Merge commit '4abe3b049d987420eb891f74a35af2cebbf52144'
authorClément Bœsch <u@pkh.me>
Thu, 23 Mar 2017 15:48:41 +0000 (16:48 +0100)
committerClément Bœsch <u@pkh.me>
Thu, 23 Mar 2017 15:48:41 +0000 (16:48 +0100)
* commit '4abe3b049d987420eb891f74a35af2cebbf52144':
  hevc: rename hevc.[ch] to hevcdec.[ch]

Merged-by: Clément Bœsch <u@pkh.me>
25 files changed:
1  2 
libavcodec/Makefile
libavcodec/dxva2_hevc.c
libavcodec/h2645_parse.c
libavcodec/hevc_cabac.c
libavcodec/hevc_data.c
libavcodec/hevc_filter.c
libavcodec/hevc_mp4toannexb_bsf.c
libavcodec/hevc_mvs.c
libavcodec/hevc_parse.c
libavcodec/hevc_parse.h
libavcodec/hevc_parser.c
libavcodec/hevc_ps.c
libavcodec/hevc_ps_enc.c
libavcodec/hevc_refs.c
libavcodec/hevc_sei.c
libavcodec/hevcdec.c
libavcodec/hevcdec.h
libavcodec/hevcdsp_template.c
libavcodec/hevcpred.c
libavcodec/qsvenc_hevc.c
libavcodec/vaapi_encode_h265.c
libavcodec/vaapi_hevc.c
libavcodec/vdpau_hevc.c
libavformat/hevc.c
libavformat/hevcdec.c

@@@ -331,16 -262,12 +331,16 @@@ OBJS-$(CONFIG_H264_OMX_ENCODER)        
  OBJS-$(CONFIG_H264_QSV_DECODER)        += qsvdec_h2645.o
  OBJS-$(CONFIG_H264_QSV_ENCODER)        += qsvenc_h264.o
  OBJS-$(CONFIG_H264_VAAPI_ENCODER)      += vaapi_encode_h264.o vaapi_encode_h26x.o
 +OBJS-$(CONFIG_H264_VIDEOTOOLBOX_ENCODER) += videotoolboxenc.o
  OBJS-$(CONFIG_HAP_DECODER)             += hapdec.o hap.o
  OBJS-$(CONFIG_HAP_ENCODER)             += hapenc.o hap.o
- OBJS-$(CONFIG_HEVC_DECODER)            += hevc.o hevc_mvs.o hevc_ps.o hevc_sei.o \
+ OBJS-$(CONFIG_HEVC_DECODER)            += hevcdec.o hevc_mvs.o hevc_ps.o hevc_sei.o \
                                            hevc_cabac.o hevc_refs.o hevcpred.o    \
                                            hevcdsp.o hevc_filter.o h2645_parse.o hevc_data.o
 +OBJS-$(CONFIG_HEVC_CUVID_DECODER)      += cuvid.o
 +OBJS-$(CONFIG_HEVC_MEDIACODEC_DECODER) += mediacodecdec.o hevc_parse.o
  OBJS-$(CONFIG_HEVC_NVENC_ENCODER)      += nvenc_hevc.o
 +OBJS-$(CONFIG_NVENC_HEVC_ENCODER)      += nvenc_hevc.o
  OBJS-$(CONFIG_HEVC_QSV_DECODER)        += qsvdec_h2645.o
  OBJS-$(CONFIG_HEVC_QSV_ENCODER)        += qsvenc_hevc.o hevc_ps_enc.o h2645_parse.o
  OBJS-$(CONFIG_HEVC_VAAPI_ENCODER)      += vaapi_encode_h265.o vaapi_encode_h26x.o
Simple merge
@@@ -26,7 -26,6 +26,7 @@@
  #include "libavutil/intreadwrite.h"
  #include "libavutil/mem.h"
  
- #include "hevc.h"
++#include "hevcdec.h"
  #include "h2645_parse.h"
  
  int ff_h2645_extract_rbsp(const uint8_t *src, int length,
Simple merge
Simple merge
  #include "libavutil/internal.h"
  
  #include "cabac_functions.h"
- #include "hevc.h"
+ #include "hevcdec.h"
  
 +#include "bit_depth_template.c"
 +
  #define LUMA 0
  #define CB 1
  #define CR 2
Simple merge
Simple merge
index cf04bc2,0000000..d6bf872
mode 100644,000000..100644
--- /dev/null
@@@ -1,134 -1,0 +1,134 @@@
- #include "hevc.h"
 +/*
 + * 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
 + */
 +
 +#include "bytestream.h"
 +#include "h2645_parse.h"
++#include "hevcdec.h"
 +#include "hevc_parse.h"
 +
 +static int hevc_decode_nal_units(const uint8_t *buf, int buf_size, HEVCParamSets *ps,
 +                                 int is_nalff, int nal_length_size, void *logctx)
 +{
 +    int i;
 +    int ret = 0;
 +    H2645Packet pkt = { 0 };
 +
 +    ret = ff_h2645_packet_split(&pkt, buf, buf_size, logctx, is_nalff, nal_length_size, AV_CODEC_ID_HEVC, 1);
 +    if (ret < 0) {
 +        goto done;
 +    }
 +
 +    for (i = 0; i < pkt.nb_nals; i++) {
 +        H2645NAL *nal = &pkt.nals[i];
 +
 +        /* ignore everything except parameter sets and VCL NALUs */
 +        switch (nal->type) {
 +        case NAL_VPS: ff_hevc_decode_nal_vps(&nal->gb, logctx, ps);    break;
 +        case NAL_SPS: ff_hevc_decode_nal_sps(&nal->gb, logctx, ps, 1); break;
 +        case NAL_PPS: ff_hevc_decode_nal_pps(&nal->gb, logctx, ps);    break;
 +        case NAL_TRAIL_R:
 +        case NAL_TRAIL_N:
 +        case NAL_TSA_N:
 +        case NAL_TSA_R:
 +        case NAL_STSA_N:
 +        case NAL_STSA_R:
 +        case NAL_BLA_W_LP:
 +        case NAL_BLA_W_RADL:
 +        case NAL_BLA_N_LP:
 +        case NAL_IDR_W_RADL:
 +        case NAL_IDR_N_LP:
 +        case NAL_CRA_NUT:
 +        case NAL_RADL_N:
 +        case NAL_RADL_R:
 +        case NAL_RASL_N:
 +        case NAL_RASL_R:
 +            av_log(logctx, AV_LOG_ERROR, "Invalid NAL unit: %d\n", nal->type);
 +            ret = AVERROR_INVALIDDATA;
 +            goto done;
 +            break;
 +        }
 +    }
 +
 +done:
 +    ff_h2645_packet_uninit(&pkt);
 +    return ret;
 +}
 +
 +int ff_hevc_decode_extradata(const uint8_t *data, int size, HEVCParamSets *ps,
 +                             int *is_nalff, int *nal_length_size,
 +                             int err_recognition, void *logctx)
 +{
 +    int ret = 0;
 +    GetByteContext gb;
 +
 +    bytestream2_init(&gb, data, size);
 +
 +    if (size > 3 && (data[0] || data[1] || data[2] > 1)) {
 +        /* It seems the extradata is encoded as hvcC format.
 +         * Temporarily, we support configurationVersion==0 until 14496-15 3rd
 +         * is finalized. When finalized, configurationVersion will be 1 and we
 +         * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
 +        int i, j, num_arrays, nal_len_size;
 +
 +        *is_nalff = 1;
 +
 +        bytestream2_skip(&gb, 21);
 +        nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
 +        num_arrays   = bytestream2_get_byte(&gb);
 +
 +        /* nal units in the hvcC always have length coded with 2 bytes,
 +         * so put a fake nal_length_size = 2 while parsing them */
 +        *nal_length_size = 2;
 +
 +        /* Decode nal units from hvcC. */
 +        for (i = 0; i < num_arrays; i++) {
 +            int type = bytestream2_get_byte(&gb) & 0x3f;
 +            int cnt  = bytestream2_get_be16(&gb);
 +
 +            for (j = 0; j < cnt; j++) {
 +                // +2 for the nal size field
 +                int nalsize = bytestream2_peek_be16(&gb) + 2;
 +                if (bytestream2_get_bytes_left(&gb) < nalsize) {
 +                    av_log(logctx, AV_LOG_ERROR,
 +                           "Invalid NAL unit size in extradata.\n");
 +                    return AVERROR_INVALIDDATA;
 +                }
 +
 +                ret = hevc_decode_nal_units(gb.buffer, nalsize, ps, *is_nalff, *nal_length_size, logctx);
 +                if (ret < 0) {
 +                    av_log(logctx, AV_LOG_ERROR,
 +                           "Decoding nal unit %d %d from hvcC failed\n",
 +                           type, i);
 +                    return ret;
 +                }
 +                bytestream2_skip(&gb, nalsize);
 +            }
 +        }
 +
 +        /* Now store right nal length size, that will be used to parse
 +         * all other nals */
 +        *nal_length_size = nal_len_size;
 +    } else {
 +        *is_nalff = 0;
 +        ret = hevc_decode_nal_units(data, size, ps, *is_nalff, *nal_length_size, logctx);
 +        if (ret < 0)
 +            return ret;
 +    }
 +
 +    return ret;
 +}
index 877356a,0000000..39bd23e
mode 100644,000000..100644
--- /dev/null
@@@ -1,33 -1,0 +1,33 @@@
- #include "hevc.h"
 +/*
 + * 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
 + */
 +
 +/**
 + * @file
 + * H.265 parser code
 + */
 +
 +#ifndef AVCODEC_HEVC_PARSE_H
 +#define AVCODEC_HEVC_PARSE_H
 +
++#include "hevcdec.h"
 +
 +int ff_hevc_decode_extradata(const uint8_t *data, int size, HEVCParamSets *ps,
 +                             int *is_nalff, int *nal_length_size,
 +                             int err_recognition, void *logctx);
 +
 +#endif /* AVCODEC_HEVC_PARSE_H */
Simple merge
@@@ -24,8 -24,9 +24,8 @@@
   */
  
  #include "libavutil/imgutils.h"
 -
  #include "golomb.h"
- #include "hevc.h"
+ #include "hevcdec.h"
  
  static const uint8_t default_scaling_list_intra[] = {
      16, 16, 16, 16, 17, 18, 21, 24,
Simple merge
Simple merge
Simple merge
index 0000000,2afd638..8893648
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,3077 +1,3422 @@@
 - * HEVC video decoder
+ /*
 - * This file is part of Libav.
++ * 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
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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
 -const uint8_t ff_hevc_qpel_extra_before[4] = { 0, 3, 3, 3 };
 -const uint8_t ff_hevc_qpel_extra_after[4]  = { 0, 4, 4, 4 };
 -const uint8_t ff_hevc_qpel_extra[4]        = { 0, 7, 7, 7 };
 -
 -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, },
 -};
 -
 -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, },
 -};
 -
 -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, },
 -};
++ * 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/attributes.h"
+ #include "libavutil/common.h"
+ #include "libavutil/display.h"
+ #include "libavutil/internal.h"
++#include "libavutil/mastering_display_metadata.h"
+ #include "libavutil/md5.h"
+ #include "libavutil/opt.h"
+ #include "libavutil/pixdesc.h"
+ #include "libavutil/stereo3d.h"
+ #include "bswapdsp.h"
+ #include "bytestream.h"
+ #include "cabac_functions.h"
+ #include "golomb.h"
+ #include "hevcdec.h"
+ #include "profiles.h"
 -    s->bs_width  = width  >> 3;
 -    s->bs_height = height >> 3;
++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
+  * specification (HLS stands for High Level Syntax).
+  */
+ /**
+  * Section 5.7
+  */
+ /* free everything allocated  by pic_arrays_init() */
+ static void pic_arrays_free(HEVCContext *s)
+ {
+     av_freep(&s->sao);
+     av_freep(&s->deblock);
+     av_freep(&s->skip_flag);
+     av_freep(&s->tab_ct_depth);
+     av_freep(&s->tab_ipm);
+     av_freep(&s->cbf_luma);
+     av_freep(&s->is_pcm);
+     av_freep(&s->qp_y_tab);
+     av_freep(&s->tab_slice_address);
+     av_freep(&s->filter_slice_edges);
+     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);
+ }
+ /* allocate arrays that depend on frame dimensions */
+ static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
+ {
+     int log2_min_cb_size = sps->log2_min_cb_size;
+     int width            = sps->width;
+     int height           = sps->height;
+     int pic_size_in_ctb  = ((width  >> log2_min_cb_size) + 1) *
+                            ((height >> log2_min_cb_size) + 1);
+     int ctb_count        = sps->ctb_width * sps->ctb_height;
+     int min_pu_size      = sps->min_pu_width * sps->min_pu_height;
 -    s->skip_flag    = av_malloc(pic_size_in_ctb);
 -    s->tab_ct_depth = av_malloc(sps->min_cb_height * sps->min_cb_width);
++    s->bs_width  = (width  >> 2) + 1;
++    s->bs_height = (height >> 2) + 1;
+     s->sao           = av_mallocz_array(ctb_count, sizeof(*s->sao));
+     s->deblock       = av_mallocz_array(ctb_count, sizeof(*s->deblock));
+     if (!s->sao || !s->deblock)
+         goto fail;
 -    s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
++    s->skip_flag    = av_malloc_array(sps->min_cb_height, sps->min_cb_width);
++    s->tab_ct_depth = av_malloc_array(sps->min_cb_height, sps->min_cb_width);
+     if (!s->skip_flag || !s->tab_ct_depth)
+         goto fail;
 -    s->is_pcm   = av_malloc(min_pu_size);
++    s->cbf_luma = av_malloc_array(sps->min_tb_width, sps->min_tb_height);
+     s->tab_ipm  = av_mallocz(min_pu_size);
 -    s->filter_slice_edges = av_malloc(ctb_count);
 -    s->tab_slice_address  = av_malloc(pic_size_in_ctb *
++    s->is_pcm   = av_malloc_array(sps->min_pu_width + 1, sps->min_pu_height + 1);
+     if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
+         goto fail;
 -    s->qp_y_tab           = av_malloc(pic_size_in_ctb *
++    s->filter_slice_edges = av_mallocz(ctb_count);
++    s->tab_slice_address  = av_malloc_array(pic_size_in_ctb,
+                                       sizeof(*s->tab_slice_address));
 -    s->horizontal_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
 -    s->vertical_bs   = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
++    s->qp_y_tab           = av_malloc_array(pic_size_in_ctb,
+                                       sizeof(*s->qp_y_tab));
+     if (!s->qp_y_tab || !s->filter_slice_edges || !s->tab_slice_address)
+         goto fail;
 -                                          av_buffer_alloc);
++    s->horizontal_bs = av_mallocz_array(s->bs_width, s->bs_height);
++    s->vertical_bs   = av_mallocz_array(s->bs_width, s->bs_height);
+     if (!s->horizontal_bs || !s->vertical_bs)
+         goto fail;
+     s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
 -    s->sh.luma_log2_weight_denom = av_clip(get_ue_golomb_long(gb), 0, 7);
++                                          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)
+         goto fail;
+     return 0;
+ fail:
+     pic_arrays_free(s);
+     return AVERROR(ENOMEM);
+ }
+ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
+ {
+     int i = 0;
+     int j = 0;
+     uint8_t luma_weight_l0_flag[16];
+     uint8_t chroma_weight_l0_flag[16];
+     uint8_t luma_weight_l1_flag[16];
+     uint8_t chroma_weight_l1_flag[16];
++    int luma_log2_weight_denom;
 -        s->sh.chroma_log2_weight_denom = av_clip(s->sh.luma_log2_weight_denom + delta, 0, 7);
++    luma_log2_weight_denom = get_ue_golomb_long(gb);
++    if (luma_log2_weight_denom < 0 || luma_log2_weight_denom > 7)
++        av_log(s->avctx, AV_LOG_ERROR, "luma_log2_weight_denom %d is invalid\n", luma_log2_weight_denom);
++    s->sh.luma_log2_weight_denom = av_clip_uintp2(luma_log2_weight_denom, 3);
+     if (s->ps.sps->chroma_format_idc != 0) {
+         int delta = get_se_golomb(gb);
 -    if (s->ps.sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
++        s->sh.chroma_log2_weight_denom = av_clip_uintp2(s->sh.luma_log2_weight_denom + delta, 3);
+     }
+     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
+         luma_weight_l0_flag[i] = get_bits1(gb);
+         if (!luma_weight_l0_flag[i]) {
+             s->sh.luma_weight_l0[i] = 1 << s->sh.luma_log2_weight_denom;
+             s->sh.luma_offset_l0[i] = 0;
+         }
+     }
 -    if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
++    if (s->ps.sps->chroma_format_idc != 0) {
+         for (i = 0; i < s->sh.nb_refs[L0]; i++)
+             chroma_weight_l0_flag[i] = get_bits1(gb);
+     } else {
+         for (i = 0; i < s->sh.nb_refs[L0]; i++)
+             chroma_weight_l0_flag[i] = 0;
+     }
+     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
+         if (luma_weight_l0_flag[i]) {
+             int delta_luma_weight_l0 = get_se_golomb(gb);
+             s->sh.luma_weight_l0[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l0;
+             s->sh.luma_offset_l0[i] = get_se_golomb(gb);
+         }
+         if (chroma_weight_l0_flag[i]) {
+             for (j = 0; j < 2; j++) {
+                 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((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
+                                                                                     >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
+             }
+         } else {
+             s->sh.chroma_weight_l0[i][0] = 1 << s->sh.chroma_log2_weight_denom;
+             s->sh.chroma_offset_l0[i][0] = 0;
+             s->sh.chroma_weight_l0[i][1] = 1 << s->sh.chroma_log2_weight_denom;
+             s->sh.chroma_offset_l0[i][1] = 0;
+         }
+     }
+     if (s->sh.slice_type == B_SLICE) {
+         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
+             luma_weight_l1_flag[i] = get_bits1(gb);
+             if (!luma_weight_l1_flag[i]) {
+                 s->sh.luma_weight_l1[i] = 1 << s->sh.luma_log2_weight_denom;
+                 s->sh.luma_offset_l1[i] = 0;
+             }
+         }
+         if (s->ps.sps->chroma_format_idc != 0) {
+             for (i = 0; i < s->sh.nb_refs[L1]; i++)
+                 chroma_weight_l1_flag[i] = get_bits1(gb);
+         } else {
+             for (i = 0; i < s->sh.nb_refs[L1]; i++)
+                 chroma_weight_l1_flag[i] = 0;
+         }
+         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
+             if (luma_weight_l1_flag[i]) {
+                 int delta_luma_weight_l1 = get_se_golomb(gb);
+                 s->sh.luma_weight_l1[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l1;
+                 s->sh.luma_offset_l1[i] = get_se_golomb(gb);
+             }
+             if (chroma_weight_l1_flag[i]) {
+                 for (j = 0; j < 2; j++) {
+                     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((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
+                                                                                         >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
+                 }
+             } else {
+                 s->sh.chroma_weight_l1[i][0] = 1 << s->sh.chroma_log2_weight_denom;
+                 s->sh.chroma_offset_l1[i][0] = 0;
+                 s->sh.chroma_weight_l1[i][1] = 1 << s->sh.chroma_log2_weight_denom;
+                 s->sh.chroma_offset_l1[i][1] = 0;
+             }
+         }
+     }
+ }
+ static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
+ {
+     const HEVCSPS *sps = s->ps.sps;
+     int max_poc_lsb    = 1 << sps->log2_max_poc_lsb;
+     int prev_delta_msb = 0;
+     unsigned int nb_sps = 0, nb_sh;
+     int i;
+     rps->nb_refs = 0;
+     if (!sps->long_term_ref_pics_present_flag)
+         return 0;
+     if (sps->num_long_term_ref_pics_sps > 0)
+         nb_sps = get_ue_golomb_long(gb);
+     nb_sh = get_ue_golomb_long(gb);
 -static int set_sps(HEVCContext *s, const HEVCSPS *sps)
++    if (nb_sh + (uint64_t)nb_sps > FF_ARRAY_ELEMS(rps->poc))
+         return AVERROR_INVALIDDATA;
+     rps->nb_refs = nb_sh + nb_sps;
+     for (i = 0; i < rps->nb_refs; i++) {
+         uint8_t delta_poc_msb_present;
+         if (i < nb_sps) {
+             uint8_t lt_idx_sps = 0;
+             if (sps->num_long_term_ref_pics_sps > 1)
+                 lt_idx_sps = get_bits(gb, av_ceil_log2(sps->num_long_term_ref_pics_sps));
+             rps->poc[i]  = sps->lt_ref_pic_poc_lsb_sps[lt_idx_sps];
+             rps->used[i] = sps->used_by_curr_pic_lt_sps_flag[lt_idx_sps];
+         } else {
+             rps->poc[i]  = get_bits(gb, sps->log2_max_poc_lsb);
+             rps->used[i] = get_bits1(gb);
+         }
+         delta_poc_msb_present = get_bits1(gb);
+         if (delta_poc_msb_present) {
+             int delta = get_ue_golomb_long(gb);
+             if (i && i != nb_sps)
+                 delta += prev_delta_msb;
+             rps->poc[i] += s->poc - delta * max_poc_lsb - s->sh.pic_order_cnt_lsb;
+             prev_delta_msb = delta;
+         }
+     }
+     return 0;
+ }
+ static void export_stream_params(AVCodecContext *avctx, const HEVCParamSets *ps,
+                                  const HEVCSPS *sps)
+ {
+     const HEVCVPS *vps = (const HEVCVPS*)ps->vps_list[sps->vps_id]->data;
+     unsigned int num = 0, den = 0;
+     avctx->pix_fmt             = sps->pix_fmt;
+     avctx->coded_width         = sps->width;
+     avctx->coded_height        = sps->height;
+     avctx->width               = sps->output_width;
+     avctx->height              = sps->output_height;
+     avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
+     avctx->profile             = sps->ptl.general_ptl.profile_idc;
+     avctx->level               = sps->ptl.general_ptl.level_idc;
+     ff_set_sar(avctx, sps->vui.sar);
+     if (sps->vui.video_signal_type_present_flag)
+         avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
+                                                             : AVCOL_RANGE_MPEG;
+     else
+         avctx->color_range = AVCOL_RANGE_MPEG;
+     if (sps->vui.colour_description_present_flag) {
+         avctx->color_primaries = sps->vui.colour_primaries;
+         avctx->color_trc       = sps->vui.transfer_characteristic;
+         avctx->colorspace      = sps->vui.matrix_coeffs;
+     } else {
+         avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
+         avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
+         avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
+     }
+     if (vps->vps_timing_info_present_flag) {
+         num = vps->vps_num_units_in_tick;
+         den = vps->vps_time_scale;
+     } else if (sps->vui.vui_timing_info_present_flag) {
+         num = sps->vui.vui_num_units_in_tick;
+         den = sps->vui.vui_time_scale;
+     }
+     if (num != 0 && den != 0)
+         av_reduce(&avctx->framerate.den, &avctx->framerate.num,
+                   num, den, 1 << 30);
+ }
 -    #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL)
++static int set_sps(HEVCContext *s, const HEVCSPS *sps, enum AVPixelFormat pix_fmt)
+ {
 -    int ret;
++    #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL + CONFIG_HEVC_VAAPI_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL)
+     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
 -    if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P ||
 -        sps->pix_fmt == AV_PIX_FMT_YUV420P10) {
++    int ret, i;
+     pic_arrays_free(s);
+     s->ps.sps = NULL;
+     s->ps.vps = NULL;
+     if (!sps)
+         return 0;
+     ret = pic_arrays_init(s, sps);
+     if (ret < 0)
+         goto fail;
+     export_stream_params(s->avctx, &s->ps, sps);
 -    }
 -    if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P) {
++    switch (sps->pix_fmt) {
++    case AV_PIX_FMT_YUV420P:
++    case AV_PIX_FMT_YUVJ420P:
+ #if CONFIG_HEVC_DXVA2_HWACCEL
+         *fmt++ = AV_PIX_FMT_DXVA2_VLD;
+ #endif
 -    *fmt++ = sps->pix_fmt;
 -    *fmt = AV_PIX_FMT_NONE;
+ #if CONFIG_HEVC_D3D11VA_HWACCEL
+         *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
+ #endif
++#if CONFIG_HEVC_VAAPI_HWACCEL
++        *fmt++ = AV_PIX_FMT_VAAPI;
++#endif
+ #if CONFIG_HEVC_VDPAU_HWACCEL
+         *fmt++ = AV_PIX_FMT_VDPAU;
+ #endif
++        break;
++    case AV_PIX_FMT_YUV420P10:
++#if CONFIG_HEVC_DXVA2_HWACCEL
++        *fmt++ = AV_PIX_FMT_DXVA2_VLD;
++#endif
++#if CONFIG_HEVC_D3D11VA_HWACCEL
++        *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
++#endif
++#if CONFIG_HEVC_VAAPI_HWACCEL
++        *fmt++ = AV_PIX_FMT_VAAPI;
++#endif
++        break;
+     }
 -    ret = ff_get_format(s->avctx, pix_fmts);
 -    if (ret < 0)
 -        goto fail;
 -    s->avctx->pix_fmt = ret;
++    if (pix_fmt == AV_PIX_FMT_NONE) {
++        *fmt++ = sps->pix_fmt;
++        *fmt = AV_PIX_FMT_NONE;
 -        av_frame_unref(s->tmp_frame);
 -        ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
 -        if (ret < 0)
 -            goto fail;
 -        s->frame = s->tmp_frame;
++        ret = ff_thread_get_format(s->avctx, pix_fmts);
++        if (ret < 0)
++            goto fail;
++        s->avctx->pix_fmt = ret;
++    }
++    else {
++        s->avctx->pix_fmt = pix_fmt;
++    }
+     ff_hevc_pred_init(&s->hpc,     sps->bit_depth);
+     ff_hevc_dsp_init (&s->hevcdsp, sps->bit_depth);
+     ff_videodsp_init (&s->vdsp,    sps->bit_depth);
++    for (i = 0; i < 3; i++) {
