint8_t ff_hevc_diag_scan8x8_y[64] = {
 -    0, 1, 0, 2,
 -    1, 0, 3, 2,
 -    1, 0, 4, 3,
 -    2, 1, 0, 5,
 -    4, 3, 2, 1,
 -    0, 6, 5, 4,
 -    3, 2, 1, 0,
 -    7, 6, 5, 4,
 -    3, 2, 1, 0,
 -    7, 6, 5, 4,
 -    3, 2, 1, 7,
 -    6, 5, 4, 3,
 -    2, 7, 6, 5,
 -    4, 3, 7, 6,
 -    5, 4, 7, 6,
 -    5, 7, 6, 7,
 -};
 -
 -static const uint8_t diag_scan8x8_inv[8][8] = {
 -    {  0,  2,  5,  9, 14, 20, 27, 35, },
 -    {  1,  4,  8, 13, 19, 26, 34, 42, },
 -    {  3,  7, 12, 18, 25, 33, 41, 48, },
 -    {  6, 11, 17, 24, 32, 40, 47, 53, },
 -    { 10, 16, 23, 31, 39, 46, 52, 57, },
 -    { 15, 22, 30, 38, 45, 51, 56, 60, },
 -    { 21, 29, 37, 44, 50, 55, 59, 62, },
 -    { 28, 36, 43, 49, 54, 58, 61, 63, },
 -};
 +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 };
  
  /**
   * NOTE: Each function hls_foo correspond to the function foo in the
@@@ -70,10 -181,6 +70,10 @@@ static void pic_arrays_free(HEVCContex
      av_freep(&s->horizontal_bs);
      av_freep(&s->vertical_bs);
  
 +    av_freep(&s->sh.entry_point_offset);
 +    av_freep(&s->sh.size);
 +    av_freep(&s->sh.offset);
 +
      av_buffer_pool_uninit(&s->tab_mvf_pool);
      av_buffer_pool_uninit(&s->rpl_tab_pool);
  }
@@@ -89,40 -196,40 +89,40 @@@ static int pic_arrays_init(HEVCContext 
      int ctb_count        = sps->ctb_width * sps->ctb_height;
      int min_pu_size      = sps->min_pu_width * sps->min_pu_height;
  
 -    s->bs_width  = width  >> 3;
 -    s->bs_height = height >> 3;
 +    s->bs_width  = (width  >> 2) + 1;
 +    s->bs_height = (height >> 2) + 1;
  
      s->sao           = av_mallocz_array(ctb_count, sizeof(*s->sao));
      s->deblock       = av_mallocz_array(ctb_count, sizeof(*s->deblock));
      if (!s->sao || !s->deblock)
          goto fail;
  
 -    s->skip_flag    = av_malloc(pic_size_in_ctb);
 -    s->tab_ct_depth = av_malloc(sps->min_cb_height * sps->min_cb_width);
 +    s->skip_flag    = av_malloc(sps->min_cb_height * sps->min_cb_width);
 +    s->tab_ct_depth = av_malloc_array(sps->min_cb_height, sps->min_cb_width);
      if (!s->skip_flag || !s->tab_ct_depth)
          goto fail;
  
 -    s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
 +    s->cbf_luma = av_malloc_array(sps->min_tb_width, sps->min_tb_height);
      s->tab_ipm  = av_mallocz(min_pu_size);
 -    s->is_pcm   = av_malloc(min_pu_size);
 +    s->is_pcm   = av_malloc((sps->min_pu_width + 1) * (sps->min_pu_height + 1));
      if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
          goto fail;
  
      s->filter_slice_edges = av_malloc(ctb_count);
 -    s->tab_slice_address  = av_malloc(pic_size_in_ctb *
 +    s->tab_slice_address  = av_malloc_array(pic_size_in_ctb,
                                        sizeof(*s->tab_slice_address));
 -    s->qp_y_tab           = av_malloc(pic_size_in_ctb *
 +    s->qp_y_tab           = av_malloc_array(pic_size_in_ctb,
                                        sizeof(*s->qp_y_tab));
      if (!s->qp_y_tab || !s->filter_slice_edges || !s->tab_slice_address)
          goto fail;
  
 -    s->horizontal_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
 -    s->vertical_bs   = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
 +    s->horizontal_bs = av_mallocz_array(s->bs_width, s->bs_height);
 +    s->vertical_bs   = av_mallocz_array(s->bs_width, s->bs_height);
      if (!s->horizontal_bs || !s->vertical_bs)
          goto fail;
  
      s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
 -                                          av_buffer_alloc);
 +                                          av_buffer_allocz);
      s->rpl_tab_pool = av_buffer_pool_init(ctb_count * sizeof(RefPicListTab),
                                            av_buffer_allocz);
      if (!s->tab_mvf_pool || !s->rpl_tab_pool)
@@@ -147,7 -254,7 +147,7 @@@ static void pred_weight_table(HEVCConte
      s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
      if (s->sps->chroma_format_idc != 0) {
          int delta = get_se_golomb(gb);
 -        s->sh.chroma_log2_weight_denom = av_clip_c(s->sh.luma_log2_weight_denom + delta, 0, 7);
 +        s->sh.chroma_log2_weight_denom = av_clip(s->sh.luma_log2_weight_denom + delta, 0, 7);
      }
  
      for (i = 0; i < s->sh.nb_refs[L0]; i++) {
              s->sh.luma_offset_l0[i] = 0;
          }
      }
 -    if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
 +    if (s->sps->chroma_format_idc != 0) {
          for (i = 0; i < s->sh.nb_refs[L0]; i++)
              chroma_weight_l0_flag[i] = get_bits1(gb);
      } else {
                  int delta_chroma_weight_l0 = get_se_golomb(gb);
                  int delta_chroma_offset_l0 = get_se_golomb(gb);
                  s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
 -                s->sh.chroma_offset_l0[i][j] = av_clip_c((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
 +                s->sh.chroma_offset_l0[i][j] = av_clip((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
                                                                                      >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
              }
          } else {
                      int delta_chroma_weight_l1 = get_se_golomb(gb);
                      int delta_chroma_offset_l1 = get_se_golomb(gb);
                      s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
 -                    s->sh.chroma_offset_l1[i][j] = av_clip_c((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
 +                    s->sh.chroma_offset_l1[i][j] = av_clip((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
                                                                                          >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
                  }
              } else {
@@@ -240,7 -347,7 +240,7 @@@ static int decode_lt_rps(HEVCContext *s
          nb_sps = get_ue_golomb_long(gb);
      nb_sh = get_ue_golomb_long(gb);
  
 -    if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
 +    if (nb_sh + (uint64_t)nb_sps > FF_ARRAY_ELEMS(rps->poc))
          return AVERROR_INVALIDDATA;
  
      rps->nb_refs = nb_sh + nb_sps;
      return 0;
  }
  
 +static int get_buffer_sao(HEVCContext *s, AVFrame *frame, const HEVCSPS *sps)
 +{
 +    int ret, i;
 +
 +    frame->width  = s->avctx->coded_width  + 2;
 +    frame->height = s->avctx->coded_height + 2;
 +    if ((ret = ff_get_buffer(s->avctx, frame, AV_GET_BUFFER_FLAG_REF)) < 0)
 +        return ret;
 +    for (i = 0; frame->data[i]; i++) {
 +        int offset = frame->linesize[i] + (1 << sps->pixel_shift);
 +        frame->data[i] += offset;
 +    }
 +    frame->width  = s->avctx->coded_width;
 +    frame->height = s->avctx->coded_height;
 +
 +    return 0;
 +}
 +
  static int set_sps(HEVCContext *s, const HEVCSPS *sps)
  {
      int ret;
  
      if (sps->sao_enabled) {
          av_frame_unref(s->tmp_frame);
 -        ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
 -        if (ret < 0)
 -            goto fail;
 -        s->frame = s->tmp_frame;
 +        ret = get_buffer_sao(s, s->tmp_frame, sps);
 +        s->sao_frame = s->tmp_frame;
      }
  
      s->sps = sps;
@@@ -364,9 -455,9 +364,9 @@@ fail
  
  static int hls_slice_header(HEVCContext *s)
  {
 -    GetBitContext *gb = &s->HEVClc.gb;
 +    GetBitContext *gb = &s->HEVClc->gb;
      SliceHeader *sh   = &s->sh;
 -    int i, ret;
 +    int i, j, ret;
  
      // Coded parameters
      sh->first_slice_in_pic_flag = get_bits1(gb);
          if (IS_IDR(s))
              ff_hevc_clear_refs(s);
      }
 +    sh->no_output_of_prior_pics_flag = 0;
      if (IS_IRAP(s))
          sh->no_output_of_prior_pics_flag = get_bits1(gb);
  
          return AVERROR_INVALIDDATA;
      }
      s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
 +    if (s->nal_unit_type == NAL_CRA_NUT && s->last_eos == 1)
 +        sh->no_output_of_prior_pics_flag = 1;
  
      if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
 +        const HEVCSPS* last_sps = s->sps;
          s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
 -
 +        if (last_sps && IS_IRAP(s) && s->nal_unit_type != NAL_CRA_NUT) {
 +            if (s->sps->width !=  last_sps->width || s->sps->height != last_sps->height ||
 +                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)
 +                sh->no_output_of_prior_pics_flag = 0;
 +        }
          ff_hevc_clear_refs(s);
          ret = set_sps(s, s->sps);
          if (ret < 0)
              sh->slice_cr_qp_offset = 0;
          }
  
 +        if (s->pps->chroma_qp_offset_list_enabled_flag)
 +            sh->cu_chroma_qp_offset_enabled_flag = get_bits1(gb);
 +        else
 +            sh->cu_chroma_qp_offset_enabled_flag = 0;
 +
          if (s->pps->deblocking_filter_control_present_flag) {
              int deblocking_filter_override_flag = 0;
  
          sh->num_entry_point_offsets = get_ue_golomb_long(gb);
          if (sh->num_entry_point_offsets > 0) {
              int offset_len = get_ue_golomb_long(gb) + 1;
 -
 -            for (i = 0; i < sh->num_entry_point_offsets; i++)
 -                skip_bits(gb, offset_len);
 -        }
 +            int segments = offset_len >> 4;
 +            int rest = (offset_len & 15);
 +            av_freep(&sh->entry_point_offset);
 +            av_freep(&sh->offset);
 +            av_freep(&sh->size);
 +            sh->entry_point_offset = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
 +            sh->offset = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
 +            sh->size = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
 +            if (!sh->entry_point_offset || !sh->offset || !sh->size) {
 +                sh->num_entry_point_offsets = 0;
 +                av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate memory\n");
 +                return AVERROR(ENOMEM);
 +            }
 +            for (i = 0; i < sh->num_entry_point_offsets; i++) {
 +                int val = 0;
 +                for (j = 0; j < segments; j++) {
 +                    val <<= 16;
 +                    val += get_bits(gb, 16);
 +                }
 +                if (rest) {
 +                    val <<= rest;
 +                    val += get_bits(gb, rest);
 +                }
 +                sh->entry_point_offset[i] = val + 1; // +1; // +1 to get the size
 +            }
 +            if (s->threads_number > 1 && (s->pps->num_tile_rows > 1 || s->pps->num_tile_columns > 1)) {
 +                s->enable_parallel_tiles = 0; // TODO: you can enable tiles in parallel here
 +                s->threads_number = 1;
 +            } else
 +                s->enable_parallel_tiles = 0;
 +        } else
 +            s->enable_parallel_tiles = 0;
      }
  
      if (s->pps->slice_header_extension_present_flag) {
          unsigned int length = get_ue_golomb_long(gb);
 +        if (length*8LL > get_bits_left(gb)) {
 +            av_log(s->avctx, AV_LOG_ERROR, "too many slice_header_extension_data_bytes\n");
 +            return AVERROR_INVALIDDATA;
 +        }
          for (i = 0; i < length; i++)
              skip_bits(gb, 8);  // slice_header_extension_data_byte
      }
  
      // Inferred parameters
 -    sh->slice_qp = 26 + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
 +    sh->slice_qp = 26U + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
      if (sh->slice_qp > 51 ||
          sh->slice_qp < -s->sps->qp_bd_offset) {
          av_log(s->avctx, AV_LOG_ERROR,
          return AVERROR_INVALIDDATA;
      }
  
 -    s->HEVClc.first_qp_group = !s->sh.dependent_slice_segment_flag;
 +    if (get_bits_left(gb) < 0) {
 +        av_log(s->avctx, AV_LOG_ERROR,
 +               "Overread slice header by %d bits\n", -get_bits_left(gb));
 +        return AVERROR_INVALIDDATA;
 +    }
 +
 +    s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
  
      if (!s->pps->cu_qp_delta_enabled_flag)
 -        s->HEVClc.qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset,
 -                                52 + s->sps->qp_bd_offset) - s->sps->qp_bd_offset;
 +        s->HEVClc->qp_y = s->sh.slice_qp;
  
      s->slice_initialized = 1;
 +    s->HEVClc->tu.cu_qp_offset_cb = 0;
 +    s->HEVClc->tu.cu_qp_offset_cr = 0;
  
      return 0;
  }
@@@ -774,9 -812,10 +774,9 @@@ do 
  
  static void hls_sao_param(HEVCContext *s, int rx, int ry)
  {
 -    HEVCLocalContext *lc    = &s->HEVClc;
 +    HEVCLocalContext *lc    = s->HEVClc;
      int sao_merge_left_flag = 0;
      int sao_merge_up_flag   = 0;
 -    int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
      SAOParams *sao          = &CTB(s->sao, rx, ry);
      int c_idx, i;
  
