Merge commit 'a1c2b48018b09d2613f075ec0748c95bd520ac00'
authorMichael Niedermayer <michaelni@gmx.at>
Mon, 19 May 2014 12:10:13 +0000 (14:10 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Mon, 19 May 2014 12:10:13 +0000 (14:10 +0200)
* commit 'a1c2b48018b09d2613f075ec0748c95bd520ac00':
  hevc: templatize intra_pred

Conflicts:
libavcodec/hevc.h

Merged-by: Michael Niedermayer <michaelni@gmx.at>
1  2 
libavcodec/hevc.c
libavcodec/hevcpred.c
libavcodec/hevcpred.h
libavcodec/hevcpred_template.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/internal.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);
  }
@@@ -124,7 -231,7 +124,7 @@@ static int pic_arrays_init(HEVCContext 
          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)
@@@ -149,7 -256,7 +149,7 @@@ 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++) {
                  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 {
@@@ -347,22 -454,11 +347,22 @@@ fail
      return ret;
  }
  
 +static int is_sps_exist(HEVCContext *s, const HEVCSPS* last_sps)
 +{
 +    int i;
 +
 +    for( i = 0; i < MAX_SPS_COUNT; i++)
 +        if(s->sps_list[i])
 +            if (last_sps == (HEVCSPS*)s->sps_list[i]->data)
 +                return 1;
 +    return 0;
 +}
 +
  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);
      }
 -    if (s->nal_unit_type >= 16 && s->nal_unit_type <= 23)
 +    sh->no_output_of_prior_pics_flag = 0;
 +    if (IS_IRAP(s))
          sh->no_output_of_prior_pics_flag = get_bits1(gb);
 +    if (s->nal_unit_type == NAL_CRA_NUT && s->last_eos == 1)
 +        sh->no_output_of_prior_pics_flag = 1;
  
      sh->pps_id = get_ue_golomb_long(gb);
      if (sh->pps_id >= MAX_PPS_COUNT || !s->pps_list[sh->pps_id]) {
      s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
  
      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) {
 +            if (is_sps_exist(s, last_sps)) {
 +                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;
 +            } else
 +                sh->no_output_of_prior_pics_flag = 0;
 +        }
          ff_hevc_clear_refs(s);
          ret = set_sps(s, s->sps);
          if (ret < 0)
              return AVERROR_INVALIDDATA;
          }
  
 +        sh->pic_output_flag = 1;
          if (s->pps->output_flag_present_flag)
              sh->pic_output_flag = get_bits1(gb);
  
          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(sh->num_entry_point_offsets * sizeof(int));
 +            sh->offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
 +            sh->size = av_malloc(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) {
      }
  
      // 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;
 +    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;
  
@@@ -754,7 -811,7 +754,7 @@@ 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);
  #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;
 -
 -        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;
 -    }
 -    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);
 -    }
 -}
 -
  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 trafo_depth, int blk_idx)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
  
      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(s, x0, y0, log2_trafo_size, 0);
+         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(s, x0, y0, log2_trafo_size - 1, 1);
-             s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 2);
+             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(s, xBase, yBase, log2_trafo_size, 1);
-             s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 2);
+             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 (lc->tt.cbf_luma)
 -            hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
 +            ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
          if (log2_trafo_size > 2) {
              if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
 -                hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
 +                ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
              if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
 -                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 - 1, scan_idx_c, 2);
          } else if (blk_idx == 3) {
              if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
 +                ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
              if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
 +                ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
          }
      }
      return 0;
@@@ -931,7 -1337,7 +931,7 @@@ static int hls_transform_tree(HEVCConte
                                int log2_cb_size, int log2_trafo_size,
                                int trafo_depth, int blk_idx)
  {
 -    HEVCLocalContext *lc = &s->HEVClc;
 +    HEVCLocalContext *lc = s->HEVClc;
      uint8_t split_transform_flag;
      int ret;
  
                  }
          }
          if (!s->sh.disable_deblocking_filter_flag) {
 -            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size,
 -                                                  lc->slice_or_tiles_up_boundary,
 -                                                  lc->slice_or_tiles_left_boundary);
 +            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
              if (s->pps->transquant_bypass_enable_flag &&
                  lc->cu.cu_transquant_bypass_flag)
                  set_deblocking_bypass(s, x0, y0, 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;
      GetBitContext gb;
      int cb_size   = 1 << log2_cb_size;
      int stride0   = s->frame->linesize[0];
      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;
 -    const uint8_t *pcm = skip_bytes(&s->HEVClc.cc, (length + 7) >> 3);
 +    const uint8_t *pcm = skip_bytes(&s->HEVClc->cc, (length + 7) >> 3);
      int ret;
  
 -    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
 -                                          lc->slice_or_tiles_up_boundary,
 -                                          lc->slice_or_tiles_left_boundary);
 +    if (!s->sh.disable_deblocking_filter_flag)
 +        ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
  
      ret = init_get_bits(&gb, pcm, length);
      if (ret < 0)
      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);
  }
  
  /**
 - * 8.5.3.2.2.2 Chroma sample interpolation process
 + * 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;
 +    DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
 +    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](tmp, MAX_PB_SIZE, src0, src0stride,
 +                                                block_h, mx0, my0, block_w);
 +    if (!weight_flag)
 +        s->hevcdsp.put_hevc_qpel_bi[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
 +                                                       block_h, mx1, my1, block_w);
 +    else
 +        s->hevcdsp.put_hevc_qpel_bi_w[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, tmp, MAX_PB_SIZE,
 +                                                         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 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)
 +{
 +    DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
 +    int tmpstride = MAX_PB_SIZE;
 +    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](tmp, tmpstride, 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, tmp, tmpstride,
 +                                                       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, tmp, tmpstride,
 +                                                         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);
 +
 +    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,
  #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);
          } 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);
  