++        av_freep(&s->sao_pixel_buffer_h[i]);
++        av_freep(&s->sao_pixel_buffer_v[i]);
++    }
++
+     if (sps->sao_enabled && !s->avctx->hwaccel) {
 -    GetBitContext *gb = &s->HEVClc.gb;
++        int c_count = (sps->chroma_format_idc != 0) ? 3 : 1;
++        int c_idx;
++
++        for(c_idx = 0; c_idx < c_count; c_idx++) {
++            int w = sps->width >> sps->hshift[c_idx];
++            int h = sps->height >> sps->vshift[c_idx];
++            s->sao_pixel_buffer_h[c_idx] =
++                av_malloc((w * 2 * sps->ctb_height) <<
++                          sps->pixel_shift);
++            s->sao_pixel_buffer_v[c_idx] =
++                av_malloc((h * 2 * sps->ctb_width) <<
++                          sps->pixel_shift);
++        }
+     }
+     s->ps.sps = sps;
+     s->ps.vps = (HEVCVPS*) s->ps.vps_list[s->ps.sps->vps_id]->data;
+     return 0;
+ fail:
+     pic_arrays_free(s);
+     s->ps.sps = NULL;
+     return ret;
+ }
+ static int hls_slice_header(HEVCContext *s)
+ {
 -
++    GetBitContext *gb = &s->HEVClc->gb;
+     SliceHeader *sh   = &s->sh;
+     int i, ret;
+     // Coded parameters
+     sh->first_slice_in_pic_flag = get_bits1(gb);
+     if ((IS_IDR(s) || IS_BLA(s)) && sh->first_slice_in_pic_flag) {
+         s->seq_decode = (s->seq_decode + 1) & 0xff;
+         s->max_ra     = INT_MAX;
+         if (IS_IDR(s))
+             ff_hevc_clear_refs(s);
+     }
++    sh->no_output_of_prior_pics_flag = 0;
+     if (IS_IRAP(s))
+         sh->no_output_of_prior_pics_flag = get_bits1(gb);
+     sh->pps_id = get_ue_golomb_long(gb);
+     if (sh->pps_id >= MAX_PPS_COUNT || !s->ps.pps_list[sh->pps_id]) {
+         av_log(s->avctx, AV_LOG_ERROR, "PPS id out of range: %d\n", sh->pps_id);
+         return AVERROR_INVALIDDATA;
+     }
+     if (!sh->first_slice_in_pic_flag &&
+         s->ps.pps != (HEVCPPS*)s->ps.pps_list[sh->pps_id]->data) {
+         av_log(s->avctx, AV_LOG_ERROR, "PPS changed between slices.\n");
+         return AVERROR_INVALIDDATA;
+     }
+     s->ps.pps = (HEVCPPS*)s->ps.pps_list[sh->pps_id]->data;
++    if (s->nal_unit_type == NAL_CRA_NUT && s->last_eos == 1)
++        sh->no_output_of_prior_pics_flag = 1;
+     if (s->ps.sps != (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data) {
++        const HEVCSPS* last_sps = s->ps.sps;
+         s->ps.sps = (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data;
 -        ret = set_sps(s, s->ps.sps);
++        if (last_sps && IS_IRAP(s) && s->nal_unit_type != NAL_CRA_NUT) {
++            if (s->ps.sps->width !=  last_sps->width || s->ps.sps->height != last_sps->height ||
++                s->ps.sps->temporal_layer[s->ps.sps->max_sub_layers - 1].max_dec_pic_buffering !=
++                last_sps->temporal_layer[last_sps->max_sub_layers - 1].max_dec_pic_buffering)
++                sh->no_output_of_prior_pics_flag = 0;
++        }
+         ff_hevc_clear_refs(s);
 -        sh->slice_segment_addr = slice_address_length ? get_bits(gb, slice_address_length) : 0;
++        ret = set_sps(s, s->ps.sps, AV_PIX_FMT_NONE);
+         if (ret < 0)
+             return ret;
+         s->seq_decode = (s->seq_decode + 1) & 0xff;
+         s->max_ra     = INT_MAX;
+     }
+     sh->dependent_slice_segment_flag = 0;
+     if (!sh->first_slice_in_pic_flag) {
+         int slice_address_length;
+         if (s->ps.pps->dependent_slice_segments_enabled_flag)
+             sh->dependent_slice_segment_flag = get_bits1(gb);
+         slice_address_length = av_ceil_log2(s->ps.sps->ctb_width *
+                                             s->ps.sps->ctb_height);
 -            sh->slice_sample_adaptive_offset_flag[1] =
 -            sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
++        sh->slice_segment_addr = get_bitsz(gb, slice_address_length);
+         if (sh->slice_segment_addr >= s->ps.sps->ctb_width * s->ps.sps->ctb_height) {
+             av_log(s->avctx, AV_LOG_ERROR,
+                    "Invalid slice segment address: %u.\n",
+                    sh->slice_segment_addr);
+             return AVERROR_INVALIDDATA;
+         }
+         if (!sh->dependent_slice_segment_flag) {
+             sh->slice_addr = sh->slice_segment_addr;
+             s->slice_idx++;
+         }
+     } else {
+         sh->slice_segment_addr = sh->slice_addr = 0;
+         s->slice_idx           = 0;
+         s->slice_initialized   = 0;
+     }
+     if (!sh->dependent_slice_segment_flag) {
+         s->slice_initialized = 0;
+         for (i = 0; i < s->ps.pps->num_extra_slice_header_bits; i++)
+             skip_bits(gb, 1);  // slice_reserved_undetermined_flag[]
+         sh->slice_type = get_ue_golomb_long(gb);
+         if (!(sh->slice_type == I_SLICE ||
+               sh->slice_type == P_SLICE ||
+               sh->slice_type == B_SLICE)) {
+             av_log(s->avctx, AV_LOG_ERROR, "Unknown slice type: %d.\n",
+                    sh->slice_type);
+             return AVERROR_INVALIDDATA;
+         }
+         if (IS_IRAP(s) && sh->slice_type != I_SLICE) {
+             av_log(s->avctx, AV_LOG_ERROR, "Inter slices in an IRAP frame.\n");
+             return AVERROR_INVALIDDATA;
+         }
+         // when flag is not present, picture is inferred to be output
+         sh->pic_output_flag = 1;
+         if (s->ps.pps->output_flag_present_flag)
+             sh->pic_output_flag = get_bits1(gb);
+         if (s->ps.sps->separate_colour_plane_flag)
+             sh->colour_plane_id = get_bits(gb, 2);
+         if (!IS_IDR(s)) {
+             int poc, pos;
+             sh->pic_order_cnt_lsb = get_bits(gb, s->ps.sps->log2_max_poc_lsb);
+             poc = ff_hevc_compute_poc(s, sh->pic_order_cnt_lsb);
+             if (!sh->first_slice_in_pic_flag && poc != s->poc) {
+                 av_log(s->avctx, AV_LOG_WARNING,
+                        "Ignoring POC change between slices: %d -> %d\n", s->poc, poc);
+                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
+                     return AVERROR_INVALIDDATA;
+                 poc = s->poc;
+             }
+             s->poc = poc;
+             sh->short_term_ref_pic_set_sps_flag = get_bits1(gb);
+             pos = get_bits_left(gb);
+             if (!sh->short_term_ref_pic_set_sps_flag) {
+                 ret = ff_hevc_decode_short_term_rps(gb, s->avctx, &sh->slice_rps, s->ps.sps, 1);
+                 if (ret < 0)
+                     return ret;
+                 sh->short_term_rps = &sh->slice_rps;
+             } else {
+                 int numbits, rps_idx;
+                 if (!s->ps.sps->nb_st_rps) {
+                     av_log(s->avctx, AV_LOG_ERROR, "No ref lists in the SPS.\n");
+                     return AVERROR_INVALIDDATA;
+                 }
+                 numbits = av_ceil_log2(s->ps.sps->nb_st_rps);
+                 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
+                 sh->short_term_rps = &s->ps.sps->st_rps[rps_idx];
+             }
+             sh->short_term_ref_pic_set_size = pos - get_bits_left(gb);
+             pos = get_bits_left(gb);
+             ret = decode_lt_rps(s, &sh->long_term_rps, gb);
+             if (ret < 0) {
+                 av_log(s->avctx, AV_LOG_WARNING, "Invalid long term RPS.\n");
+                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
+                     return AVERROR_INVALIDDATA;
+             }
+             sh->long_term_ref_pic_set_size = pos - get_bits_left(gb);
+             if (s->ps.sps->sps_temporal_mvp_enabled_flag)
+                 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
+             else
+                 sh->slice_temporal_mvp_enabled_flag = 0;
+         } else {
+             s->sh.short_term_rps = NULL;
+             s->poc               = 0;
+         }
+         /* 8.3.1 */
+         if (s->temporal_id == 0 &&
+             s->nal_unit_type != NAL_TRAIL_N &&
+             s->nal_unit_type != NAL_TSA_N   &&
+             s->nal_unit_type != NAL_STSA_N  &&
+             s->nal_unit_type != NAL_RADL_N  &&
+             s->nal_unit_type != NAL_RADL_R  &&
+             s->nal_unit_type != NAL_RASL_N  &&
+             s->nal_unit_type != NAL_RASL_R)
+             s->pocTid0 = s->poc;
+         if (s->ps.sps->sao_enabled) {
+             sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
 -        sh->num_entry_point_offsets = get_ue_golomb_long(gb);
++            if (s->ps.sps->chroma_format_idc) {
++                sh->slice_sample_adaptive_offset_flag[1] =
++                sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
++            }
+         } else {
+             sh->slice_sample_adaptive_offset_flag[0] = 0;
+             sh->slice_sample_adaptive_offset_flag[1] = 0;
+             sh->slice_sample_adaptive_offset_flag[2] = 0;
+         }
+         sh->nb_refs[L0] = sh->nb_refs[L1] = 0;
+         if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
+             int nb_refs;
+             sh->nb_refs[L0] = s->ps.pps->num_ref_idx_l0_default_active;
+             if (sh->slice_type == B_SLICE)
+                 sh->nb_refs[L1] = s->ps.pps->num_ref_idx_l1_default_active;
+             if (get_bits1(gb)) { // num_ref_idx_active_override_flag
+                 sh->nb_refs[L0] = get_ue_golomb_long(gb) + 1;
+                 if (sh->slice_type == B_SLICE)
+                     sh->nb_refs[L1] = get_ue_golomb_long(gb) + 1;
+             }
+             if (sh->nb_refs[L0] > MAX_REFS || sh->nb_refs[L1] > MAX_REFS) {
+                 av_log(s->avctx, AV_LOG_ERROR, "Too many refs: %d/%d.\n",
+                        sh->nb_refs[L0], sh->nb_refs[L1]);
+                 return AVERROR_INVALIDDATA;
+             }
+             sh->rpl_modification_flag[0] = 0;
+             sh->rpl_modification_flag[1] = 0;
+             nb_refs = ff_hevc_frame_nb_refs(s);
+             if (!nb_refs) {
+                 av_log(s->avctx, AV_LOG_ERROR, "Zero refs for a frame with P or B slices.\n");
+                 return AVERROR_INVALIDDATA;
+             }
+             if (s->ps.pps->lists_modification_present_flag && nb_refs > 1) {
+                 sh->rpl_modification_flag[0] = get_bits1(gb);
+                 if (sh->rpl_modification_flag[0]) {
+                     for (i = 0; i < sh->nb_refs[L0]; i++)
+                         sh->list_entry_lx[0][i] = get_bits(gb, av_ceil_log2(nb_refs));
+                 }
+                 if (sh->slice_type == B_SLICE) {
+                     sh->rpl_modification_flag[1] = get_bits1(gb);
+                     if (sh->rpl_modification_flag[1] == 1)
+                         for (i = 0; i < sh->nb_refs[L1]; i++)
+                             sh->list_entry_lx[1][i] = get_bits(gb, av_ceil_log2(nb_refs));
+                 }
+             }
+             if (sh->slice_type == B_SLICE)
+                 sh->mvd_l1_zero_flag = get_bits1(gb);
+             if (s->ps.pps->cabac_init_present_flag)
+                 sh->cabac_init_flag = get_bits1(gb);
+             else
+                 sh->cabac_init_flag = 0;
+             sh->collocated_ref_idx = 0;
+             if (sh->slice_temporal_mvp_enabled_flag) {
+                 sh->collocated_list = L0;
+                 if (sh->slice_type == B_SLICE)
+                     sh->collocated_list = !get_bits1(gb);
+                 if (sh->nb_refs[sh->collocated_list] > 1) {
+                     sh->collocated_ref_idx = get_ue_golomb_long(gb);
+                     if (sh->collocated_ref_idx >= sh->nb_refs[sh->collocated_list]) {
+                         av_log(s->avctx, AV_LOG_ERROR,
+                                "Invalid collocated_ref_idx: %d.\n",
+                                sh->collocated_ref_idx);
+                         return AVERROR_INVALIDDATA;
+                     }
+                 }
+             }
+             if ((s->ps.pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
+                 (s->ps.pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
+                 pred_weight_table(s, gb);
+             }
+             sh->max_num_merge_cand = 5 - get_ue_golomb_long(gb);
+             if (sh->max_num_merge_cand < 1 || sh->max_num_merge_cand > 5) {
+                 av_log(s->avctx, AV_LOG_ERROR,
+                        "Invalid number of merging MVP candidates: %d.\n",
+                        sh->max_num_merge_cand);
+                 return AVERROR_INVALIDDATA;
+             }
+         }
+         sh->slice_qp_delta = get_se_golomb(gb);
+         if (s->ps.pps->pic_slice_level_chroma_qp_offsets_present_flag) {
+             sh->slice_cb_qp_offset = get_se_golomb(gb);
+             sh->slice_cr_qp_offset = get_se_golomb(gb);
+         } else {
+             sh->slice_cb_qp_offset = 0;
+             sh->slice_cr_qp_offset = 0;
+         }
++        if (s->ps.pps->chroma_qp_offset_list_enabled_flag)
++            sh->cu_chroma_qp_offset_enabled_flag = get_bits1(gb);
++        else
++            sh->cu_chroma_qp_offset_enabled_flag = 0;
++
+         if (s->ps.pps->deblocking_filter_control_present_flag) {
+             int deblocking_filter_override_flag = 0;
+             if (s->ps.pps->deblocking_filter_override_enabled_flag)
+                 deblocking_filter_override_flag = get_bits1(gb);
+             if (deblocking_filter_override_flag) {
+                 sh->disable_deblocking_filter_flag = get_bits1(gb);
+                 if (!sh->disable_deblocking_filter_flag) {
+                     sh->beta_offset = get_se_golomb(gb) * 2;
+                     sh->tc_offset   = get_se_golomb(gb) * 2;
+                 }
+             } else {
+                 sh->disable_deblocking_filter_flag = s->ps.pps->disable_dbf;
+                 sh->beta_offset                    = s->ps.pps->beta_offset;
+                 sh->tc_offset                      = s->ps.pps->tc_offset;
+             }
+         } else {
+             sh->disable_deblocking_filter_flag = 0;
+             sh->beta_offset                    = 0;
+             sh->tc_offset                      = 0;
+         }
+         if (s->ps.pps->seq_loop_filter_across_slices_enabled_flag &&
+             (sh->slice_sample_adaptive_offset_flag[0] ||
+              sh->slice_sample_adaptive_offset_flag[1] ||
+              !sh->disable_deblocking_filter_flag)) {
+             sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
+         } else {
+             sh->slice_loop_filter_across_slices_enabled_flag = s->ps.pps->seq_loop_filter_across_slices_enabled_flag;
+         }
+     } else if (!s->slice_initialized) {
+         av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
+         return AVERROR_INVALIDDATA;
+     }
+     sh->num_entry_point_offsets = 0;
+     if (s->ps.pps->tiles_enabled_flag || s->ps.pps->entropy_coding_sync_enabled_flag) {
 -            for (i = 0; i < sh->num_entry_point_offsets; i++)
 -                skip_bits(gb, offset_len);
 -        }
++        unsigned num_entry_point_offsets = get_ue_golomb_long(gb);
++        // It would be possible to bound this tighter but this here is simpler
++        if (num_entry_point_offsets > get_bits_left(gb)) {
++            av_log(s->avctx, AV_LOG_ERROR, "num_entry_point_offsets %d is invalid\n", num_entry_point_offsets);
++            return AVERROR_INVALIDDATA;
++        }
++
++        sh->num_entry_point_offsets = num_entry_point_offsets;
+         if (sh->num_entry_point_offsets > 0) {
+             int offset_len = get_ue_golomb_long(gb) + 1;
 -    sh->slice_qp = 26 + s->ps.pps->pic_init_qp_minus26 + sh->slice_qp_delta;
++            if (offset_len < 1 || offset_len > 32) {
++                sh->num_entry_point_offsets = 0;
++                av_log(s->avctx, AV_LOG_ERROR, "offset_len %d is invalid\n", offset_len);
++                return AVERROR_INVALIDDATA;
++            }
++
++            av_freep(&sh->entry_point_offset);
++            av_freep(&sh->offset);
++            av_freep(&sh->size);
++            sh->entry_point_offset = av_malloc_array(sh->num_entry_point_offsets, sizeof(unsigned));
++            sh->offset = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
++            sh->size = av_malloc_array(sh->num_entry_point_offsets, sizeof(int));
++            if (!sh->entry_point_offset || !sh->offset || !sh->size) {
++                sh->num_entry_point_offsets = 0;
++                av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate memory\n");
++                return AVERROR(ENOMEM);
++            }
++            for (i = 0; i < sh->num_entry_point_offsets; i++) {
++                unsigned val = get_bits_long(gb, offset_len);
++                sh->entry_point_offset[i] = val + 1; // +1; // +1 to get the size
++            }
++            if (s->threads_number > 1 && (s->ps.pps->num_tile_rows > 1 || s->ps.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->ps.pps->slice_header_extension_present_flag) {
+         unsigned int length = get_ue_golomb_long(gb);
++        if (length*8LL > get_bits_left(gb)) {
++            av_log(s->avctx, AV_LOG_ERROR, "too many slice_header_extension_data_bytes\n");
++            return AVERROR_INVALIDDATA;
++        }
+         for (i = 0; i < length; i++)
+             skip_bits(gb, 8);  // slice_header_extension_data_byte
+     }
+     // Inferred parameters
 -    s->HEVClc.first_qp_group = !s->sh.dependent_slice_segment_flag;
++    sh->slice_qp = 26U + s->ps.pps->pic_init_qp_minus26 + sh->slice_qp_delta;
+     if (sh->slice_qp > 51 ||
+         sh->slice_qp < -s->ps.sps->qp_bd_offset) {
+         av_log(s->avctx, AV_LOG_ERROR,
+                "The slice_qp %d is outside the valid range "
+                "[%d, 51].\n",
+                sh->slice_qp,
+                -s->ps.sps->qp_bd_offset);
+         return AVERROR_INVALIDDATA;
+     }
+     sh->slice_ctb_addr_rs = sh->slice_segment_addr;
+     if (!s->sh.slice_ctb_addr_rs && s->sh.dependent_slice_segment_flag) {
+         av_log(s->avctx, AV_LOG_ERROR, "Impossible slice segment.\n");
+         return AVERROR_INVALIDDATA;
+     }
 -        s->HEVClc.qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->ps.sps->qp_bd_offset,
 -                                52 + s->ps.sps->qp_bd_offset) - s->ps.sps->qp_bd_offset;
++    if (get_bits_left(gb) < 0) {
++        av_log(s->avctx, AV_LOG_ERROR,
++               "Overread slice header by %d bits\n", -get_bits_left(gb));
++        return AVERROR_INVALIDDATA;
++    }
++
++    s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
+     if (!s->ps.pps->cu_qp_delta_enabled_flag)
 -    HEVCLocalContext *lc    = &s->HEVClc;
++        s->HEVClc->qp_y = s->sh.slice_qp;
+     s->slice_initialized = 1;
++    s->HEVClc->tu.cu_qp_offset_cb = 0;
++    s->HEVClc->tu.cu_qp_offset_cr = 0;
++
++    s->no_rasl_output_flag = IS_IDR(s) || IS_BLA(s) || (s->nal_unit_type == NAL_CRA_NUT && s->last_eos);
+     return 0;
+ }
+ #define CTB(tab, x, y) ((tab)[(y) * s->ps.sps->ctb_width + (x)])
+ #define SET_SAO(elem, value)                            \
+ do {                                                    \
+     if (!sao_merge_up_flag && !sao_merge_left_flag)     \
+         sao->elem = value;                              \
+     else if (sao_merge_left_flag)                       \
+         sao->elem = CTB(s->sao, rx-1, ry).elem;         \
+     else if (sao_merge_up_flag)                         \
+         sao->elem = CTB(s->sao, rx, ry-1).elem;         \
+     else                                                \
+         sao->elem = 0;                                  \
+ } while (0)
+ static void hls_sao_param(HEVCContext *s, int rx, int ry)
+ {
 -    int shift               = s->ps.sps->bit_depth - FFMIN(s->ps.sps->bit_depth, 10);
++    HEVCLocalContext *lc    = s->HEVClc;
+     int sao_merge_left_flag = 0;
+     int sao_merge_up_flag   = 0;
 -    for (c_idx = 0; c_idx < 3; c_idx++) {
+     SAOParams *sao          = &CTB(s->sao, rx, ry);
+     int c_idx, i;
+     if (s->sh.slice_sample_adaptive_offset_flag[0] ||
+         s->sh.slice_sample_adaptive_offset_flag[1]) {
+         if (rx > 0) {
+             if (lc->ctb_left_flag)
+                 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
+         }
+         if (ry > 0 && !sao_merge_left_flag) {
+             if (lc->ctb_up_flag)
+                 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
+         }
+     }
 -            sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
++    for (c_idx = 0; c_idx < (s->ps.sps->chroma_format_idc ? 3 : 1); c_idx++) {
++        int log2_sao_offset_scale = c_idx == 0 ? s->ps.pps->log2_sao_offset_scale_luma :
++                                                 s->ps.pps->log2_sao_offset_scale_chroma;
++
+         if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
+             sao->type_idx[c_idx] = SAO_NOT_APPLIED;
+             continue;
+         }
+         if (c_idx == 2) {
+             sao->type_idx[2] = sao->type_idx[1];
+             sao->eo_class[2] = sao->eo_class[1];
+         } else {
+             SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
+         }
+         if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
+             continue;
+         for (i = 0; i < 4; i++)
+             SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
+         if (sao->type_idx[c_idx] == SAO_BAND) {
+             for (i = 0; i < 4; i++) {
+                 if (sao->offset_abs[c_idx][i]) {
+                     SET_SAO(offset_sign[c_idx][i],
+                             ff_hevc_sao_offset_sign_decode(s));
+                 } else {
+                     sao->offset_sign[c_idx][i] = 0;
+                 }
+             }
+             SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
+         } else if (c_idx != 2) {
+             SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
+         }
+         // Inferred parameters
+         sao->offset_val[c_idx][0] = 0;
+         for (i = 0; i < 4; i++) {
 -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->ps.sps->hshift[c_idx];
 -    int vshift       = s->ps.sps->vshift[c_idx];
 -    uint8_t *dst     = &s->frame->data[c_idx][(y0 >> vshift) * stride +
 -                                              ((x0 >> hshift) << s->ps.sps->pixel_shift)];
 -    DECLARE_ALIGNED(32, 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;
 -    static 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->ps.sps->qp_bd_offset;
 -        } else {
 -            int qp_i, offset;
 -
 -            if (c_idx == 1)
 -                offset = s->ps.pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
 -            else
 -                offset = s->ps.pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
 -
 -            qp_i = av_clip(qp_y + offset, -s->ps.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->ps.sps->qp_bd_offset;
 -        }
 -
 -        shift    = s->ps.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->ps.sps->scaling_list_enable_flag) {
 -            const ScalingList *sl = s->ps.pps->scaling_list_data_present_flag ?