      }
  
      for (c_idx = 0; c_idx < 3; c_idx++) {
 +        int log2_sao_offset_scale = c_idx == 0 ? s->pps->log2_sao_offset_scale_luma :
 +                                                 s->pps->log2_sao_offset_scale_chroma;
 +
          if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
              sao->type_idx[c_idx] = SAO_NOT_APPLIED;
              continue;
          // Inferred parameters
          sao->offset_val[c_idx][0] = 0;
          for (i = 0; i < 4; i++) {
 -            sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
 +            sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i];
              if (sao->type_idx[c_idx] == SAO_EDGE) {
                  if (i > 1)
                      sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
              } else if (sao->offset_sign[c_idx][i]) {
                  sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
              }
 +            sao->offset_val[c_idx][i + 1] <<= log2_sao_offset_scale;
          }
      }
  }
  #undef SET_SAO
  #undef CTB
  
 -static void hls_residual_coding(HEVCContext *s, int x0, int y0,
 -                                int log2_trafo_size, enum ScanType scan_idx,
 -                                int c_idx)
 -{
 -#define GET_COORD(offset, n)                                    \
 -    do {                                                        \
 -        x_c = (scan_x_cg[offset >> 4] << 2) + scan_x_off[n];    \
 -        y_c = (scan_y_cg[offset >> 4] << 2) + scan_y_off[n];    \
 -    } while (0)
 -    HEVCLocalContext *lc    = &s->HEVClc;
 -    int transform_skip_flag = 0;
 -
 -    int last_significant_coeff_x, last_significant_coeff_y;
 -    int last_scan_pos;
 -    int n_end;
 -    int num_coeff    = 0;
 -    int greater1_ctx = 1;
 -
 -    int num_last_subset;
 -    int x_cg_last_sig, y_cg_last_sig;
 -
 -    const uint8_t *scan_x_cg, *scan_y_cg, *scan_x_off, *scan_y_off;
 -
 -    ptrdiff_t stride = s->frame->linesize[c_idx];
 -    int hshift       = s->sps->hshift[c_idx];
 -    int vshift       = s->sps->vshift[c_idx];
 -    uint8_t *dst     = &s->frame->data[c_idx][(y0 >> vshift) * stride +
 -                                              ((x0 >> hshift) << s->sps->pixel_shift)];
 -    DECLARE_ALIGNED(16, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
 -    DECLARE_ALIGNED(8, uint8_t, significant_coeff_group_flag[8][8]) = { { 0 } };
 -
 -    int trafo_size = 1 << log2_trafo_size;
 -    int i, qp, shift, add, scale, scale_m;
 -    const uint8_t level_scale[] = { 40, 45, 51, 57, 64, 72 };
 -    const uint8_t *scale_matrix;
 -    uint8_t dc_scale;
 -
 -    // Derive QP for dequant
 -    if (!lc->cu.cu_transquant_bypass_flag) {
 -        static const int qp_c[] = {
 -            29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
 -        };
 -
 -        static const uint8_t rem6[51 + 2 * 6 + 1] = {
 -            0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
 -            3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
 -            0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
 -        };
 -
 -        static const uint8_t div6[51 + 2 * 6 + 1] = {
 -            0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2,  3,  3,  3,
 -            3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6,  6,  6,  6,
 -            7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
 -        };
 -        int qp_y = lc->qp_y;
 -
 -        if (c_idx == 0) {
 -            qp = qp_y + s->sps->qp_bd_offset;
 -        } else {
 -            int qp_i, offset;
 -
 -            if (c_idx == 1)
 -                offset = s->pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
 -            else
 -                offset = s->pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
 -
 -            qp_i = av_clip_c(qp_y + offset, -s->sps->qp_bd_offset, 57);
 -            if (qp_i < 30)
 -                qp = qp_i;
 -            else if (qp_i > 43)
 -                qp = qp_i - 6;
 -            else
 -                qp = qp_c[qp_i - 30];
 -
 -            qp += s->sps->qp_bd_offset;
 -        }
 -
 -        shift    = s->sps->bit_depth + log2_trafo_size - 5;
 -        add      = 1 << (shift - 1);
 -        scale    = level_scale[rem6[qp]] << (div6[qp]);
 -        scale_m  = 16; // default when no custom scaling lists.
 -        dc_scale = 16;
 -
 -        if (s->sps->scaling_list_enable_flag) {
 -            const ScalingList *sl = s->pps->scaling_list_data_present_flag ?
 -                                    &s->pps->scaling_list : &s->sps->scaling_list;
 -            int matrix_id = lc->cu.pred_mode != MODE_INTRA;
 -
 -            if (log2_trafo_size != 5)
 -                matrix_id = 3 * matrix_id + c_idx;
 -
 -            scale_matrix = sl->sl[log2_trafo_size - 2][matrix_id];
 -            if (log2_trafo_size >= 4)
 -                dc_scale = sl->sl_dc[log2_trafo_size - 4][matrix_id];
 -        }
 -    }
 -
 -    if (s->pps->transform_skip_enabled_flag &&
 -        !lc->cu.cu_transquant_bypass_flag   &&
 -        log2_trafo_size == 2) {
 -        transform_skip_flag = ff_hevc_transform_skip_flag_decode(s, c_idx);
 -    }
 -
 -    last_significant_coeff_x =
 -        ff_hevc_last_significant_coeff_x_prefix_decode(s, c_idx, log2_trafo_size);
 -    last_significant_coeff_y =
 -        ff_hevc_last_significant_coeff_y_prefix_decode(s, c_idx, log2_trafo_size);
 -
 -    if (last_significant_coeff_x > 3) {
 -        int suffix = ff_hevc_last_significant_coeff_suffix_decode(s, last_significant_coeff_x);
 -        last_significant_coeff_x = (1 << ((last_significant_coeff_x >> 1) - 1)) *
 -                                   (2 + (last_significant_coeff_x & 1)) +
 -                                   suffix;
 -    }
 -
 -    if (last_significant_coeff_y > 3) {
 -        int suffix = ff_hevc_last_significant_coeff_suffix_decode(s, last_significant_coeff_y);
 -        last_significant_coeff_y = (1 << ((last_significant_coeff_y >> 1) - 1)) *
 -                                   (2 + (last_significant_coeff_y & 1)) +
 -                                   suffix;
 -    }
 -
 -    if (scan_idx == SCAN_VERT)
 -        FFSWAP(int, last_significant_coeff_x, last_significant_coeff_y);
 -
 -    x_cg_last_sig = last_significant_coeff_x >> 2;
 -    y_cg_last_sig = last_significant_coeff_y >> 2;
 -
 -    switch (scan_idx) {
 -    case SCAN_DIAG: {
 -        int last_x_c = last_significant_coeff_x & 3;
 -        int last_y_c = last_significant_coeff_y & 3;
 +static int hls_cross_component_pred(HEVCContext *s, int idx) {
 +    HEVCLocalContext *lc    = s->HEVClc;
 +    int log2_res_scale_abs_plus1 = ff_hevc_log2_res_scale_abs(s, idx);
  
 -        scan_x_off = ff_hevc_diag_scan4x4_x;
 -        scan_y_off = ff_hevc_diag_scan4x4_y;
 -        num_coeff  = diag_scan4x4_inv[last_y_c][last_x_c];
 -        if (trafo_size == 4) {
 -            scan_x_cg = scan_1x1;
 -            scan_y_cg = scan_1x1;
 -        } else if (trafo_size == 8) {
 -            num_coeff += diag_scan2x2_inv[y_cg_last_sig][x_cg_last_sig] << 4;
 -            scan_x_cg  = diag_scan2x2_x;
 -            scan_y_cg  = diag_scan2x2_y;
 -        } else if (trafo_size == 16) {
 -            num_coeff += diag_scan4x4_inv[y_cg_last_sig][x_cg_last_sig] << 4;
 -            scan_x_cg  = ff_hevc_diag_scan4x4_x;
 -            scan_y_cg  = ff_hevc_diag_scan4x4_y;
 -        } else { // trafo_size == 32
 -            num_coeff += diag_scan8x8_inv[y_cg_last_sig][x_cg_last_sig] << 4;
 -            scan_x_cg  = ff_hevc_diag_scan8x8_x;
 -            scan_y_cg  = ff_hevc_diag_scan8x8_y;
 -        }
 -        break;
 -    }
 -    case SCAN_HORIZ:
 -        scan_x_cg  = horiz_scan2x2_x;
 -        scan_y_cg  = horiz_scan2x2_y;
 -        scan_x_off = horiz_scan4x4_x;
 -        scan_y_off = horiz_scan4x4_y;
 -        num_coeff  = horiz_scan8x8_inv[last_significant_coeff_y][last_significant_coeff_x];
 -        break;
 -    default: //SCAN_VERT
 -        scan_x_cg  = horiz_scan2x2_y;
 -        scan_y_cg  = horiz_scan2x2_x;
 -        scan_x_off = horiz_scan4x4_y;
 -        scan_y_off = horiz_scan4x4_x;
 -        num_coeff  = horiz_scan8x8_inv[last_significant_coeff_x][last_significant_coeff_y];
 -        break;
 +    if (log2_res_scale_abs_plus1 !=  0) {
 +        int res_scale_sign_flag = ff_hevc_res_scale_sign_flag(s, idx);
 +        lc->tu.res_scale_val = (1 << (log2_res_scale_abs_plus1 - 1)) *
 +                               (1 - 2 * res_scale_sign_flag);
 +    } else {
 +        lc->tu.res_scale_val = 0;
      }
 -    num_coeff++;
 -    num_last_subset = (num_coeff - 1) >> 4;
 -
 -    for (i = num_last_subset; i >= 0; i--) {
 -        int n, m;
 -        int x_cg, y_cg, x_c, y_c;
 -        int implicit_non_zero_coeff = 0;
 -        int64_t trans_coeff_level;
 -        int prev_sig = 0;
 -        int offset   = i << 4;
 -
 -        uint8_t significant_coeff_flag_idx[16];
 -        uint8_t nb_significant_coeff_flag = 0;
 -
 -        x_cg = scan_x_cg[i];
 -        y_cg = scan_y_cg[i];
 -
 -        if (i < num_last_subset && i > 0) {
 -            int ctx_cg = 0;
 -            if (x_cg < (1 << (log2_trafo_size - 2)) - 1)
 -                ctx_cg += significant_coeff_group_flag[x_cg + 1][y_cg];
 -            if (y_cg < (1 << (log2_trafo_size - 2)) - 1)
 -                ctx_cg += significant_coeff_group_flag[x_cg][y_cg + 1];
 -
 -            significant_coeff_group_flag[x_cg][y_cg] =
 -                ff_hevc_significant_coeff_group_flag_decode(s, c_idx, ctx_cg);
 -            implicit_non_zero_coeff = 1;
 -        } else {
 -            significant_coeff_group_flag[x_cg][y_cg] =
 -                ((x_cg == x_cg_last_sig && y_cg == y_cg_last_sig) ||
 -                 (x_cg == 0 && y_cg == 0));
 -        }
 -
 -        last_scan_pos = num_coeff - offset - 1;
  