                      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);
 +                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,
                      lc->pu.mvd.x = 0;
                      lc->pu.mvd.y = 0;
                  } else {
 -                    hls_mvd_coding(s, x0, y0, 1);
 +                    ff_hevc_hls_mvd_coding(s, x0, y0, 1);
                  }
  
 -                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,
          }
      }
  
 -    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;
          }
      }
  
@@@ -1687,7 -2014,7 +1687,7 @@@ static av_always_inline void set_ct_dep
  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;
@@@ -1728,7 -2055,7 +1728,7 @@@ 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 x, y, ret;
 +    int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
  
      lc->cu.x                = x0;
      lc->cu.y                = y0;
      if (s->sh.slice_type != I_SLICE) {
          uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
  
 -        lc->cu.pred_mode = MODE_SKIP;
          x = y_cb * min_cb_width + x_cb;
          for (y = 0; y < length; y++) {
              memset(&s->skip_flag[x], skip_flag, length);
          intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
  
          if (!s->sh.disable_deblocking_filter_flag)
 -            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
 -                                                  lc->slice_or_tiles_up_boundary,
 -                                                  lc->slice_or_tiles_left_boundary);
 +            ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
      } else {
          if (s->sh.slice_type != I_SLICE)
              lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
                      return ret;
              } else {
                  if (!s->sh.disable_deblocking_filter_flag)
 -                    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
 -                                                          lc->slice_or_tiles_up_boundary,
 -                                                          lc->slice_or_tiles_left_boundary);
 +                    ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
              }
          }
      }
          x += min_cb_width;
      }
  
 +    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;
  
      lc->ct.depth = cb_depth;
      if (x0 + cb_size <= s->sps->width  &&
          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;
 +
 +        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;
  
 -        SUBDIVIDE(x0, y0);
 -        SUBDIVIDE(x1, y0);
 -        SUBDIVIDE(x0, y1);
 -        SUBDIVIDE(x1, y1);
 +        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->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
  
      if (s->pps->tiles_enabled_flag) {
 -        tile_left_boundary  = x_ctb > 0 &&
 -                              s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]];
 +        tile_left_boundary = x_ctb > 0 &&
 +                             s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
          slice_left_boundary = x_ctb > 0 &&
 -                              s->tab_slice_address[ctb_addr_rs] == s->tab_slice_address[ctb_addr_rs - 1];
 +                              s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
          tile_up_boundary  = y_ctb > 0 &&
 -                            s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
 +                            s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
          slice_up_boundary = y_ctb > 0 &&
 -                            s->tab_slice_address[ctb_addr_rs] == s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
 +                            s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
      } else {
 -        tile_left_boundary  =
 -        tile_up_boundary    = 1;
 -        slice_left_boundary = ctb_addr_in_slice > 0;
 -        slice_up_boundary   = ctb_addr_in_slice >= s->sps->ctb_width;
 -    }
 -    lc->slice_or_tiles_left_boundary = (!slice_left_boundary) + (!tile_left_boundary << 1);
 -    lc->slice_or_tiles_up_boundary   = (!slice_up_boundary + (!tile_up_boundary << 1));
 -    lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && tile_left_boundary);
 -    lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && tile_up_boundary);
 -    lc->ctb_up_right_flag = ((y_ctb > 0)  && (ctb_addr_in_slice+1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->sps->ctb_width]]));
 -    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]]));
 +        tile_left_boundary =
 +        tile_up_boundary   = 0;
 +        slice_left_boundary = ctb_addr_in_slice <= 0;
 +        slice_up_boundary   = ctb_addr_in_slice < s->sps->ctb_width;
 +    }
 +    lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
 +    lc->slice_or_tiles_up_boundary   = slice_up_boundary   + (tile_up_boundary   << 1);
 +    lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0)                  && !tile_left_boundary);
 +    lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
 +    lc->ctb_up_right_flag = ((y_ctb > 0)                 && (ctb_addr_in_slice+1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->sps->ctb_width]]));
 +    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);
      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);
 +            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((s->sh.num_entry_point_offsets + 1) * sizeof(int));
 +    int *arg = av_malloc((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_malloc(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)
@@@ -2342,18 -2461,15 +2342,18 @@@ static int set_side_data(HEVCContext *s
  
  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;
 +    AVFrame *cur_frame;
  
      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->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->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 (ret < 0)
          goto fail;
  
 +    cur_frame = s->sps->sao_enabled ? s->sao_frame : s->frame;
 +    cur_frame->pict_type = 3 - s->sh.slice_type;
 +
      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 ||
@@@ -2537,13 -2647,12 +2537,13 @@@ 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;
@@@ -2649,7 -2747,6 +2649,7 @@@ 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;
@@@ -2862,7 -2939,6 +2862,7 @@@ static int hevc_decode_frame(AVCodecCon
      s->is_md5 = 0;
  
      if (s->is_decoded) {
 +        s->ref->frame->key_frame = IS_IRAP(s);
          av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
          s->is_decoded = 0;
      }
  
  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;
  