 -                                    &s->ps.pps->scaling_list : &s->ps.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->ps.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;
++            sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i];
+             if (sao->type_idx[c_idx] == SAO_EDGE) {
+                 if (i > 1)
+                     sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
+             } else if (sao->offset_sign[c_idx][i]) {
+                 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
+             }
++            sao->offset_val[c_idx][i + 1] *= 1 << log2_sao_offset_scale;
+         }
+     }
+ }
+ #undef SET_SAO
+ #undef CTB
 -        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;
++static int hls_cross_component_pred(HEVCContext *s, int idx) {
++    HEVCLocalContext *lc    = s->HEVClc;
++    int log2_res_scale_abs_plus1 = ff_hevc_log2_res_scale_abs(s, idx);
 -    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 (log2_res_scale_abs_plus1 !=  0) {
++        int res_scale_sign_flag = ff_hevc_res_scale_sign_flag(s, idx);
++        lc->tu.res_scale_val = (1 << (log2_res_scale_abs_plus1 - 1)) *
++                               (1 - 2 * res_scale_sign_flag);
++    } else {
++        lc->tu.res_scale_val = 0;
+     }
 -            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->ps.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->ps.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->ps.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) {
 -        if (transform_skip_flag)
 -            s->hevcdsp.dequant(coeffs);
 -        else if (lc->cu.pred_mode == MODE_INTRA && c_idx == 0 &&
 -                 log2_trafo_size == 2)
 -            s->hevcdsp.transform_4x4_luma(coeffs);
 -        else {
 -            int max_xy = FFMAX(last_significant_coeff_x, last_significant_coeff_y);
 -            if (max_xy == 0)
 -                s->hevcdsp.idct_dc[log2_trafo_size - 2](coeffs);
 -            else {
 -                int col_limit = last_significant_coeff_x + last_significant_coeff_y + 4;
 -                if (max_xy < 4)
 -                    col_limit = FFMIN(4, col_limit);
 -                else if (max_xy < 8)
 -                    col_limit = FFMIN(8, col_limit);
 -                else if (max_xy < 12)
 -                    col_limit = FFMIN(24, col_limit);
 -                s->hevcdsp.idct[log2_trafo_size - 2](coeffs, col_limit);
 -            }
 -        }
 -    }
 -    s->hevcdsp.add_residual[log2_trafo_size - 2](dst, coeffs, stride);
 -                              int blk_idx, int cbf_luma, int cbf_cb, int cbf_cr)
++    return 0;
+ }
+ static int hls_transform_unit(HEVCContext *s, int x0, int y0,
+                               int xBase, int yBase, int cb_xBase, int cb_yBase,
+                               int log2_cb_size, int log2_trafo_size,
 -    HEVCLocalContext *lc = &s->HEVClc;
++                              int blk_idx, int cbf_luma, int *cbf_cb, int *cbf_cr)
+ {
 -        if (log2_trafo_size > 2) {
 -            trafo_size = trafo_size << (s->ps.sps->hshift[1] - 1);
 -            ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
 -            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 1);
 -            s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 2);
 -        } else if (blk_idx == 3) {
 -            trafo_size = trafo_size << s->ps.sps->hshift[1];
 -            ff_hevc_set_neighbour_available(s, xBase, yBase,
 -                                            trafo_size, trafo_size);
 -            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
 -            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
 -        }
++    HEVCLocalContext *lc = s->HEVClc;
++    const int log2_trafo_size_c = log2_trafo_size - s->ps.sps->hshift[1];
++    int i;
+     if (lc->cu.pred_mode == MODE_INTRA) {
+         int trafo_size = 1 << log2_trafo_size;
+         ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
+         s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
 -    if (cbf_luma || cbf_cb || cbf_cr) {
+     }
 -            ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
++    if (cbf_luma || cbf_cb[0] || cbf_cr[0] ||
++        (s->ps.sps->chroma_format_idc == 2 && (cbf_cb[1] || cbf_cr[1]))) {
+         int scan_idx   = SCAN_DIAG;
+         int scan_idx_c = SCAN_DIAG;
++        int cbf_chroma = cbf_cb[0] || cbf_cr[0] ||
++                         (s->ps.sps->chroma_format_idc == 2 &&
++                         (cbf_cb[1] || cbf_cr[1]));
+         if (s->ps.pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
+             lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
+             if (lc->tu.cu_qp_delta != 0)
+                 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
+                     lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
+             lc->tu.is_cu_qp_delta_coded = 1;
+             if (lc->tu.cu_qp_delta < -(26 + s->ps.sps->qp_bd_offset / 2) ||
+                 lc->tu.cu_qp_delta >  (25 + s->ps.sps->qp_bd_offset / 2)) {
+                 av_log(s->avctx, AV_LOG_ERROR,
+                        "The cu_qp_delta %d is outside the valid range "
+                        "[%d, %d].\n",
+                        lc->tu.cu_qp_delta,
+                        -(26 + s->ps.sps->qp_bd_offset / 2),
+                         (25 + s->ps.sps->qp_bd_offset / 2));
+                 return AVERROR_INVALIDDATA;
+             }
 -            if (lc->tu.cur_intra_pred_mode >= 6 &&
 -                lc->tu.cur_intra_pred_mode <= 14) {
++            ff_hevc_set_qPy(s, cb_xBase, cb_yBase, log2_cb_size);
++        }
++
++        if (s->sh.cu_chroma_qp_offset_enabled_flag && cbf_chroma &&
++            !lc->cu.cu_transquant_bypass_flag  &&  !lc->tu.is_cu_chroma_qp_offset_coded) {
++            int cu_chroma_qp_offset_flag = ff_hevc_cu_chroma_qp_offset_flag(s);
++            if (cu_chroma_qp_offset_flag) {
++                int cu_chroma_qp_offset_idx  = 0;
++                if (s->ps.pps->chroma_qp_offset_list_len_minus1 > 0) {
++                    cu_chroma_qp_offset_idx = ff_hevc_cu_chroma_qp_offset_idx(s);
++                    av_log(s->avctx, AV_LOG_ERROR,
++                        "cu_chroma_qp_offset_idx not yet tested.\n");
++                }
++                lc->tu.cu_qp_offset_cb = s->ps.pps->cb_qp_offset_list[cu_chroma_qp_offset_idx];
++                lc->tu.cu_qp_offset_cr = s->ps.pps->cr_qp_offset_list[cu_chroma_qp_offset_idx];
++            } else {
++                lc->tu.cu_qp_offset_cb = 0;
++                lc->tu.cu_qp_offset_cr = 0;
++            }
++            lc->tu.is_cu_chroma_qp_offset_coded = 1;
+         }
+         if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
 -            } else if (lc->tu.cur_intra_pred_mode >= 22 &&
 -                       lc->tu.cur_intra_pred_mode <= 30) {
++            if (lc->tu.intra_pred_mode >= 6 &&
++                lc->tu.intra_pred_mode <= 14) {
+                 scan_idx = SCAN_VERT;
 -            if (lc->pu.intra_pred_mode_c >=  6 &&
 -                lc->pu.intra_pred_mode_c <= 14) {
++            } else if (lc->tu.intra_pred_mode >= 22 &&
++                       lc->tu.intra_pred_mode <= 30) {
+                 scan_idx = SCAN_HORIZ;
+             }
 -            } else if (lc->pu.intra_pred_mode_c >= 22 &&
 -                       lc->pu.intra_pred_mode_c <= 30) {
++            if (lc->tu.intra_pred_mode_c >=  6 &&
++                lc->tu.intra_pred_mode_c <= 14) {
+                 scan_idx_c = SCAN_VERT;
 -            hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
 -        if (log2_trafo_size > 2) {
 -            if (cbf_cb)
 -                hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
 -            if (cbf_cr)
 -                hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
++            } else if (lc->tu.intra_pred_mode_c >= 22 &&
++                       lc->tu.intra_pred_mode_c <= 30) {
+                 scan_idx_c = SCAN_HORIZ;
+             }
+         }
++        lc->tu.cross_pf = 0;
++
+         if (cbf_luma)
 -            if (cbf_cb)
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
 -            if (cbf_cr)
 -                hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
++            ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
++        if (s->ps.sps->chroma_format_idc && (log2_trafo_size > 2 || s->ps.sps->chroma_format_idc == 3)) {
++            int trafo_size_h = 1 << (log2_trafo_size_c + s->ps.sps->hshift[1]);
++            int trafo_size_v = 1 << (log2_trafo_size_c + s->ps.sps->vshift[1]);
++            lc->tu.cross_pf  = (s->ps.pps->cross_component_prediction_enabled_flag && cbf_luma &&
++                                (lc->cu.pred_mode == MODE_INTER ||
++                                 (lc->tu.chroma_mode_c ==  4)));
++
++            if (lc->tu.cross_pf) {
++                hls_cross_component_pred(s, 0);
++            }
++            for (i = 0; i < (s->ps.sps->chroma_format_idc == 2 ? 2 : 1); i++) {
++                if (lc->cu.pred_mode == MODE_INTRA) {
++                    ff_hevc_set_neighbour_available(s, x0, y0 + (i << log2_trafo_size_c), trafo_size_h, trafo_size_v);
++                    s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (i << log2_trafo_size_c), 1);
++                }
++                if (cbf_cb[i])
++                    ff_hevc_hls_residual_coding(s, x0, y0 + (i << log2_trafo_size_c),
++                                                log2_trafo_size_c, scan_idx_c, 1);
++                else
++                    if (lc->tu.cross_pf) {
++                        ptrdiff_t stride = s->frame->linesize[1];
++                        int hshift = s->ps.sps->hshift[1];
++                        int vshift = s->ps.sps->vshift[1];
++                        int16_t *coeffs_y = (int16_t*)lc->edge_emu_buffer;
++                        int16_t *coeffs   = (int16_t*)lc->edge_emu_buffer2;
++                        int size = 1 << log2_trafo_size_c;
++
++                        uint8_t *dst = &s->frame->data[1][(y0 >> vshift) * stride +
++                                                              ((x0 >> hshift) << s->ps.sps->pixel_shift)];
++                        for (i = 0; i < (size * size); i++) {
++                            coeffs[i] = ((lc->tu.res_scale_val * coeffs_y[i]) >> 3);
++                        }
++                        s->hevcdsp.add_residual[log2_trafo_size_c-2](dst, coeffs, stride);
++                    }
++            }
++
++            if (lc->tu.cross_pf) {
++                hls_cross_component_pred(s, 1);
++            }
++            for (i = 0; i < (s->ps.sps->chroma_format_idc == 2 ? 2 : 1); i++) {
++                if (lc->cu.pred_mode == MODE_INTRA) {
++                    ff_hevc_set_neighbour_available(s, x0, y0 + (i << log2_trafo_size_c), trafo_size_h, trafo_size_v);
++                    s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (i << log2_trafo_size_c), 2);
++                }
++                if (cbf_cr[i])
++                    ff_hevc_hls_residual_coding(s, x0, y0 + (i << log2_trafo_size_c),
++                                                log2_trafo_size_c, scan_idx_c, 2);
++                else
++                    if (lc->tu.cross_pf) {
++                        ptrdiff_t stride = s->frame->linesize[2];
++                        int hshift = s->ps.sps->hshift[2];
++                        int vshift = s->ps.sps->vshift[2];
++                        int16_t *coeffs_y = (int16_t*)lc->edge_emu_buffer;
++                        int16_t *coeffs   = (int16_t*)lc->edge_emu_buffer2;
++                        int size = 1 << log2_trafo_size_c;
++
++                        uint8_t *dst = &s->frame->data[2][(y0 >> vshift) * stride +
++                                                          ((x0 >> hshift) << s->ps.sps->pixel_shift)];
++                        for (i = 0; i < (size * size); i++) {
++                            coeffs[i] = ((lc->tu.res_scale_val * coeffs_y[i]) >> 3);
++                        }
++                        s->hevcdsp.add_residual[log2_trafo_size_c-2](dst, coeffs, stride);
++                    }
++            }
++        } else if (s->ps.sps->chroma_format_idc && blk_idx == 3) {
++            int trafo_size_h = 1 << (log2_trafo_size + 1);
++            int trafo_size_v = 1 << (log2_trafo_size + s->ps.sps->vshift[1]);
++            for (i = 0; i < (s->ps.sps->chroma_format_idc == 2 ? 2 : 1); i++) {
++                if (lc->cu.pred_mode == MODE_INTRA) {
++                    ff_hevc_set_neighbour_available(s, xBase, yBase + (i << log2_trafo_size),
++                                                    trafo_size_h, trafo_size_v);
++                    s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (i << log2_trafo_size), 1);
++                }
++                if (cbf_cb[i])
++                    ff_hevc_hls_residual_coding(s, xBase, yBase + (i << log2_trafo_size),
++                                                log2_trafo_size, scan_idx_c, 1);
++            }
++            for (i = 0; i < (s->ps.sps->chroma_format_idc == 2 ? 2 : 1); i++) {
++                if (lc->cu.pred_mode == MODE_INTRA) {
++                    ff_hevc_set_neighbour_available(s, xBase, yBase + (i << log2_trafo_size),
++                                                trafo_size_h, trafo_size_v);
++                    s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (i << log2_trafo_size), 2);
++                }
++                if (cbf_cr[i])
++                    ff_hevc_hls_residual_coding(s, xBase, yBase + (i << log2_trafo_size),
++                                                log2_trafo_size, scan_idx_c, 2);
++            }
++        }
++    } else if (s->ps.sps->chroma_format_idc && lc->cu.pred_mode == MODE_INTRA) {
++        if (log2_trafo_size > 2 || s->ps.sps->chroma_format_idc == 3) {
++            int trafo_size_h = 1 << (log2_trafo_size_c + s->ps.sps->hshift[1]);
++            int trafo_size_v = 1 << (log2_trafo_size_c + s->ps.sps->vshift[1]);
++            ff_hevc_set_neighbour_available(s, x0, y0, trafo_size_h, trafo_size_v);
++            s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0, 1);
++            s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0, 2);
++            if (s->ps.sps->chroma_format_idc == 2) {
++                ff_hevc_set_neighbour_available(s, x0, y0 + (1 << log2_trafo_size_c),
++                                                trafo_size_h, trafo_size_v);
++                s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (1 << log2_trafo_size_c), 1);
++                s->hpc.intra_pred[log2_trafo_size_c - 2](s, x0, y0 + (1 << log2_trafo_size_c), 2);
++            }
+         } else if (blk_idx == 3) {
 -                              int cbf_cb, int cbf_cr)
++            int trafo_size_h = 1 << (log2_trafo_size + 1);
++            int trafo_size_v = 1 << (log2_trafo_size + s->ps.sps->vshift[1]);
++            ff_hevc_set_neighbour_available(s, xBase, yBase,
++                                            trafo_size_h, trafo_size_v);
++            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
++            s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
++            if (s->ps.sps->chroma_format_idc == 2) {
++                ff_hevc_set_neighbour_available(s, xBase, yBase + (1 << (log2_trafo_size)),
++                                                trafo_size_h, trafo_size_v);
++                s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (1 << (log2_trafo_size)), 1);
++                s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase + (1 << (log2_trafo_size)), 2);
++            }
+         }
+     }
++
+     return 0;
+ }
+ static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
+ {
+     int cb_size          = 1 << log2_cb_size;
+     int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
+     int min_pu_width     = s->ps.sps->min_pu_width;
+     int x_end = FFMIN(x0 + cb_size, s->ps.sps->width);
+     int y_end = FFMIN(y0 + cb_size, s->ps.sps->height);
+     int i, j;
+     for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
+         for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
+             s->is_pcm[i + j * min_pu_width] = 2;
+ }
+ static int hls_transform_tree(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;
++                              const int *base_cbf_cb, const int *base_cbf_cr)
+ {
 -        if (trafo_depth == 1)
 -            lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
++    HEVCLocalContext *lc = s->HEVClc;
+     uint8_t split_transform_flag;
++    int cbf_cb[2];
++    int cbf_cr[2];
+     int ret;
++    cbf_cb[0] = base_cbf_cb[0];
++    cbf_cb[1] = base_cbf_cb[1];
++    cbf_cr[0] = base_cbf_cr[0];
++    cbf_cr[1] = base_cbf_cr[1];
++
+     if (lc->cu.intra_split_flag) {
 -        lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
++        if (trafo_depth == 1) {
++            lc->tu.intra_pred_mode   = lc->pu.intra_pred_mode[blk_idx];
++            if (s->ps.sps->chroma_format_idc == 3) {
++                lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[blk_idx];
++                lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[blk_idx];
++            } else {
++                lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[0];
++                lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[0];
++            }
++        }
+     } else {
 -    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cb))
 -        cbf_cb = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 -    else if (log2_trafo_size > 2 || trafo_depth == 0)
 -        cbf_cb = 0;
 -    if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cr))
 -        cbf_cr = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
 -    else if (log2_trafo_size > 2 || trafo_depth == 0)
 -        cbf_cr = 0;
++        lc->tu.intra_pred_mode   = lc->pu.intra_pred_mode[0];
++        lc->tu.intra_pred_mode_c = lc->pu.intra_pred_mode_c[0];
++        lc->tu.chroma_mode_c     = lc->pu.chroma_mode_c[0];
+     }
+     if (log2_trafo_size <= s->ps.sps->log2_max_trafo_size &&
+         log2_trafo_size >  s->ps.sps->log2_min_tb_size    &&
+         trafo_depth     < lc->cu.max_trafo_depth       &&
+         !(lc->cu.intra_split_flag && trafo_depth == 0)) {
+         split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
+     } else {
+         int inter_split = s->ps.sps->max_transform_hierarchy_depth_inter == 0 &&
+                           lc->cu.pred_mode == MODE_INTER &&
+                           lc->cu.part_mode != PART_2Nx2N &&
+                           trafo_depth == 0;
+         split_transform_flag = log2_trafo_size > s->ps.sps->log2_max_trafo_size ||
+                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
+                                inter_split;
+     }
 -            cbf_cb || cbf_cr)
++    if (s->ps.sps->chroma_format_idc && (log2_trafo_size > 2 || s->ps.sps->chroma_format_idc == 3)) {
++        if (trafo_depth == 0 || cbf_cb[0]) {
++            cbf_cb[0] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
++            if (s->ps.sps->chroma_format_idc == 2 && (!split_transform_flag || log2_trafo_size == 3)) {
++                cbf_cb[1] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
++            }
++        }
++
++        if (trafo_depth == 0 || cbf_cr[0]) {
++            cbf_cr[0] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
++            if (s->ps.sps->chroma_format_idc == 2 && (!split_transform_flag || log2_trafo_size == 3)) {
++                cbf_cr[1] = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
++            }
++        }
++    }
+     if (split_transform_flag) {
+         const int trafo_size_split = 1 << (log2_trafo_size - 1);
+         const int x1 = x0 + trafo_size_split;
+         const int y1 = y0 + trafo_size_split;
+ #define SUBDIVIDE(x, y, idx)                                                    \
+ do {                                                                            \
+     ret = hls_transform_tree(s, x, y, x0, y0, cb_xBase, cb_yBase, log2_cb_size, \
+                              log2_trafo_size - 1, trafo_depth + 1, idx,         \
+                              cbf_cb, cbf_cr);                                   \
+     if (ret < 0)                                                                \
+         return ret;                                                             \
+ } while (0)
+         SUBDIVIDE(x0, y0, 0);
+         SUBDIVIDE(x1, y0, 1);
+         SUBDIVIDE(x0, y1, 2);
+         SUBDIVIDE(x1, y1, 3);
+ #undef SUBDIVIDE
+     } else {
+         int min_tu_size      = 1 << s->ps.sps->log2_min_tb_size;
+         int log2_min_tu_size = s->ps.sps->log2_min_tb_size;
+         int min_tu_width     = s->ps.sps->min_tb_width;
+         int cbf_luma         = 1;
+         if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
 -    //TODO: non-4:2:0 support
 -    HEVCLocalContext *lc = &s->HEVClc;
++            cbf_cb[0] || cbf_cr[0] ||
++            (s->ps.sps->chroma_format_idc == 2 && (cbf_cb[1] || cbf_cr[1]))) {
+             cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
++        }
+         ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
+                                  log2_cb_size, log2_trafo_size,
+                                  blk_idx, cbf_luma, cbf_cb, cbf_cr);
+         if (ret < 0)
+             return ret;
+         // TODO: store cbf_luma somewhere else
+         if (cbf_luma) {
+             int i, j;
+             for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
+                 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
+                     int x_tu = (x0 + j) >> log2_min_tu_size;
+                     int y_tu = (y0 + i) >> log2_min_tu_size;
+                     s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
+                 }
+         }
+         if (!s->sh.disable_deblocking_filter_flag) {
+             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
+             if (s->ps.pps->transquant_bypass_enable_flag &&
+                 lc->cu.cu_transquant_bypass_flag)
+                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
+         }
+     }
+     return 0;
+ }
+ static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
+ {
 -    int length         = cb_size * cb_size * s->ps.sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->ps.sps->pcm.bit_depth_chroma;
++    HEVCLocalContext *lc = s->HEVClc;
+     GetBitContext gb;
+     int cb_size   = 1 << log2_cb_size;
+     ptrdiff_t stride0 = s->frame->linesize[0];
+     ptrdiff_t stride1 = s->frame->linesize[1];
+     ptrdiff_t stride2 = s->frame->linesize[2];
+     uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->ps.sps->pixel_shift)];
+     uint8_t *dst1 = &s->frame->data[1][(y0 >> s->ps.sps->vshift[1]) * stride1 + ((x0 >> s->ps.sps->hshift[1]) << s->ps.sps->pixel_shift)];
+     uint8_t *dst2 = &s->frame->data[2][(y0 >> s->ps.sps->vshift[2]) * stride2 + ((x0 >> s->ps.sps->hshift[2]) << s->ps.sps->pixel_shift)];
 -    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->ps.sps->pcm.bit_depth);
 -    s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->ps.sps->pcm.bit_depth_chroma);
 -    s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->ps.sps->pcm.bit_depth_chroma);
 -    return 0;
 -}
 -
 -static void hls_mvd_coding(HEVCContext *s, int x0, int y0, int log2_cb_size)
 -{
 -    HEVCLocalContext *lc = &s->HEVClc;
 -    int x = ff_hevc_abs_mvd_greater0_flag_decode(s);
 -    int y = ff_hevc_abs_mvd_greater0_flag_decode(s);
 -
 -    if (x)
 -        x += ff_hevc_abs_mvd_greater1_flag_decode(s);
 -    if (y)
 -        y += ff_hevc_abs_mvd_greater1_flag_decode(s);
 -
 -    switch (x) {
 -    case 2: lc->pu.mvd.x = ff_hevc_mvd_decode(s);           break;
 -    case 1: lc->pu.mvd.x = ff_hevc_mvd_sign_flag_decode(s); break;
 -    case 0: lc->pu.mvd.x = 0;                               break;
++    int length         = cb_size * cb_size * s->ps.sps->pcm.bit_depth +
++                         (((cb_size >> s->ps.sps->hshift[1]) * (cb_size >> s->ps.sps->vshift[1])) +
++                          ((cb_size >> s->ps.sps->hshift[2]) * (cb_size >> s->ps.sps->vshift[2]))) *
++                          s->ps.sps->pcm.bit_depth_chroma;
+     const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
+     int ret;
+     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 ret;
 -    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;
 -    }
++    s->hevcdsp.put_pcm(dst0, stride0, cb_size, cb_size,     &gb, s->ps.sps->pcm.bit_depth);
++    if (s->ps.sps->chroma_format_idc) {
++        s->hevcdsp.put_pcm(dst1, stride1,
++                           cb_size >> s->ps.sps->hshift[1],
++                           cb_size >> s->ps.sps->vshift[1],
++                           &gb, s->ps.sps->pcm.bit_depth_chroma);
++        s->hevcdsp.put_pcm(dst2, stride2,
++                           cb_size >> s->ps.sps->hshift[2],
++                           cb_size >> s->ps.sps->vshift[2],
++                           &gb, s->ps.sps->pcm.bit_depth_chroma);
+     }
 - * 8.5.3.2.2.1 Luma sample interpolation process
++    return 0;
+ }
+ /**
 -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, int pred_idx)
++ * 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 dststride stride of the dst buffer
+  * @param ref reference picture buffer at origin (0, 0)
+  * @param mv motion vector (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 luma_weight weighting factor applied to the luma prediction
++ * @param luma_offset additive offset applied to the luma prediction value
+  */
 -    HEVCLocalContext *lc = &s->HEVClc;
++
++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)
+ {
 -
 -    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];
++    HEVCLocalContext *lc = s->HEVClc;
+     uint8_t *src         = ref->data[0];
+     ptrdiff_t srcstride  = ref->linesize[0];
+     int pic_width        = s->ps.sps->width;
+     int pic_height       = s->ps.sps->height;
 -    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]) {