 -        if (i == num_last_subset) {
 -            n_end                         = last_scan_pos - 1;
 -            significant_coeff_flag_idx[0] = last_scan_pos;
 -            nb_significant_coeff_flag     = 1;
 -        } else {
 -            n_end = 15;
 -        }
 -
 -        if (x_cg < ((1 << log2_trafo_size) - 1) >> 2)
 -            prev_sig = significant_coeff_group_flag[x_cg + 1][y_cg];
 -        if (y_cg < ((1 << log2_trafo_size) - 1) >> 2)
 -            prev_sig += significant_coeff_group_flag[x_cg][y_cg + 1] << 1;
 -
 -        for (n = n_end; n >= 0; n--) {
 -            GET_COORD(offset, n);
 -
 -            if (significant_coeff_group_flag[x_cg][y_cg] &&
 -                (n > 0 || implicit_non_zero_coeff == 0)) {
 -                if (ff_hevc_significant_coeff_flag_decode(s, c_idx, x_c, y_c,
 -                                                          log2_trafo_size,
 -                                                          scan_idx,
 -                                                          prev_sig) == 1) {
 -                    significant_coeff_flag_idx[nb_significant_coeff_flag] = n;
 -                    nb_significant_coeff_flag++;
 -                    implicit_non_zero_coeff = 0;
 -                }
 -            } else {
 -                int last_cg = (x_c == (x_cg << 2) && y_c == (y_cg << 2));
 -                if (last_cg && implicit_non_zero_coeff && significant_coeff_group_flag[x_cg][y_cg]) {
 -                    significant_coeff_flag_idx[nb_significant_coeff_flag] = n;
 -                    nb_significant_coeff_flag++;
 -                }
 -            }
 -        }
 -
 -        n_end = nb_significant_coeff_flag;
 -
 -        if (n_end) {
 -            int first_nz_pos_in_cg = 16;
 -            int last_nz_pos_in_cg = -1;
 -            int c_rice_param = 0;
 -            int first_greater1_coeff_idx = -1;
 -            uint8_t coeff_abs_level_greater1_flag[16] = { 0 };
 -            uint16_t coeff_sign_flag;
 -            int sum_abs = 0;
 -            int sign_hidden = 0;
 -
 -            // initialize first elem of coeff_bas_level_greater1_flag
 -            int ctx_set = (i > 0 && c_idx == 0) ? 2 : 0;
 -
 -            if (!(i == num_last_subset) && greater1_ctx == 0)
 -                ctx_set++;
 -            greater1_ctx      = 1;
 -            last_nz_pos_in_cg = significant_coeff_flag_idx[0];
 -
 -            for (m = 0; m < (n_end > 8 ? 8 : n_end); m++) {
 -                int n_idx = significant_coeff_flag_idx[m];
 -                int inc   = (ctx_set << 2) + greater1_ctx;
 -                coeff_abs_level_greater1_flag[n_idx] =
 -                    ff_hevc_coeff_abs_level_greater1_flag_decode(s, c_idx, inc);
 -                if (coeff_abs_level_greater1_flag[n_idx]) {
 -                    greater1_ctx = 0;
 -                } else if (greater1_ctx > 0 && greater1_ctx < 3) {
 -                    greater1_ctx++;
 -                }
 -
 -                if (coeff_abs_level_greater1_flag[n_idx] &&
 -                    first_greater1_coeff_idx == -1)
 -                    first_greater1_coeff_idx = n_idx;
 -            }
 -            first_nz_pos_in_cg = significant_coeff_flag_idx[n_end - 1];
 -            sign_hidden        = last_nz_pos_in_cg - first_nz_pos_in_cg >= 4 &&
 -                                 !lc->cu.cu_transquant_bypass_flag;
 -
 -            if (first_greater1_coeff_idx != -1) {
 -                coeff_abs_level_greater1_flag[first_greater1_coeff_idx] += ff_hevc_coeff_abs_level_greater2_flag_decode(s, c_idx, ctx_set);
 -            }
 -            if (!s->pps->sign_data_hiding_flag || !sign_hidden) {
 -                coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag) << (16 - nb_significant_coeff_flag);
 -            } else {
 -                coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag - 1) << (16 - (nb_significant_coeff_flag - 1));
 -            }
 -
 -            for (m = 0; m < n_end; m++) {
 -                n = significant_coeff_flag_idx[m];
 -                GET_COORD(offset, n);
 -                trans_coeff_level = 1 + coeff_abs_level_greater1_flag[n];
 -                if (trans_coeff_level == ((m < 8) ?
 -                                          ((n == first_greater1_coeff_idx) ? 3 : 2) : 1)) {
 -                    int last_coeff_abs_level_remaining = ff_hevc_coeff_abs_level_remaining(s, trans_coeff_level, c_rice_param);
 -
 -                    trans_coeff_level += last_coeff_abs_level_remaining;
 -                    if ((trans_coeff_level) > (3 * (1 << c_rice_param)))
 -                        c_rice_param = FFMIN(c_rice_param + 1, 4);
 -                }
 -                if (s->pps->sign_data_hiding_flag && sign_hidden) {
 -                    sum_abs += trans_coeff_level;
 -                    if (n == first_nz_pos_in_cg && ((sum_abs & 1) == 1))
 -                        trans_coeff_level = -trans_coeff_level;
 -                }
 -                if (coeff_sign_flag >> 15)
 -                    trans_coeff_level = -trans_coeff_level;
 -                coeff_sign_flag <<= 1;
 -                if (!lc->cu.cu_transquant_bypass_flag) {
 -                    if (s->sps->scaling_list_enable_flag) {
 -                        if (y_c || x_c || log2_trafo_size < 4) {
 -                            int pos;
 -                            switch (log2_trafo_size) {
 -                            case 3:  pos = (y_c        << 3) +  x_c;       break;
 -                            case 4:  pos = ((y_c >> 1) << 3) + (x_c >> 1); break;
 -                            case 5:  pos = ((y_c >> 2) << 3) + (x_c >> 2); break;
 -                            default: pos = (y_c        << 2) +  x_c;
 -                            }
 -                            scale_m = scale_matrix[pos];
 -                        } else {
 -                            scale_m = dc_scale;
 -                        }
 -                    }
 -                    trans_coeff_level = (trans_coeff_level * (int64_t)scale * (int64_t)scale_m + add) >> shift;
 -                    if(trans_coeff_level < 0) {
 -                        if((~trans_coeff_level) & 0xFffffffffff8000)
 -                            trans_coeff_level = -32768;
 -                    } else {
 -                        if (trans_coeff_level & 0xffffffffffff8000)
 -                            trans_coeff_level = 32767;
 -                    }
 -                }
 -                coeffs[y_c * trafo_size + x_c] = trans_coeff_level;
 -            }
 -        }
 -    }
  
 -    if (lc->cu.cu_transquant_bypass_flag) {
 -        s->hevcdsp.transquant_bypass[log2_trafo_size - 2](dst, coeffs, stride);
 -    } else {
 -        if (transform_skip_flag)
 -            s->hevcdsp.transform_skip(dst, coeffs, stride);
 -        else if (lc->cu.pred_mode == MODE_INTRA && c_idx == 0 &&
 -                 log2_trafo_size == 2)
 -            s->hevcdsp.transform_4x4_luma_add(dst, coeffs, stride);
 -        else
 -            s->hevcdsp.transform_add[log2_trafo_size - 2](dst, coeffs, stride);
 -    }
 +    return 0;
  }
  
  static int hls_transform_unit(HEVCContext *s, int x0, int y0,
                                int xBase, int yBase, int cb_xBase, int cb_yBase,
                                int log2_cb_size, int log2_trafo_size,
 -                              int blk_idx, int cbf_luma, int cbf_cb, int cbf_cr)
 +                              int blk_idx, int cbf_luma, int *cbf_cb, int *cbf_cr)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
 +    const int log2_trafo_size_c = log2_trafo_size - s->sps->hshift[1];
 +    int i;
  
      if (lc->cu.pred_mode == MODE_INTRA) {
          int trafo_size = 1 << log2_trafo_size;
          ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
  
          s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
 -        if (log2_trafo_size > 2) {
 -            trafo_size = trafo_size << (s->sps->hshift[1] - 1);
 -            ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
 -            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 1);
 -            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 2);
 -        } else if (blk_idx == 3) {
 -            trafo_size = trafo_size << s->sps->hshift[1];
 -            ff_hevc_set_neighbour_available(s, xBase, yBase,
 -                                            trafo_size, trafo_size);
 -            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
 -            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
 -        }
      }
  
 -    if (cbf_luma || cbf_cb || cbf_cr) {
 +    if (cbf_luma || cbf_cb[0] || cbf_cr[0] ||
 +        (s->sps->chroma_format_idc == 2 && (cbf_cb[1] || cbf_cr[1]))) {
          int scan_idx   = SCAN_DIAG;
          int scan_idx_c = SCAN_DIAG;
 +        int cbf_chroma = cbf_cb[0] || cbf_cr[0] ||
 +                         (s->sps->chroma_format_idc == 2 &&
 +                         (cbf_cb[1] || cbf_cr[1]));
  
          if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
              lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
                  return AVERROR_INVALIDDATA;
              }
  
 -            ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
 +            ff_hevc_set_qPy(s, cb_xBase, cb_yBase, log2_cb_size);
 +        }
 +
 +        if (s->sh.cu_chroma_qp_offset_enabled_flag && cbf_chroma &&
 +            !lc->cu.cu_transquant_bypass_flag  &&  !lc->tu.is_cu_chroma_qp_offset_coded) {
 +            int cu_chroma_qp_offset_flag = ff_hevc_cu_chroma_qp_offset_flag(s);
 +            if (cu_chroma_qp_offset_flag) {
 +                int cu_chroma_qp_offset_idx  = 0;
 +                if (s->pps->chroma_qp_offset_list_len_minus1 > 0) {
 +                    cu_chroma_qp_offset_idx = ff_hevc_cu_chroma_qp_offset_idx(s);
 +                    av_log(s->avctx, AV_LOG_ERROR,
 +                        "cu_chroma_qp_offset_idx not yet tested.\n");
 +                }
 +                lc->tu.cu_qp_offset_cb = s->pps->cb_qp_offset_list[cu_chroma_qp_offset_idx];
 +                lc->tu.cu_qp_offset_cr = s->pps->cr_qp_offset_list[cu_chroma_qp_offset_idx];
 +            } else {
 +                lc->tu.cu_qp_offset_cb = 0;
 +                lc->tu.cu_qp_offset_cr = 0;
 +            }
 +            lc->tu.is_cu_chroma_qp_offset_coded = 1;
          }
  
          if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
 -            if (lc->tu.cur_intra_pred_mode >= 6 &&
 -                lc->tu.cur_intra_pred_mode <= 14) {
 +            if (lc->tu.intra_pred_mode >= 6 &&
 +                lc->tu.intra_pred_mode <= 14) {
                  scan_idx = SCAN_VERT;
 -            } else if (lc->tu.cur_intra_pred_mode >= 22 &&
 -                       lc->tu.cur_intra_pred_mode <= 30) {
 +            } else if (lc->tu.intra_pred_mode >= 22 &&
 +                       lc->tu.intra_pred_mode <= 30) {
                  scan_idx = SCAN_HORIZ;
              }
  
 -            if (lc->pu.intra_pred_mode_c >=  6 &&
 -                lc->pu.intra_pred_mode_c <= 14) {
 +            if (lc->tu.intra_pred_mode_c >=  6 &&
 +                lc->tu.intra_pred_mode_c <= 14) {
                  scan_idx_c = SCAN_VERT;
 -            } else if (lc->pu.intra_pred_mode_c >= 22 &&
 -                       lc->pu.intra_pred_mode_c <= 30) {
 +            } else if (lc->tu.intra_pred_mode_c >= 22 &&
 +                       lc->tu.intra_pred_mode_c <= 30) {
                  scan_idx_c = SCAN_HORIZ;
              }
          }
  
 +        lc->tu.cross_pf = 0;
 +
          if (cbf_luma)
 -            hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
 -        if (log2_trafo_size > 2) {
 -            if (cbf_cb)
 -                hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
 -            if (cbf_cr)
 -                hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
 +            ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
 +        if (log2_trafo_size > 2 || s->sps->chroma_format_idc == 3) {
 +            int trafo_size_h = 1 << (log2_trafo_size_c + s->sps->hshift[1]);
 +            int trafo_size_v = 1 << (log2_trafo_size_c + s->sps->vshift[1]);
 +            lc->tu.cross_pf  = (s->pps->cross_component_prediction_enabled_flag && cbf_luma &&
 +                                (lc->cu.pred_mode == MODE_INTER ||
 +                                 (lc->tu.chroma_mode_c ==  4)));
 +
 +            if (lc->tu.cross_pf) {
 +                hls_cross_component_pred(s, 0);
 +            }
 +            for (i = 0; i < (s->sps->chroma_format_idc == 2 ? 2 : 1); i++) {
 +                if (lc->cu.pred_mode == MODE_INTRA) {
 +                    ff_hevc_set_neighbour_available(s, x0, y0 + (i << log2_trafo_size_c), trafo_size_h, trafo_size_v);
 +                    s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (i << log2_trafo_size_c), 1);
 +                }
 +                if (cbf_cb[i])
 +                    ff_hevc_hls_residual_coding(s, x0, y0 + (i << log2_trafo_size_c),
 +                                                log2_trafo_size_c, scan_idx_c, 1);
 +                else
 +                    if (lc->tu.cross_pf) {
 +                        ptrdiff_t stride = s->frame->linesize[1];
 +                        int hshift = s->sps->hshift[1];
 +                        int vshift = s->sps->vshift[1];
 +                        int16_t *coeffs_y = (int16_t*)lc->edge_emu_buffer;
 +                        int16_t *coeffs   = (int16_t*)lc->edge_emu_buffer2;
 +                        int size = 1 << log2_trafo_size_c;
 +
 +                        uint8_t *dst = &s->frame->data[1][(y0 >> vshift) * stride +
 +                                                              ((x0 >> hshift) << s->sps->pixel_shift)];
 +                        for (i = 0; i < (size * size); i++) {
 +                            coeffs[i] = ((lc->tu.res_scale_val * coeffs_y[i]) >> 3);
 +                        }
 +                        s->hevcdsp.transform_add[log2_trafo_size_c-2](dst, coeffs, stride);
 +                    }
 +            }
 +
 +            if (lc->tu.cross_pf) {
 +                hls_cross_component_pred(s, 1);
 +            }
 +            for (i = 0; i < (s->sps->chroma_format_idc == 2 ? 2 : 1); i++) {
 +                if (lc->cu.pred_mode == MODE_INTRA) {
 +                    ff_hevc_set_neighbour_available(s, x0, y0 + (i << log2_trafo_size_c), trafo_size_h, trafo_size_v);
 +                    s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (i << log2_trafo_size_c), 2);
 +                }
 +                if (cbf_cr[i])
 +                    ff_hevc_hls_residual_coding(s, x0, y0 + (i << log2_trafo_size_c),
 +                                                log2_trafo_size_c, scan_idx_c, 2);
 +                else
 +                    if (lc->tu.cross_pf) {
 +                        ptrdiff_t stride = s->frame->linesize[2];
 +                        int hshift = s->sps->hshift[2];
 +                        int vshift = s->sps->vshift[2];
 +                        int16_t *coeffs_y = (int16_t*)lc->edge_emu_buffer;
 +                        int16_t *coeffs   = (int16_t*)lc->edge_emu_buffer2;
 +                        int size = 1 << log2_trafo_size_c;
 +
 +                        uint8_t *dst = &s->frame->data[2][(y0 >> vshift) * stride +
 +                                                          ((x0 >> hshift) << s->sps->pixel_shift)];
 +                        for (i = 0; i < (size * size); i++) {
 +                            coeffs[i] = ((lc->tu.res_scale_val * coeffs_y[i]) >> 3);
 +                        }
 +                        s->hevcdsp.transform_add[log2_trafo_size_c-2](dst, coeffs, stride);
 +                    }
 +            }
          } else if (blk_idx == 3) {
 -            if (cbf_cb)
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
 -            if (cbf_cr)
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
 +            int trafo_size_h = 1 << (log2_trafo_size + 1);
 +            int trafo_size_v = 1 << (log2_trafo_size + s->sps->vshift[1]);
 +            for (i = 0; i < (s->sps->chroma_format_idc == 2 ? 2 : 1); i++) {
 +                if (lc->cu.pred_mode == MODE_INTRA) {
 +                    ff_hevc_set_neighbour_available(s, xBase, yBase + (i << log2_trafo_size),
 +                                                    trafo_size_h, trafo_size_v);
 +                    s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (i << log2_trafo_size), 1);
 +                }
 +                if (cbf_cb[i])
 +                    ff_hevc_hls_residual_coding(s, xBase, yBase + (i << log2_trafo_size),
 +                                                log2_trafo_size, scan_idx_c, 1);
 +            }
 +            for (i = 0; i < (s->sps->chroma_format_idc == 2 ? 2 : 1); i++) {
 +                if (lc->cu.pred_mode == MODE_INTRA) {
 +                    ff_hevc_set_neighbour_available(s, xBase, yBase + (i << log2_trafo_size),
 +                                                trafo_size_h, trafo_size_v);
 +                    s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (i << log2_trafo_size), 2);
 +                }
 +                if (cbf_cr[i])
 +                    ff_hevc_hls_residual_coding(s, xBase, yBase + (i << log2_trafo_size),
 +                                                log2_trafo_size, scan_idx_c, 2);
 +            }
 +        }
 +    } else if (lc->cu.pred_mode == MODE_INTRA) {
 +        if (log2_trafo_size > 2 || s->sps->chroma_format_idc == 3) {
 +            int trafo_size_h = 1 << (log2_trafo_size_c + s->sps->hshift[1]);
 +            int trafo_size_v = 1 << (log2_trafo_size_c + s->sps->vshift[1]);
 +            ff_hevc_set_neighbour_available(s, x0, y0, trafo_size_h, trafo_size_v);
 +            s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0, 1);
 +            s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0, 2);
 +            if (s->sps->chroma_format_idc == 2) {
 +                ff_hevc_set_neighbour_available(s, x0, y0 + (1 << log2_trafo_size_c),
 +                                                trafo_size_h, trafo_size_v);
 +                s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (1 << log2_trafo_size_c), 1);
 +                s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (1 << log2_trafo_size_c), 2);
 +            }
 +        } else if (blk_idx == 3) {
 +            int trafo_size_h = 1 << (log2_trafo_size + 1);
 +            int trafo_size_v = 1 << (log2_trafo_size + s->sps->vshift[1]);
 +            ff_hevc_set_neighbour_available(s, xBase, yBase,
 +                                            trafo_size_h, trafo_size_v);
 +            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
 +            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
 +            if (s->sps->chroma_format_idc == 2) {
 +                ff_hevc_set_neighbour_available(s, xBase, yBase + (1 << (log2_trafo_size)),
 +                                                trafo_size_h, trafo_size_v);
 +                s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (1 << (log2_trafo_size)), 1);
 +                s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (1 << (log2_trafo_size)), 2);
 +            }
          }
      }
 +
      return 0;
  }
  