@@@ -2912,22 -2986,12 +2912,22 @@@ fail
  static av_cold int hevc_decode_free(AVCodecContext *avctx)
  {
      HEVCContext       *s = avctx->priv_data;
 +    HEVCLocalContext *lc = s->HEVClc;
      int i;
  
      pic_arrays_free(s);
  
      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);
  
      for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
          av_buffer_unref(&s->pps_list[i]);
  
 +    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++) {
 +        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);
      av_freep(&s->nals);
@@@ -2973,16 -3022,6 +2973,16 @@@ 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_dsputil_init(&s->dsp, avctx);
  
      s->context_initialized = 1;
 +    s->eos = 0;
  
      return 0;
  
@@@ -3072,14 -3110,10 +3072,14 @@@ static int hevc_update_thread_context(A
      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;
@@@ -3165,14 -3199,6 +3165,14 @@@ 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;
  }
  
@@@ -3223,8 -3244,6 +3223,8 @@@ static const AVProfile profiles[] = 
  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 },
  };
  
@@@ -3249,6 -3268,6 +3249,6 @@@ 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),
  };
diff --combined libavcodec/hevcpred.c
@@@ -1,29 -1,27 +1,29 @@@
  /*
 - * HEVC video decoder
 + * HEVC video Decoder
   *
   * Copyright (C) 2012 - 2013 Guillaume Martres
   *
 - * 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 "hevc.h"
  
 +#include "hevcpred.h"
 +
  #define BIT_DEPTH 8
  #include "hevcpred_template.c"
  #undef BIT_DEPTH
@@@ -42,7 -40,10 +42,10 @@@ void ff_hevc_pred_init(HEVCPredContext 
  #define FUNC(a, depth) a ## _ ## depth
  
  #define HEVC_PRED(depth)                                \
-     hpc->intra_pred      = FUNC(intra_pred, depth);     \
+     hpc->intra_pred[0]   = FUNC(intra_pred_2, depth);   \
+     hpc->intra_pred[1]   = FUNC(intra_pred_3, depth);   \
+     hpc->intra_pred[2]   = FUNC(intra_pred_4, depth);   \
+     hpc->intra_pred[3]   = FUNC(intra_pred_5, depth);   \
      hpc->pred_planar[0]  = FUNC(pred_planar_0, depth);  \
      hpc->pred_planar[1]  = FUNC(pred_planar_1, depth);  \
      hpc->pred_planar[2]  = FUNC(pred_planar_2, depth);  \
diff --combined libavcodec/hevcpred.h
index 4dead2e,0000000..7f14a76
mode 100644,000000..100644
--- /dev/null
@@@ -1,43 -1,0 +1,45 @@@
-     void (*intra_pred)(struct HEVCContext *s, int x0, int y0, int log2_size, int c_idx);
 +/*
 + * HEVC video Decoder
 + *
 + * Copyright (C) 2012 - 2013 Guillaume Martres
 + *
 + * This file is part of FFmpeg.
 + *
 + * 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.
 + *
 + * 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 FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#ifndef AVCODEC_HEVCPRED_H
 +#define AVCODEC_HEVCPRED_H
 +
 +#include <stddef.h>
 +#include <stdint.h>
 +
 +struct HEVCContext;
 +
 +typedef struct HEVCPredContext {
-     void(*pred_planar[4])(uint8_t *src, const uint8_t *top, const uint8_t *left, ptrdiff_t stride);
-     void(*pred_dc)(uint8_t *src, const uint8_t *top, const uint8_t *left, ptrdiff_t stride,
-                    int log2_size, int c_idx);
-     void(*pred_angular[4])(uint8_t *src, const uint8_t *top, const uint8_t *left, ptrdiff_t stride,
-                          int c_idx, int mode);
++    void (*intra_pred[4])(struct HEVCContext *s, int x0, int y0, int c_idx);
 +
++    void (*pred_planar[4])(uint8_t *src, const uint8_t *top,
++                           const uint8_t *left, ptrdiff_t stride);
++    void (*pred_dc)(uint8_t *src, const uint8_t *top, const uint8_t *left,
++                    ptrdiff_t stride, int log2_size, int c_idx);
++    void (*pred_angular[4])(uint8_t *src, const uint8_t *top,
++                            const uint8_t *left, ptrdiff_t stride,
++                            int c_idx, int mode);
 +} HEVCPredContext;
 +
 +void ff_hevc_pred_init(HEVCPredContext *hpc, int bit_depth);
 +
 +#endif /* AVCODEC_HEVCPRED_H */
@@@ -3,31 -3,33 +3,32 @@@
   *
   * Copyright (C) 2012 - 2013 Guillaume Martres
   *
 - * 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/pixdesc.h"
  
 -#include "hevc.h"
 -
  #include "bit_depth_template.c"
 +#include "hevcpred.h"
  
  #define POS(x, y) src[(x) + stride * (y)]
  