++    int mx               = mv->x & 3;
++    int my               = mv->y & 3;
++    int weight_flag      = (s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
++                           (s->sh.slice_type == B_SLICE && s->ps.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 * (1 << s->ps.sps->pixel_shift));
 -        int offset = extra_top * srcstride + (extra_left << s->ps.sps->pixel_shift);
 -        int buf_offset = extra_top *
 -                         edge_emu_stride + (extra_left << s->ps.sps->pixel_shift);
++    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 ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
 -                                 block_w + ff_hevc_qpel_extra[mx],
 -                                 block_h + ff_hevc_qpel_extra[my],
 -                                 x_off - extra_left, y_off - extra_top,
++        int offset     = QPEL_EXTRA_BEFORE * srcstride       + (QPEL_EXTRA_BEFORE << s->ps.sps->pixel_shift);
++        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->ps.sps->pixel_shift);
+         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
+                                  edge_emu_stride, srcstride,
 -    s->hevcdsp.put_hevc_qpel[!!my][!!mx][pred_idx](dst, dststride, src, srcstride,
 -                                                   block_h, mx, my, lc->mc_buffer);
++                                 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;
+     }
 - * 8.5.3.2.2.2 Chroma sample interpolation process
++
++    if (!weight_flag)
++        s->hevcdsp.put_hevc_qpel_uni[idx][!!my][!!mx](dst, dststride, src, srcstride,
++                                                      block_h, mx, my, block_w);
++    else
++        s->hevcdsp.put_hevc_qpel_uni_w[idx][!!my][!!mx](dst, dststride, src, srcstride,
++                                                        block_h, s->sh.luma_log2_weight_denom,
++                                                        luma_weight, luma_offset, mx, my, block_w);
++}
++
++/**
++ * 8.5.3.2.2.1 Luma sample bidirectional interpolation process
++ *
++ * @param s HEVC decoding context
++ * @param dst target buffer for block data at block position
++ * @param dststride stride of the dst buffer
++ * @param ref0 reference picture0 buffer at origin (0, 0)
++ * @param mv0 motion vector0 (relative to block position) to get pixel data from
++ * @param x_off horizontal position of block from origin (0, 0)
++ * @param y_off vertical position of block from origin (0, 0)
++ * @param block_w width of block
++ * @param block_h height of block
++ * @param ref1 reference picture1 buffer at origin (0, 0)
++ * @param mv1 motion vector1 (relative to block position) to get pixel data from
++ * @param current_mv current motion vector structure
++ */
++ static void luma_mc_bi(HEVCContext *s, uint8_t *dst, ptrdiff_t dststride,
++                       AVFrame *ref0, const Mv *mv0, int x_off, int y_off,
++                       int block_w, int block_h, AVFrame *ref1, const Mv *mv1, struct MvField *current_mv)
++{
++    HEVCLocalContext *lc = s->HEVClc;
++    ptrdiff_t src0stride  = ref0->linesize[0];
++    ptrdiff_t src1stride  = ref1->linesize[0];
++    int pic_width        = s->ps.sps->width;
++    int pic_height       = s->ps.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->ps.pps->weighted_pred_flag) ||
++                           (s->sh.slice_type == B_SLICE && s->ps.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->ps.sps->pixel_shift);
++    uint8_t *src1  = ref1->data[0] + y_off1 * src1stride + (int)((unsigned)x_off1 << s->ps.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 ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
++        int offset     = QPEL_EXTRA_BEFORE * src0stride       + (QPEL_EXTRA_BEFORE << s->ps.sps->pixel_shift);
++        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->ps.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 ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
++        int offset     = QPEL_EXTRA_BEFORE * src1stride       + (QPEL_EXTRA_BEFORE << s->ps.sps->pixel_shift);
++        int buf_offset = QPEL_EXTRA_BEFORE * edge_emu_stride + (QPEL_EXTRA_BEFORE << s->ps.sps->pixel_shift);
++
++        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src1 - offset,
++                                 edge_emu_stride, src1stride,
++                                 block_w + QPEL_EXTRA,
++                                 block_h + QPEL_EXTRA,
++                                 x_off1 - QPEL_EXTRA_BEFORE, y_off1 - QPEL_EXTRA_BEFORE,
++                                 pic_width, pic_height);
++        src1 = lc->edge_emu_buffer2 + buf_offset;
++        src1stride = edge_emu_stride;
++    }
++
++    s->hevcdsp.put_hevc_qpel[idx][!!my0][!!mx0](lc->tmp, src0, src0stride,
++                                                block_h, mx0, my0, block_w);
++    if (!weight_flag)
++        s->hevcdsp.put_hevc_qpel_bi[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, lc->tmp,
++                                                       block_h, mx1, my1, block_w);
++    else
++        s->hevcdsp.put_hevc_qpel_bi_w[idx][!!my1][!!mx1](dst, dststride, src1, src1stride, lc->tmp,
++                                                         block_h, s->sh.luma_log2_weight_denom,
++                                                         s->sh.luma_weight_l0[current_mv->ref_idx[0]],
++                                                         s->sh.luma_weight_l1[current_mv->ref_idx[1]],
++                                                         s->sh.luma_offset_l0[current_mv->ref_idx[0]],
++                                                         s->sh.luma_offset_l1[current_mv->ref_idx[1]],
++                                                         mx1, my1, block_w);
++
+ }
+ /**
 -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, int pred_idx)
++ * 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 dst2 target buffer for block data at block position (V plane)
+  * @param dststride stride of the dst1 and dst2 buffers
+  * @param ref reference picture buffer at origin (0, 0)
+  * @param mv motion vector (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 chroma_weight weighting factor applied to the chroma prediction
++ * @param chroma_offset additive offset applied to the chroma prediction value
+  */
 -    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->ps.sps->width >> 1;
 -    int pic_height       = s->ps.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 * (1 << s->ps.sps->pixel_shift));
 -    src2  += y_off * src2stride + (x_off * (1 << s->ps.sps->pixel_shift));
++
++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)
+ {
 -        const ptrdiff_t edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
++    HEVCLocalContext *lc = s->HEVClc;
++    int pic_width        = s->ps.sps->width >> s->ps.sps->hshift[1];
++    int pic_height       = s->ps.sps->height >> s->ps.sps->vshift[1];
++    const Mv *mv         = &current_mv->mv[reflist];
++    int weight_flag      = (s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
++                           (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag);
++    int idx              = ff_hevc_pel_weight[block_w];
++    int hshift           = s->ps.sps->hshift[1];
++    int vshift           = s->ps.sps->vshift[1];
++    intptr_t mx          = av_mod_uintp2(mv->x, 2 + hshift);
++    intptr_t my          = av_mod_uintp2(mv->y, 2 + vshift);
++    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 * (1 << s->ps.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) {
 -        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->ps.sps->pixel_shift));
 -        int buf_offset2 = EPEL_EXTRA_BEFORE *
 -                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
++        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->ps.sps->pixel_shift;
++        int offset0 = EPEL_EXTRA_BEFORE * (srcstride + (1 << s->ps.sps->pixel_shift));
++        int buf_offset0 = EPEL_EXTRA_BEFORE *
++                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
++        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src0 - offset0,
++                                 edge_emu_stride, srcstride,
++                                 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
++                                 x_off - EPEL_EXTRA_BEFORE,
++                                 y_off - EPEL_EXTRA_BEFORE,
++                                 pic_width, pic_height);
++
++        src0 = lc->edge_emu_buffer + buf_offset0;
++        srcstride = edge_emu_stride;
++    }
++    if (!weight_flag)
++        s->hevcdsp.put_hevc_epel_uni[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
++                                                  block_h, _mx, _my, block_w);
++    else
++        s->hevcdsp.put_hevc_epel_uni_w[idx][!!my][!!mx](dst0, dststride, src0, srcstride,
++                                                        block_h, s->sh.chroma_log2_weight_denom,
++                                                        chroma_weight, chroma_offset, _mx, _my, block_w);
++}
++
++/**
++ * 8.5.3.2.2.2 Chroma sample bidirectional interpolation process
++ *
++ * @param s HEVC decoding context
++ * @param dst target buffer for block data at block position
++ * @param dststride stride of the dst buffer
++ * @param ref0 reference picture0 buffer at origin (0, 0)
++ * @param mv0 motion vector0 (relative to block position) to get pixel data from
++ * @param x_off horizontal position of block from origin (0, 0)
++ * @param y_off vertical position of block from origin (0, 0)
++ * @param block_w width of block
++ * @param block_h height of block
++ * @param ref1 reference picture1 buffer at origin (0, 0)
++ * @param mv1 motion vector1 (relative to block position) to get pixel data from
++ * @param current_mv current motion vector structure
++ * @param cidx chroma component(cb, cr)
++ */
++static void chroma_mc_bi(HEVCContext *s, uint8_t *dst0, ptrdiff_t dststride, AVFrame *ref0, AVFrame *ref1,
++                         int x_off, int y_off, int block_w, int block_h, struct MvField *current_mv, int cidx)
++{
++    HEVCLocalContext *lc = s->HEVClc;
++    uint8_t *src1        = ref0->data[cidx+1];
++    uint8_t *src2        = ref1->data[cidx+1];
++    ptrdiff_t src1stride = ref0->linesize[cidx+1];
++    ptrdiff_t src2stride = ref1->linesize[cidx+1];
++    int weight_flag      = (s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
++                           (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag);
++    int pic_width        = s->ps.sps->width >> s->ps.sps->hshift[1];
++    int pic_height       = s->ps.sps->height >> s->ps.sps->vshift[1];
++    Mv *mv0              = &current_mv->mv[0];
++    Mv *mv1              = &current_mv->mv[1];
++    int hshift = s->ps.sps->hshift[1];
++    int vshift = s->ps.sps->vshift[1];
++
++    intptr_t mx0 = av_mod_uintp2(mv0->x, 2 + hshift);
++    intptr_t my0 = av_mod_uintp2(mv0->y, 2 + vshift);
++    intptr_t mx1 = av_mod_uintp2(mv1->x, 2 + hshift);
++    intptr_t my1 = av_mod_uintp2(mv1->y, 2 + vshift);
++    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->ps.sps->pixel_shift);
++    src2  += y_off1 * src2stride + (int)((unsigned)x_off1 << s->ps.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->ps.sps->pixel_shift;
+         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->ps.sps->pixel_shift));
+         int buf_offset1 = EPEL_EXTRA_BEFORE *
+                           (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
 -                                 x_off - EPEL_EXTRA_BEFORE,
 -                                 y_off - EPEL_EXTRA_BEFORE,
+         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
+                                  edge_emu_stride, src1stride,
+                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst1, dststride, src1, src1stride,
 -                                                       block_h, mx, my, lc->mc_buffer);
++                                 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->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
++    }
 -                                 x_off - EPEL_EXTRA_BEFORE,
 -                                 y_off - EPEL_EXTRA_BEFORE,
++    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->ps.sps->pixel_shift;
++        int offset1 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->ps.sps->pixel_shift));
++        int buf_offset1 = EPEL_EXTRA_BEFORE *
++                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
++
++        s->vdsp.emulated_edge_mc(lc->edge_emu_buffer2, src2 - offset1,
+                                  edge_emu_stride, src2stride,
+                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
 -        src2 = lc->edge_emu_buffer + buf_offset2;
 -        src2stride = edge_emu_stride;
++                                 x_off1 - EPEL_EXTRA_BEFORE,
++                                 y_off1 - EPEL_EXTRA_BEFORE,
+                                  pic_width, pic_height);
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst2, dststride, src2, src2stride,
 -                                                       block_h, mx, my, lc->mc_buffer);
 -    } else {
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst1, dststride, src1, src1stride,
 -                                                       block_h, mx, my, lc->mc_buffer);
 -        s->hevcdsp.put_hevc_epel[!!my][!!mx][pred_idx](dst2, dststride, src2, src2stride,
 -                                                       block_h, mx, my, lc->mc_buffer);
 -    int y = (mv->y >> 2) + y0 + height + 9;
 -    ff_thread_await_progress(&ref->tf, y, 0);
++        src2 = lc->edge_emu_buffer2 + buf_offset1;
++        src2stride = edge_emu_stride;
+     }
++
++    s->hevcdsp.put_hevc_epel[idx][!!my0][!!mx0](lc->tmp, src1, src1stride,
++                                                block_h, _mx0, _my0, block_w);
++    if (!weight_flag)
++        s->hevcdsp.put_hevc_epel_bi[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
++                                                       src2, src2stride, lc->tmp,
++                                                       block_h, _mx1, _my1, block_w);
++    else
++        s->hevcdsp.put_hevc_epel_bi_w[idx][!!my1][!!mx1](dst0, s->frame->linesize[cidx+1],
++                                                         src2, src2stride, lc->tmp,
++                                                         block_h,
++                                                         s->sh.chroma_log2_weight_denom,
++                                                         s->sh.chroma_weight_l0[current_mv->ref_idx[0]][cidx],
++                                                         s->sh.chroma_weight_l1[current_mv->ref_idx[1]][cidx],
++                                                         s->sh.chroma_offset_l0[current_mv->ref_idx[0]][cidx],
++                                                         s->sh.chroma_offset_l1[current_mv->ref_idx[1]][cidx],
++                                                         _mx1, _my1, block_w);
+ }
+ static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
+                                 const Mv *mv, int y0, int height)
+ {
 -static void hevc_luma_mv_mpv_mode(HEVCContext *s, int x0, int y0, int nPbW,
++    int y = FFMAX(0, (mv->y >> 2) + y0 + height + 9);
++
++    if (s->threads_type == FF_THREAD_FRAME )
++        ff_thread_await_progress(&ref->tf, y, 0);
+ }
 -    HEVCLocalContext *lc             = &s->HEVClc;
++static void hevc_luma_mv_mvp_mode(HEVCContext *s, int x0, int y0, int nPbW,
+                                   int nPbH, int log2_cb_size, int part_idx,
+                                   int merge_idx, MvField *mv)
+ {
 -        mv->pred_flag[0] = 1;
 -        hls_mvd_coding(s, x0, y0, 0);
++    HEVCLocalContext *lc = s->HEVClc;
+     enum InterPredIdc inter_pred_idc = PRED_L0;
+     int mvp_flag;
+     ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
++    mv->pred_flag = 0;
+     if (s->sh.slice_type == B_SLICE)
+         inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
+     if (inter_pred_idc != PRED_L1) {
+         if (s->sh.nb_refs[L0])
+             mv->ref_idx[0]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
 -            hls_mvd_coding(s, x0, y0, 1);
++        mv->pred_flag = PF_L0;
++        ff_hevc_hls_mvd_coding(s, x0, y0, 0);
+         mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
+         ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                  part_idx, merge_idx, mv, mvp_flag, 0);
+         mv->mv[0].x += lc->pu.mvd.x;
+         mv->mv[0].y += lc->pu.mvd.y;
+     }
+     if (inter_pred_idc != PRED_L0) {
+         if (s->sh.nb_refs[L1])
+             mv->ref_idx[1]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
+         if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
+             AV_ZERO32(&lc->pu.mvd);
+         } else {
 -        mv->pred_flag[1] = 1;
++            ff_hevc_hls_mvd_coding(s, x0, y0, 1);
+         }
 -                                int log2_cb_size, int partIdx)
++        mv->pred_flag += PF_L1;
+         mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
+         ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                  part_idx, merge_idx, mv, mvp_flag, 1);
+         mv->mv[1].x += lc->pu.mvd.x;
+         mv->mv[1].y += lc->pu.mvd.y;
+     }
+ }
+ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
+                                 int nPbW, int nPbH,
 -    static const int pred_indices[] = {
 -        [4] = 0, [8] = 1, [12] = 2, [16] = 3, [24] = 4, [32] = 5, [48] = 6, [64] = 7,
 -    };
 -    const int pred_idx = pred_indices[nPbW];
 -
++                                int log2_cb_size, int partIdx, int idx)
+ {
 -    HEVCLocalContext *lc = &s->HEVClc;
+ #define POS(c_idx, x, y)                                                              \
+     &s->frame->data[c_idx][((y) >> s->ps.sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
+                            (((x) >> s->ps.sps->hshift[c_idx]) << s->ps.sps->pixel_shift)]
 -    HEVCFrame *ref0, *ref1;
 -
 -    ptrdiff_t tmpstride = MAX_PB_SIZE * sizeof(int16_t);
 -
++    HEVCLocalContext *lc = s->HEVClc;
+     int merge_idx = 0;
+     struct MvField current_mv = {{{ 0 }}};
+     int min_pu_width = s->ps.sps->min_pu_width;
+     MvField *tab_mvf = s->ref->tab_mvf;
+     RefPicList  *refPicList = s->ref->refPicList;
 -        hevc_luma_mv_mpv_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
++    HEVCFrame *ref0 = NULL, *ref1 = NULL;
+     uint8_t *dst0 = POS(0, x0, y0);
+     uint8_t *dst1 = POS(1, x0, y0);
+     uint8_t *dst2 = POS(2, x0, y0);
+     int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
+     int min_cb_width     = s->ps.sps->min_cb_width;
+     int x_cb             = x0 >> log2_min_cb_size;
+     int y_cb             = y0 >> log2_min_cb_size;
+     int x_pu, y_pu;
+     int i, j;
+     int skip_flag = SAMPLE_CTB(s->skip_flag, x_cb, y_cb);
+     if (!skip_flag)
+         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
+     if (skip_flag || lc->pu.merge_flag) {
+         if (s->sh.max_num_merge_cand > 1)
+             merge_idx = ff_hevc_merge_idx_decode(s);
+         else
+             merge_idx = 0;
+         ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                                    partIdx, merge_idx, &current_mv);
+     } else {
 -    if (current_mv.pred_flag[0]) {
++        hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
+                               partIdx, merge_idx, &current_mv);
+     }
+     x_pu = x0 >> s->ps.sps->log2_min_pu_size;
+     y_pu = y0 >> s->ps.sps->log2_min_pu_size;
+     for (j = 0; j < nPbH >> s->ps.sps->log2_min_pu_size; j++)
+         for (i = 0; i < nPbW >> s->ps.sps->log2_min_pu_size; i++)
+             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
 -    if (current_mv.pred_flag[1]) {
++    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[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 (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);
+     }
 -        luma_mc(s, tmp, tmpstride, ref0->frame,
 -                &current_mv.mv[0], x0, y0, nPbW, nPbH, pred_idx);
++    if (current_mv.pred_flag == PF_L0) {
++        int x0_c = x0 >> s->ps.sps->hshift[1];
++        int y0_c = y0 >> s->ps.sps->vshift[1];
++        int nPbW_c = nPbW >> s->ps.sps->hshift[1];
++        int nPbH_c = nPbH >> s->ps.sps->vshift[1];
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred[pred_idx](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, nPbH);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred[pred_idx](dst0, s->frame->linesize[0], tmp, tmpstride, nPbH);
 -        }
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
 -                  &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2, pred_idx);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_chroma[pred_idx](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,
 -                                                      nPbH / 2);
 -            s->hevcdsp.weighted_pred_chroma[pred_idx](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,
 -                                                      nPbH / 2);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst1, s->frame->linesize[1], tmp,  tmpstride, nPbH / 2);
 -            s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst2, s->frame->linesize[2], tmp2, tmpstride, 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]);
 -
 -        luma_mc(s, tmp, tmpstride, ref1->frame,
 -                &current_mv.mv[1], x0, y0, nPbW, nPbH, pred_idx);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred[pred_idx](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,
 -                                               nPbH);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred[pred_idx](dst0, s->frame->linesize[0], tmp, tmpstride, nPbH);
 -        }
 -
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
 -                  &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2, pred_idx);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_chroma[pred_idx](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, nPbH/2);
 -            s->hevcdsp.weighted_pred_chroma[pred_idx](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, nPbH/2);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst1, s->frame->linesize[1], tmp,  tmpstride, nPbH / 2);
 -            s->hevcdsp.put_unweighted_pred_chroma[pred_idx](dst2, s->frame->linesize[2], tmp2, tmpstride, 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]);
 -
 -        luma_mc(s, tmp, tmpstride, ref0->frame,
 -                &current_mv.mv[0], x0, y0, nPbW, nPbH, pred_idx);
 -        luma_mc(s, tmp2, tmpstride, ref1->frame,
 -                &current_mv.mv[1], x0, y0, nPbW, nPbH, pred_idx);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_avg[pred_idx](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, nPbH);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred_avg[pred_idx](dst0, s->frame->linesize[0],
 -                                                         tmp, tmp2, tmpstride, nPbH);
 -        }
 -
 -        chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
 -                  &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2, pred_idx);
 -        chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
 -                  &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2, pred_idx);
 -
 -        if ((s->sh.slice_type == P_SLICE && s->ps.pps->weighted_pred_flag) ||
 -            (s->sh.slice_type == B_SLICE && s->ps.pps->weighted_bipred_flag)) {
 -            s->hevcdsp.weighted_pred_avg_chroma[pred_idx](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, nPbH / 2);
 -            s->hevcdsp.weighted_pred_avg_chroma[pred_idx](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, nPbH / 2);
 -        } else {
 -            s->hevcdsp.put_unweighted_pred_avg_chroma[pred_idx](dst1, s->frame->linesize[1], tmp, tmp3,  tmpstride, nPbH/2);
 -            s->hevcdsp.put_unweighted_pred_avg_chroma[pred_idx](dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbH/2);
++        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]]);
 -    HEVCLocalContext *lc = &s->HEVClc;
++        if (s->ps.sps->chroma_format_idc) {
++            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->ps.sps->hshift[1];
++        int y0_c = y0 >> s->ps.sps->vshift[1];
++        int nPbW_c = nPbW >> s->ps.sps->hshift[1];
++        int nPbH_c = nPbH >> s->ps.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]]);
++
++        if (s->ps.sps->chroma_format_idc) {
++            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->ps.sps->hshift[1];
++        int y0_c = y0 >> s->ps.sps->vshift[1];
++        int nPbW_c = nPbW >> s->ps.sps->hshift[1];
++        int nPbH_c = nPbH >> s->ps.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);
++
++        if (s->ps.sps->chroma_format_idc) {
++            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);
+         }
+     }
+ }
+ /**
+  * 8.4.1
+  */
+ static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
+                                 int prev_intra_luma_pred_flag)
+ {
 -    int x0b              = x0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
 -    int y0b              = y0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
++    HEVCLocalContext *lc = s->HEVClc;
+     int x_pu             = x0 >> s->ps.sps->log2_min_pu_size;
+     int y_pu             = y0 >> s->ps.sps->log2_min_pu_size;
+     int min_pu_width     = s->ps.sps->min_pu_width;
+     int size_in_pus      = pu_size >> s->ps.sps->log2_min_pu_size;
 -            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;
++    int x0b              = av_mod_uintp2(x0, s->ps.sps->log2_ctb_size);
++    int y0b              = av_mod_uintp2(y0, s->ps.sps->log2_ctb_size);
+     int cand_up   = (lc->ctb_up_flag || y0b) ?
+                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
+     int cand_left = (lc->ctb_left_flag || x0b) ?