@@@ -1087,34 -1335,17 +1087,34 @@@ static int hls_transform_tree(HEVCConte
                                int xBase, int yBase, int cb_xBase, int cb_yBase,
                                int log2_cb_size, int log2_trafo_size,
                                int trafo_depth, int blk_idx,
 -                              int cbf_cb, int cbf_cr)
 +                              const int *base_cbf_cb, const int *base_cbf_cr)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      uint8_t split_transform_flag;
 +    int cbf_cb[2];
 +    int cbf_cr[2];
      int ret;
  
 +    cbf_cb[0] = base_cbf_cb[0];
 +    cbf_cb[1] = base_cbf_cb[1];
 +    cbf_cr[0] = base_cbf_cr[0];
 +    cbf_cr[1] = base_cbf_cr[1];
 +
      if (lc->cu.intra_split_flag) {
 -        if (trafo_depth == 1)
 -            lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
 +        if (trafo_depth == 1) {
 +            lc->tu.intra_pred_mode   = lc->pu.intra_pred_mode[blk_idx];
 +            if (s->sps->chroma_format_idc == 3) {
 +                lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[blk_idx];
 +                lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[blk_idx];
 +            } else {
 +                lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[0];
 +                lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[0];
 +            }
 +        }
      } else {
 -        lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
 +        lc->tu.intra_pred_mode   = lc->pu.intra_pred_mode[0];
 +        lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[0];
 +        lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[0];
      }
  
      if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
                                 inter_split;
      }
  
 -    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cb))
 -        cbf_cb = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 -    else if (log2_trafo_size > 2 || trafo_depth == 0)
 -        cbf_cb = 0;
 -    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cr))
 -        cbf_cr = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 -    else if (log2_trafo_size > 2 || trafo_depth == 0)
 -        cbf_cr = 0;
 +    if (log2_trafo_size > 2 || s->sps->chroma_format_idc == 3) {
 +        if (trafo_depth == 0 || cbf_cb[0]) {
 +            cbf_cb[0] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 +            if (s->sps->chroma_format_idc == 2 && (!split_transform_flag || log2_trafo_size == 3)) {
 +                cbf_cb[1] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 +            }
 +        }
 +
 +        if (trafo_depth == 0 || cbf_cr[0]) {
 +            cbf_cr[0] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 +            if (s->sps->chroma_format_idc == 2 && (!split_transform_flag || log2_trafo_size == 3)) {
 +                cbf_cr[1] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 +            }
 +        }
 +    }
  
      if (split_transform_flag) {
          const int trafo_size_split = 1 << (log2_trafo_size - 1);
@@@ -1176,10 -1400,8 +1176,10 @@@ do 
          int cbf_luma         = 1;
  
          if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
 -            cbf_cb || cbf_cr)
 +            cbf_cb[0] || cbf_cr[0] ||
 +            (s->sps->chroma_format_idc == 2 && (cbf_cb[1] || cbf_cr[1]))) {
              cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
 +        }
  
          ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
                                   log2_cb_size, log2_trafo_size,
  
  static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
  {
 -    //TODO: non-4:2:0 support
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      GetBitContext gb;
      int cb_size   = 1 << log2_cb_size;
      int stride0   = s->frame->linesize[0];
      int   stride2 = s->frame->linesize[2];
      uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
  
 -    int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
 +    int length         = cb_size * cb_size * s->sps->pcm.bit_depth +
 +                         (((cb_size >> s->sps->hshift[1]) * (cb_size >> s->sps->vshift[1])) +
 +                          ((cb_size >> s->sps->hshift[2]) * (cb_size >> s->sps->vshift[2]))) *
 +                          s->sps->pcm.bit_depth_chroma;
      const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
      int ret;
  
      if (ret < 0)
          return ret;
  
 -    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
 -    s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
 -    s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
 +    s->hevcdsp.put_pcm(dst0, stride0, cb_size, cb_size,     &gb, s->sps->pcm.bit_depth);
 +    s->hevcdsp.put_pcm(dst1, stride1,
 +                       cb_size >> s->sps->hshift[1],
 +                       cb_size >> s->sps->vshift[1],
 +                       &gb, s->sps->pcm.bit_depth_chroma);
 +    s->hevcdsp.put_pcm(dst2, stride2,
 +                       cb_size >> s->sps->hshift[2],
 +                       cb_size >> s->sps->vshift[2],
 +                       &gb, s->sps->pcm.bit_depth_chroma);
      return 0;
  }
  
 -static void hls_mvd_coding(HEVCContext *s, int x0, int y0, int log2_cb_size)
 -{
 -    HEVCLocalContext *lc = &s->HEVClc;
 -    int x = ff_hevc_abs_mvd_greater0_flag_decode(s);
 -    int y = ff_hevc_abs_mvd_greater0_flag_decode(s);
 -
 -    if (x)
 -        x += ff_hevc_abs_mvd_greater1_flag_decode(s);
 -    if (y)
 -        y += ff_hevc_abs_mvd_greater1_flag_decode(s);
 -
 -    switch (x) {
 -    case 2: lc->pu.mvd.x = ff_hevc_mvd_decode(s);           break;
 -    case 1: lc->pu.mvd.x = ff_hevc_mvd_sign_flag_decode(s); break;
 -    case 0: lc->pu.mvd.x = 0;                               break;
 -    }
 -
 -    switch (y) {
 -    case 2: lc->pu.mvd.y = ff_hevc_mvd_decode(s);           break;
 -    case 1: lc->pu.mvd.y = ff_hevc_mvd_sign_flag_decode(s); break;
 -    case 0: lc->pu.mvd.y = 0;                               break;
 -    }
 -}
 -
  /**
 - * 8.5.3.2.2.1 Luma sample interpolation process
 + * 8.5.3.2.2.1 Luma sample unidirectional interpolation process
   *
   * @param s HEVC decoding context
   * @param dst target buffer for block data at block position
   * @param y_off vertical position of block from origin (0, 0)
   * @param block_w width of block
   * @param block_h height of block
 + * @param luma_weight weighting factor applied to the luma prediction
 + * @param luma_offset additive offset applied to the luma prediction value
   */
 -static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
 -                    AVFrame *ref, const Mv *mv, int x_off, int y_off,
 -                    int block_w, int block_h)
 +
 +static void luma_mc_uni(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
 +                        AVFrame *ref, const Mv *mv, int x_off, int y_off,
 +                        int block_w, int block_h, int luma_weight, int luma_offset)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      uint8_t *src         = ref->data[0];
      ptrdiff_t srcstride  = ref->linesize[0];
      int pic_width        = s->sps->width;
      int pic_height       = s->sps->height;
 -
 -    int mx         = mv->x & 3;
 -    int my         = mv->y & 3;
 -    int extra_left = ff_hevc_qpel_extra_before[mx];
 -    int extra_top  = ff_hevc_qpel_extra_before[my];
 +    int mx               = mv->x & 3;
 +    int my               = mv->y & 3;
 +    int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 +                           (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
 +    int idx              = ff_hevc_pel_weight[block_w];
  
      x_off += mv->x >> 2;
      y_off += mv->y >> 2;
      src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
  
 -    if (x_off < extra_left || y_off < extra_top ||
 -        x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
 -        y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
 +    if (x_off < QPEL_EXTRA_BEFORE || y_off < QPEL_EXTRA_AFTER ||
 +        x_off >= pic_width - block_w - QPEL_EXTRA_AFTER ||
 +        y_off >= pic_height - block_h - QPEL_EXTRA_AFTER) {
          const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
 -        int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
 -        int buf_offset = extra_top *
 -                         edge_emu_stride + (extra_left << s->sps->pixel_shift);
 +        int offset     = QPEL_EXTRA_BEFORE * srcstride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
 +        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
  
          s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
                                   edge_emu_stride, srcstride,
 -                                 block_w + ff_hevc_qpel_extra[mx],
 -                                 block_h + ff_hevc_qpel_extra[my],
 -                                 x_off - extra_left, y_off - extra_top,
 +                                 block_w + QPEL_EXTRA,
 +                                 block_h + QPEL_EXTRA,
 +                                 x_off - QPEL_EXTRA_BEFORE, y_off - QPEL_EXTRA_BEFORE,
                                   pic_width, pic_height);
          src = lc->edge_emu_buffer + buf_offset;
          srcstride = edge_emu_stride;
      }
 -    s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
 -                                     block_h, lc->mc_buffer);
 +
 +    if (!weight_flag)
 +        s->hevcdsp.put_hevc_qpel_uni[idx][!!my][!!mx](dst, dststride, src, srcstride,
 +                                                      block_h, mx, my, block_w);
 +    else
 +        s->hevcdsp.put_hevc_qpel_uni_w[idx][!!my][!!mx](dst, dststride, src, srcstride,
 +                                                        block_h, s->sh.luma_log2_weight_denom,
 +                                                        luma_weight, luma_offset, mx, my, block_w);
 +}
 +
 +/**
 + * 8.5.3.2.2.1 Luma sample bidirectional interpolation process
 + *
 + * @param s HEVC decoding context
 + * @param dst target buffer for block data at block position
 + * @param dststride stride of the dst buffer
 + * @param ref0 reference picture0 buffer at origin (0, 0)
 + * @param mv0 motion vector0 (relative to block position) to get pixel data from
 + * @param x_off horizontal position of block from origin (0, 0)
 + * @param y_off vertical position of block from origin (0, 0)
 + * @param block_w width of block
 + * @param block_h height of block
 + * @param ref1 reference picture1 buffer at origin (0, 0)
 + * @param mv1 motion vector1 (relative to block position) to get pixel data from
 + * @param current_mv current motion vector structure
 + */
 + static void luma_mc_bi(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
 +                       AVFrame *ref0, const Mv *mv0, int x_off, int y_off,
 +                       int block_w, int block_h, AVFrame *ref1, const Mv *mv1, struct MvField *current_mv)
 +{
 +    HEVCLocalContext *lc = s->HEVClc;
 +    ptrdiff_t src0stride  = ref0->linesize[0];
 +    ptrdiff_t src1stride  = ref1->linesize[0];
 +    int pic_width        = s->sps->width;
 +    int pic_height       = s->sps->height;
 +    int mx0              = mv0->x & 3;
 +    int my0              = mv0->y & 3;
 +    int mx1              = mv1->x & 3;
 +    int my1              = mv1->y & 3;
 +    int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 +                           (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
 +    int x_off0           = x_off + (mv0->x >> 2);
 +    int y_off0           = y_off + (mv0->y >> 2);
 +    int x_off1           = x_off + (mv1->x >> 2);
 +    int y_off1           = y_off + (mv1->y >> 2);
 +    int idx              = ff_hevc_pel_weight[block_w];
 +
 +    uint8_t *src0  = ref0->data[0] + y_off0 * src0stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
 +    uint8_t *src1  = ref1->data[0] + y_off1 * src1stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
 +
 +    if (x_off0 < QPEL_EXTRA_BEFORE || y_off0 < QPEL_EXTRA_AFTER ||
 +        x_off0 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
 +        y_off0 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
 +        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
 +        int offset     = QPEL_EXTRA_BEFORE * src0stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
 +        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
 +
 +        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset,
 +                                 edge_emu_stride, src0stride,
 +                                 block_w + QPEL_EXTRA,
 +                                 block_h + QPEL_EXTRA,
 +                                 x_off0 - QPEL_EXTRA_BEFORE, y_off0 - QPEL_EXTRA_BEFORE,
 +                                 pic_width, pic_height);
 +        src0 = lc->edge_emu_buffer + buf_offset;
 +        src0stride = edge_emu_stride;
 +    }
 +
 +    if (x_off1 < QPEL_EXTRA_BEFORE || y_off1 < QPEL_EXTRA_AFTER ||
 +        x_off1 >= pic_width - block_w - QPEL_EXTRA_AFTER ||
 +        y_off1 >= pic_height - block_h - QPEL_EXTRA_AFTER) {
 +        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
 +        int offset     = QPEL_EXTRA_BEFORE * src1stride       + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
 +        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->sps->pixel_shift);
 +
 +        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src1 - offset,
 +                                 edge_emu_stride, src1stride,
 +                                 block_w + QPEL_EXTRA,
 +                                 block_h + QPEL_EXTRA,
 +                                 x_off1 - QPEL_EXTRA_BEFORE, y_off1 - QPEL_EXTRA_BEFORE,
 +                                 pic_width, pic_height);
 +        src1 = lc->edge_emu_buffer2 + buf_offset;
 +        src1stride = edge_emu_stride;
 +    }
 +
 +    s->hevcdsp.put_hevc_qpel[idx][!!my0][!!mx0](lc->tmp, src0, src0stride,
 +                                                block_h, mx0, my0, block_w);
 +    if (!weight_flag)
 +        s->hevcdsp.put_hevc_qpel_bi[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, lc->tmp,
 +                                                       block_h, mx1, my1, block_w);
 +    else
 +        s->hevcdsp.put_hevc_qpel_bi_w[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, lc->tmp,
 +                                                         block_h, s->sh.luma_log2_weight_denom,
 +                                                         s->sh.luma_weight_l0[current_mv->ref_idx[0]],
 +                                                         s->sh.luma_weight_l1[current_mv->ref_idx[1]],
 +                                                         s->sh.luma_offset_l0[current_mv->ref_idx[0]],
 +                                                         s->sh.luma_offset_l1[current_mv->ref_idx[1]],
 +                                                         mx1, my1, block_w);
 +
  }
  