- static void FUNC(intra_pred)(HEVCContext *s, int x0, int y0, int log2_size, int c_idx)
+ static av_always_inline void FUNC(intra_pred)(HEVCContext *s, int x0, int y0,
+                                               int log2_size, int c_idx)
  {
  #define PU(x) \
      ((x) >> s->sps->log2_min_pu_size)
  #define MVF_PU(x, y) \
      MVF(PU(x0 + ((x) << hshift)), PU(y0 + ((y) << vshift)))
  #define IS_INTRA(x, y) \
 -    MVF_PU(x, y).is_intra
 +    (MVF_PU(x, y).pred_flag == PF_INTRA)
  #define MIN_TB_ADDR_ZS(x, y) \
      s->pps->min_tb_addr_zs[(y) * s->sps->min_tb_width + (x)]
 -#define EXTEND_LEFT(ptr, start, length) \
 -        for (i = (start); i > (start) - (length); i--) \
 -            ptr[i - 1] = ptr[i]
 -#define EXTEND_RIGHT(ptr, start, length) \
 -        for (i = (start); i < (start) + (length); i++) \
 -            ptr[i] = ptr[i - 1]
 -#define EXTEND_UP(ptr, start, length)   EXTEND_LEFT(ptr, start, length)
 -#define EXTEND_DOWN(ptr, start, length) EXTEND_RIGHT(ptr, start, length)
 +#define EXTEND(ptr, start, length)                                             \
 +        for (i = start; i < (start) + (length); i += 4)                        \
 +            AV_WN4P(&(ptr[i]), a)
 +#define EXTEND_RIGHT_CIP(ptr, start, length)                                   \
 +        for (i = start; i < (start) + (length); i += 4)                        \
 +            if (!IS_INTRA(i, -1))                                              \
 +                AV_WN4P(&ptr[i], a);                                           \
 +            else                                                               \
 +                a = PIXEL_SPLAT_X4(ptr[i+3])
  #define EXTEND_LEFT_CIP(ptr, start, length) \
 -        for (i = (start); i > (start) - (length); i--) \
 +        for (i = start; i > (start) - (length); i--) \
              if (!IS_INTRA(i - 1, -1)) \
                  ptr[i - 1] = ptr[i]
 -#define EXTEND_RIGHT_CIP(ptr, start, length) \
 -        for (i = (start); i < (start) + (length); i++) \
 -            if (!IS_INTRA(i, -1)) \
 -                ptr[i] = ptr[i - 1]
 -#define EXTEND_UP_CIP(ptr, start, length) \
 -        for (i = (start); i > (start) - (length); i--) \
 -            if (!IS_INTRA(-1, i - 1)) \
 -                ptr[i - 1] = ptr[i]
 -#define EXTEND_UP_CIP_0(ptr, start, length) \
 -        for (i = (start); i > (start) - (length); i--) \
 -            ptr[i - 1] = ptr[i]
 -#define EXTEND_DOWN_CIP(ptr, start, length) \
 -        for (i = (start); i < (start) + (length); i++) \
 -            if (!IS_INTRA(-1, i)) \
 -                ptr[i] = ptr[i - 1]
 -    HEVCLocalContext *lc = &s->HEVClc;
 +#define EXTEND_UP_CIP(ptr, start, length)                                      \
 +        for (i = (start); i > (start) - (length); i -= 4)                      \
 +            if (!IS_INTRA(-1, i - 3))                                          \
 +                AV_WN4P(&ptr[i - 3], a);                                       \
 +            else                                                               \
 +                a = PIXEL_SPLAT_X4(ptr[i - 3])
 +#define EXTEND_DOWN_CIP(ptr, start, length)                                    \
 +        for (i = start; i < (start) + (length); i += 4)                        \
 +            if (!IS_INTRA(-1, i))                                              \
 +                AV_WN4P(&ptr[i], a);                                           \
 +            else                                                               \
 +                a = PIXEL_SPLAT_X4(ptr[i + 3])
 +
 +    HEVCLocalContext *lc = s->HEVClc;
      int i;
      int hshift = s->sps->hshift[c_idx];
      int vshift = s->sps->vshift[c_idx];
      int size = (1 << log2_size);
 -    int size_in_luma = size << hshift;
 -    int size_in_tbs = size_in_luma >> s->sps->log2_min_tb_size;
 +    int size_in_luma_h = size << hshift;
 +    int size_in_tbs_h  = size_in_luma_h >> s->sps->log2_min_tb_size;
 +    int size_in_luma_v = size << vshift;
 +    int size_in_tbs_v  = size_in_luma_v >> s->sps->log2_min_tb_size;
      int x = x0 >> hshift;
      int y = y0 >> vshift;
      int x_tb = x0 >> s->sps->log2_min_tb_size;
  
      enum IntraPredMode mode = c_idx ? lc->pu.intra_pred_mode_c :
                                lc->tu.cur_intra_pred_mode;
 +    pixel4 a;
 +    pixel  left_array[2 * MAX_TB_SIZE + 1];
 +    pixel  filtered_left_array[2 * MAX_TB_SIZE + 1];
 +    pixel  top_array[2 * MAX_TB_SIZE + 1];
 +    pixel  filtered_top_array[2 * MAX_TB_SIZE + 1];
  
 -    pixel left_array[2 * MAX_TB_SIZE + 1];
 -    pixel filtered_left_array[2 * MAX_TB_SIZE + 1];
 -    pixel top_array[2 * MAX_TB_SIZE + 1];
 -    pixel filtered_top_array[2 * MAX_TB_SIZE + 1];
 -
 -    pixel *left          = left_array + 1;
 -    pixel *top           = top_array  + 1;
 -    pixel *filtered_left = filtered_left_array + 1;
 -    pixel *filtered_top  = filtered_top_array  + 1;
 +    pixel  *left          = left_array + 1;
 +    pixel  *top           = top_array  + 1;
 +    pixel  *filtered_left = filtered_left_array + 1;
 +    pixel  *filtered_top  = filtered_top_array  + 1;
  