+                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
+     int y_ctb = (y0 >> (s->ps.sps->log2_ctb_size)) << (s->ps.sps->log2_ctb_size);
+     MvField *tab_mvf = s->ref->tab_mvf;
+     int intra_pred_mode;
+     int candidate[3];
+     int i, j;
+     // intra_pred_mode prediction does not cross vertical CTB boundaries
+     if ((y0 - 1) < y_ctb)
+         cand_up = INTRA_DC;
+     if (cand_left == cand_up) {
+         if (cand_left < 2) {
+             candidate[0] = INTRA_PLANAR;
+             candidate[1] = INTRA_DC;
+             candidate[2] = INTRA_ANGULAR_26;
+         } else {
+             candidate[0] = cand_left;
+             candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
+             candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
+         }
+     } else {
+         candidate[0] = cand_left;
+         candidate[1] = cand_up;
+         if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
+             candidate[2] = INTRA_PLANAR;
+         } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
+             candidate[2] = INTRA_DC;
+         } else {
+             candidate[2] = INTRA_ANGULAR_26;
+         }
+     }
+     if (prev_intra_luma_pred_flag) {
+         intra_pred_mode = candidate[lc->pu.mpm_idx];
+     } else {
+         if (candidate[0] > candidate[1])
+             FFSWAP(uint8_t, candidate[0], candidate[1]);
+         if (candidate[0] > candidate[2])
+             FFSWAP(uint8_t, candidate[0], candidate[2]);
+         if (candidate[1] > candidate[2])
+             FFSWAP(uint8_t, candidate[1], candidate[2]);
+         intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
+         for (i = 0; i < 3; i++)
+             if (intra_pred_mode >= candidate[i])
+                 intra_pred_mode++;
+     }
+     /* write the intra prediction units into the mv array */
+     if (!size_in_pus)
+         size_in_pus = 1;
+     for (i = 0; i < size_in_pus; i++) {
+         memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
+                intra_pred_mode, size_in_pus);
+         for (j = 0; j < size_in_pus; j++) {
 -    HEVCLocalContext *lc = &s->HEVClc;
++            tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag = PF_INTRA;
+         }
+     }
+     return intra_pred_mode;
+ }
+ static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
+                                           int log2_cb_size, int ct_depth)
+ {
+     int length = (1 << log2_cb_size) >> s->ps.sps->log2_min_cb_size;
+     int x_cb   = x0 >> s->ps.sps->log2_min_cb_size;
+     int y_cb   = y0 >> s->ps.sps->log2_min_cb_size;
+     int y;
+     for (y = 0; y < length; y++)
+         memset(&s->tab_ct_depth[(y_cb + y) * s->ps.sps->min_cb_width + x_cb],
+                ct_depth, length);
+ }
++static const uint8_t tab_mode_idx[] = {
++     0,  1,  2,  2,  2,  2,  3,  5,  7,  8, 10, 12, 13, 15, 17, 18, 19, 20,
++    21, 22, 23, 23, 24, 24, 25, 25, 26, 27, 27, 28, 28, 29, 29, 30, 31};
++
+ static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
+                                   int log2_cb_size)
+ {
 -    chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
 -    if (chroma_mode != 4) {
 -        if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
 -            lc->pu.intra_pred_mode_c = 34;
 -        else
 -            lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
 -    } else {
 -        lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
++    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;
+     int pb_size = (1 << log2_cb_size) >> split;
+     int side    = split + 1;
+     int chroma_mode;
+     int i, j;
+     for (i = 0; i < side; i++)
+         for (j = 0; j < side; j++)
+             prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
+     for (i = 0; i < side; i++) {
+         for (j = 0; j < side; j++) {
+             if (prev_intra_luma_pred_flag[2 * i + j])
+                 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
+             else
+                 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
+             lc->pu.intra_pred_mode[2 * i + j] =
+                 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
+                                      prev_intra_luma_pred_flag[2 * i + j]);
+         }
+     }
 -    HEVCLocalContext *lc = &s->HEVClc;
++    if (s->ps.sps->chroma_format_idc == 3) {
++        for (i = 0; i < side; i++) {
++            for (j = 0; j < side; j++) {
++                lc->pu.chroma_mode_c[2 * i + j] = chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
++                if (chroma_mode != 4) {
++                    if (lc->pu.intra_pred_mode[2 * i + j] == intra_chroma_table[chroma_mode])
++                        lc->pu.intra_pred_mode_c[2 * i + j] = 34;
++                    else
++                        lc->pu.intra_pred_mode_c[2 * i + j] = intra_chroma_table[chroma_mode];
++                } else {
++                    lc->pu.intra_pred_mode_c[2 * i + j] = lc->pu.intra_pred_mode[2 * i + j];
++                }
++            }
++        }
++    } else if (s->ps.sps->chroma_format_idc == 2) {
++        int mode_idx;
++        lc->pu.chroma_mode_c[0] = chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
++        if (chroma_mode != 4) {
++            if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
++                mode_idx = 34;
++            else
++                mode_idx = intra_chroma_table[chroma_mode];
++        } else {
++            mode_idx = lc->pu.intra_pred_mode[0];
++        }
++        lc->pu.intra_pred_mode_c[0] = tab_mode_idx[mode_idx];
++    } else if (s->ps.sps->chroma_format_idc != 0) {
++        chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
++        if (chroma_mode != 4) {
++            if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
++                lc->pu.intra_pred_mode_c[0] = 34;
++            else
++                lc->pu.intra_pred_mode_c[0] = intra_chroma_table[chroma_mode];
++        } else {
++            lc->pu.intra_pred_mode_c[0] = lc->pu.intra_pred_mode[0];
++        }
+     }
+ }
+ static void intra_prediction_unit_default_value(HEVCContext *s,
+                                                 int x0, int y0,
+                                                 int log2_cb_size)
+ {
 -    for (j = 0; j < size_in_pus; j++) {
++    HEVCLocalContext *lc = s->HEVClc;
+     int pb_size          = 1 << log2_cb_size;
+     int size_in_pus      = pb_size >> s->ps.sps->log2_min_pu_size;
+     int min_pu_width     = s->ps.sps->min_pu_width;
+     MvField *tab_mvf     = s->ref->tab_mvf;
+     int x_pu             = x0 >> s->ps.sps->log2_min_pu_size;
+     int y_pu             = y0 >> s->ps.sps->log2_min_pu_size;
+     int j, k;
+     if (size_in_pus == 0)
+         size_in_pus = 1;
 -        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;
 -    }
++    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);
 -    HEVCLocalContext *lc = &s->HEVClc;
++    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;
 -        hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
++    HEVCLocalContext *lc = s->HEVClc;
+     int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
+     int length           = cb_size >> log2_min_cb_size;
+     int min_cb_width     = s->ps.sps->min_cb_width;
+     int x_cb             = x0 >> log2_min_cb_size;
+     int y_cb             = y0 >> log2_min_cb_size;
++    int idx              = log2_cb_size - 2;
++    int qp_block_mask    = (1<<(s->ps.sps->log2_ctb_size - s->ps.pps->diff_cu_qp_delta_depth)) - 1;
+     int x, y, ret;
+     lc->cu.x                = x0;
+     lc->cu.y                = y0;
+     lc->cu.pred_mode        = MODE_INTRA;
+     lc->cu.part_mode        = PART_2Nx2N;
+     lc->cu.intra_split_flag = 0;
+     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
+     for (x = 0; x < 4; x++)
+         lc->pu.intra_pred_mode[x] = 1;
+     if (s->ps.pps->transquant_bypass_enable_flag) {
+         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
+         if (lc->cu.cu_transquant_bypass_flag)
+             set_deblocking_bypass(s, x0, y0, log2_cb_size);
+     } else
+         lc->cu.cu_transquant_bypass_flag = 0;
+     if (s->sh.slice_type != I_SLICE) {
+         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
+         x = y_cb * min_cb_width + x_cb;
+         for (y = 0; y < length; y++) {
+             memset(&s->skip_flag[x], skip_flag, length);
+             x += min_cb_width;
+         }
+         lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
++    } else {
++        x = y_cb * min_cb_width + x_cb;
++        for (y = 0; y < length; y++) {
++            memset(&s->skip_flag[x], 0, length);
++            x += min_cb_width;
++        }
+     }
+     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
 -                hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
++        hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0, idx);
+         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
+         if (!s->sh.disable_deblocking_filter_flag)
+             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
+     } else {
+         int pcm_flag = 0;
+         if (s->sh.slice_type != I_SLICE)
+             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
+         if (lc->cu.pred_mode != MODE_INTRA ||
+             log2_cb_size == s->ps.sps->log2_min_cb_size) {
+             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
+             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
+                                       lc->cu.pred_mode == MODE_INTRA;
+         }
+         if (lc->cu.pred_mode == MODE_INTRA) {
+             if (lc->cu.part_mode == PART_2Nx2N && s->ps.sps->pcm_enabled_flag &&
+                 log2_cb_size >= s->ps.sps->pcm.log2_min_pcm_cb_size &&
+                 log2_cb_size <= s->ps.sps->pcm.log2_max_pcm_cb_size) {
+                 pcm_flag = ff_hevc_pcm_flag_decode(s);
+             }
+             if (pcm_flag) {
+                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
+                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
+                 if (s->ps.sps->pcm.loop_filter_disable_flag)
+                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
+                 if (ret < 0)
+                     return ret;
+             } else {
+                 intra_prediction_unit(s, x0, y0, log2_cb_size);
+             }
+         } else {
+             intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
+             switch (lc->cu.part_mode) {
+             case PART_2Nx2N:
 -                hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
++                hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0, idx);
+                 break;
+             case PART_2NxN:
 -                hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
++                hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0, idx);
++                hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1, idx);
+                 break;
+             case PART_Nx2N:
 -                hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
++                hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0, idx - 1);
++                hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1, idx - 1);
+                 break;
+             case PART_2NxnU:
 -                hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
++                hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0, idx);
++                hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1, idx);
+                 break;
+             case PART_2NxnD:
 -                hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
++                hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0, idx);
++                hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1, idx);
+                 break;
+             case PART_nLx2N:
 -                hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
++                hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0, idx - 2);
++                hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1, idx - 2);
+                 break;
+             case PART_nRx2N:
 -                hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
 -                hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
 -                hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
++                hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0, idx - 2);
++                hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1, idx - 2);
+                 break;
+             case PART_NxN:
 -                                         log2_cb_size, 0, 0, 0, 0);
++                hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0, idx - 1);
++                hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1, idx - 1);
++                hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2, idx - 1);
++                hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3, idx - 1);
+                 break;
+             }
+         }
+         if (!pcm_flag) {
+             int rqt_root_cbf = 1;
+             if (lc->cu.pred_mode != MODE_INTRA &&
+                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
+                 rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
+             }
+             if (rqt_root_cbf) {
++                const static int cbf[2] = { 0 };
+                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
+                                          s->ps.sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
+                                          s->ps.sps->max_transform_hierarchy_depth_inter;
+                 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
+                                          log2_cb_size,
 -        ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
++                                         log2_cb_size, 0, 0, cbf, cbf);
+                 if (ret < 0)
+                     return ret;
+             } else {
+                 if (!s->sh.disable_deblocking_filter_flag)
+                     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
+             }
+         }
+     }
+     if (s->ps.pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
 -    set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
++        ff_hevc_set_qPy(s, x0, y0, log2_cb_size);
+     x = y_cb * min_cb_width + x_cb;
+     for (y = 0; y < length; y++) {
+         memset(&s->qp_y_tab[x], lc->qp_y, length);
+         x += min_cb_width;
+     }
 -    HEVCLocalContext *lc = &s->HEVClc;
++    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)
+ {
 -    lc->ct.depth = cb_depth;
++    HEVCLocalContext *lc = s->HEVClc;
+     const int cb_size    = 1 << log2_cb_size;
++    int ret;
+     int split_cu;
 -        log2_cb_size--;
 -        cb_depth++;
++    lc->ct_depth = cb_depth;
+     if (x0 + cb_size <= s->ps.sps->width  &&
+         y0 + cb_size <= s->ps.sps->height &&
+         log2_cb_size > s->ps.sps->log2_min_cb_size) {
+         split_cu = ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
+     } else {
+         split_cu = (log2_cb_size > s->ps.sps->log2_min_cb_size);
+     }
+     if (s->ps.pps->cu_qp_delta_enabled_flag &&
+         log2_cb_size >= s->ps.sps->log2_ctb_size - s->ps.pps->diff_cu_qp_delta_depth) {
+         lc->tu.is_cu_qp_delta_coded = 0;
+         lc->tu.cu_qp_delta          = 0;
+     }
++    if (s->sh.cu_chroma_qp_offset_enabled_flag &&
++        log2_cb_size >= s->ps.sps->log2_ctb_size - s->ps.pps->diff_cu_chroma_qp_offset_depth) {
++        lc->tu.is_cu_chroma_qp_offset_coded = 0;
++    }
++
+     if (split_cu) {
++        int qp_block_mask = (1<<(s->ps.sps->log2_ctb_size - s->ps.pps->diff_cu_qp_delta_depth)) - 1;
+         const int cb_size_split = cb_size >> 1;
+         const int x1 = x0 + cb_size_split;
+         const int y1 = y0 + cb_size_split;
 -#define SUBDIVIDE(x, y)                                                \
 -do {                                                                   \
 -    if (x < s->ps.sps->width && y < s->ps.sps->height) {                     \
 -        int ret = hls_coding_quadtree(s, x, y, log2_cb_size, cb_depth);\
 -        if (ret < 0)                                                   \
 -            return ret;                                                \
 -    }                                                                  \
 -} while (0)
++        int more_data = 0;
 -        SUBDIVIDE(x0, y0);
 -        SUBDIVIDE(x1, y0);
 -        SUBDIVIDE(x0, y1);
 -        SUBDIVIDE(x1, y1);
++        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->ps.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->ps.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->ps.sps->width &&
++            y1 < s->ps.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;
 -        int ret = hls_coding_unit(s, x0, y0, log2_cb_size);
++        if (more_data)
++            return ((x1 + cb_size_split) < s->ps.sps->width ||
++                    (y1 + cb_size_split) < s->ps.sps->height);
++        else
++            return 0;
+     } else {
 -    HEVCLocalContext *lc  = &s->HEVClc;
++        ret = hls_coding_unit(s, x0, y0, log2_cb_size);
+         if (ret < 0)
+             return ret;
++        if ((!((x0 + cb_size) %
++               (1 << (s->ps.sps->log2_ctb_size))) ||
++             (x0 + cb_size >= s->ps.sps->width)) &&
++            (!((y0 + cb_size) %
++               (1 << (s->ps.sps->log2_ctb_size))) ||
++             (y0 + cb_size >= s->ps.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)
+ {
 -            lc->start_of_tiles_x = x_ctb;
++    HEVCLocalContext *lc  = s->HEVClc;
+     int ctb_size          = 1 << s->ps.sps->log2_ctb_size;
+     int ctb_addr_rs       = s->ps.pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
+     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
+     if (s->ps.pps->entropy_coding_sync_enabled_flag) {
+         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
+             lc->first_qp_group = 1;
+         lc->end_of_tiles_x = s->ps.sps->width;
+     } else if (s->ps.pps->tiles_enabled_flag) {
+         if (ctb_addr_ts && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[ctb_addr_ts - 1]) {
+             int idxX = s->ps.pps->col_idxX[x_ctb >> s->ps.sps->log2_ctb_size];
 -        if (!ctb_addr_in_slice)
+             lc->end_of_tiles_x   = x_ctb + (s->ps.pps->column_width[idxX] << s->ps.sps->log2_ctb_size);
+             lc->first_qp_group   = 1;
+         }
+     } else {
+         lc->end_of_tiles_x = s->ps.sps->width;
+     }
+     lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->ps.sps->height);
+     lc->boundary_flags = 0;
+     if (s->ps.pps->tiles_enabled_flag) {
+         if (x_ctb > 0 && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]])
+             lc->boundary_flags |= BOUNDARY_LEFT_TILE;
+         if (x_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1])
+             lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
+         if (y_ctb > 0 && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->ps.sps->ctb_width]])
+             lc->boundary_flags |= BOUNDARY_UPPER_TILE;
+         if (y_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->ps.sps->ctb_width])
+             lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
+     } else {
 -static int hls_slice_data(HEVCContext *s)
++        if (ctb_addr_in_slice <= 0)
+             lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
+         if (ctb_addr_in_slice < s->ps.sps->ctb_width)
+             lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
+     }
+     lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && !(lc->boundary_flags & BOUNDARY_LEFT_TILE));
+     lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->ps.sps->ctb_width) && !(lc->boundary_flags & BOUNDARY_UPPER_TILE));
+     lc->ctb_up_right_flag = ((y_ctb > 0)  && (ctb_addr_in_slice+1 >= s->ps.sps->ctb_width) && (s->ps.pps->tile_id[ctb_addr_ts] == s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->ps.sps->ctb_width]]));
+     lc->ctb_up_left_flag = ((x_ctb > 0) && (y_ctb > 0)  && (ctb_addr_in_slice-1 >= s->ps.sps->ctb_width) && (s->ps.pps->tile_id[ctb_addr_ts] == s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs-1 - s->ps.sps->ctb_width]]));
+ }
 -    int ret;
++static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
+ {
++    HEVCContext *s  = avctxt->priv_data;
+     int ctb_size    = 1 << s->ps.sps->log2_ctb_size;
+     int more_data   = 1;
+     int x_ctb       = 0;
+     int y_ctb       = 0;
+     int ctb_addr_ts = s->ps.pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
 -        ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->ps.sps->log2_ctb_size, 0);
 -        if (ret < 0)
 -            return ret;
 -        more_data = !ff_hevc_end_of_slice_flag_decode(s);
++
++    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->ps.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->ps.sps->ctb_size) {
+         int ctb_addr_rs = s->ps.pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+         x_ctb = (ctb_addr_rs % ((s->ps.sps->width + ctb_size - 1) >> s->ps.sps->log2_ctb_size)) << s->ps.sps->log2_ctb_size;
+         y_ctb = (ctb_addr_rs / ((s->ps.sps->width + ctb_size - 1) >> s->ps.sps->log2_ctb_size)) << s->ps.sps->log2_ctb_size;
+         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
+         ff_hevc_cabac_init(s, ctb_addr_ts);
+         hls_sao_param(s, x_ctb >> s->ps.sps->log2_ctb_size, y_ctb >> s->ps.sps->log2_ctb_size);
+         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
+         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;
 -        ff_hevc_hls_filter(s, x_ctb, y_ctb);
++        more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->ps.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_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
+     }
+     if (x_ctb + ctb_size >= s->ps.sps->width &&
+         y_ctb + ctb_size >= s->ps.sps->height)
 -static void restore_tqb_pixels(HEVCContext *s)
++        ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
+     return ctb_addr_ts;
+ }
 -    int min_pu_size = 1 << s->ps.sps->log2_min_pu_size;
 -    int x, y, c_idx;
 -
 -    for (c_idx = 0; c_idx < 3; c_idx++) {
 -        ptrdiff_t stride = s->frame->linesize[c_idx];
 -        int hshift       = s->ps.sps->hshift[c_idx];
 -        int vshift       = s->ps.sps->vshift[c_idx];
 -        for (y = 0; y < s->ps.sps->min_pu_height; y++) {
 -            for (x = 0; x < s->ps.sps->min_pu_width; x++) {
 -                if (s->is_pcm[y * s->ps.sps->min_pu_width + x]) {
 -                    int n;
 -                    int len      = min_pu_size >> hshift;
 -                    uint8_t *src = &s->frame->data[c_idx][((y << s->ps.sps->log2_min_pu_size) >> vshift) * stride + (((x << s->ps.sps->log2_min_pu_size) >> hshift) << s->ps.sps->pixel_shift)];
 -                    uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->ps.sps->log2_min_pu_size) >> vshift) * stride + (((x << s->ps.sps->log2_min_pu_size) >> hshift) << s->ps.sps->pixel_shift)];
 -                    for (n = 0; n < (min_pu_size >> vshift); n++) {
 -                        memcpy(dst, src, len);
 -                        src += stride;
 -                        dst += stride;
 -                    }
 -                }
++static int hls_slice_data(HEVCContext *s)
+ {
 -    HEVCLocalContext *lc = &s->HEVClc;
++    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->ps.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->ps.sps->width + ctb_size - 1) >> s1->ps.sps->log2_ctb_size);
++    int ctb_addr_ts = s1->ps.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->ps.sps->ctb_size) {
++        int x_ctb = (ctb_addr_rs % s->ps.sps->ctb_width) << s->ps.sps->log2_ctb_size;
++        int y_ctb = (ctb_addr_rs / s->ps.sps->ctb_width) << s->ps.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 (atomic_load(&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->ps.sps->log2_ctb_size, y_ctb >> s->ps.sps->log2_ctb_size);
++        more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->ps.sps->log2_ctb_size, 0);
++
++        if (more_data < 0) {
++            s->tab_slice_address[ctb_addr_rs] = -1;
++            atomic_store(&s1->wpp_err, 1);
++            ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
++            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->ps.sps->width && ctb_row != s->sh.num_entry_point_offsets) {
++            atomic_store(&s1->wpp_err, 1);
++            ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
++            return 0;
++        }
++
++        if ((x_ctb+ctb_size) >= s->ps.sps->width && (y_ctb+ctb_size) >= s->ps.sps->height ) {
++            ff_hevc_hls_filter(s, x_ctb, y_ctb, ctb_size);
++            ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
++            return ctb_addr_ts;
++        }
++        ctb_addr_rs       = s->ps.pps->ctb_addr_ts_to_rs[ctb_addr_ts];
++        x_ctb+=ctb_size;
++
++        if(x_ctb >= s->ps.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 H2645NAL *nal)
++{
++    const uint8_t *data = nal->data;
++    int length          = nal->size;
++    HEVCLocalContext *lc = s->HEVClc;
++    int *ret = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
++    int *arg = av_malloc_array(s->sh.num_entry_point_offsets + 1, sizeof(int));
++    int64_t offset;
++    int64_t startheader, cmpt = 0;
++    int i, j, res = 0;
++
++    if (!ret || !arg) {
++        av_free(ret);
++        av_free(arg);
++        return AVERROR(ENOMEM);
++    }
++
++    if (s->sh.slice_ctb_addr_rs + s->sh.num_entry_point_offsets * s->ps.sps->ctb_width >= s->ps.sps->ctb_width * s->ps.sps->ctb_height) {
++        av_log(s->avctx, AV_LOG_ERROR, "WPP ctb addresses are wrong (%d %d %d %d)\n",
++            s->sh.slice_ctb_addr_rs, s->sh.num_entry_point_offsets,
++            s->ps.sps->ctb_width, s->ps.sps->ctb_height
++        );
++        res = AVERROR_INVALIDDATA;
++        goto error;
++    }
++
++    ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
++
++    if (!s->sList[1]) {
++        for (i = 1; i < s->threads_number; i++) {
++            s->sList[i] = av_malloc(sizeof(HEVCContext));
++            memcpy(s->sList[i], s, sizeof(HEVCContext));
++            s->HEVClcList[i] = av_mallocz(sizeof(HEVCLocalContext));
++            s->sList[i]->HEVClc = s->HEVClcList[i];
++        }
++    }
++
++    offset = (lc->gb.index >> 3);
++
++    for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < nal->skipped_bytes; j++) {
++        if (nal->skipped_bytes_pos[j] >= offset && nal->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 < nal->skipped_bytes; j++) {
++            if (nal->skipped_bytes_pos[j] >= offset && nal->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;
++        if (length < offset) {
++            av_log(s->avctx, AV_LOG_ERROR, "entry_point_offset table is corrupted\n");
++            res = AVERROR_INVALIDDATA;
++            goto error;
++        }
++        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 = data;
++
++    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];
+     }
++
++    atomic_store(&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->ps.pps->entropy_coding_sync_enabled_flag)
++        s->avctx->execute2(s->avctx, 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];
++error:
++    av_free(ret);
++    av_free(arg);
++    return res;
+ }
+ static int set_side_data(HEVCContext *s)
+ {
+     AVFrame *out = s->ref->frame;
+     if (s->sei_frame_packing_present &&
+         s->frame_packing_arrangement_type >= 3 &&
+         s->frame_packing_arrangement_type <= 5 &&
+         s->content_interpretation_type > 0 &&
+         s->content_interpretation_type < 3) {
+         AVStereo3D *stereo = av_stereo3d_create_side_data(out);
+         if (!stereo)
+             return AVERROR(ENOMEM);
+         switch (s->frame_packing_arrangement_type) {
+         case 3:
+             if (s->quincunx_subsampling)
+                 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
+             else
+                 stereo->type = AV_STEREO3D_SIDEBYSIDE;
+             break;
+         case 4:
+             stereo->type = AV_STEREO3D_TOPBOTTOM;
+             break;
+         case 5:
+             stereo->type = AV_STEREO3D_FRAMESEQUENCE;
+             break;
+         }
+         if (s->content_interpretation_type == 2)
+             stereo->flags = AV_STEREO3D_FLAG_INVERT;
+     }
+     if (s->sei_display_orientation_present &&
+         (s->sei_anticlockwise_rotation || s->sei_hflip || s->sei_vflip)) {
+         double angle = s->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
+         AVFrameSideData *rotation = av_frame_new_side_data(out,
+                                                            AV_FRAME_DATA_DISPLAYMATRIX,
+                                                            sizeof(int32_t) * 9);
+         if (!rotation)
+             return AVERROR(ENOMEM);
+         av_display_rotation_set((int32_t *)rotation->data, angle);
+         av_display_matrix_flip((int32_t *)rotation->data,
+                                s->sei_hflip, s->sei_vflip);
+     }
++    // Decrement the mastering display flag when IRAP frame has no_rasl_output_flag=1
++    // so the side data persists for the entire coded video sequence.