  /**
 - * 8.5.3.2.2.2 Chroma sample interpolation process
 + * 8.5.3.2.2.2 Chroma sample uniprediction interpolation process
   *
   * @param s HEVC decoding context
   * @param dst1 target buffer for block data at block position (U plane)
   * @param y_off vertical position of block from origin (0, 0)
   * @param block_w width of block
   * @param block_h height of block
 + * @param chroma_weight weighting factor applied to the chroma prediction
 + * @param chroma_offset additive offset applied to the chroma prediction value
   */
 -static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
 -                      ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
 -                      int x_off, int y_off, int block_w, int block_h)
 +
 +static void chroma_mc_uni(HEVCContext *s, uint8_t *dst0,
 +                          ptrdiff_t dststride, uint8_t *src0, ptrdiff_t srcstride, int reflist,
 +                          int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int chroma_weight, int chroma_offset)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 -    uint8_t *src1        = ref->data[1];
 -    uint8_t *src2        = ref->data[2];
 -    ptrdiff_t src1stride = ref->linesize[1];
 -    ptrdiff_t src2stride = ref->linesize[2];
 -    int pic_width        = s->sps->width >> 1;
 -    int pic_height       = s->sps->height >> 1;
 -
 -    int mx = mv->x & 7;
 -    int my = mv->y & 7;
 -
 -    x_off += mv->x >> 3;
 -    y_off += mv->y >> 3;
 -    src1  += y_off * src1stride + (x_off << s->sps->pixel_shift);
 -    src2  += y_off * src2stride + (x_off << s->sps->pixel_shift);
 +    HEVCLocalContext *lc = s->HEVClc;
 +    int pic_width        = s->sps->width >> s->sps->hshift[1];
 +    int pic_height       = s->sps->height >> s->sps->vshift[1];
 +    const Mv *mv         = &current_mv->mv[reflist];
 +    int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 +                           (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
 +    int idx              = ff_hevc_pel_weight[block_w];
 +    int hshift           = s->sps->hshift[1];
 +    int vshift           = s->sps->vshift[1];
 +    intptr_t mx          = mv->x & ((1 << (2 + hshift)) - 1);
 +    intptr_t my          = mv->y & ((1 << (2 + vshift)) - 1);
 +    intptr_t _mx         = mx << (1 - hshift);
 +    intptr_t _my         = my << (1 - vshift);
 +
 +    x_off += mv->x >> (2 + hshift);
 +    y_off += mv->y >> (2 + vshift);
 +    src0  += y_off * srcstride + (x_off << s->sps->pixel_shift);
  
      if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
          x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
          y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
          const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
 +        int offset0 = EPEL_EXTRA_BEFORE * (srcstride + (1 << s->sps->pixel_shift));
 +        int buf_offset0 = EPEL_EXTRA_BEFORE *
 +                          (edge_emu_stride + (1 << s->sps->pixel_shift));
 +        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset0,
 +                                 edge_emu_stride, srcstride,
 +                                 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
 +                                 x_off - EPEL_EXTRA_BEFORE,
 +                                 y_off - EPEL_EXTRA_BEFORE,
 +                                 pic_width, pic_height);
 +
 +        src0 = lc->edge_emu_buffer + buf_offset0;
 +        srcstride = edge_emu_stride;
 +    }
 +    if (!weight_flag)
 +        s->hevcdsp.put_hevc_epel_uni[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
 +                                                  block_h, _mx, _my, block_w);
 +    else
 +        s->hevcdsp.put_hevc_epel_uni_w[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
 +                                                        block_h, s->sh.chroma_log2_weight_denom,
 +                                                        chroma_weight, chroma_offset, _mx, _my, block_w);
 +}
 +
 +/**
 + * 8.5.3.2.2.2 Chroma sample bidirectional interpolation process
 + *
 + * @param s HEVC decoding context
 + * @param dst target buffer for block data at block position
 + * @param dststride stride of the dst buffer
 + * @param ref0 reference picture0 buffer at origin (0, 0)
 + * @param mv0 motion vector0 (relative to block position) to get pixel data from
 + * @param x_off horizontal position of block from origin (0, 0)
 + * @param y_off vertical position of block from origin (0, 0)
 + * @param block_w width of block
 + * @param block_h height of block
 + * @param ref1 reference picture1 buffer at origin (0, 0)
 + * @param mv1 motion vector1 (relative to block position) to get pixel data from
 + * @param current_mv current motion vector structure
 + * @param cidx chroma component(cb, cr)
 + */
 +static void chroma_mc_bi(HEVCContext *s, uint8_t *dst0, ptrdiff_t dststride, AVFrame *ref0, AVFrame *ref1,
 +                         int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int cidx)
 +{
 +    HEVCLocalContext *lc = s->HEVClc;
 +    uint8_t *src1        = ref0->data[cidx+1];
 +    uint8_t *src2        = ref1->data[cidx+1];
 +    ptrdiff_t src1stride = ref0->linesize[cidx+1];
 +    ptrdiff_t src2stride = ref1->linesize[cidx+1];
 +    int weight_flag      = (s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 +                           (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag);
 +    int pic_width        = s->sps->width >> s->sps->hshift[1];
 +    int pic_height       = s->sps->height >> s->sps->vshift[1];
 +    Mv *mv0              = &current_mv->mv[0];
 +    Mv *mv1              = &current_mv->mv[1];
 +    int hshift = s->sps->hshift[1];
 +    int vshift = s->sps->vshift[1];
 +
 +    intptr_t mx0 = mv0->x & ((1 << (2 + hshift)) - 1);
 +    intptr_t my0 = mv0->y & ((1 << (2 + vshift)) - 1);
 +    intptr_t mx1 = mv1->x & ((1 << (2 + hshift)) - 1);
 +    intptr_t my1 = mv1->y & ((1 << (2 + vshift)) - 1);
 +    intptr_t _mx0 = mx0 << (1 - hshift);
 +    intptr_t _my0 = my0 << (1 - vshift);
 +    intptr_t _mx1 = mx1 << (1 - hshift);
 +    intptr_t _my1 = my1 << (1 - vshift);
 +
 +    int x_off0 = x_off + (mv0->x >> (2 + hshift));
 +    int y_off0 = y_off + (mv0->y >> (2 + vshift));
 +    int x_off1 = x_off + (mv1->x >> (2 + hshift));
 +    int y_off1 = y_off + (mv1->y >> (2 + vshift));
 +    int idx = ff_hevc_pel_weight[block_w];
 +    src1  += y_off0 * src1stride + (int)((unsigned)x_off0 << s->sps->pixel_shift);
 +    src2  += y_off1 * src2stride + (int)((unsigned)x_off1 << s->sps->pixel_shift);
 +
 +    if (x_off0 < EPEL_EXTRA_BEFORE || y_off0 < EPEL_EXTRA_AFTER ||
 +        x_off0 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
 +        y_off0 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
 +        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
          int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
          int buf_offset1 = EPEL_EXTRA_BEFORE *
                            (edge_emu_stride + (1 << s->sps->pixel_shift));
 -        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
 -        int buf_offset2 = EPEL_EXTRA_BEFORE *
 -                          (edge_emu_stride + (1 << s->sps->pixel_shift));
  
          s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
                                   edge_emu_stride, src1stride,
                                   block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
 -                                 x_off - EPEL_EXTRA_BEFORE,
 -                                 y_off - EPEL_EXTRA_BEFORE,
 +                                 x_off0 - EPEL_EXTRA_BEFORE,
 +                                 y_off0 - EPEL_EXTRA_BEFORE,
                                   pic_width, pic_height);
  
          src1 = lc->edge_emu_buffer + buf_offset1;
          src1stride = edge_emu_stride;
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
 -                                             block_w, block_h, mx, my, lc->mc_buffer);
 +    }
 +
 +    if (x_off1 < EPEL_EXTRA_BEFORE || y_off1 < EPEL_EXTRA_AFTER ||
 +        x_off1 >= pic_width - block_w - EPEL_EXTRA_AFTER ||
 +        y_off1 >= pic_height - block_h - EPEL_EXTRA_AFTER) {
 +        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
 +        int offset1 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
 +        int buf_offset1 = EPEL_EXTRA_BEFORE *
 +                          (edge_emu_stride + (1 << s->sps->pixel_shift));
  
 -        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
 +        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src2 - offset1,
                                   edge_emu_stride, src2stride,
                                   block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
 -                                 x_off - EPEL_EXTRA_BEFORE,
 -                                 y_off - EPEL_EXTRA_BEFORE,
 +                                 x_off1 - EPEL_EXTRA_BEFORE,
 +                                 y_off1 - EPEL_EXTRA_BEFORE,
                                   pic_width, pic_height);
 -        src2 = lc->edge_emu_buffer + buf_offset2;
 -        src2stride = edge_emu_stride;
  
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
 -                                             block_w, block_h, mx, my,
 -                                             lc->mc_buffer);
 -    } else {
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
 -                                             block_w, block_h, mx, my,
 -                                             lc->mc_buffer);
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
 -                                             block_w, block_h, mx, my,
 -                                             lc->mc_buffer);
 +        src2 = lc->edge_emu_buffer2 + buf_offset1;
 +        src2stride = edge_emu_stride;
      }
 +
 +    s->hevcdsp.put_hevc_epel[idx][!!my0][!!mx0](lc->tmp, src1, src1stride,
 +                                                block_h, _mx0, _my0, block_w);
 +    if (!weight_flag)
 +        s->hevcdsp.put_hevc_epel_bi[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
 +                                                       src2, src2stride, lc->tmp,
 +                                                       block_h, _mx1, _my1, block_w);
 +    else
 +        s->hevcdsp.put_hevc_epel_bi_w[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
 +                                                         src2, src2stride, lc->tmp,
 +                                                         block_h,
 +                                                         s->sh.chroma_log2_weight_denom,
 +                                                         s->sh.chroma_weight_l0[current_mv->ref_idx[0]][cidx],
 +                                                         s->sh.chroma_weight_l1[current_mv->ref_idx[1]][cidx],
 +                                                         s->sh.chroma_offset_l0[current_mv->ref_idx[0]][cidx],
 +                                                         s->sh.chroma_offset_l1[current_mv->ref_idx[1]][cidx],
 +                                                         _mx1, _my1, block_w);
  }
  
  static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
                                  const Mv *mv, int y0, int height)
  {
 -    int y = (mv->y >> 2) + y0 + height + 9;
 -    ff_thread_await_progress(&ref->tf, y, 0);
 +    int y = FFMAX(0, (mv->y >> 2) + y0 + height + 9);
 +
 +    if (s->threads_type == FF_THREAD_FRAME )
 +        ff_thread_await_progress(&ref->tf, y, 0);
  }
  
  static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                                  int nPbW, int nPbH,
 -                                int log2_cb_size, int partIdx)
 +                                int log2_cb_size, int partIdx, int idx)
  {
  #define POS(c_idx, x, y)                                                              \
      &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
                             (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      int merge_idx = 0;
      struct MvField current_mv = {{{ 0 }}};
  
      MvField *tab_mvf = s->ref->tab_mvf;
      RefPicList  *refPicList = s->ref->refPicList;
      HEVCFrame *ref0, *ref1;
 -
 -    int tmpstride = MAX_PB_SIZE;
 -
      uint8_t *dst0 = POS(0, x0, y0);
      uint8_t *dst1 = POS(1, x0, y0);
      uint8_t *dst2 = POS(2, x0, y0);
      int x_pu, y_pu;
      int i, j;
  