 -    int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs);
 +    int cand_bottom_left = lc->na.cand_bottom_left && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb - 1, y_tb + size_in_tbs_v);
      int cand_left        = lc->na.cand_left;
      int cand_up_left     = lc->na.cand_up_left;
      int cand_up          = lc->na.cand_up;
 -    int cand_up_right    = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs, y_tb - 1);
 +    int cand_up_right    = lc->na.cand_up_right && cur_tb_addr > MIN_TB_ADDR_ZS(x_tb + size_in_tbs_h, y_tb - 1);
  
 -    int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma, s->sps->height) -
 -                            (y0 + size_in_luma)) >> vshift;
 -    int top_right_size   = (FFMIN(x0 + 2 * size_in_luma, s->sps->width) -
 -                            (x0 + size_in_luma)) >> hshift;
 +    int bottom_left_size = (FFMIN(y0 + 2 * size_in_luma_v, s->sps->height) -
 +                           (y0 + size_in_luma_v)) >> vshift;
 +    int top_right_size   = (FFMIN(x0 + 2 * size_in_luma_h, s->sps->width) -
 +                           (x0 + size_in_luma_h)) >> hshift;
  
      if (s->pps->constrained_intra_pred_flag == 1) {
 -        int size_in_luma_pu = PU(size_in_luma);
 +        int size_in_luma_pu_v = PU(size_in_luma_v);
 +        int size_in_luma_pu_h = PU(size_in_luma_h);
          int on_pu_edge_x    = !(x0 & ((1 << s->sps->log2_min_pu_size) - 1));
          int on_pu_edge_y    = !(y0 & ((1 << s->sps->log2_min_pu_size) - 1));
 -        if (!size_in_luma_pu)
 -            size_in_luma_pu++;
 +        if (!size_in_luma_pu_h)
 +            size_in_luma_pu_h++;
          if (cand_bottom_left == 1 && on_pu_edge_x) {
              int x_left_pu   = PU(x0 - 1);
 -            int y_bottom_pu = PU(y0 + size_in_luma);
 -            int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_bottom_pu);
 +            int y_bottom_pu = PU(y0 + size_in_luma_v);
 +            int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_bottom_pu);
              cand_bottom_left = 0;
 -            for (i = 0; i < max; i++)
 -                cand_bottom_left |= MVF(x_left_pu, y_bottom_pu + i).is_intra;
 +            for (i = 0; i < max; i += 2)
 +                cand_bottom_left |= (MVF(x_left_pu, y_bottom_pu + i).pred_flag == PF_INTRA);
          }
          if (cand_left == 1 && on_pu_edge_x) {
              int x_left_pu   = PU(x0 - 1);
              int y_left_pu   = PU(y0);
 -            int max = FFMIN(size_in_luma_pu, s->sps->min_pu_height - y_left_pu);
 +            int max = FFMIN(size_in_luma_pu_v, s->sps->min_pu_height - y_left_pu);
              cand_left = 0;
 -            for (i = 0; i < max; i++)
 -                cand_left |= MVF(x_left_pu, y_left_pu + i).is_intra;
 +            for (i = 0; i < max; i += 2)
 +                cand_left |= (MVF(x_left_pu, y_left_pu + i).pred_flag == PF_INTRA);
          }
          if (cand_up_left == 1) {
              int x_left_pu   = PU(x0 - 1);
              int y_top_pu    = PU(y0 - 1);
 -            cand_up_left = MVF(x_left_pu, y_top_pu).is_intra;
 +            cand_up_left = MVF(x_left_pu, y_top_pu).pred_flag == PF_INTRA;
          }
          if (cand_up == 1 && on_pu_edge_y) {
              int x_top_pu    = PU(x0);
              int y_top_pu    = PU(y0 - 1);
 -            int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_top_pu);
 +            int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_top_pu);
              cand_up = 0;
 -            for (i = 0; i < max; i++)
 -                cand_up |= MVF(x_top_pu + i, y_top_pu).is_intra;
 +            for (i = 0; i < max; i += 2)
 +                cand_up |= (MVF(x_top_pu + i, y_top_pu).pred_flag == PF_INTRA);
          }
          if (cand_up_right == 1 && on_pu_edge_y) {
              int y_top_pu    = PU(y0 - 1);
 -            int x_right_pu  = PU(x0 + size_in_luma);
 -            int max = FFMIN(size_in_luma_pu, s->sps->min_pu_width - x_right_pu);
 +            int x_right_pu  = PU(x0 + size_in_luma_h);
 +            int max = FFMIN(size_in_luma_pu_h, s->sps->min_pu_width - x_right_pu);
              cand_up_right = 0;
 -            for (i = 0; i < max; i++)
 -                cand_up_right |= MVF(x_right_pu + i, y_top_pu).is_intra;
 -        }
 -        for (i = 0; i < 2 * MAX_TB_SIZE; i++) {
 -            left[i] = 128;
 -            top[i]  = 128;
 +            for (i = 0; i < max; i += 2)
 +                cand_up_right |= (MVF(x_right_pu + i, y_top_pu).pred_flag == PF_INTRA);
          }
 +        memset(left, 128, 2 * MAX_TB_SIZE*sizeof(pixel));
 +        memset(top , 128, 2 * MAX_TB_SIZE*sizeof(pixel));
 +        top[-1] = 128;
      }
 -    if (cand_bottom_left) {
 -        for (i = size + bottom_left_size; i < (size << 1); i++)
 -            if (IS_INTRA(-1, size + bottom_left_size - 1) ||
 -                !s->pps->constrained_intra_pred_flag)
 -                left[i] = POS(-1, size + bottom_left_size - 1);
 -        for (i = size + bottom_left_size - 1; i >= size; i--)
 -            if (IS_INTRA(-1, i) || !s->pps->constrained_intra_pred_flag)
 -                left[i] = POS(-1, i);
 +    if (cand_up_left) {
 +        left[-1] = POS(-1, -1);
 +        top[-1]  = left[-1];
      }
 -    if (cand_left)
 -        for (i = size - 1; i >= 0; i--)
 -            if (IS_INTRA(-1, i) || !s->pps->constrained_intra_pred_flag)
 -                left[i] = POS(-1, i);
 -    if (cand_up_left)
 -        if (IS_INTRA(-1, -1) || !s->pps->constrained_intra_pred_flag) {
 -            left[-1] = POS(-1, -1);
 -            top[-1]  = left[-1];
 -        }
      if (cand_up)
 -        for (i = size - 1; i >= 0; i--)
 -            if (IS_INTRA(i, -1) || !s->pps->constrained_intra_pred_flag)
 -                top[i] = POS(i, -1);
 +        for (i = 0; i <size; i+=4)
 +            AV_WN4P(&top[i], AV_RN4P(&POS(i, -1)));
 +
      if (cand_up_right) {
 -        for (i = size + top_right_size; i < (size << 1); i++)
 -            if (IS_INTRA(size + top_right_size - 1, -1) ||
 -                !s->pps->constrained_intra_pred_flag)
 -                top[i] = POS(size + top_right_size - 1, -1);
 -        for (i = size + top_right_size - 1; i >= size; i--)
 -            if (IS_INTRA(i, -1) || !s->pps->constrained_intra_pred_flag)
 -                top[i] = POS(i, -1);
 +        a = PIXEL_SPLAT_X4(POS(size + top_right_size - 1, -1));
 +        for (i = size + top_right_size; i < (size << 1); i += 4)
 +            AV_WN4P(&top[i], a);
 +        for (i = size ; i < size+top_right_size; i+=4)
 +            AV_WN4P(&top[i], AV_RN4P(&POS(i, -1)));
 +    }
 +    if (cand_left)
 +        for (i = 0; i < size; i++)
 +            left[i] = POS(-1, i);
 +    if (cand_bottom_left) {
 +        for (i = size ; i < size+bottom_left_size; i++)
 +            left[i] = POS(-1, i);
 +        a = PIXEL_SPLAT_X4(POS(-1, size + bottom_left_size - 1));
 +        for (i = size + bottom_left_size; i < (size << 1); i+=4)
 +            AV_WN4P(&left[i], a);
      }
  