++    if (s->sei_mastering_display_info_present > 0 &&
++        IS_IRAP(s) && s->no_rasl_output_flag) {
++        s->sei_mastering_display_info_present--;
++    }
++    if (s->sei_mastering_display_info_present) {
++        // HEVC uses a g,b,r ordering, which we convert to a more natural r,g,b
++        const int mapping[3] = {2, 0, 1};
++        const int chroma_den = 50000;
++        const int luma_den = 10000;
++        int i;
++        AVMasteringDisplayMetadata *metadata =
++            av_mastering_display_metadata_create_side_data(out);
++        if (!metadata)
++            return AVERROR(ENOMEM);
++
++        for (i = 0; i < 3; i++) {
++            const int j = mapping[i];
++            metadata->display_primaries[i][0].num = s->display_primaries[j][0];
++            metadata->display_primaries[i][0].den = chroma_den;
++            metadata->display_primaries[i][1].num = s->display_primaries[j][1];
++            metadata->display_primaries[i][1].den = chroma_den;
++        }
++        metadata->white_point[0].num = s->white_point[0];
++        metadata->white_point[0].den = chroma_den;
++        metadata->white_point[1].num = s->white_point[1];
++        metadata->white_point[1].den = chroma_den;
++
++        metadata->max_luminance.num = s->max_mastering_luminance;
++        metadata->max_luminance.den = luma_den;
++        metadata->min_luminance.num = s->min_mastering_luminance;
++        metadata->min_luminance.den = luma_den;
++        metadata->has_luminance = 1;
++        metadata->has_primaries = 1;
++
++        av_log(s->avctx, AV_LOG_DEBUG, "Mastering Display Metadata:\n");
++        av_log(s->avctx, AV_LOG_DEBUG,
++               "r(%5.4f,%5.4f) g(%5.4f,%5.4f) b(%5.4f %5.4f) wp(%5.4f, %5.4f)\n",
++               av_q2d(metadata->display_primaries[0][0]),
++               av_q2d(metadata->display_primaries[0][1]),
++               av_q2d(metadata->display_primaries[1][0]),
++               av_q2d(metadata->display_primaries[1][1]),
++               av_q2d(metadata->display_primaries[2][0]),
++               av_q2d(metadata->display_primaries[2][1]),
++               av_q2d(metadata->white_point[0]), av_q2d(metadata->white_point[1]));
++        av_log(s->avctx, AV_LOG_DEBUG,
++               "min_luminance=%f, max_luminance=%f\n",
++               av_q2d(metadata->min_luminance), av_q2d(metadata->max_luminance));
++    }
++
++    if (s->a53_caption) {
++        AVFrameSideData* sd = av_frame_new_side_data(out,
++                                                     AV_FRAME_DATA_A53_CC,
++                                                     s->a53_caption_size);
++        if (sd)
++            memcpy(sd->data, s->a53_caption, s->a53_caption_size);
++        av_freep(&s->a53_caption);
++        s->a53_caption_size = 0;
++        s->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
++    }
++
+     return 0;
+ }
+ static int hevc_frame_start(HEVCContext *s)
+ {
 -    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));
++    HEVCLocalContext *lc = s->HEVClc;
++    int pic_size_in_ctb  = ((s->ps.sps->width  >> s->ps.sps->log2_min_cb_size) + 1) *
++                           ((s->ps.sps->height >> s->ps.sps->log2_min_cb_size) + 1);
+     int ret;
 -    memset(s->is_pcm,        0, s->ps.sps->min_pu_width * s->ps.sps->min_pu_height);
++    memset(s->horizontal_bs, 0, s->bs_width * s->bs_height);
++    memset(s->vertical_bs,   0, s->bs_width * s->bs_height);
+     memset(s->cbf_luma,      0, s->ps.sps->min_tb_width * s->ps.sps->min_tb_height);
 -    lc->start_of_tiles_x = 0;
++    memset(s->is_pcm,        0, (s->ps.sps->min_pu_width + 1) * (s->ps.sps->min_pu_height + 1));
++    memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
 -    ret = ff_hevc_set_new_ref(s, s->ps.sps->sao_enabled ? &s->sao_frame : &s->frame,
 -                              s->poc);
+     s->is_decoded        = 0;
+     s->first_nal_type    = s->nal_unit_type;
+     if (s->ps.pps->tiles_enabled_flag)
+         lc->end_of_tiles_x = s->ps.pps->column_width[0] << s->ps.sps->log2_ctb_size;
 -    ff_thread_finish_setup(s->avctx);
++    ret = ff_hevc_set_new_ref(s, &s->frame, s->poc);
+     if (ret < 0)
+         goto fail;
+     ret = ff_hevc_frame_rps(s);
+     if (ret < 0) {
+         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
+         goto fail;
+     }
+     s->ref->frame->key_frame = IS_IRAP(s);
+     ret = set_side_data(s);
+     if (ret < 0)
+         goto fail;
++    s->frame->pict_type = 3 - s->sh.slice_type;
++
++    if (!IS_IRAP(s))
++        ff_hevc_bump_frame(s);
++
+     av_frame_unref(s->output_frame);
+     ret = ff_hevc_output_frame(s, s->output_frame, 0);
+     if (ret < 0)
+         goto fail;
 -    HEVCLocalContext *lc = &s->HEVClc;
++    if (!s->avctx->hwaccel)
++        ff_thread_finish_setup(s->avctx);
+     return 0;
+ fail:
+     if (s->ref)
+         ff_hevc_unref_frame(s, s->ref, ~0);
+     s->ref = NULL;
+     return ret;
+ }
+ static int decode_nal_unit(HEVCContext *s, const H2645NAL *nal)
+ {
 -            ctb_addr_ts = hls_slice_data(s);
++    HEVCLocalContext *lc = s->HEVClc;
+     GetBitContext *gb    = &lc->gb;
+     int ctb_addr_ts, ret;
+     *gb              = nal->gb;
+     s->nal_unit_type = nal->type;
+     s->temporal_id   = nal->temporal_id;
+     switch (s->nal_unit_type) {
+     case NAL_VPS:
+         ret = ff_hevc_decode_nal_vps(gb, s->avctx, &s->ps);
+         if (ret < 0)
+             goto fail;
+         break;
+     case NAL_SPS:
+         ret = ff_hevc_decode_nal_sps(gb, s->avctx, &s->ps,
+                                      s->apply_defdispwin);
+         if (ret < 0)
+             goto fail;
+         break;
+     case NAL_PPS:
+         ret = ff_hevc_decode_nal_pps(gb, s->avctx, &s->ps);
+         if (ret < 0)
+             goto fail;
+         break;
+     case NAL_SEI_PREFIX:
+     case NAL_SEI_SUFFIX:
+         ret = ff_hevc_decode_nal_sei(s);
+         if (ret < 0)
+             goto fail;
+         break;
+     case NAL_TRAIL_R:
+     case NAL_TRAIL_N:
+     case NAL_TSA_N:
+     case NAL_TSA_R:
+     case NAL_STSA_N:
+     case NAL_STSA_R:
+     case NAL_BLA_W_LP:
+     case NAL_BLA_W_RADL:
+     case NAL_BLA_N_LP:
+     case NAL_IDR_W_RADL:
+     case NAL_IDR_N_LP:
+     case NAL_CRA_NUT:
+     case NAL_RADL_N:
+     case NAL_RADL_R:
+     case NAL_RASL_N:
+     case NAL_RASL_R:
+         ret = hls_slice_header(s);
+         if (ret < 0)
+             return ret;
+         if (s->max_ra == INT_MAX) {
+             if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
+                 s->max_ra = s->poc;
+             } else {
+                 if (IS_IDR(s))
+                     s->max_ra = INT_MIN;
+             }
+         }
+         if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
+             s->poc <= s->max_ra) {
+             s->is_decoded = 0;
+             break;
+         } else {
+             if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
+                 s->max_ra = INT_MIN;
+         }
+         if (s->sh.first_slice_in_pic_flag) {
+             ret = hevc_frame_start(s);
+             if (ret < 0)
+                 return ret;
+         } else if (!s->ref) {
+             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
+             goto fail;
+         }
+         if (s->nal_unit_type != s->first_nal_type) {
+             av_log(s->avctx, AV_LOG_ERROR,
+                    "Non-matching NAL types of the VCL NALUs: %d %d\n",
+                    s->first_nal_type, s->nal_unit_type);
+             return AVERROR_INVALIDDATA;
+         }
+         if (!s->sh.dependent_slice_segment_flag &&
+             s->sh.slice_type != I_SLICE) {
+             ret = ff_hevc_slice_rpl(s);
+             if (ret < 0) {
+                 av_log(s->avctx, AV_LOG_WARNING,
+                        "Error constructing the reference lists for the current slice.\n");
+                 goto fail;
+             }
+         }
+         if (s->sh.first_slice_in_pic_flag && s->avctx->hwaccel) {
+             ret = s->avctx->hwaccel->start_frame(s->avctx, NULL, 0);
+             if (ret < 0)
+                 goto fail;
+         }
+         if (s->avctx->hwaccel) {
+             ret = s->avctx->hwaccel->decode_slice(s->avctx, nal->raw_data, nal->raw_size);
+             if (ret < 0)
+                 goto fail;
+         } else {
 -                if ((s->ps.pps->transquant_bypass_enable_flag ||
 -                     (s->ps.sps->pcm.loop_filter_disable_flag && s->ps.sps->pcm_enabled_flag)) &&
 -                    s->ps.sps->sao_enabled)
 -                    restore_tqb_pixels(s);
++            if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
++                ctb_addr_ts = hls_slice_data_wpp(s, nal);
++            else
++                ctb_addr_ts = hls_slice_data(s);
+             if (ctb_addr_ts >= (s->ps.sps->ctb_width * s->ps.sps->ctb_height)) {
+                 s->is_decoded = 1;
 -                                s->nal_length_size, s->avctx->codec_id);
+             }
+             if (ctb_addr_ts < 0) {
+                 ret = ctb_addr_ts;
+                 goto fail;
+             }
+         }
+         break;
+     case NAL_EOS_NUT:
+     case NAL_EOB_NUT:
+         s->seq_decode = (s->seq_decode + 1) & 0xff;
+         s->max_ra     = INT_MAX;
+         break;
+     case NAL_AUD:
+     case NAL_FD_NUT:
+         break;
+     default:
+         av_log(s->avctx, AV_LOG_INFO,
+                "Skipping NAL unit %d\n", s->nal_unit_type);
+     }
+     return 0;
+ fail:
+     if (s->avctx->err_recognition & AV_EF_EXPLODE)
+         return ret;
+     return 0;
+ }
+ static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
+ {
+     int i, 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
+      * number of slices in the frame */
+     ret = ff_h2645_packet_split(&s->pkt, buf, length, s->avctx, s->is_nalff,
 -    if (s->ref)
++                                s->nal_length_size, s->avctx->codec_id, 1);
+     if (ret < 0) {
+         av_log(s->avctx, AV_LOG_ERROR,
+                "Error splitting the input into NAL units.\n");
+         return ret;
+     }
+     for (i = 0; i < s->pkt.nb_nals; i++) {
+         if (s->pkt.nals[i].type == NAL_EOB_NUT ||
+             s->pkt.nals[i].type == NAL_EOS_NUT)
+             s->eos = 1;
+     }
+     /* decode the NAL units */
+     for (i = 0; i < s->pkt.nb_nals; i++) {
+         ret = decode_nal_unit(s, &s->pkt.nals[i]);
+         if (ret < 0) {
+             av_log(s->avctx, AV_LOG_WARNING,
+                    "Error parsing NAL unit #%d.\n", i);
+             goto fail;
+         }
+     }
+ fail:
 -        if (s->ref && avctx->hwaccel->end_frame(avctx) < 0)
++    if (s->ref && s->threads_type == FF_THREAD_FRAME)
+         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
+     return ret;
+ }
+ static void print_md5(void *log_ctx, int level, uint8_t md5[16])
+ {
+     int i;
+     for (i = 0; i < 16; i++)
+         av_log(log_ctx, level, "%02"PRIx8, md5[i]);
+ }
+ static int verify_md5(HEVCContext *s, AVFrame *frame)
+ {
+     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
+     int pixel_shift;
+     int i, j;
+     if (!desc)
+         return AVERROR(EINVAL);
+     pixel_shift = desc->comp[0].depth > 8;
+     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
+            s->poc);
+     /* the checksums are LE, so we have to byteswap for >8bpp formats
+      * on BE arches */
+ #if HAVE_BIGENDIAN
+     if (pixel_shift && !s->checksum_buf) {
+         av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
+                        FFMAX3(frame->linesize[0], frame->linesize[1],
+                               frame->linesize[2]));
+         if (!s->checksum_buf)
+             return AVERROR(ENOMEM);
+     }
+ #endif
+     for (i = 0; frame->data[i]; i++) {
+         int width  = s->avctx->coded_width;
+         int height = s->avctx->coded_height;
+         int w = (i == 1 || i == 2) ? (width  >> desc->log2_chroma_w) : width;
+         int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
+         uint8_t md5[16];
+         av_md5_init(s->md5_ctx);
+         for (j = 0; j < h; j++) {
+             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
+ #if HAVE_BIGENDIAN
+             if (pixel_shift) {
+                 s->bdsp.bswap16_buf((uint16_t *) s->checksum_buf,
+                                     (const uint16_t *) src, w);
+                 src = s->checksum_buf;
+             }
+ #endif
+             av_md5_update(s->md5_ctx, src, w << pixel_shift);
+         }
+         av_md5_final(s->md5_ctx, md5);
+         if (!memcmp(md5, s->md5[i], 16)) {
+             av_log   (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
+             print_md5(s->avctx, AV_LOG_DEBUG, md5);
+             av_log   (s->avctx, AV_LOG_DEBUG, "; ");
+         } else {
+             av_log   (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
+             print_md5(s->avctx, AV_LOG_ERROR, md5);
+             av_log   (s->avctx, AV_LOG_ERROR, " != ");
+             print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
+             av_log   (s->avctx, AV_LOG_ERROR, "\n");
+             return AVERROR_INVALIDDATA;
+         }
+     }
+     av_log(s->avctx, AV_LOG_DEBUG, "\n");
+     return 0;
+ }
++static int hevc_decode_extradata(HEVCContext *s, uint8_t *buf, int length)
++{
++    AVCodecContext *avctx = s->avctx;
++    GetByteContext gb;
++    int ret, i;
++
++    bytestream2_init(&gb, buf, length);
++
++    if (length > 3 && (buf[0] || buf[1] || buf[2] > 1)) {
++        /* It seems the extradata is encoded as hvcC format.
++         * Temporarily, we support configurationVersion==0 until 14496-15 3rd
++         * is finalized. When finalized, configurationVersion will be 1 and we
++         * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
++        int i, j, num_arrays, nal_len_size;
++
++        s->is_nalff = 1;
++
++        bytestream2_skip(&gb, 21);
++        nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
++        num_arrays   = bytestream2_get_byte(&gb);
++
++        /* nal units in the hvcC always have length coded with 2 bytes,
++         * so put a fake nal_length_size = 2 while parsing them */
++        s->nal_length_size = 2;
++
++        /* Decode nal units from hvcC. */
++        for (i = 0; i < num_arrays; i++) {
++            int type = bytestream2_get_byte(&gb) & 0x3f;
++            int cnt  = bytestream2_get_be16(&gb);
++
++            for (j = 0; j < cnt; j++) {
++                // +2 for the nal size field
++                int nalsize = bytestream2_peek_be16(&gb) + 2;
++                if (bytestream2_get_bytes_left(&gb) < nalsize) {
++                    av_log(s->avctx, AV_LOG_ERROR,
++                           "Invalid NAL unit size in extradata.\n");
++                    return AVERROR_INVALIDDATA;
++                }
++
++                ret = decode_nal_units(s, gb.buffer, nalsize);
++                if (ret < 0) {
++                    av_log(avctx, AV_LOG_ERROR,
++                           "Decoding nal unit %d %d from hvcC failed\n",
++                           type, i);
++                    return ret;
++                }
++                bytestream2_skip(&gb, nalsize);
++            }
++        }
++
++        /* Now store right nal length size, that will be used to parse
++         * all other nals */
++        s->nal_length_size = nal_len_size;
++    } else {
++        s->is_nalff = 0;
++        ret = decode_nal_units(s, buf, length);
++        if (ret < 0)
++            return ret;
++    }
++
++    /* export stream parameters from the first SPS */
++    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
++        if (s->ps.sps_list[i]) {
++            const HEVCSPS *sps = (const HEVCSPS*)s->ps.sps_list[i]->data;
++            export_stream_params(s->avctx, &s->ps, sps);
++            break;
++        }
++    }
++
++    return 0;
++}
++
+ static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
+                              AVPacket *avpkt)
+ {
+     int ret;
++    int new_extradata_size;
++    uint8_t *new_extradata;
+     HEVCContext *s = avctx->priv_data;
+     if (!avpkt->size) {
+         ret = ff_hevc_output_frame(s, data, 1);
+         if (ret < 0)
+             return ret;
+         *got_output = ret;
+         return 0;
+     }
++    new_extradata = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
++                                            &new_extradata_size);
++    if (new_extradata && new_extradata_size > 0) {
++        ret = hevc_decode_extradata(s, new_extradata, new_extradata_size);
++        if (ret < 0)
++            return ret;
++    }
++
+     s->ref = NULL;
+     ret    = decode_nal_units(s, avpkt->data, avpkt->size);
+     if (ret < 0)
+         return ret;
+     if (avctx->hwaccel) {
 -    int ret = ff_thread_ref_frame(&dst->tf, &src->tf);
++        if (s->ref && (ret = avctx->hwaccel->end_frame(avctx)) < 0) {
+             av_log(avctx, AV_LOG_ERROR,
+                    "hardware accelerator failed to decode picture\n");
++            ff_hevc_unref_frame(s, s->ref, ~0);
++            return ret;
++        }
+     } else {
+         /* verify the SEI checksum */
+         if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
+             s->is_md5) {
+             ret = verify_md5(s, s->ref->frame);
+             if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
+                 ff_hevc_unref_frame(s, s->ref, ~0);
+                 return ret;
+             }
+         }
+     }
+     s->is_md5 = 0;
+     if (s->is_decoded) {
+         av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
+         s->is_decoded = 0;
+     }
+     if (s->output_frame->buf[0]) {
+         av_frame_move_ref(data, s->output_frame);
+         *got_output = 1;
+     }
+     return avpkt->size;
+ }
+ static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
+ {
 -    av_frame_free(&s->tmp_frame);
++    int ret;
++
++    ret = ff_thread_ref_frame(&dst->tf, &src->tf);
+     if (ret < 0)
+         return ret;
+     dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
+     if (!dst->tab_mvf_buf)
+         goto fail;
+     dst->tab_mvf = src->tab_mvf;
+     dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
+     if (!dst->rpl_tab_buf)
+         goto fail;
+     dst->rpl_tab = src->rpl_tab;
+     dst->rpl_buf = av_buffer_ref(src->rpl_buf);
+     if (!dst->rpl_buf)
+         goto fail;
+     dst->poc        = src->poc;
+     dst->ctb_count  = src->ctb_count;
+     dst->window     = src->window;
+     dst->flags      = src->flags;
+     dst->sequence   = src->sequence;
+     if (src->hwaccel_picture_private) {
+         dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
+         if (!dst->hwaccel_priv_buf)
+             goto fail;
+         dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
+     }
+     return 0;
+ fail:
+     ff_hevc_unref_frame(s, dst, ~0);
+     return AVERROR(ENOMEM);
+ }
+ static av_cold int hevc_decode_free(AVCodecContext *avctx)
+ {
+     HEVCContext       *s = avctx->priv_data;
+     int i;
+     pic_arrays_free(s);
+     av_freep(&s->md5_ctx);
 -    s->tmp_frame = av_frame_alloc();
 -    if (!s->tmp_frame)
++    av_freep(&s->cabac_state);
++
++    for (i = 0; i < 3; i++) {
++        av_freep(&s->sao_pixel_buffer_h[i]);
++        av_freep(&s->sao_pixel_buffer_v[i]);
++    }
+     av_frame_free(&s->output_frame);
+     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
+         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
+         av_frame_free(&s->DPB[i].frame);
+     }
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.vps_list); i++)
+         av_buffer_unref(&s->ps.vps_list[i]);
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++)
+         av_buffer_unref(&s->ps.sps_list[i]);
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.pps_list); i++)
+         av_buffer_unref(&s->ps.pps_list[i]);
++    s->ps.sps = NULL;
++    s->ps.pps = NULL;
++    s->ps.vps = NULL;
++
++    av_freep(&s->sh.entry_point_offset);
++    av_freep(&s->sh.offset);
++    av_freep(&s->sh.size);
++
++    for (i = 1; i < s->threads_number; i++) {
++        HEVCLocalContext *lc = s->HEVClcList[i];
++        if (lc) {
++            av_freep(&s->HEVClcList[i]);
++            av_freep(&s->sList[i]);
++        }
++    }
++    if (s->HEVClc == s->HEVClcList[0])
++        s->HEVClc = NULL;
++    av_freep(&s->HEVClcList[0]);
+     ff_h2645_packet_uninit(&s->pkt);
+     return 0;
+ }
+ static av_cold int hevc_init_context(AVCodecContext *avctx)
+ {
+     HEVCContext *s = avctx->priv_data;
+     int i;
+     s->avctx = avctx;
 -        ret = set_sps(s, s0->ps.sps);
++    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->output_frame = av_frame_alloc();
+     if (!s->output_frame)
+         goto fail;
+     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
+         s->DPB[i].frame = av_frame_alloc();
+         if (!s->DPB[i].frame)
+             goto fail;
+         s->DPB[i].tf.f = s->DPB[i].frame;
+     }
+     s->max_ra = INT_MAX;
+     s->md5_ctx = av_md5_alloc();
+     if (!s->md5_ctx)
+         goto fail;
+     ff_bswapdsp_init(&s->bdsp);
+     s->context_initialized = 1;
++    s->eos = 0;
++
++    ff_hevc_reset_sei(s);
+     return 0;
+ fail:
+     hevc_decode_free(avctx);
+     return AVERROR(ENOMEM);
+ }
+ static int hevc_update_thread_context(AVCodecContext *dst,
+                                       const AVCodecContext *src)
+ {
+     HEVCContext *s  = dst->priv_data;
+     HEVCContext *s0 = src->priv_data;
+     int i, ret;
+     if (!s->context_initialized) {
+         ret = hevc_init_context(dst);
+         if (ret < 0)
+             return ret;
+     }
+     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
+         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
+         if (s0->DPB[i].frame->buf[0]) {
+             ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
+             if (ret < 0)
+                 return ret;
+         }
+     }
++    if (s->ps.sps != s0->ps.sps)
++        s->ps.sps = NULL;
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.vps_list); i++) {
+         av_buffer_unref(&s->ps.vps_list[i]);
+         if (s0->ps.vps_list[i]) {
+             s->ps.vps_list[i] = av_buffer_ref(s0->ps.vps_list[i]);
+             if (!s->ps.vps_list[i])
+                 return AVERROR(ENOMEM);
+         }
+     }
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
+         av_buffer_unref(&s->ps.sps_list[i]);
+         if (s0->ps.sps_list[i]) {
+             s->ps.sps_list[i] = av_buffer_ref(s0->ps.sps_list[i]);
+             if (!s->ps.sps_list[i])
+                 return AVERROR(ENOMEM);
+         }
+     }
+     for (i = 0; i < FF_ARRAY_ELEMS(s->ps.pps_list); i++) {
+         av_buffer_unref(&s->ps.pps_list[i]);
+         if (s0->ps.pps_list[i]) {
+             s->ps.pps_list[i] = av_buffer_ref(s0->ps.pps_list[i]);
+             if (!s->ps.pps_list[i])
+                 return AVERROR(ENOMEM);
+         }
+     }
+     if (s->ps.sps != s0->ps.sps)
 -static int hevc_decode_extradata(HEVCContext *s)
 -{
 -    AVCodecContext *avctx = s->avctx;
 -    GetByteContext gb;
 -    int ret, i;
 -
 -    bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
 -
 -    if (avctx->extradata_size > 3 &&
 -        (avctx->extradata[0] || avctx->extradata[1] ||
 -         avctx->extradata[2] > 1)) {
 -        /* It seems the extradata is encoded as hvcC format.
 -         * Temporarily, we support configurationVersion==0 until 14496-15 3rd
 -         * is finalized. When finalized, configurationVersion will be 1 and we
 -         * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
 -        int i, j, num_arrays, nal_len_size;
 -
 -        s->is_nalff = 1;
 -
 -        bytestream2_skip(&gb, 21);
 -        nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
 -        num_arrays   = bytestream2_get_byte(&gb);
 -
 -        /* nal units in the hvcC always have length coded with 2 bytes,
 -         * so put a fake nal_length_size = 2 while parsing them */
 -        s->nal_length_size = 2;
 -
 -        /* Decode nal units from hvcC. */
 -        for (i = 0; i < num_arrays; i++) {
 -            int type = bytestream2_get_byte(&gb) & 0x3f;
 -            int cnt  = bytestream2_get_be16(&gb);
 -
 -            for (j = 0; j < cnt; j++) {
 -                // +2 for the nal size field
 -                int nalsize = bytestream2_peek_be16(&gb) + 2;
 -                if (bytestream2_get_bytes_left(&gb) < nalsize) {
 -                    av_log(s->avctx, AV_LOG_ERROR,
 -                           "Invalid NAL unit size in extradata.\n");
 -                    return AVERROR_INVALIDDATA;
 -                }
 -
 -                ret = decode_nal_units(s, gb.buffer, nalsize);
 -                if (ret < 0) {
 -                    av_log(avctx, AV_LOG_ERROR,
 -                           "Decoding nal unit %d %d from hvcC failed\n",
 -                           type, i);
 -                    return ret;
 -                }
 -                bytestream2_skip(&gb, nalsize);
 -            }
 -        }
 -
 -        /* Now store right nal length size, that will be used to parse
 -         * all other nals */
 -        s->nal_length_size = nal_len_size;
 -    } else {
 -        s->is_nalff = 0;
 -        ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
 -        if (ret < 0)
 -            return ret;
 -    }
 -
 -    /* export stream parameters from the first SPS */
 -    for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
 -        if (s->ps.sps_list[i]) {
 -            const HEVCSPS *sps = (const HEVCSPS*)s->ps.sps_list[i]->data;
 -            export_stream_params(s->avctx, &s->ps, sps);
 -            break;
 -        }
 -    }
 -
 -    return 0;
 -}
 -
++        if ((ret = set_sps(s, s0->ps.sps, src->pix_fmt)) < 0)
++            return ret;
+     s->seq_decode = s0->seq_decode;
+     s->seq_output = s0->seq_output;
+     s->pocTid0    = s0->pocTid0;
+     s->max_ra     = s0->max_ra;
++    s->eos        = s0->eos;
++    s->no_rasl_output_flag = s0->no_rasl_output_flag;
+     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;
+     }
+     return 0;
+ }
 -        ret = hevc_decode_extradata(s);
+ static av_cold int hevc_decode_init(AVCodecContext *avctx)
+ {
+     HEVCContext *s = avctx->priv_data;
+     int ret;
+     avctx->internal->allocate_progress = 1;
+     ret = hevc_init_context(avctx);
+     if (ret < 0)
+         return ret;
++    s->enable_parallel_tiles = 0;
++    s->picture_struct = 0;
++    s->eos = 1;
++
++    atomic_init(&s->wpp_err, 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) {
 -        AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
++        ret = hevc_decode_extradata(s, avctx->extradata, avctx->extradata_size);
+         if (ret < 0) {
+             hevc_decode_free(avctx);
+             return ret;
+         }
+     }
++    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;
+ }
+ static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
+ {
+     HEVCContext *s = avctx->priv_data;
+     int ret;
+     memset(s, 0, sizeof(*s));
+     ret = hevc_init_context(avctx);
+     if (ret < 0)
+         return ret;
+     return 0;
+ }
+ static void hevc_decode_flush(AVCodecContext *avctx)
+ {
+     HEVCContext *s = avctx->priv_data;
+     ff_hevc_flush_dpb(s);
+     s->max_ra = INT_MAX;
++    s->eos = 1;
+ }
+ #define OFFSET(x) offsetof(HEVCContext, x)
+ #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
+ static const AVOption options[] = {
+     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
 -                             AV_CODEC_CAP_FRAME_THREADS,
++        AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, PAR },
++    { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
++        AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, PAR },
+     { NULL },
+ };
+ static const AVClass hevc_decoder_class = {
+     .class_name = "HEVC decoder",
+     .item_name  = av_default_item_name,
+     .option     = options,
+     .version    = LIBAVUTIL_VERSION_INT,
+ };
+ AVCodec ff_hevc_decoder = {
+     .name                  = "hevc",
+     .long_name             = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
+     .type                  = AVMEDIA_TYPE_VIDEO,
+     .id                    = AV_CODEC_ID_HEVC,
+     .priv_data_size        = sizeof(HEVCContext),
+     .priv_class            = &hevc_decoder_class,
+     .init                  = hevc_decode_init,
+     .close                 = hevc_decode_free,
+     .decode                = hevc_decode_frame,
+     .flush                 = hevc_decode_flush,
+     .update_thread_context = hevc_update_thread_context,
+     .init_thread_copy      = hevc_init_thread_copy,
+     .capabilities          = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
++                             AV_CODEC_CAP_SLICE_THREADS | AV_CODEC_CAP_FRAME_THREADS,
++    .caps_internal         = FF_CODEC_CAP_INIT_THREADSAFE,
+     .profiles              = NULL_IF_CONFIG_SMALL(ff_hevc_profiles),
+ };
index 0000000,ba261e6..519c7a6
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,1016 +1,1086 @@@
 - * This file is part of Libav.
+ /*
+  * HEVC video decoder
+  *
+  * Copyright (C) 2012 - 2013 Guillaume Martres
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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
 -#include <stddef.h>
 -#include <stdint.h>
++ * 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_HEVCDEC_H
+ #define AVCODEC_HEVCDEC_H
 -#define MAX_PB_SIZE 64
++#include <stdatomic.h>
+ #include "libavutil/buffer.h"
+ #include "libavutil/md5.h"
+ #include "avcodec.h"
+ #include "bswapdsp.h"
+ #include "cabac.h"
+ #include "get_bits.h"
++#include "hevcpred.h"
+ #include "h2645_parse.h"
+ #include "hevcdsp.h"
+ #include "internal.h"
+ #include "thread.h"
+ #include "videodsp.h"
+ #define MAX_DPB_SIZE 16 // A.4.1
+ #define MAX_REFS 16
++#define MAX_NB_THREADS 16
++#define SHIFT_CTB_WPP 2
++
+ /**
+  * 7.4.2.1
+  */
+ #define MAX_SUB_LAYERS 7
+ #define MAX_VPS_COUNT 16
+ #define MAX_SPS_COUNT 32
+ #define MAX_PPS_COUNT 256
+ #define MAX_SHORT_TERM_RPS_COUNT 64
+ #define MAX_CU_SIZE 128
+ //TODO: check if this is really the maximum
+ #define MAX_TRANSFORM_DEPTH 5
+ #define MAX_TB_SIZE 32
 -#define HEVC_CONTEXTS 183
+ #define MAX_LOG2_CTB_SIZE 6
+ #define MAX_QP 51
+ #define DEFAULT_INTRA_TC_OFFSET 2
 -#define IS_IDR(s) (s->nal_unit_type == NAL_IDR_W_RADL || s->nal_unit_type == NAL_IDR_N_LP)
 -#define IS_BLA(s) (s->nal_unit_type == NAL_BLA_W_RADL || s->nal_unit_type == NAL_BLA_W_LP || \
 -                   s->nal_unit_type == NAL_BLA_N_LP)
 -#define IS_IRAP(s) (s->nal_unit_type >= 16 && s->nal_unit_type <= 23)
 -
 -#define FFUDIV(a,b) (((a) > 0 ? (a) : (a) - (b) + 1) / (b))
 -#define FFUMOD(a,b) ((a) - (b) * FFUDIV(a,b))
++#define HEVC_CONTEXTS 199
+ #define MRG_MAX_NUM_CANDS     5
+ #define L0 0
+ #define L1 1
+ #define EPEL_EXTRA_BEFORE 1
+ #define EPEL_EXTRA_AFTER  2
+ #define EPEL_EXTRA        3
++#define QPEL_EXTRA_BEFORE 3
++#define QPEL_EXTRA_AFTER  4
++#define QPEL_EXTRA        7
+ #define EDGE_EMU_BUFFER_STRIDE 80
+ /**
+  * Value of the luma sample at position (x, y) in the 2D array tab.