-     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
+     int skip_flag = SAMPLE_CTB(s->skip_flag, x_cb, y_cb);
+     if (!skip_flag)
+         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
+     if (skip_flag || lc->pu.merge_flag) {
          if (s->sh.max_num_merge_cand > 1)
              merge_idx = ff_hevc_merge_idx_decode(s);
          else
              merge_idx = 0;
  
-         ff_hevc_luma_mv_merge_mode(s, x0, y0,
-                                    1 << log2_cb_size,
-                                    1 << log2_cb_size,
-                                    log2_cb_size, partIdx,
-                                    merge_idx, &current_mv);
+         ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                    partIdx, merge_idx, &current_mv);
          x_pu = x0 >> s->sps->log2_min_pu_size;
          y_pu = y0 >> s->sps->log2_min_pu_size;
  
          for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
              for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
                  tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
-     } else { /* MODE_INTER */
-         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
-         if (lc->pu.merge_flag) {
-             if (s->sh.max_num_merge_cand > 1)
-                 merge_idx = ff_hevc_merge_idx_decode(s);
-             else
-                 merge_idx = 0;
-             ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                        partIdx, merge_idx, &current_mv);
-             x_pu = x0 >> s->sps->log2_min_pu_size;
-             y_pu = y0 >> s->sps->log2_min_pu_size;
-             for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
-                 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
-                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
-         } else {
-             enum InterPredIdc inter_pred_idc = PRED_L0;
-             ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
-             current_mv.pred_flag = 0;
-             if (s->sh.slice_type == B_SLICE)
-                 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
-             if (inter_pred_idc != PRED_L1) {
-                 if (s->sh.nb_refs[L0]) {
-                     ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
-                     current_mv.ref_idx[0] = ref_idx[0];
-                 }
-                 current_mv.pred_flag = PF_L0;
-                 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
-                 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
-                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                          partIdx, merge_idx, &current_mv,
-                                          mvp_flag[0], 0);
-                 current_mv.mv[0].x += lc->pu.mvd.x;
-                 current_mv.mv[0].y += lc->pu.mvd.y;
+     } else {
+         enum InterPredIdc inter_pred_idc = PRED_L0;
+         ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
++        current_mv.pred_flag = 0;
+         if (s->sh.slice_type == B_SLICE)
+             inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
+         if (inter_pred_idc != PRED_L1) {
+             if (s->sh.nb_refs[L0]) {
+                 ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
+                 current_mv.ref_idx[0] = ref_idx[0];
              }
 -            current_mv.pred_flag[0] = 1;
 -            hls_mvd_coding(s, x0, y0, 0);
 +
-             if (inter_pred_idc != PRED_L0) {
-                 if (s->sh.nb_refs[L1]) {
-                     ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
-                     current_mv.ref_idx[1] = ref_idx[1];
-                 }
-                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
-                     AV_ZERO32(&lc->pu.mvd);
-                 } else {
-                     ff_hevc_hls_mvd_coding(s, x0, y0, 1);
-                 }
++            current_mv.pred_flag = PF_L0;
++            ff_hevc_hls_mvd_coding(s, x0, y0, 0);
+             mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
+             ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                      partIdx, merge_idx, &current_mv,
+                                      mvp_flag[0], 0);
+             current_mv.mv[0].x += lc->pu.mvd.x;
+             current_mv.mv[0].y += lc->pu.mvd.y;
+         }
  
-                 current_mv.pred_flag += PF_L1;
-                 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
-                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
-                                          partIdx, merge_idx, &current_mv,
-                                          mvp_flag[1], 1);
-                 current_mv.mv[1].x += lc->pu.mvd.x;
-                 current_mv.mv[1].y += lc->pu.mvd.y;
+         if (inter_pred_idc != PRED_L0) {
+             if (s->sh.nb_refs[L1]) {
+                 ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
+                 current_mv.ref_idx[1] = ref_idx[1];
              }
  
-             x_pu = x0 >> s->sps->log2_min_pu_size;
-             y_pu = y0 >> s->sps->log2_min_pu_size;
+             if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
+                 AV_ZERO32(&lc->pu.mvd);
+             } else {
 -                hls_mvd_coding(s, x0, y0, 1);
++                ff_hevc_hls_mvd_coding(s, x0, y0, 1);
+             }
  
-             for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
-                 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
-                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
 -            current_mv.pred_flag[1] = 1;
++            current_mv.pred_flag += PF_L1;
+             mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
+             ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                      partIdx, merge_idx, &current_mv,
+                                      mvp_flag[1], 1);
+             current_mv.mv[1].x += lc->pu.mvd.x;
+             current_mv.mv[1].y += lc->pu.mvd.y;
          }
+         x_pu = x0 >> s->sps->log2_min_pu_size;
+         y_pu = y0 >> s->sps->log2_min_pu_size;
+         for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
+             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
+                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
      }
  
 -    if (current_mv.pred_flag[0]) {
 +    if (current_mv.pred_flag & PF_L0) {
          ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
          if (!ref0)
              return;
          hevc_await_progress(s, ref0, &current_mv.mv[0], y0, nPbH);
      }
 -    if (current_mv.pred_flag[1]) {
 +    if (current_mv.pred_flag & PF_L1) {
          ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
          if (!ref1)
              return;
          hevc_await_progress(s, ref1, &current_mv.mv[1], y0, nPbH);
      }
  
 -    if (current_mv.pred_flag[0] && !current_mv.pred_flag[1]) {
 -        DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
 -        DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
 -
 -        luma_mc(s, tmp, tmpstride, ref0->frame,
 -                &current_mv.mv[0], x0, y0, nPbW, nPbH);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
 -                                     s->sh.luma_weight_l0[current_mv.ref_idx[0]],
 -                                     s->sh.luma_offset_l0[current_mv.ref_idx[0]],
 -                                     dst0, s->frame->linesize[0], tmp,
 -                                     tmpstride, nPbW, nPbH);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
 -        }
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
 -                  &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
 -                                     s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
 -                                     s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
 -                                     dst1, s->frame->linesize[1], tmp, tmpstride,
 -                                     nPbW / 2, nPbH / 2);
 -            s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
 -                                     s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
 -                                     s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
 -                                     dst2, s->frame->linesize[2], tmp2, tmpstride,
 -                                     nPbW / 2, nPbH / 2);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
 -            s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
 -        }
 -    } else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
 -        DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
 -        DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
 -
 -        if (!ref1)
 -            return;
 -
 -        luma_mc(s, tmp, tmpstride, ref1->frame,
 -                &current_mv.mv[1], x0, y0, nPbW, nPbH);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
 -                                      s->sh.luma_weight_l1[current_mv.ref_idx[1]],
 -                                      s->sh.luma_offset_l1[current_mv.ref_idx[1]],
 -                                      dst0, s->frame->linesize[0], tmp, tmpstride,
 -                                      nPbW, nPbH);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
 -        }
 -
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
 -                  &current_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
 -                                     s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
 -                                     s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
 -                                     dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
 -            s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
 -                                     s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
 -                                     s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
 -                                     dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
 -            s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
 -        }
 -    } else if (current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
 -        DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
 -        DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
 -        DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
 -        DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
 -        HEVCFrame *ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
 -        HEVCFrame *ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
 -
 -        if (!ref0 || !ref1)
 -            return;
 -
 -        luma_mc(s, tmp, tmpstride, ref0->frame,
 -                &current_mv.mv[0], x0, y0, nPbW, nPbH);
 -        luma_mc(s, tmp2, tmpstride, ref1->frame,
 -                &current_mv.mv[1], x0, y0, nPbW, nPbH);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
 -                                         s->sh.luma_weight_l0[current_mv.ref_idx[0]],
 -                                         s->sh.luma_weight_l1[current_mv.ref_idx[1]],
 -                                         s->sh.luma_offset_l0[current_mv.ref_idx[0]],
 -                                         s->sh.luma_offset_l1[current_mv.ref_idx[1]],
 -                                         dst0, s->frame->linesize[0],
 -                                         tmp, tmp2, tmpstride, nPbW, nPbH);
 -        } else {
 -            s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
 -                                             tmp, tmp2, tmpstride, nPbW, nPbH);
 -        }
 -
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
 -                  &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 -        chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
 -                  &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
 -                                         s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
 -                                         s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
 -                                         s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
 -                                         s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
 -                                         dst1, s->frame->linesize[1], tmp, tmp3,
 -                                         tmpstride, nPbW / 2, nPbH / 2);
 -            s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
 -                                         s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
 -                                         s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
 -                                         s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
 -                                         s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
 -                                         dst2, s->frame->linesize[2], tmp2, tmp4,
 -                                         tmpstride, nPbW / 2, nPbH / 2);
 -        } else {
 -            s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
 -            s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
 -        }
 +    if (current_mv.pred_flag == PF_L0) {
 +        int x0_c = x0 >> s->sps->hshift[1];
 +        int y0_c = y0 >> s->sps->vshift[1];
 +        int nPbW_c = nPbW >> s->sps->hshift[1];
 +        int nPbH_c = nPbH >> s->sps->vshift[1];
 +
 +        luma_mc_uni(s, dst0, s->frame->linesize[0], ref0->frame,
 +                    &current_mv.mv[0], x0, y0, nPbW, nPbH,
 +                    s->sh.luma_weight_l0[current_mv.ref_idx[0]],
 +                    s->sh.luma_offset_l0[current_mv.ref_idx[0]]);
 +
 +        chroma_mc_uni(s, dst1, s->frame->linesize[1], ref0->frame->data[1], ref0->frame->linesize[1],
 +                      0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
 +                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0]);
 +        chroma_mc_uni(s, dst2, s->frame->linesize[2], ref0->frame->data[2], ref0->frame->linesize[2],
 +                      0, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
 +                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1], s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1]);
 +    } else if (current_mv.pred_flag == PF_L1) {
 +        int x0_c = x0 >> s->sps->hshift[1];
 +        int y0_c = y0 >> s->sps->vshift[1];
 +        int nPbW_c = nPbW >> s->sps->hshift[1];
 +        int nPbH_c = nPbH >> s->sps->vshift[1];
 +
 +        luma_mc_uni(s, dst0, s->frame->linesize[0], ref1->frame,
 +                    &current_mv.mv[1], x0, y0, nPbW, nPbH,
 +                    s->sh.luma_weight_l1[current_mv.ref_idx[1]],
 +                    s->sh.luma_offset_l1[current_mv.ref_idx[1]]);
 +
 +        chroma_mc_uni(s, dst1, s->frame->linesize[1], ref1->frame->data[1], ref1->frame->linesize[1],
 +                      1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
 +                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0]);
 +
 +        chroma_mc_uni(s, dst2, s->frame->linesize[2], ref1->frame->data[2], ref1->frame->linesize[2],
 +                      1, x0_c, y0_c, nPbW_c, nPbH_c, &current_mv,
 +                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1], s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1]);
 +    } else if (current_mv.pred_flag == PF_BI) {
 +        int x0_c = x0 >> s->sps->hshift[1];
 +        int y0_c = y0 >> s->sps->vshift[1];
 +        int nPbW_c = nPbW >> s->sps->hshift[1];
 +        int nPbH_c = nPbH >> s->sps->vshift[1];
 +
 +        luma_mc_bi(s, dst0, s->frame->linesize[0], ref0->frame,
 +                   &current_mv.mv[0], x0, y0, nPbW, nPbH,
 +                   ref1->frame, &current_mv.mv[1], &current_mv);
 +
 +        chroma_mc_bi(s, dst1, s->frame->linesize[1], ref0->frame, ref1->frame,
 +                     x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 0);
 +
 +        chroma_mc_bi(s, dst2, s->frame->linesize[2], ref0->frame, ref1->frame,
 +                     x0_c, y0_c, nPbW_c, nPbH_c, &current_mv, 1);
      }
  }
  
  static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
                                  int prev_intra_luma_pred_flag)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      int x_pu             = x0 >> s->sps->log2_min_pu_size;
      int y_pu             = y0 >> s->sps->log2_min_pu_size;
      int min_pu_width     = s->sps->min_pu_width;
                 intra_pred_mode, size_in_pus);
  
          for (j = 0; j < size_in_pus; j++) {
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].is_intra     = 1;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[0] = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[1] = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[0]   = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[1]   = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].x      = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].y      = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].x      = 0;
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].y      = 0;
 +            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag = PF_INTRA;
          }
      }
  
@@@ -1851,14 -1970,10 +1837,14 @@@ static av_always_inline void set_ct_dep
                 ct_depth, length);
  }
  
 +static const uint8_t tab_mode_idx[] = {
 +     0,  1,  2,  2,  2,  2,  3,  5,  7,  8, 10, 12, 13, 15, 17, 18, 19, 20,
 +    21, 22, 23, 23, 24, 24, 25, 25, 26, 27, 27, 28, 28, 29, 29, 30, 31};
 +
  static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
                                    int log2_cb_size)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
      uint8_t prev_intra_luma_pred_flag[4];
      int split   = lc->cu.part_mode == PART_NxN;
          }
      }
  