      if (s->pps->constrained_intra_pred_flag == 1) {
                  left[-1] = top[-1];
                  j        = 0;
              }
 +            left[-1] = top[-1];
              if (cand_bottom_left || cand_left) {
 -                EXTEND_DOWN_CIP(left, j, size_max_y - j);
 +                a = PIXEL_SPLAT_X4(left[-1]);
 +                EXTEND_DOWN_CIP(left, 0, size_max_y);
              }
              if (!cand_left) {
 -                EXTEND_DOWN(left, 0, size);
 +                a = PIXEL_SPLAT_X4(left[-1]);
 +                EXTEND(left, 0, size);
              }
              if (!cand_bottom_left) {
 -                EXTEND_DOWN(left, size, size);
 +                a = PIXEL_SPLAT_X4(left[size - 1]);
 +                EXTEND(left, size, size);
              }
              if (x0 != 0 && y0 != 0) {
 +                a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
                  EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);
 +                if (!IS_INTRA(-1, - 1))
 +                    left[-1] = left[0];
              } else if (x0 == 0) {
 -                EXTEND_UP_CIP_0(left, size_max_y - 1, size_max_y);
 +                a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
 +                EXTEND(left, 0, size_max_y);
              } else {
 -                EXTEND_UP_CIP(left, size_max_y - 1, size_max_y - 1);
 +                a = PIXEL_SPLAT_X4(left[size_max_y - 1]);
 +                EXTEND_UP_CIP(left, size_max_y - 1, size_max_y);
              }
              top[-1] = left[-1];
              if (y0 != 0) {
 +                a = PIXEL_SPLAT_X4(left[-1]);
                  EXTEND_RIGHT_CIP(top, 0, size_max_x);
              }
          }
      // Infer the unavailable samples
      if (!cand_bottom_left) {
          if (cand_left) {
 -            EXTEND_DOWN(left, size, size);
 +            a = PIXEL_SPLAT_X4(left[size-1]);
 +            EXTEND(left, size, size);
          } else if (cand_up_left) {
 -            EXTEND_DOWN(left, 0, 2 * size);
 +            a = PIXEL_SPLAT_X4(left[-1]);
 +            EXTEND(left, 0, 2 * size);
              cand_left = 1;
          } else if (cand_up) {
              left[-1] = top[0];
 -            EXTEND_DOWN(left, 0, 2 * size);
 +            a = PIXEL_SPLAT_X4(left[-1]);
 +            EXTEND(left, 0, 2 * size);
              cand_up_left = 1;
              cand_left    = 1;
          } else if (cand_up_right) {
 -            EXTEND_LEFT(top, size, size);
 -            left[-1] = top[0];
 -            EXTEND_DOWN(left, 0, 2 * size);
 +            left[-1] = top[size];
 +            a = PIXEL_SPLAT_X4(left[-1]);
 +            EXTEND(top, 0, size);
 +            EXTEND(left, 0, 2 * size);
              cand_up      = 1;
              cand_up_left = 1;
              cand_left    = 1;
          } else { // No samples available
 -            top[0] = left[-1] = (1 << (BIT_DEPTH - 1));
 -            EXTEND_RIGHT(top, 1, 2 * size - 1);
 -            EXTEND_DOWN(left, 0, 2 * size);
 +            left[-1] = (1 << (BIT_DEPTH - 1));
 +            a = PIXEL_SPLAT_X4(left[-1]);
 +            EXTEND(top, 0, 2 * size);
 +            EXTEND(left, 0, 2 * size);
          }
      }
  