+  */
+ #define SAMPLE(tab, x, y) ((tab)[(y) * s->sps->width + (x)])
+ #define SAMPLE_CTB(tab, x, y) ((tab)[(y) * min_cb_width + (x)])
 -    int vps_id;
++#define IS_IDR(s) ((s)->nal_unit_type == NAL_IDR_W_RADL || (s)->nal_unit_type == NAL_IDR_N_LP)
++#define IS_BLA(s) ((s)->nal_unit_type == NAL_BLA_W_RADL || (s)->nal_unit_type == NAL_BLA_W_LP || \
++                   (s)->nal_unit_type == NAL_BLA_N_LP)
++#define IS_IRAP(s) ((s)->nal_unit_type >= 16 && (s)->nal_unit_type <= 23)
+ /**
+  * Table 7-3: NAL unit type codes
+  */
+ enum NALUnitType {
+     NAL_TRAIL_N    = 0,
+     NAL_TRAIL_R    = 1,
+     NAL_TSA_N      = 2,
+     NAL_TSA_R      = 3,
+     NAL_STSA_N     = 4,
+     NAL_STSA_R     = 5,
+     NAL_RADL_N     = 6,
+     NAL_RADL_R     = 7,
+     NAL_RASL_N     = 8,
+     NAL_RASL_R     = 9,
+     NAL_BLA_W_LP   = 16,
+     NAL_BLA_W_RADL = 17,
+     NAL_BLA_N_LP   = 18,
+     NAL_IDR_W_RADL = 19,
+     NAL_IDR_N_LP   = 20,
+     NAL_CRA_NUT    = 21,
+     NAL_VPS        = 32,
+     NAL_SPS        = 33,
+     NAL_PPS        = 34,
+     NAL_AUD        = 35,
+     NAL_EOS_NUT    = 36,
+     NAL_EOB_NUT    = 37,
+     NAL_FD_NUT     = 38,
+     NAL_SEI_PREFIX = 39,
+     NAL_SEI_SUFFIX = 40,
+ };
+ enum RPSType {
+     ST_CURR_BEF = 0,
+     ST_CURR_AFT,
+     ST_FOLL,
+     LT_CURR,
+     LT_FOLL,
+     NB_RPS_TYPE,
+ };
+ enum SliceType {
+     B_SLICE = 0,
+     P_SLICE = 1,
+     I_SLICE = 2,
+ };
+ enum SyntaxElement {
+     SAO_MERGE_FLAG = 0,
+     SAO_TYPE_IDX,
+     SAO_EO_CLASS,
+     SAO_BAND_POSITION,
+     SAO_OFFSET_ABS,
+     SAO_OFFSET_SIGN,
+     END_OF_SLICE_FLAG,
+     SPLIT_CODING_UNIT_FLAG,
+     CU_TRANSQUANT_BYPASS_FLAG,
+     SKIP_FLAG,
+     CU_QP_DELTA,
+     PRED_MODE_FLAG,
+     PART_MODE,
+     PCM_FLAG,
+     PREV_INTRA_LUMA_PRED_FLAG,
+     MPM_IDX,
+     REM_INTRA_LUMA_PRED_MODE,
+     INTRA_CHROMA_PRED_MODE,
+     MERGE_FLAG,
+     MERGE_IDX,
+     INTER_PRED_IDC,
+     REF_IDX_L0,
+     REF_IDX_L1,
+     ABS_MVD_GREATER0_FLAG,
+     ABS_MVD_GREATER1_FLAG,
+     ABS_MVD_MINUS2,
+     MVD_SIGN_FLAG,
+     MVP_LX_FLAG,
+     NO_RESIDUAL_DATA_FLAG,
+     SPLIT_TRANSFORM_FLAG,
+     CBF_LUMA,
+     CBF_CB_CR,
+     TRANSFORM_SKIP_FLAG,
++    EXPLICIT_RDPCM_FLAG,
++    EXPLICIT_RDPCM_DIR_FLAG,
+     LAST_SIGNIFICANT_COEFF_X_PREFIX,
+     LAST_SIGNIFICANT_COEFF_Y_PREFIX,
+     LAST_SIGNIFICANT_COEFF_X_SUFFIX,
+     LAST_SIGNIFICANT_COEFF_Y_SUFFIX,
+     SIGNIFICANT_COEFF_GROUP_FLAG,
+     SIGNIFICANT_COEFF_FLAG,
+     COEFF_ABS_LEVEL_GREATER1_FLAG,
+     COEFF_ABS_LEVEL_GREATER2_FLAG,
+     COEFF_ABS_LEVEL_REMAINING,
+     COEFF_SIGN_FLAG,
++    LOG2_RES_SCALE_ABS,
++    RES_SCALE_SIGN_FLAG,
++    CU_CHROMA_QP_OFFSET_FLAG,
++    CU_CHROMA_QP_OFFSET_IDX,
+ };
+ enum PartMode {
+     PART_2Nx2N = 0,
+     PART_2NxN  = 1,
+     PART_Nx2N  = 2,
+     PART_NxN   = 3,
+     PART_2NxnU = 4,
+     PART_2NxnD = 5,
+     PART_nLx2N = 6,
+     PART_nRx2N = 7,
+ };
+ enum PredMode {
+     MODE_INTER = 0,
+     MODE_INTRA,
+     MODE_SKIP,
+ };
+ enum InterPredIdc {
+     PRED_L0 = 0,
+     PRED_L1,
+     PRED_BI,
+ };
++enum PredFlag {
++    PF_INTRA = 0,
++    PF_L0,
++    PF_L1,
++    PF_BI,
++};
++
+ enum IntraPredMode {
+     INTRA_PLANAR = 0,
+     INTRA_DC,
+     INTRA_ANGULAR_2,
+     INTRA_ANGULAR_3,
+     INTRA_ANGULAR_4,
+     INTRA_ANGULAR_5,
+     INTRA_ANGULAR_6,
+     INTRA_ANGULAR_7,
+     INTRA_ANGULAR_8,
+     INTRA_ANGULAR_9,
+     INTRA_ANGULAR_10,
+     INTRA_ANGULAR_11,
+     INTRA_ANGULAR_12,
+     INTRA_ANGULAR_13,
+     INTRA_ANGULAR_14,
+     INTRA_ANGULAR_15,
+     INTRA_ANGULAR_16,
+     INTRA_ANGULAR_17,
+     INTRA_ANGULAR_18,
+     INTRA_ANGULAR_19,
+     INTRA_ANGULAR_20,
+     INTRA_ANGULAR_21,
+     INTRA_ANGULAR_22,
+     INTRA_ANGULAR_23,
+     INTRA_ANGULAR_24,
+     INTRA_ANGULAR_25,
+     INTRA_ANGULAR_26,
+     INTRA_ANGULAR_27,
+     INTRA_ANGULAR_28,
+     INTRA_ANGULAR_29,
+     INTRA_ANGULAR_30,
+     INTRA_ANGULAR_31,
+     INTRA_ANGULAR_32,
+     INTRA_ANGULAR_33,
+     INTRA_ANGULAR_34,
+ };
+ enum SAOType {
+     SAO_NOT_APPLIED = 0,
+     SAO_BAND,
+     SAO_EDGE,
++    SAO_APPLIED
+ };
+ enum SAOEOClass {
+     SAO_EO_HORIZ = 0,
+     SAO_EO_VERT,
+     SAO_EO_135D,
+     SAO_EO_45D,
+ };
+ enum ScanType {
+     SCAN_DIAG = 0,
+     SCAN_HORIZ,
+     SCAN_VERT,
+ };
+ typedef struct ShortTermRPS {
+     unsigned int num_negative_pics;
+     int num_delta_pocs;
+     int rps_idx_num_delta_pocs;
+     int32_t delta_poc[32];
+     uint8_t used[32];
+ } ShortTermRPS;
+ typedef struct LongTermRPS {
+     int     poc[32];
+     uint8_t used[32];
+     uint8_t nb_refs;
+ } LongTermRPS;
+ typedef struct RefPicList {
+     struct HEVCFrame *ref[MAX_REFS];
+     int list[MAX_REFS];
+     int isLongTerm[MAX_REFS];
+     int nb_refs;
+ } RefPicList;
+ typedef struct RefPicListTab {
+     RefPicList refPicList[2];
+ } RefPicListTab;
+ typedef struct HEVCWindow {
+     unsigned int left_offset;
+     unsigned int right_offset;
+     unsigned int top_offset;
+     unsigned int bottom_offset;
+ } HEVCWindow;
+ typedef struct VUI {
+     AVRational sar;
+     int overscan_info_present_flag;
+     int overscan_appropriate_flag;
+     int video_signal_type_present_flag;
+     int video_format;
+     int video_full_range_flag;
+     int colour_description_present_flag;
+     uint8_t colour_primaries;
+     uint8_t transfer_characteristic;
+     uint8_t matrix_coeffs;
+     int chroma_loc_info_present_flag;
+     int chroma_sample_loc_type_top_field;
+     int chroma_sample_loc_type_bottom_field;
+     int neutra_chroma_indication_flag;
+     int field_seq_flag;
+     int frame_field_info_present_flag;
+     int default_display_window_flag;
+     HEVCWindow def_disp_win;
+     int vui_timing_info_present_flag;
+     uint32_t vui_num_units_in_tick;
+     uint32_t vui_time_scale;
+     int vui_poc_proportional_to_timing_flag;
+     int vui_num_ticks_poc_diff_one_minus1;
+     int vui_hrd_parameters_present_flag;
+     int bitstream_restriction_flag;
+     int tiles_fixed_structure_flag;
+     int motion_vectors_over_pic_boundaries_flag;
+     int restricted_ref_pic_lists_flag;
+     int min_spatial_segmentation_idc;
+     int max_bytes_per_pic_denom;
+     int max_bits_per_min_cu_denom;
+     int log2_max_mv_length_horizontal;
+     int log2_max_mv_length_vertical;
+ } VUI;
+ typedef struct PTLCommon {
+     uint8_t profile_space;
+     uint8_t tier_flag;
+     uint8_t profile_idc;
+     uint8_t profile_compatibility_flag[32];
+     uint8_t level_idc;
+     uint8_t progressive_source_flag;
+     uint8_t interlaced_source_flag;
+     uint8_t non_packed_constraint_flag;
+     uint8_t frame_only_constraint_flag;
+ } PTLCommon;
+ typedef struct PTL {
+     PTLCommon general_ptl;
+     PTLCommon sub_layer_ptl[MAX_SUB_LAYERS];
+     uint8_t sub_layer_profile_present_flag[MAX_SUB_LAYERS];
+     uint8_t sub_layer_level_present_flag[MAX_SUB_LAYERS];
+ } PTL;
+ typedef struct HEVCVPS {
+     uint8_t vps_temporal_id_nesting_flag;
+     int vps_max_layers;
+     int vps_max_sub_layers; ///< vps_max_temporal_layers_minus1 + 1
+     PTL ptl;
+     int vps_sub_layer_ordering_info_present_flag;
+     unsigned int vps_max_dec_pic_buffering[MAX_SUB_LAYERS];
+     unsigned int vps_num_reorder_pics[MAX_SUB_LAYERS];
+     unsigned int vps_max_latency_increase[MAX_SUB_LAYERS];
+     int vps_max_layer_id;
+     int vps_num_layer_sets; ///< vps_num_layer_sets_minus1 + 1
+     uint8_t vps_timing_info_present_flag;
+     uint32_t vps_num_units_in_tick;
+     uint32_t vps_time_scale;
+     uint8_t vps_poc_proportional_to_timing_flag;
+     int vps_num_ticks_poc_diff_one; ///< vps_num_ticks_poc_diff_one_minus1 + 1
+     int vps_num_hrd_parameters;
++
++    uint8_t data[4096];
++    int data_size;
+ } HEVCVPS;
+ typedef struct ScalingList {
+     /* This is a little wasteful, since sizeID 0 only needs 8 coeffs,
+      * and size ID 3 only has 2 arrays, not 6. */
+     uint8_t sl[4][6][64];
+     uint8_t sl_dc[2][6];
+ } ScalingList;
+ typedef struct HEVCSPS {
 -typedef struct CodingTree {
 -    int depth; ///< ctDepth
 -} CodingTree;
 -
++    unsigned vps_id;
+     int chroma_format_idc;
+     uint8_t separate_colour_plane_flag;
+     ///< output (i.e. cropped) values
+     int output_width, output_height;
+     HEVCWindow output_window;
+     HEVCWindow pic_conf_win;
+     int bit_depth;
+     int pixel_shift;
+     enum AVPixelFormat pix_fmt;
+     unsigned int log2_max_poc_lsb;
+     int pcm_enabled_flag;
+     int max_sub_layers;
+     struct {
+         int max_dec_pic_buffering;
+         int num_reorder_pics;
+         int max_latency_increase;
+     } temporal_layer[MAX_SUB_LAYERS];
+     VUI vui;
+     PTL ptl;
+     uint8_t scaling_list_enable_flag;
+     ScalingList scaling_list;
+     unsigned int nb_st_rps;
+     ShortTermRPS st_rps[MAX_SHORT_TERM_RPS_COUNT];
+     uint8_t amp_enabled_flag;
+     uint8_t sao_enabled;
+     uint8_t long_term_ref_pics_present_flag;
+     uint16_t lt_ref_pic_poc_lsb_sps[32];
+     uint8_t used_by_curr_pic_lt_sps_flag[32];
+     uint8_t num_long_term_ref_pics_sps;
+     struct {
+         uint8_t bit_depth;
+         uint8_t bit_depth_chroma;
+         unsigned int log2_min_pcm_cb_size;
+         unsigned int log2_max_pcm_cb_size;
+         uint8_t loop_filter_disable_flag;
+     } pcm;
+     uint8_t sps_temporal_mvp_enabled_flag;
+     uint8_t sps_strong_intra_smoothing_enable_flag;
+     unsigned int log2_min_cb_size;
+     unsigned int log2_diff_max_min_coding_block_size;
+     unsigned int log2_min_tb_size;
+     unsigned int log2_max_trafo_size;
+     unsigned int log2_ctb_size;
+     unsigned int log2_min_pu_size;
+     int max_transform_hierarchy_depth_inter;
+     int max_transform_hierarchy_depth_intra;
++    int transform_skip_rotation_enabled_flag;
++    int transform_skip_context_enabled_flag;
++    int implicit_rdpcm_enabled_flag;
++    int explicit_rdpcm_enabled_flag;
++    int intra_smoothing_disabled_flag;
++    int persistent_rice_adaptation_enabled_flag;
++
+     ///< coded frame dimension in various units
+     int width;
+     int height;
+     int ctb_width;
+     int ctb_height;
+     int ctb_size;
+     int min_cb_width;
+     int min_cb_height;
+     int min_tb_width;
+     int min_tb_height;
+     int min_pu_width;
+     int min_pu_height;
++    int tb_mask;
+     int hshift[3];
+     int vshift[3];
+     int qp_bd_offset;
++
++    uint8_t data[4096];
++    int data_size;
+ } HEVCSPS;
+ typedef struct HEVCPPS {
+     unsigned int sps_id; ///< seq_parameter_set_id
+     uint8_t sign_data_hiding_flag;
+     uint8_t cabac_init_present_flag;
+     int num_ref_idx_l0_default_active; ///< num_ref_idx_l0_default_active_minus1 + 1
+     int num_ref_idx_l1_default_active; ///< num_ref_idx_l1_default_active_minus1 + 1
+     int pic_init_qp_minus26;
+     uint8_t constrained_intra_pred_flag;
+     uint8_t transform_skip_enabled_flag;
+     uint8_t cu_qp_delta_enabled_flag;
+     int diff_cu_qp_delta_depth;
+     int cb_qp_offset;
+     int cr_qp_offset;
+     uint8_t pic_slice_level_chroma_qp_offsets_present_flag;
+     uint8_t weighted_pred_flag;
+     uint8_t weighted_bipred_flag;
+     uint8_t output_flag_present_flag;
+     uint8_t transquant_bypass_enable_flag;
+     uint8_t dependent_slice_segments_enabled_flag;
+     uint8_t tiles_enabled_flag;
+     uint8_t entropy_coding_sync_enabled_flag;
+     int num_tile_columns;   ///< num_tile_columns_minus1 + 1
+     int num_tile_rows;      ///< num_tile_rows_minus1 + 1
+     uint8_t uniform_spacing_flag;
+     uint8_t loop_filter_across_tiles_enabled_flag;
+     uint8_t seq_loop_filter_across_slices_enabled_flag;
+     uint8_t deblocking_filter_control_present_flag;
+     uint8_t deblocking_filter_override_enabled_flag;
+     uint8_t disable_dbf;
+     int beta_offset;    ///< beta_offset_div2 * 2
+     int tc_offset;      ///< tc_offset_div2 * 2
+     uint8_t scaling_list_data_present_flag;
+     ScalingList scaling_list;
+     uint8_t lists_modification_present_flag;
+     int log2_parallel_merge_level; ///< log2_parallel_merge_level_minus2 + 2
+     int num_extra_slice_header_bits;
+     uint8_t slice_header_extension_present_flag;
++    uint8_t log2_max_transform_skip_block_size;
++    uint8_t cross_component_prediction_enabled_flag;
++    uint8_t chroma_qp_offset_list_enabled_flag;
++    uint8_t diff_cu_chroma_qp_offset_depth;
++    uint8_t chroma_qp_offset_list_len_minus1;
++    int8_t  cb_qp_offset_list[5];
++    int8_t  cr_qp_offset_list[5];
++    uint8_t log2_sao_offset_scale_luma;
++    uint8_t log2_sao_offset_scale_chroma;
+     // Inferred parameters
+     unsigned int *column_width;  ///< ColumnWidth
+     unsigned int *row_height;    ///< RowHeight
+     unsigned int *col_bd;        ///< ColBd
+     unsigned int *row_bd;        ///< RowBd
+     int *col_idxX;
+     int *ctb_addr_rs_to_ts; ///< CtbAddrRSToTS
+     int *ctb_addr_ts_to_rs; ///< CtbAddrTSToRS
+     int *tile_id;           ///< TileId
+     int *tile_pos_rs;       ///< TilePosRS
+     int *min_tb_addr_zs;    ///< MinTbAddrZS
++    int *min_tb_addr_zs_tab;///< MinTbAddrZS
++
++    uint8_t data[4096];
++    int data_size;
+ } HEVCPPS;
+ typedef struct HEVCParamSets {
+     AVBufferRef *vps_list[MAX_VPS_COUNT];
+     AVBufferRef *sps_list[MAX_SPS_COUNT];
+     AVBufferRef *pps_list[MAX_PPS_COUNT];
+     /* currently active parameter sets */
+     const HEVCVPS *vps;
+     const HEVCSPS *sps;
+     const HEVCPPS *pps;
+ } HEVCParamSets;
+ typedef struct SliceHeader {
+     unsigned int pps_id;
+     ///< address (in raster order) of the first block in the current slice segment
+     unsigned int   slice_segment_addr;
+     ///< address (in raster order) of the first block in the current slice
+     unsigned int   slice_addr;
+     enum SliceType slice_type;
+     int pic_order_cnt_lsb;
+     uint8_t first_slice_in_pic_flag;
+     uint8_t dependent_slice_segment_flag;
+     uint8_t pic_output_flag;
+     uint8_t colour_plane_id;
+     ///< RPS coded in the slice header itself is stored here
+     int short_term_ref_pic_set_sps_flag;
+     int short_term_ref_pic_set_size;
+     ShortTermRPS slice_rps;
+     const ShortTermRPS *short_term_rps;
+     int long_term_ref_pic_set_size;
+     LongTermRPS long_term_rps;
+     unsigned int list_entry_lx[2][32];
+     uint8_t rpl_modification_flag[2];
+     uint8_t no_output_of_prior_pics_flag;
+     uint8_t slice_temporal_mvp_enabled_flag;
+     unsigned int nb_refs[2];
+     uint8_t slice_sample_adaptive_offset_flag[3];
+     uint8_t mvd_l1_zero_flag;
+     uint8_t cabac_init_flag;
+     uint8_t disable_deblocking_filter_flag; ///< slice_header_disable_deblocking_filter_flag
+     uint8_t slice_loop_filter_across_slices_enabled_flag;
+     uint8_t collocated_list;
+     unsigned int collocated_ref_idx;
+     int slice_qp_delta;
+     int slice_cb_qp_offset;
+     int slice_cr_qp_offset;
++    uint8_t cu_chroma_qp_offset_enabled_flag;
++
+     int beta_offset;    ///< beta_offset_div2 * 2
+     int tc_offset;      ///< tc_offset_div2 * 2
+     unsigned int max_num_merge_cand; ///< 5 - 5_minus_max_num_merge_cand
++    unsigned *entry_point_offset;
++    int * offset;
++    int * size;
+     int num_entry_point_offsets;
+     int8_t slice_qp;
+     uint8_t luma_log2_weight_denom;
+     int16_t chroma_log2_weight_denom;
+     int16_t luma_weight_l0[16];
+     int16_t chroma_weight_l0[16][2];
+     int16_t chroma_weight_l1[16][2];
+     int16_t luma_weight_l1[16];
+     int16_t luma_offset_l0[16];
+     int16_t chroma_offset_l0[16][2];
+     int16_t luma_offset_l1[16];
+     int16_t chroma_offset_l1[16][2];
+     int slice_ctb_addr_rs;
+ } SliceHeader;
 -    int8_t pred_flag[2];
 -    uint8_t is_intra;
+ typedef struct CodingUnit {
+     int x;
+     int y;
+     enum PredMode pred_mode;    ///< PredMode
+     enum PartMode part_mode;    ///< PartMode
+     // Inferred parameters
+     uint8_t intra_split_flag;   ///< IntraSplitFlag
+     uint8_t max_trafo_depth;    ///< MaxTrafoDepth
+     uint8_t cu_transquant_bypass_flag;
+ } CodingUnit;
+ typedef struct Mv {
+     int16_t x;  ///< horizontal component of motion vector
+     int16_t y;  ///< vertical component of motion vector
+ } Mv;
+ typedef struct MvField {
+     DECLARE_ALIGNED(4, Mv, mv)[2];
+     int8_t ref_idx[2];
 -    uint8_t intra_pred_mode_c;
++    int8_t pred_flag;
+ } MvField;
+ typedef struct NeighbourAvailable {
+     int cand_bottom_left;
+     int cand_left;
+     int cand_up;
+     int cand_up_left;
+     int cand_up_right;
+     int cand_up_right_sap;
+ } NeighbourAvailable;
+ typedef struct PredictionUnit {
+     int mpm_idx;
+     int rem_intra_luma_pred_mode;
+     uint8_t intra_pred_mode[4];
+     Mv mvd;
+     uint8_t merge_flag;
 -    int cur_intra_pred_mode;
++    uint8_t intra_pred_mode_c[4];
++    uint8_t chroma_mode_c[4];
+ } PredictionUnit;
+ typedef struct TransformUnit {
+     int cu_qp_delta;
++    int res_scale_val;
++
+     // Inferred parameters;
 -struct HEVCContext;
 -
 -typedef struct HEVCPredContext {
 -    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;
 -
++    int intra_pred_mode;
++    int intra_pred_mode_c;
++    int chroma_mode_c;
+     uint8_t is_cu_qp_delta_coded;
++    uint8_t is_cu_chroma_qp_offset_coded;
++    int8_t  cu_qp_offset_cb;
++    int8_t  cu_qp_offset_cr;
++    uint8_t cross_pf;
+ } TransformUnit;
+ typedef struct DBParams {
+     int beta_offset;
+     int tc_offset;
+ } DBParams;
+ #define HEVC_FRAME_FLAG_OUTPUT    (1 << 0)
+ #define HEVC_FRAME_FLAG_SHORT_REF (1 << 1)
+ #define HEVC_FRAME_FLAG_LONG_REF  (1 << 2)
++#define HEVC_FRAME_FLAG_BUMPING   (1 << 3)
+ typedef struct HEVCFrame {
+     AVFrame *frame;
+     ThreadFrame tf;
+     MvField *tab_mvf;
+     RefPicList *refPicList;
+     RefPicListTab **rpl_tab;
+     int ctb_count;
+     int poc;
+     struct HEVCFrame *collocated_ref;
+     HEVCWindow window;
+     AVBufferRef *tab_mvf_buf;
+     AVBufferRef *rpl_tab_buf;
+     AVBufferRef *rpl_buf;
+     AVBufferRef *hwaccel_priv_buf;
+     void *hwaccel_picture_private;
+     /**
+      * A sequence counter, so that old frames are output first
+      * after a POC reset
+      */
+     uint16_t sequence;
+     /**
+      * A combination of HEVC_FRAME_FLAG_*
+      */
+     uint8_t flags;
+ } HEVCFrame;
 -    DECLARE_ALIGNED(16, int16_t, mc_buffer[(MAX_PB_SIZE + 24) * MAX_PB_SIZE]);
+ typedef struct HEVCLocalContext {
 -    int     start_of_tiles_x;
+     uint8_t cabac_state[HEVC_CONTEXTS];
++    uint8_t stat_coeff[4];
++
+     uint8_t first_qp_group;
+     GetBitContext gb;
+     CABACContext cc;
+     int8_t qp_y;
+     int8_t curr_qp_y;
++    int qPy_pred;
++
+     TransformUnit tu;
+     uint8_t ctb_left_flag;
+     uint8_t ctb_up_flag;
+     uint8_t ctb_up_right_flag;
+     uint8_t ctb_up_left_flag;
 -    CodingTree ct;
+     int     end_of_tiles_x;
+     int     end_of_tiles_y;
+     /* +7 is for subpixel interpolation, *2 for high bit depths */
+     DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
 -    HEVCLocalContext HEVClc;
++    /* The extended size between the new edge emu buffer is abused by SAO */
++    DECLARE_ALIGNED(32, uint8_t, edge_emu_buffer2)[(MAX_PB_SIZE + 7) * EDGE_EMU_BUFFER_STRIDE * 2];
++    DECLARE_ALIGNED(32, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
++
++    int ct_depth;
+     CodingUnit cu;
+     PredictionUnit pu;
+     NeighbourAvailable na;
+ #define BOUNDARY_LEFT_SLICE     (1 << 0)
+ #define BOUNDARY_LEFT_TILE      (1 << 1)
+ #define BOUNDARY_UPPER_SLICE    (1 << 2)
+ #define BOUNDARY_UPPER_TILE     (1 << 3)
+     /* properties of the boundary of the current CTB for the purposes
+      * of the deblocking filter */
+     int boundary_flags;
+ } HEVCLocalContext;
+ typedef struct HEVCContext {
+     const AVClass *c;  // needed by private avoptions
+     AVCodecContext *avctx;
 -    uint8_t cabac_state[HEVC_CONTEXTS];
++    struct HEVCContext  *sList[MAX_NB_THREADS];
 -    AVFrame *sao_frame;
 -    AVFrame *tmp_frame;
++    HEVCLocalContext    *HEVClcList[MAX_NB_THREADS];
++    HEVCLocalContext    *HEVClc;
++
++    uint8_t             threads_type;
++    uint8_t             threads_number;
++
++    int                 width;
++    int                 height;
++
++    uint8_t *cabac_state;
+     /** 1 if the independent slice segment header was successfully parsed */
+     uint8_t slice_initialized;
+     AVFrame *frame;
 -int ff_hevc_abs_mvd_greater0_flag_decode(HEVCContext *s);
 -int ff_hevc_abs_mvd_greater1_flag_decode(HEVCContext *s);
 -int ff_hevc_mvd_decode(HEVCContext *s);
 -int ff_hevc_mvd_sign_flag_decode(HEVCContext *s);
+     AVFrame *output_frame;
++    uint8_t *sao_pixel_buffer_h[3];
++    uint8_t *sao_pixel_buffer_v[3];
+     HEVCParamSets ps;
+     AVBufferPool *tab_mvf_pool;
+     AVBufferPool *rpl_tab_pool;
+     ///< candidate references for the current frame
+     RefPicList rps[5];
+     SliceHeader sh;
+     SAOParams *sao;
+     DBParams *deblock;
+     enum NALUnitType nal_unit_type;
+     int temporal_id;  ///< temporal_id_plus1 - 1
+     HEVCFrame *ref;
+     HEVCFrame DPB[32];
+     int poc;
+     int pocTid0;
+     int slice_idx; ///< number of the slice being currently decoded
+     int eos;       ///< current packet contains an EOS/EOB NAL
++    int last_eos;  ///< last packet contains an EOS/EOB NAL
+     int max_ra;
+     int bs_width;
+     int bs_height;
+     int is_decoded;
++    int no_rasl_output_flag;
+     HEVCPredContext hpc;
+     HEVCDSPContext hevcdsp;
+     VideoDSPContext vdsp;
+     BswapDSPContext bdsp;
+     int8_t *qp_y_tab;
+     uint8_t *horizontal_bs;
+     uint8_t *vertical_bs;
+     int32_t *tab_slice_address;
+     //  CU
+     uint8_t *skip_flag;
+     uint8_t *tab_ct_depth;
+     // PU
+     uint8_t *tab_ipm;
+     uint8_t *cbf_luma; // cbf_luma of colocated TU
+     uint8_t *is_pcm;
+     // CTB-level flags affecting loop filter operation
+     uint8_t *filter_slice_edges;
+     /** used on BE to byteswap the lines for checksumming */
+     uint8_t *checksum_buf;
+     int      checksum_buf_size;
+     /**
+      * Sequence counters for decoded and output frames, so that old
+      * frames are output first after a POC reset
+      */
+     uint16_t seq_decode;
+     uint16_t seq_output;
++    int enable_parallel_tiles;
++    atomic_int wpp_err;
++
++    const uint8_t *data;
++
+     H2645Packet pkt;
+     // type of the first VCL NAL of the current frame
+     enum NALUnitType first_nal_type;
+     // for checking the frame checksums
+     struct AVMD5 *md5_ctx;
+     uint8_t       md5[3][16];
+     uint8_t is_md5;
+     uint8_t context_initialized;
+     uint8_t is_nalff;       ///< this flag is != 0 if bitstream is encapsulated
+                             ///< as a format defined in 14496-15
+     int apply_defdispwin;
++    int active_seq_parameter_set_id;
++
+     int nal_length_size;    ///< Number of bytes used for nal length (1, 2 or 4)
+     int nuh_layer_id;
+     /** frame packing arrangement variables */
+     int sei_frame_packing_present;
+     int frame_packing_arrangement_type;
+     int content_interpretation_type;
+     int quincunx_subsampling;
+     /** display orientation */
+     int sei_display_orientation_present;
+     int sei_anticlockwise_rotation;
+     int sei_hflip, sei_vflip;
++
++    int picture_struct;
++
++    uint8_t* a53_caption;
++    int a53_caption_size;
++
++    /** mastering display */
++    int sei_mastering_display_info_present;
++    uint16_t display_primaries[3][2];
++    uint16_t white_point[2];
++    uint32_t max_mastering_luminance;
++    uint32_t min_mastering_luminance;
++
+ } HEVCContext;
+ int ff_hevc_decode_short_term_rps(GetBitContext *gb, AVCodecContext *avctx,
+                                   ShortTermRPS *rps, const HEVCSPS *sps, int is_slice_header);
+ /**
+  * Parse the SPS from the bitstream into the provided HEVCSPS struct.