 -    chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
 -    if (chroma_mode != 4) {
 -        if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
 -            lc->pu.intra_pred_mode_c = 34;
 -        else
 -            lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
 -    } else {
 -        lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
 +    if (s->sps->chroma_format_idc == 3) {
 +        for (i = 0; i < side; i++) {
 +            for (j = 0; j < side; j++) {
 +                lc->pu.chroma_mode_c[2 * i + j] = chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
 +                if (chroma_mode != 4) {
 +                    if (lc->pu.intra_pred_mode[2 * i + j] == intra_chroma_table[chroma_mode])
 +                        lc->pu.intra_pred_mode_c[2 * i + j] = 34;
 +                    else
 +                        lc->pu.intra_pred_mode_c[2 * i + j] = intra_chroma_table[chroma_mode];
 +                } else {
 +                    lc->pu.intra_pred_mode_c[2 * i + j] = lc->pu.intra_pred_mode[2 * i + j];
 +                }
 +            }
 +        }
 +    } else if (s->sps->chroma_format_idc == 2) {
 +        int mode_idx;
 +        lc->pu.chroma_mode_c[0] = chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
 +        if (chroma_mode != 4) {
 +            if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
 +                mode_idx = 34;
 +            else
 +                mode_idx = intra_chroma_table[chroma_mode];
 +        } else {
 +            mode_idx = lc->pu.intra_pred_mode[0];
 +        }
 +        lc->pu.intra_pred_mode_c[0] = tab_mode_idx[mode_idx];
 +    } else if (s->sps->chroma_format_idc != 0) {
 +        chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
 +        if (chroma_mode != 4) {
 +            if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
 +                lc->pu.intra_pred_mode_c[0] = 34;
 +            else
 +                lc->pu.intra_pred_mode_c[0] = intra_chroma_table[chroma_mode];
 +        } else {
 +            lc->pu.intra_pred_mode_c[0] = lc->pu.intra_pred_mode[0];
 +        }
      }
  }
  
@@@ -1927,7 -2014,7 +1913,7 @@@ static void intra_prediction_unit_defau
                                                  int x0, int y0,
                                                  int log2_cb_size)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      int pb_size          = 1 << log2_cb_size;
      int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
      int min_pu_width     = s->sps->min_pu_width;
  
      if (size_in_pus == 0)
          size_in_pus = 1;
 -    for (j = 0; j < size_in_pus; j++) {
 +    for (j = 0; j < size_in_pus; j++)
          memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
 -        for (k = 0; k < size_in_pus; k++)
 -            tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].is_intra = lc->cu.pred_mode == MODE_INTRA;
 -    }
 +    if (lc->cu.pred_mode == MODE_INTRA)
 +        for (j = 0; j < size_in_pus; j++)
 +            for (k = 0; k < size_in_pus; k++)
 +                tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].pred_flag = PF_INTRA;
  }
  
  static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
  {
      int cb_size          = 1 << log2_cb_size;
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      int log2_min_cb_size = s->sps->log2_min_cb_size;
      int length           = cb_size >> log2_min_cb_size;
      int min_cb_width     = s->sps->min_cb_width;
      int x_cb             = x0 >> log2_min_cb_size;
      int y_cb             = y0 >> log2_min_cb_size;
 +    int idx              = log2_cb_size - 2;
 +    int qp_block_mask    = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
      int x, y, ret;
  
      lc->cu.x                = x0;
              x += min_cb_width;
          }
          lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
 +    } else {
 +        x = y_cb * min_cb_width + x_cb;
 +        for (y = 0; y < length; y++) {
 +            memset(&s->skip_flag[x], 0, length);
 +            x += min_cb_width;
 +        }
      }
  
      if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
 -        hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
 +        hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0, idx);
          intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
  
          if (!s->sh.disable_deblocking_filter_flag)
              intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
              switch (lc->cu.part_mode) {
              case PART_2Nx2N:
 -                hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
 +                hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0, idx);
                  break;
              case PART_2NxN:
 -                hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0, idx);
 +                hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1, idx);
                  break;
              case PART_Nx2N:
 -                hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0, idx - 1);
 +                hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1, idx - 1);
                  break;
              case PART_2NxnU:
 -                hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0, idx);
 +                hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1, idx);
                  break;
              case PART_2NxnD:
 -                hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0, idx);
 +                hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1, idx);
                  break;
              case PART_nLx2N:
 -                hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0, idx - 2);
 +                hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1, idx - 2);
                  break;
              case PART_nRx2N:
 -                hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
 +                hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0, idx - 2);
 +                hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1, idx - 2);
                  break;
              case PART_NxN:
 -                hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
 -                hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
 +                hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0, idx - 1);
 +                hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1, idx - 1);
 +                hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2, idx - 1);
 +                hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3, idx - 1);
                  break;
              }
          }
                  rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
              }
              if (rqt_root_cbf) {
 +                const static int cbf[2] = { 0 };
                  lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
                                           s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
                                           s->sps->max_transform_hierarchy_depth_inter;
                  ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
                                           log2_cb_size,
 -                                         log2_cb_size, 0, 0, 0, 0);
 +                                         log2_cb_size, 0, 0, cbf, cbf);
                  if (ret < 0)
                      return ret;
              } else {
      }
  
      if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
 -        ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
 +        ff_hevc_set_qPy(s, x0, y0, log2_cb_size);
  
      x = y_cb * min_cb_width + x_cb;
      for (y = 0; y < length; y++) {
          x += min_cb_width;
      }
  
 -    set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
 +    if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
 +       ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0) {
 +        lc->qPy_pred = lc->qp_y;
 +    }
 +
 +    set_ct_depth(s, x0, y0, log2_cb_size, lc->ct_depth);
  
      return 0;
  }
  static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
                                 int log2_cb_size, int cb_depth)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      const int cb_size    = 1 << log2_cb_size;
 +    int ret;
 +    int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
      int split_cu;
  
 -    lc->ct.depth = cb_depth;
 +    lc->ct_depth = cb_depth;
      if (x0 + cb_size <= s->sps->width  &&
          y0 + cb_size <= s->sps->height &&
          log2_cb_size > s->sps->log2_min_cb_size) {
          lc->tu.cu_qp_delta          = 0;
      }
  
 +    if (s->sh.cu_chroma_qp_offset_enabled_flag &&
 +        log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_chroma_qp_offset_depth) {
 +        lc->tu.is_cu_chroma_qp_offset_coded = 0;
 +    }
 +
      if (split_cu) {
          const int cb_size_split = cb_size >> 1;
          const int x1 = x0 + cb_size_split;
          const int y1 = y0 + cb_size_split;
  
 -        log2_cb_size--;
 -        cb_depth++;
 +        int more_data = 0;
  
 -#define SUBDIVIDE(x, y)                                                \
 -do {                                                                   \
 -    if (x < s->sps->width && y < s->sps->height) {                     \
 -        int ret = hls_coding_quadtree(s, x, y, log2_cb_size, cb_depth);\
 -        if (ret < 0)                                                   \
 -            return ret;                                                \
 -    }                                                                  \
 -} while (0)
 +        more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
 +        if (more_data < 0)
 +            return more_data;
  
 -        SUBDIVIDE(x0, y0);
 -        SUBDIVIDE(x1, y0);
 -        SUBDIVIDE(x0, y1);
 -        SUBDIVIDE(x1, y1);
 +        if (more_data && x1 < s->sps->width) {
 +            more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
 +            if (more_data < 0)
 +                return more_data;
 +        }
 +        if (more_data && y1 < s->sps->height) {
 +            more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
 +            if (more_data < 0)
 +                return more_data;
 +        }
 +        if (more_data && x1 < s->sps->width &&
 +            y1 < s->sps->height) {
 +            more_data = hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
 +            if (more_data < 0)
 +                return more_data;
 +        }
 +
 +        if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
 +            ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0)
 +            lc->qPy_pred = lc->qp_y;
 +
 +        if (more_data)
 +            return ((x1 + cb_size_split) < s->sps->width ||
 +                    (y1 + cb_size_split) < s->sps->height);
 +        else
 +            return 0;
      } else {
 -        int ret = hls_coding_unit(s, x0, y0, log2_cb_size);
 +        ret = hls_coding_unit(s, x0, y0, log2_cb_size);
          if (ret < 0)
              return ret;
 +        if ((!((x0 + cb_size) %
 +               (1 << (s->sps->log2_ctb_size))) ||
 +             (x0 + cb_size >= s->sps->width)) &&
 +            (!((y0 + cb_size) %
 +               (1 << (s->sps->log2_ctb_size))) ||
 +             (y0 + cb_size >= s->sps->height))) {
 +            int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
 +            return !end_of_slice_flag;
 +        } else {
 +            return 1;
 +        }
      }
  
      return 0;
  static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
                                   int ctb_addr_ts)
  {
 -    HEVCLocalContext *lc  = &s->HEVClc;
 +    HEVCLocalContext *lc  = s->HEVClc;
      int ctb_size          = 1 << s->sps->log2_ctb_size;
      int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
      int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
      } else if (s->pps->tiles_enabled_flag) {
          if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
              int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
 -            lc->start_of_tiles_x = x_ctb;
              lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
              lc->first_qp_group   = 1;
          }
      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]]));
  }
  
 -static int hls_slice_data(HEVCContext *s)
 +static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
  {
 +    HEVCContext *s  = avctxt->priv_data;
      int ctb_size    = 1 << s->sps->log2_ctb_size;
      int more_data   = 1;
      int x_ctb       = 0;
      int y_ctb       = 0;
      int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
 -    int ret;
 +
 +    if (!ctb_addr_ts && s->sh.dependent_slice_segment_flag) {
 +        av_log(s->avctx, AV_LOG_ERROR, "Impossible initial tile.\n");
 +        return AVERROR_INVALIDDATA;
 +    }
 +
 +    if (s->sh.dependent_slice_segment_flag) {
 +        int prev_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts - 1];
 +        if (s->tab_slice_address[prev_rs] != s->sh.slice_addr) {
 +            av_log(s->avctx, AV_LOG_ERROR, "Previous slice segment missing\n");
 +            return AVERROR_INVALIDDATA;
 +        }
 +    }
  
      while (more_data && ctb_addr_ts < s->sps->ctb_size) {
          int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
          s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
          s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
  
 -        ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
 -        if (ret < 0)
 -            return ret;
 -        more_data = !ff_hevc_end_of_slice_flag_decode(s);
 +        more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
 +        if (more_data < 0) {
 +            s->tab_slice_address[ctb_addr_rs] = -1;
 +            return more_data;
 +        }
 +
  
          ctb_addr_ts++;
          ff_hevc_save_states(s, ctb_addr_ts);
  
      if (x_ctb + ctb_size >= s->sps->width &&
          y_ctb + ctb_size >= s->sps->height)
 -        ff_hevc_hls_filter(s, x_ctb, y_ctb);
 +        ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
  
      return ctb_addr_ts;
  }
  
 +static int hls_slice_data(HEVCContext *s)
 +{
 +    int arg[2];
 +    int ret[2];
 +
 +    arg[0] = 0;
 +    arg[1] = 1;
 +
 +    s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
 +    return ret[0];
 +}
 +static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
 +{
 +    HEVCContext *s1  = avctxt->priv_data, *s;
 +    HEVCLocalContext *lc;
 +    int ctb_size    = 1<< s1->sps->log2_ctb_size;
 +    int more_data   = 1;
 +    int *ctb_row_p    = input_ctb_row;
 +    int ctb_row = ctb_row_p[job];
 +    int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
 +    int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
 +    int thread = ctb_row % s1->threads_number;
 +    int ret;
 +
 +    s = s1->sList[self_id];
 +    lc = s->HEVClc;
 +
 +    if(ctb_row) {
 +        ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
 +
 +        if (ret < 0)
 +            return ret;
 +        ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
 +    }
 +
 +    while(more_data && ctb_addr_ts < s->sps->ctb_size) {
 +        int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
 +        int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
 +
 +        hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
 +
 +        ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
 +
 +        if (avpriv_atomic_int_get(&s1->wpp_err)){
 +            ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
 +            return 0;
 +        }
 +
 +        ff_hevc_cabac_init(s, ctb_addr_ts);
 +        hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
 +        more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
 +
 +        if (more_data < 0) {
 +            s->tab_slice_address[ctb_addr_rs] = -1;
 +            return more_data;
 +        }
 +
 +        ctb_addr_ts++;
 +
 +        ff_hevc_save_states(s, ctb_addr_ts);
 +        ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
 +        ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
 +
 +        if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
 +            avpriv_atomic_int_set(&s1->wpp_err,  1);
 +            ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
 +            return 0;
 +        }
 +
 +        if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
 +            ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
 +            ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
 +            return ctb_addr_ts;
 +        }
 +        ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
 +        x_ctb+=ctb_size;
 +
 +        if(x_ctb >= s->sps->width) {
 +            break;
 +        }
 +    }
 +    ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
 +
 +    return 0;
 +}
 +
 +static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
 +{
 +    HEVCLocalContext *lc = s->HEVClc;
 +    int *ret = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
 +    int *arg = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
 +    int offset;
 +    int startheader, cmpt = 0;
 +    int i, j, res = 0;
 +
 +
 +    if (!s->sList[1]) {
 +        ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
 +
 +
 +        for (i = 1; i < s->threads_number; i++) {
 +            s->sList[i] = av_malloc(sizeof(HEVCContext));
 +            memcpy(s->sList[i], s, sizeof(HEVCContext));
 +            s->HEVClcList[i] = av_mallocz(sizeof(HEVCLocalContext));
 +            s->sList[i]->HEVClc = s->HEVClcList[i];
 +        }
 +    }
 +
 +    offset = (lc->gb.index >> 3);
 +
 +    for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
 +        if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
 +            startheader--;
 +            cmpt++;
 +        }
 +    }
 +
 +    for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
 +        offset += (s->sh.entry_point_offset[i - 1] - cmpt);
 +        for (j = 0, cmpt = 0, startheader = offset
 +             + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
 +            if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
 +                startheader--;
 +                cmpt++;
 +            }
 +        }
 +        s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
 +        s->sh.offset[i - 1] = offset;
 +
 +    }
 +    if (s->sh.num_entry_point_offsets != 0) {
 +        offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
 +        s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
 +        s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
 +
 +    }
 +    s->data = nal;
 +
 +    for (i = 1; i < s->threads_number; i++) {
 +        s->sList[i]->HEVClc->first_qp_group = 1;
 +        s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
 +        memcpy(s->sList[i], s, sizeof(HEVCContext));
 +        s->sList[i]->HEVClc = s->HEVClcList[i];
 +    }
 +
 +    avpriv_atomic_int_set(&s->wpp_err, 0);
 +    ff_reset_entries(s->avctx);
 +
 +    for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
 +        arg[i] = i;
 +        ret[i] = 0;
 +    }
 +
 +    if (s->pps->entropy_coding_sync_enabled_flag)
 +        s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
 +
 +    for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
 +        res += ret[i];
 +    av_free(ret);
 +    av_free(arg);
 +    return res;
 +}
 +
  /**
   * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
   * 0 if the unit should be skipped, 1 otherwise
   */
  static int hls_nal_unit(HEVCContext *s)
  {
 -    GetBitContext *gb = &s->HEVClc.gb;
 +    GetBitContext *gb = &s->HEVClc->gb;
      int nuh_layer_id;
  