      if (!cand_left) {
 -        EXTEND_UP(left, size, size);
 +        a = PIXEL_SPLAT_X4(left[size]);
 +        EXTEND(left, 0, size);
      }
      if (!cand_up_left) {
          left[-1] = left[0];
      }
      if (!cand_up) {
 -        top[0] = left[-1];
 -        EXTEND_RIGHT(top, 1, size - 1);
 +        a = PIXEL_SPLAT_X4(left[-1]);
 +        EXTEND(top, 0, size);
      }
      if (!cand_up_right) {
 -        EXTEND_RIGHT(top, size, size);
 +        a = PIXEL_SPLAT_X4(top[size-1]);
 +        EXTEND(top, size, size);
      }
  
      top[-1] = left[-1];
  
      // Filtering process
 -    if (c_idx == 0 && mode != INTRA_DC && size != 4) {
 -        int intra_hor_ver_dist_thresh[] = { 7, 1, 0 };
 -        int min_dist_vert_hor = FFMIN(FFABS((int)mode - 26),
 -                                      FFABS((int)mode - 10));
 -        if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) {
 -            int threshold = 1 << (BIT_DEPTH - 5);
 -            if (s->sps->sps_strong_intra_smoothing_enable_flag &&
 -                log2_size == 5 &&
 -                FFABS(top[-1]  + top[63]  - 2 * top[31])  < threshold &&
 -                FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) {
 -                // We can't just overwrite values in top because it could be
 -                // a pointer into src
 -                filtered_top[-1] = top[-1];
 -                filtered_top[63] = top[63];
 -                for (i = 0; i < 63; i++)
 -                    filtered_top[i] = ((64 - (i + 1)) * top[-1] +
 -                                             (i + 1)  * top[63] + 32) >> 6;
 -                for (i = 0; i < 63; i++)
 -                    left[i] = ((64 - (i + 1)) * left[-1] +
 -                                     (i + 1)  * left[63] + 32) >> 6;
 -                top = filtered_top;
 -            } else {
 -                filtered_left[2 * size - 1] = left[2 * size - 1];
 -                filtered_top[2 * size - 1]  = top[2 * size - 1];
 -                for (i = 2 * size - 2; i >= 0; i--)
 -                    filtered_left[i] = (left[i + 1] + 2 * left[i] +
 -                                        left[i - 1] + 2) >> 2;
 -                filtered_top[-1]  =
 -                filtered_left[-1] = (left[0] + 2 * left[-1] + top[0] + 2) >> 2;
 -                for (i = 2 * size - 2; i >= 0; i--)
 -                    filtered_top[i] = (top[i + 1] + 2 * top[i] +
 -                                       top[i - 1] + 2) >> 2;
 -                left = filtered_left;
 -                top  = filtered_top;
 +    if (c_idx == 0) {
 +        if (mode != INTRA_DC && size != 4){
 +            int intra_hor_ver_dist_thresh[] = { 7, 1, 0 };
 +            int min_dist_vert_hor = FFMIN(FFABS((int)(mode - 26U)),
 +                                          FFABS((int)(mode - 10U)));
 +            if (min_dist_vert_hor > intra_hor_ver_dist_thresh[log2_size - 3]) {
 +                int threshold = 1 << (BIT_DEPTH - 5);
 +                if (s->sps->sps_strong_intra_smoothing_enable_flag && c_idx == 0 &&
 +                    log2_size == 5 &&
 +                    FFABS(top[-1]  + top[63]  - 2 * top[31])  < threshold &&
 +                    FFABS(left[-1] + left[63] - 2 * left[31]) < threshold) {
 +                    // We can't just overwrite values in top because it could be
 +                    // a pointer into src
 +                    filtered_top[-1] = top[-1];
 +                    filtered_top[63] = top[63];
 +                    for (i = 0; i < 63; i++)
 +                        filtered_top[i] = ((64 - (i + 1)) * top[-1] +
 +                                           (i + 1)  * top[63] + 32) >> 6;
 +                    for (i = 0; i < 63; i++)
 +                        left[i] = ((64 - (i + 1)) * left[-1] +
 +                                   (i + 1)  * left[63] + 32) >> 6;
 +                    top = filtered_top;
 +                } else {
 +                    filtered_left[2 * size - 1] = left[2 * size - 1];
 +                    filtered_top[2 * size - 1]  = top[2 * size - 1];
 +                    for (i = 2 * size - 2; i >= 0; i--)
 +                        filtered_left[i] = (left[i + 1] + 2 * left[i] +
 +                                            left[i - 1] + 2) >> 2;
 +                    filtered_top[-1]  =
 +                    filtered_left[-1] = (left[0] + 2 * left[-1] + top[0] + 2) >> 2;
 +                    for (i = 2 * size - 2; i >= 0; i--)
 +                        filtered_top[i] = (top[i + 1] + 2 * top[i] +
 +                                           top[i - 1] + 2) >> 2;
 +                    left = filtered_left;
 +                    top  = filtered_top;
 +                }
              }
          }
      }
      }
  }
  