      if (get_bits1(gb) != 0)
          return AVERROR_INVALIDDATA;
  
      av_log(s->avctx, AV_LOG_DEBUG,
 -           "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
 +           "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
             s->nal_unit_type, nuh_layer_id, s->temporal_id);
  
      return nuh_layer_id == 0;
  }
  
 -static void restore_tqb_pixels(HEVCContext *s)
 -{
 -    int min_pu_size = 1 << s->sps->log2_min_pu_size;
 -    int x, y, c_idx;
 -
 -    for (c_idx = 0; c_idx < 3; c_idx++) {
 -        ptrdiff_t stride = s->frame->linesize[c_idx];
 -        int hshift       = s->sps->hshift[c_idx];
 -        int vshift       = s->sps->vshift[c_idx];
 -        for (y = 0; y < s->sps->min_pu_height; y++) {
 -            for (x = 0; x < s->sps->min_pu_width; x++) {
 -                if (s->is_pcm[y * s->sps->min_pu_width + x]) {
 -                    int n;
 -                    int len      = min_pu_size >> hshift;
 -                    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)];
 -                    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)];
 -                    for (n = 0; n < (min_pu_size >> vshift); n++) {
 -                        memcpy(dst, src, len);
 -                        src += stride;
 -                        dst += stride;
 -                    }
 -                }
 -            }
 -        }
 -    }
 -}
 -
  static int set_side_data(HEVCContext *s)
  {
      AVFrame *out = s->ref->frame;
  
  static int hevc_frame_start(HEVCContext *s)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
 +    int pic_size_in_ctb  = ((s->sps->width  >> s->sps->log2_min_cb_size) + 1) *
 +                           ((s->sps->height >> s->sps->log2_min_cb_size) + 1);
      int ret;
  
 -    memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
 -    memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
 +    memset(s->horizontal_bs, 0, s->bs_width * s->bs_height);
 +    memset(s->vertical_bs,   0, s->bs_width * s->bs_height);
      memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
 -    memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
 +    memset(s->is_pcm,        0, (s->sps->min_pu_width + 1) * (s->sps->min_pu_height + 1));
 +    memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
  
 -    lc->start_of_tiles_x = 0;
      s->is_decoded        = 0;
      s->first_nal_type    = s->nal_unit_type;
  
      if (s->pps->tiles_enabled_flag)
          lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
  
 -    ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
 -                              s->poc);
 +    ret = ff_hevc_set_new_ref(s, &s->frame, s->poc);
      if (ret < 0)
          goto fail;
  
      if (ret < 0)
          goto fail;
  
 +    s->frame->pict_type = 3 - s->sh.slice_type;
 +
 +    if (!IS_IRAP(s))
 +        ff_hevc_bump_frame(s);
 +
      av_frame_unref(s->output_frame);
      ret = ff_hevc_output_frame(s, s->output_frame, 0);
      if (ret < 0)
      return 0;
  
  fail:
 -    if (s->ref)
 +    if (s->ref && s->threads_type == FF_THREAD_FRAME)
          ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
      s->ref = NULL;
      return ret;
  
  static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      GetBitContext *gb    = &lc->gb;
      int ctb_addr_ts, ret;
  
              }
          }
  
 -        ctb_addr_ts = hls_slice_data(s);
 +        if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
 +            ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
 +        else
 +            ctb_addr_ts = hls_slice_data(s);
          if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
              s->is_decoded = 1;
 -            if ((s->pps->transquant_bypass_enable_flag ||
 -                 (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
 -                s->sps->sao_enabled)
 -                restore_tqb_pixels(s);
          }
  
          if (ctb_addr_ts < 0) {
@@@ -2732,13 -2615,12 +2718,13 @@@ fail
  
  /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
   * between these functions would be nice. */
 -static int extract_rbsp(const uint8_t *src, int length,
 -                        HEVCNAL *nal)
 +int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
 +                         HEVCNAL *nal)
  {
      int i, si, di;
      uint8_t *dst;
  
 +    s->skipped_bytes = 0;
  #define STARTCODE_TEST                                                  \
          if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
              if (src[i + 2] != 3) {                                      \
                  dst[di++] = 0;
                  si       += 3;
  
 +                s->skipped_bytes++;
 +                if (s->skipped_bytes_pos_size < s->skipped_bytes) {
 +                    s->skipped_bytes_pos_size *= 2;
 +                    av_reallocp_array(&s->skipped_bytes_pos,
 +                            s->skipped_bytes_pos_size,
 +                            sizeof(*s->skipped_bytes_pos));
 +                    if (!s->skipped_bytes_pos)
 +                        return AVERROR(ENOMEM);
 +                }
 +                if (s->skipped_bytes_pos)
 +                    s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
                  continue;
              } else // next start code
                  goto nsc;
@@@ -2844,7 -2715,6 +2830,7 @@@ static int decode_nal_units(HEVCContex
      int i, consumed, ret = 0;
  
      s->ref = NULL;
 +    s->last_eos = s->eos;
      s->eos = 0;
  
      /* split the input packet into NAL units, so we know the upper bound on the
                  goto fail;
              }
          } else {
 -            if (buf[2] == 0) {
 -                length--;
 -                buf++;
 -                continue;
 -            }
 -            if (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
 -                ret = AVERROR_INVALIDDATA;
 -                goto fail;
 +            /* search start code */
 +            while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
 +                ++buf;
 +                --length;
 +                if (length < 4) {
 +                    av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
 +                    ret = AVERROR_INVALIDDATA;
 +                    goto fail;
 +                }
              }
  
              buf           += 3;
              length        -= 3;
 -            extract_length = length;
          }
  
 +        if (!s->is_nalff)
 +            extract_length = length;
 +
          if (s->nals_allocated < s->nb_nals + 1) {
              int new_size = s->nals_allocated + 1;
              HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
              s->nals = tmp;
              memset(s->nals + s->nals_allocated, 0,
                     (new_size - s->nals_allocated) * sizeof(*tmp));
 +            av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
 +            av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
 +            av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
 +            s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
 +            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));
              s->nals_allocated = new_size;
          }
 -        nal = &s->nals[s->nb_nals++];
 +        s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
 +        s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
 +        nal = &s->nals[s->nb_nals];
 +
 +        consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
 +
 +        s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
 +        s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
 +        s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
 +
  
 -        consumed = extract_rbsp(buf, extract_length, nal);
          if (consumed < 0) {
              ret = consumed;
              goto fail;
          }
  
 -        ret = init_get_bits8(&s->HEVClc.gb, nal->data, nal->size);
 +        ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
          if (ret < 0)
              goto fail;
          hls_nal_unit(s);
  
      /* parse the NAL units */
      for (i = 0; i < s->nb_nals; i++) {
 -        int ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
 +        int ret;
 +        s->skipped_bytes = s->skipped_bytes_nal[i];
 +        s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
 +
 +        ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
          if (ret < 0) {
              av_log(s->avctx, AV_LOG_WARNING,
                     "Error parsing NAL unit #%d.\n", i);
      }
  
  fail:
 -    if (s->ref)
 +    if (s->ref && s->threads_type == FF_THREAD_FRAME)
          ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
  
      return ret;
@@@ -3071,9 -2921,7 +3057,9 @@@ static int hevc_decode_frame(AVCodecCon
  
  static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
  {
 -    int ret = ff_thread_ref_frame(&dst->tf, &src->tf);
 +    int ret;
 +
 +    ret = ff_thread_ref_frame(&dst->tf, &src->tf);
      if (ret < 0)
          return ret;
  
@@@ -3112,15 -2960,6 +3098,15 @@@ static av_cold int hevc_decode_free(AVC
  
      av_freep(&s->md5_ctx);
  
 +    for(i=0; i < s->nals_allocated; i++) {
 +        av_freep(&s->skipped_bytes_pos_nal[i]);
 +    }
 +    av_freep(&s->skipped_bytes_pos_size_nal);
 +    av_freep(&s->skipped_bytes_nal);
 +    av_freep(&s->skipped_bytes_pos_nal);
 +
 +    av_freep(&s->cabac_state);
 +
      av_frame_free(&s->tmp_frame);
      av_frame_free(&s->output_frame);
  
          av_buffer_unref(&s->sps_list[i]);
      for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
          av_buffer_unref(&s->pps_list[i]);
 +    s->sps = NULL;
 +    s->pps = NULL;
 +    s->vps = NULL;
 +
 +    av_buffer_unref(&s->current_sps);
 +
 +    av_freep(&s->sh.entry_point_offset);
 +    av_freep(&s->sh.offset);
 +    av_freep(&s->sh.size);
 +
 +    for (i = 1; i < s->threads_number; i++) {
 +        HEVCLocalContext *lc = s->HEVClcList[i];
 +        if (lc) {
 +            av_freep(&s->HEVClcList[i]);
 +            av_freep(&s->sList[i]);
 +        }
 +    }
 +    if (s->HEVClc == s->HEVClcList[0])
 +        s->HEVClc = NULL;
 +    av_freep(&s->HEVClcList[0]);
  
      for (i = 0; i < s->nals_allocated; i++)
          av_freep(&s->nals[i].rbsp_buffer);
@@@ -3171,16 -2990,6 +3157,16 @@@ static av_cold int hevc_init_context(AV
  
      s->avctx = avctx;
  
 +    s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
 +    if (!s->HEVClc)
 +        goto fail;
 +    s->HEVClcList[0] = s->HEVClc;
 +    s->sList[0] = s;
 +
 +    s->cabac_state = av_malloc(HEVC_CONTEXTS);
 +    if (!s->cabac_state)
 +        goto fail;
 +
      s->tmp_frame = av_frame_alloc();
      if (!s->tmp_frame)
          goto fail;
      ff_bswapdsp_init(&s->bdsp);
  
      s->context_initialized = 1;
 +    s->eos = 0;
  
      return 0;
  
@@@ -3236,8 -3044,6 +3222,8 @@@ static int hevc_update_thread_context(A
          }
      }
  
 +    if (s->sps != s0->sps)
 +        s->sps = NULL;
      for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
          av_buffer_unref(&s->vps_list[i]);
          if (s0->vps_list[i]) {
          }
      }
  
 +    av_buffer_unref(&s->current_sps);
 +    if (s0->current_sps) {
 +        s->current_sps = av_buffer_ref(s0->current_sps);
 +        if (!s->current_sps)
 +            return AVERROR(ENOMEM);
 +    }
 +
      if (s->sps != s0->sps)
 -        ret = set_sps(s, s0->sps);
 +        if ((ret = set_sps(s, s0->sps)) < 0)
 +            return ret;
  
      s->seq_decode = s0->seq_decode;
      s->seq_output = s0->seq_output;
      s->pocTid0    = s0->pocTid0;
      s->max_ra     = s0->max_ra;
 +    s->eos        = s0->eos;
  
      s->is_nalff        = s0->is_nalff;
      s->nal_length_size = s0->nal_length_size;
  
 +    s->threads_number      = s0->threads_number;
 +    s->threads_type        = s0->threads_type;
 +
      if (s0->eos) {
          s->seq_decode = (s->seq_decode + 1) & 0xff;
          s->max_ra = INT_MAX;
@@@ -3373,14 -3167,6 +3359,14 @@@ static av_cold int hevc_decode_init(AVC
      if (ret < 0)
          return ret;
  
 +    s->enable_parallel_tiles = 0;
 +    s->picture_struct = 0;
 +
 +    if(avctx->active_thread_type & FF_THREAD_SLICE)
 +        s->threads_number = avctx->thread_count;
 +    else
 +        s->threads_number = 1;
 +
      if (avctx->extradata_size > 0 && avctx->extradata) {
          ret = hevc_decode_extradata(s);
          if (ret < 0) {
          }
      }
  
 +    if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
 +            s->threads_type = FF_THREAD_FRAME;
 +        else
 +            s->threads_type = FF_THREAD_SLICE;
 +
      return 0;
  }
  
@@@ -3425,15 -3206,12 +3411,15 @@@ static const AVProfile profiles[] = 
      { FF_PROFILE_HEVC_MAIN,                 "Main"                },
      { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
      { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
 +    { FF_PROFILE_HEVC_REXT,                 "Rext"  },
      { FF_PROFILE_UNKNOWN },
  };
  
  static const AVOption options[] = {
      { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
          AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
 +    { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
 +        AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
      { NULL },
  };
  
@@@ -3458,6 -3236,6 +3444,6 @@@ AVCodec ff_hevc_decoder = 
      .update_thread_context = hevc_update_thread_context,
      .init_thread_copy      = hevc_init_thread_copy,
      .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
 -                             CODEC_CAP_FRAME_THREADS,
 +                             CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
      .profiles              = NULL_IF_CONFIG_SMALL(profiles),
  };