+ #define INTRA_PRED(size)                                                            \
+ static void FUNC(intra_pred_ ## size)(HEVCContext *s, int x0, int y0, int c_idx)    \
+ {                                                                                   \
+     FUNC(intra_pred)(s, x0, y0, size, c_idx);                                       \
+ }
+ INTRA_PRED(2)
+ INTRA_PRED(3)
+ INTRA_PRED(4)
+ INTRA_PRED(5)
+ #undef INTRA_PRED
  static av_always_inline void FUNC(pred_planar)(uint8_t *_src, const uint8_t *_top,
                                    const uint8_t *_left, ptrdiff_t stride,
                                    int trafo_size)
@@@ -410,8 -411,8 +424,8 @@@ static void FUNC(pred_dc)(uint8_t *_src
      a = PIXEL_SPLAT_X4(dc);
  
      for (i = 0; i < size; i++)
 -        for (j = 0; j < size / 4; j++)
 -            AV_WN4PA(&POS(j * 4, i), a);
 +        for (j = 0; j < size; j+=4)
 +            AV_WN4P(&POS(j, i), a);
  
      if (c_idx == 0 && size < 32) {
          POS(0, 0) = (left[0] + 2 * dc + top[0] + 2) >> 2;
@@@ -443,7 -444,7 +457,7 @@@ static av_always_inline void FUNC(pred_
      };
  
      int angle = intra_pred_angle[mode - 2];
 -    pixel ref_array[3 * MAX_TB_SIZE + 1];
 +    pixel ref_array[3 * MAX_TB_SIZE + 4];
      pixel *ref_tmp = ref_array + size;
      const pixel *ref;
      int last = (size * angle) >> 5;
      if (mode >= 18) {
          ref = top - 1;
          if (angle < 0 && last < -1) {
 -            for (x = 0; x <= size; x++)
 -                ref_tmp[x] = top[x - 1];
 +            for (x = 0; x <= size; x += 4)
 +                AV_WN4P(&ref_tmp[x], AV_RN4P(&top[x - 1]));
              for (x = last; x <= -1; x++)
                  ref_tmp[x] = left[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
              ref = ref_tmp;
              int idx  = ((y + 1) * angle) >> 5;
              int fact = ((y + 1) * angle) & 31;
              if (fact) {
 -                for (x = 0; x < size; x++) {
 -                    POS(x, y) = ((32 - fact) * ref[x + idx + 1] +
 -                                       fact  * ref[x + idx + 2] + 16) >> 5;
 +                for (x = 0; x < size; x += 4) {
 +                    POS(x    , y) = ((32 - fact) * ref[x + idx + 1] +
 +                                           fact  * ref[x + idx + 2] + 16) >> 5;
 +                    POS(x + 1, y) = ((32 - fact) * ref[x + 1 + idx + 1] +
 +                                           fact  * ref[x + 1 + idx + 2] + 16) >> 5;
 +                    POS(x + 2, y) = ((32 - fact) * ref[x + 2 + idx + 1] +
 +                                           fact  * ref[x + 2 + idx + 2] + 16) >> 5;
 +                    POS(x + 3, y) = ((32 - fact) * ref[x + 3 + idx + 1] +
 +                                           fact  * ref[x + 3 + idx + 2] + 16) >> 5;
                  }
              } else {
 -                for (x = 0; x < size; x++)
 -                    POS(x, y) = ref[x + idx + 1];
 +                for (x = 0; x < size; x += 4)
 +                    AV_WN4P(&POS(x, y), AV_RN4P(&ref[x + idx + 1]));
              }
          }
          if (mode == 26 && c_idx == 0 && size < 32) {
      } else {
          ref = left - 1;
          if (angle < 0 && last < -1) {
 -            for (x = 0; x <= size; x++)
 -                ref_tmp[x] = left[x - 1];
 +            for (x = 0; x <= size; x += 4)
 +                AV_WN4P(&ref_tmp[x], AV_RN4P(&left[x - 1]));
              for (x = last; x <= -1; x++)
                  ref_tmp[x] = top[-1 + ((x * inv_angle[mode - 11] + 128) >> 8)];
              ref = ref_tmp;
              }
          }
          if (mode == 10 && c_idx == 0 && size < 32) {
 -            for (x = 0; x < size; x++)
 -                POS(x, 0) = av_clip_pixel(left[0] + ((top[x] - top[-1]) >> 1));
 +            for (x = 0; x < size; x += 4) {
 +                POS(x,     0) = av_clip_pixel(left[0] + ((top[x    ] - top[-1]) >> 1));
 +                POS(x + 1, 0) = av_clip_pixel(left[0] + ((top[x + 1] - top[-1]) >> 1));
 +                POS(x + 2, 0) = av_clip_pixel(left[0] + ((top[x + 2] - top[-1]) >> 1));
 +                POS(x + 3, 0) = av_clip_pixel(left[0] + ((top[x + 3] - top[-1]) >> 1));
 +            }
          }
      }
  }
@@@ -551,6 -542,9 +565,6 @@@ static void FUNC(pred_angular_3)(uint8_
  #undef MVF_PU
  #undef MVF
  #undef PU
 -#undef EXTEND_LEFT
 -#undef EXTEND_RIGHT
 -#undef EXTEND_UP
 -#undef EXTEND_DOWN
 +#undef EXTEND
  #undef MIN_TB_ADDR_ZS
  #undef POS