+  *
+  * @param sps_id the SPS id will be written here
+  * @param apply_defdispwin if set 1, the default display window from the VUI
+  *                         will be applied to the video dimensions
+  * @param vps_list if non-NULL, this function will validate that the SPS refers
+  *                 to an existing VPS
+  */
+ int ff_hevc_parse_sps(HEVCSPS *sps, GetBitContext *gb, unsigned int *sps_id,
+                       int apply_defdispwin, AVBufferRef **vps_list, AVCodecContext *avctx);
+ int ff_hevc_decode_nal_vps(GetBitContext *gb, AVCodecContext *avctx,
+                            HEVCParamSets *ps);
+ int ff_hevc_decode_nal_sps(GetBitContext *gb, AVCodecContext *avctx,
+                            HEVCParamSets *ps, int apply_defdispwin);
+ int ff_hevc_decode_nal_pps(GetBitContext *gb, AVCodecContext *avctx,
+                            HEVCParamSets *ps);
+ int ff_hevc_decode_nal_sei(HEVCContext *s);
+ /**
+  * Mark all frames in DPB as unused for reference.
+  */
+ void ff_hevc_clear_refs(HEVCContext *s);
+ /**
+  * Drop all frames currently in DPB.
+  */
+ void ff_hevc_flush_dpb(HEVCContext *s);
+ /**
+  * Compute POC of the current frame and return it.
+  */
+ int ff_hevc_compute_poc(HEVCContext *s, int poc_lsb);
+ RefPicList *ff_hevc_get_ref_list(HEVCContext *s, HEVCFrame *frame,
+                                  int x0, int y0);
+ /**
+  * Construct the reference picture sets for the current frame.
+  */
+ int ff_hevc_frame_rps(HEVCContext *s);
+ /**
+  * Construct the reference picture list(s) for the current slice.
+  */
+ int ff_hevc_slice_rpl(HEVCContext *s);
+ void ff_hevc_save_states(HEVCContext *s, int ctb_addr_ts);
+ void ff_hevc_cabac_init(HEVCContext *s, int ctb_addr_ts);
+ int ff_hevc_sao_merge_flag_decode(HEVCContext *s);
+ int ff_hevc_sao_type_idx_decode(HEVCContext *s);
+ int ff_hevc_sao_band_position_decode(HEVCContext *s);
+ int ff_hevc_sao_offset_abs_decode(HEVCContext *s);
+ int ff_hevc_sao_offset_sign_decode(HEVCContext *s);
+ int ff_hevc_sao_eo_class_decode(HEVCContext *s);
+ int ff_hevc_end_of_slice_flag_decode(HEVCContext *s);
+ int ff_hevc_cu_transquant_bypass_flag_decode(HEVCContext *s);
+ int ff_hevc_skip_flag_decode(HEVCContext *s, int x0, int y0,
+                              int x_cb, int y_cb);
+ int ff_hevc_pred_mode_decode(HEVCContext *s);
+ int ff_hevc_split_coding_unit_flag_decode(HEVCContext *s, int ct_depth,
+                                           int x0, int y0);
+ int ff_hevc_part_mode_decode(HEVCContext *s, int log2_cb_size);
+ int ff_hevc_pcm_flag_decode(HEVCContext *s);
+ int ff_hevc_prev_intra_luma_pred_flag_decode(HEVCContext *s);
+ int ff_hevc_mpm_idx_decode(HEVCContext *s);
+ int ff_hevc_rem_intra_luma_pred_mode_decode(HEVCContext *s);
+ int ff_hevc_intra_chroma_pred_mode_decode(HEVCContext *s);
+ int ff_hevc_merge_idx_decode(HEVCContext *s);
+ int ff_hevc_merge_flag_decode(HEVCContext *s);
+ int ff_hevc_inter_pred_idc_decode(HEVCContext *s, int nPbW, int nPbH);
+ int ff_hevc_ref_idx_lx_decode(HEVCContext *s, int num_ref_idx_lx);
+ int ff_hevc_mvp_lx_flag_decode(HEVCContext *s);
+ int ff_hevc_no_residual_syntax_flag_decode(HEVCContext *s);
 -int ff_hevc_transform_skip_flag_decode(HEVCContext *s, int c_idx);
 -int ff_hevc_last_significant_coeff_x_prefix_decode(HEVCContext *s, int c_idx,
 -                                                   int log2_size);
 -int ff_hevc_last_significant_coeff_y_prefix_decode(HEVCContext *s, int c_idx,
 -                                                   int log2_size);
 -int ff_hevc_last_significant_coeff_suffix_decode(HEVCContext *s,
 -                                                 int last_significant_coeff_prefix);
 -int ff_hevc_significant_coeff_group_flag_decode(HEVCContext *s, int c_idx,
 -                                                int ctx_cg);
 -int ff_hevc_significant_coeff_flag_decode(HEVCContext *s, int c_idx, int x_c,
 -                                          int y_c, int log2_trafo_size,
 -                                          int scan_idx, int prev_sig);
 -int ff_hevc_coeff_abs_level_greater1_flag_decode(HEVCContext *s, int c_idx,
 -                                                 int ctx_set);
 -int ff_hevc_coeff_abs_level_greater2_flag_decode(HEVCContext *s, int c_idx,
 -                                                 int inc);
 -int ff_hevc_coeff_abs_level_remaining(HEVCContext *s, int base_level,
 -                                      int rc_rice_param);
 -int ff_hevc_coeff_sign_flag(HEVCContext *s, uint8_t nb);
+ int ff_hevc_split_transform_flag_decode(HEVCContext *s, int log2_trafo_size);
+ int ff_hevc_cbf_cb_cr_decode(HEVCContext *s, int trafo_depth);
+ int ff_hevc_cbf_luma_decode(HEVCContext *s, int trafo_depth);
 -void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC, int xBase, int yBase,
++int ff_hevc_log2_res_scale_abs(HEVCContext *s, int idx);
++int ff_hevc_res_scale_sign_flag(HEVCContext *s, int idx);
+ /**
+  * Get the number of candidate references for the current frame.
+  */
+ int ff_hevc_frame_nb_refs(HEVCContext *s);
+ int ff_hevc_set_new_ref(HEVCContext *s, AVFrame **frame, int poc);
+ /**
+  * Find next frame in output order and put a reference to it in frame.
+  * @return 1 if a frame was output, 0 otherwise
+  */
+ int ff_hevc_output_frame(HEVCContext *s, AVFrame *frame, int flush);
++void ff_hevc_bump_frame(HEVCContext *s);
++
+ void ff_hevc_unref_frame(HEVCContext *s, HEVCFrame *frame, int flags);
+ void ff_hevc_set_neighbour_available(HEVCContext *s, int x0, int y0,
+                                      int nPbW, int nPbH);
+ void ff_hevc_luma_mv_merge_mode(HEVCContext *s, int x0, int y0,
+                                 int nPbW, int nPbH, int log2_cb_size,
+                                 int part_idx, int merge_idx, MvField *mv);
+ void ff_hevc_luma_mv_mvp_mode(HEVCContext *s, int x0, int y0,
+                               int nPbW, int nPbH, int log2_cb_size,
+                               int part_idx, int merge_idx,
+                               MvField *mv, int mvp_lx_flag, int LX);
 -void ff_hevc_hls_filter(HEVCContext *s, int x, int y);
++void ff_hevc_set_qPy(HEVCContext *s, int xBase, int yBase,
+                      int log2_cb_size);
+ void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
+                                            int log2_trafo_size);
+ int ff_hevc_cu_qp_delta_sign_flag(HEVCContext *s);
+ int ff_hevc_cu_qp_delta_abs(HEVCContext *s);
 -void ff_hevc_pps_free(HEVCPPS **ppps);
++int ff_hevc_cu_chroma_qp_offset_flag(HEVCContext *s);
++int ff_hevc_cu_chroma_qp_offset_idx(HEVCContext *s);
++void ff_hevc_hls_filter(HEVCContext *s, int x, int y, int ctb_size);
+ void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size);
++void ff_hevc_hls_residual_coding(HEVCContext *s, int x0, int y0,
++                                 int log2_trafo_size, enum ScanType scan_idx,
++                                 int c_idx);
 -void ff_hevc_pred_init(HEVCPredContext *hpc, int bit_depth);
++void ff_hevc_hls_mvd_coding(HEVCContext *s, int x0, int y0, int log2_cb_size);
+ int ff_hevc_encode_nal_vps(HEVCVPS *vps, unsigned int id,
+                            uint8_t *buf, int buf_size);
++/**
++ * Reset SEI values that are stored on the Context.
++ * e.g. Caption data that was extracted during NAL
++ * parsing.
++ *
++ * @param s HEVCContext.
++ */
++void ff_hevc_reset_sei(HEVCContext *s);
++
+ extern const uint8_t ff_hevc_qpel_extra_before[4];
+ extern const uint8_t ff_hevc_qpel_extra_after[4];
+ extern const uint8_t ff_hevc_qpel_extra[4];
+ extern const uint8_t ff_hevc_diag_scan4x4_x[16];
+ extern const uint8_t ff_hevc_diag_scan4x4_y[16];
+ extern const uint8_t ff_hevc_diag_scan8x8_x[64];
+ extern const uint8_t ff_hevc_diag_scan8x8_y[64];
+ #endif /* AVCODEC_HEVCDEC_H */
   */
  
  #include "get_bits.h"
- #include "hevc.h"
+ #include "hevcdec.h"
  
  #include "bit_depth_template.c"
 +#include "hevcdsp.h"
  
 -static void FUNC(put_pcm)(uint8_t *_dst, ptrdiff_t stride, int size,
 +static void FUNC(put_pcm)(uint8_t *_dst, ptrdiff_t stride, int width, int height,
                            GetBitContext *gb, int pcm_bit_depth)
  {
      int x, y;
   * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
   */
  
- #include "hevc.h"
+ #include "hevcdec.h"
  
 +#include "hevcpred.h"
 +
  #define BIT_DEPTH 8
  #include "hevcpred_template.c"
  #undef BIT_DEPTH
Simple merge
Simple merge
index 34d3279,0000000..6dd35db
mode 100644,000000..100644
--- /dev/null
@@@ -1,439 -1,0 +1,439 @@@
- #include "hevc.h"
 +/*
 + * HEVC HW decode acceleration through VA API
 + *
 + * Copyright (C) 2015 Timo Rothenpieler <timo@rothenpieler.org>
 + *
 + * 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
 + */
 +
 +#include <va/va.h>
 +#include <va/va_dec_hevc.h>
 +
 +#include "avcodec.h"
++#include "hevcdec.h"
 +#include "hwaccel.h"
 +#include "vaapi_decode.h"
 +
 +typedef struct VAAPIDecodePictureHEVC {
 +    VAPictureParameterBufferHEVC pic_param;
 +    VASliceParameterBufferHEVC last_slice_param;
 +    const uint8_t *last_buffer;
 +    size_t         last_size;
 +
 +    VAAPIDecodePicture pic;
 +} VAAPIDecodePictureHEVC;
 +
 +static void init_vaapi_pic(VAPictureHEVC *va_pic)
 +{
 +    va_pic->picture_id    = VA_INVALID_ID;
 +    va_pic->flags         = VA_PICTURE_HEVC_INVALID;
 +    va_pic->pic_order_cnt = 0;
 +}
 +
 +static void fill_vaapi_pic(VAPictureHEVC *va_pic, const HEVCFrame *pic, int rps_type)
 +{
 +    va_pic->picture_id    = ff_vaapi_get_surface_id(pic->frame);
 +    va_pic->pic_order_cnt = pic->poc;
 +    va_pic->flags         = rps_type;
 +
 +    if (pic->flags & HEVC_FRAME_FLAG_LONG_REF)
 +        va_pic->flags |= VA_PICTURE_HEVC_LONG_TERM_REFERENCE;
 +
 +    if (pic->frame->interlaced_frame) {
 +        va_pic->flags |= VA_PICTURE_HEVC_FIELD_PIC;
 +
 +        if (!pic->frame->top_field_first)
 +            va_pic->flags |= VA_PICTURE_HEVC_BOTTOM_FIELD;
 +    }
 +}
 +
 +static int find_frame_rps_type(const HEVCContext *h, const HEVCFrame *pic)
 +{
 +    VASurfaceID pic_surf = ff_vaapi_get_surface_id(pic->frame);
 +    int i;
 +
 +    for (i = 0; i < h->rps[ST_CURR_BEF].nb_refs; i++) {
 +        if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_BEF].ref[i]->frame))
 +            return VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE;
 +    }
 +
 +    for (i = 0; i < h->rps[ST_CURR_AFT].nb_refs; i++) {
 +        if (pic_surf == ff_vaapi_get_surface_id(h->rps[ST_CURR_AFT].ref[i]->frame))
 +            return VA_PICTURE_HEVC_RPS_ST_CURR_AFTER;
 +    }
 +
 +    for (i = 0; i < h->rps[LT_CURR].nb_refs; i++) {
 +        if (pic_surf == ff_vaapi_get_surface_id(h->rps[LT_CURR].ref[i]->frame))
 +            return VA_PICTURE_HEVC_RPS_LT_CURR;
 +    }
 +
 +    return 0;
 +}
 +
 +static void fill_vaapi_reference_frames(const HEVCContext *h, VAPictureParameterBufferHEVC *pp)
 +{
 +    const HEVCFrame *current_picture = h->ref;
 +    int i, j, rps_type;
 +
 +    for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); i++) {
 +        const HEVCFrame *frame = NULL;
 +
 +        while (!frame && j < FF_ARRAY_ELEMS(h->DPB)) {
 +            if (&h->DPB[j] != current_picture && (h->DPB[j].flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF)))
 +                frame = &h->DPB[j];
 +            j++;
 +        }
 +
 +        init_vaapi_pic(&pp->ReferenceFrames[i]);
 +
 +        if (frame) {
 +            rps_type = find_frame_rps_type(h, frame);
 +            fill_vaapi_pic(&pp->ReferenceFrames[i], frame, rps_type);
 +        }
 +    }
 +}
 +
 +static int vaapi_hevc_start_frame(AVCodecContext          *avctx,
 +                                  av_unused const uint8_t *buffer,
 +                                  av_unused uint32_t       size)
 +{
 +    const HEVCContext        *h = avctx->priv_data;
 +    VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private;
 +    const HEVCSPS          *sps = h->ps.sps;
 +    const HEVCPPS          *pps = h->ps.pps;
 +
 +    const ScalingList *scaling_list = NULL;
 +    int err, i;
 +
 +    pic->pic.output_surface = ff_vaapi_get_surface_id(h->ref->frame);
 +
 +    pic->pic_param = (VAPictureParameterBufferHEVC) {
 +        .pic_fields.value                             = 0,
 +        .slice_parsing_fields.value                   = 0,
 +        .pic_width_in_luma_samples                    = sps->width,
 +        .pic_height_in_luma_samples                   = sps->height,
 +        .log2_min_luma_coding_block_size_minus3       = sps->log2_min_cb_size - 3,
 +        .sps_max_dec_pic_buffering_minus1             = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1,
 +        .log2_diff_max_min_luma_coding_block_size     = sps->log2_diff_max_min_coding_block_size,
 +        .log2_min_transform_block_size_minus2         = sps->log2_min_tb_size - 2,
 +        .log2_diff_max_min_transform_block_size       = sps->log2_max_trafo_size  - sps->log2_min_tb_size,
 +        .max_transform_hierarchy_depth_inter          = sps->max_transform_hierarchy_depth_inter,
 +        .max_transform_hierarchy_depth_intra          = sps->max_transform_hierarchy_depth_intra,
 +        .num_short_term_ref_pic_sets                  = sps->nb_st_rps,
 +        .num_long_term_ref_pic_sps                    = sps->num_long_term_ref_pics_sps,
 +        .num_ref_idx_l0_default_active_minus1         = pps->num_ref_idx_l0_default_active - 1,
 +        .num_ref_idx_l1_default_active_minus1         = pps->num_ref_idx_l1_default_active - 1,
 +        .init_qp_minus26                              = pps->pic_init_qp_minus26,
 +        .pps_cb_qp_offset                             = pps->cb_qp_offset,
 +        .pps_cr_qp_offset                             = pps->cr_qp_offset,
 +        .pcm_sample_bit_depth_luma_minus1             = sps->pcm.bit_depth - 1,
 +        .pcm_sample_bit_depth_chroma_minus1           = sps->pcm.bit_depth_chroma - 1,
 +        .log2_min_pcm_luma_coding_block_size_minus3   = sps->pcm.log2_min_pcm_cb_size - 3,
 +        .log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size,
 +        .diff_cu_qp_delta_depth                       = pps->diff_cu_qp_delta_depth,
 +        .pps_beta_offset_div2                         = pps->beta_offset / 2,
 +        .pps_tc_offset_div2                           = pps->tc_offset / 2,
 +        .log2_parallel_merge_level_minus2             = pps->log2_parallel_merge_level - 2,
 +        .bit_depth_luma_minus8                        = sps->bit_depth - 8,
 +        .bit_depth_chroma_minus8                      = sps->bit_depth - 8,
 +        .log2_max_pic_order_cnt_lsb_minus4            = sps->log2_max_poc_lsb - 4,
 +        .num_extra_slice_header_bits                  = pps->num_extra_slice_header_bits,
 +        .pic_fields.bits = {
 +            .chroma_format_idc                          = sps->chroma_format_idc,
 +            .tiles_enabled_flag                         = pps->tiles_enabled_flag,
 +            .separate_colour_plane_flag                 = sps->separate_colour_plane_flag,
 +            .pcm_enabled_flag                           = sps->pcm_enabled_flag,
 +            .scaling_list_enabled_flag                  = sps->scaling_list_enable_flag,
 +            .transform_skip_enabled_flag                = pps->transform_skip_enabled_flag,
 +            .amp_enabled_flag                           = sps->amp_enabled_flag,
 +            .strong_intra_smoothing_enabled_flag        = sps->sps_strong_intra_smoothing_enable_flag,
 +            .sign_data_hiding_enabled_flag              = pps->sign_data_hiding_flag,
 +            .constrained_intra_pred_flag                = pps->constrained_intra_pred_flag,
 +            .cu_qp_delta_enabled_flag                   = pps->cu_qp_delta_enabled_flag,
 +            .weighted_pred_flag                         = pps->weighted_pred_flag,
 +            .weighted_bipred_flag                       = pps->weighted_bipred_flag,
 +            .transquant_bypass_enabled_flag             = pps->transquant_bypass_enable_flag,
 +            .entropy_coding_sync_enabled_flag           = pps->entropy_coding_sync_enabled_flag,
 +            .pps_loop_filter_across_slices_enabled_flag = pps->seq_loop_filter_across_slices_enabled_flag,
 +            .loop_filter_across_tiles_enabled_flag      = pps->loop_filter_across_tiles_enabled_flag,
 +            .pcm_loop_filter_disabled_flag              = sps->pcm.loop_filter_disable_flag,
 +        },
 +        .slice_parsing_fields.bits = {
 +            .lists_modification_present_flag             = pps->lists_modification_present_flag,
 +            .long_term_ref_pics_present_flag             = sps->long_term_ref_pics_present_flag,
 +            .sps_temporal_mvp_enabled_flag               = sps->sps_temporal_mvp_enabled_flag,
 +            .cabac_init_present_flag                     = pps->cabac_init_present_flag,
 +            .output_flag_present_flag                    = pps->output_flag_present_flag,
 +            .dependent_slice_segments_enabled_flag       = pps->dependent_slice_segments_enabled_flag,
 +            .pps_slice_chroma_qp_offsets_present_flag    = pps->pic_slice_level_chroma_qp_offsets_present_flag,
 +            .sample_adaptive_offset_enabled_flag         = sps->sao_enabled,
 +            .deblocking_filter_override_enabled_flag     = pps->deblocking_filter_override_enabled_flag,
 +            .pps_disable_deblocking_filter_flag          = pps->disable_dbf,
 +            .slice_segment_header_extension_present_flag = pps->slice_header_extension_present_flag,
 +            .RapPicFlag                                  = IS_IRAP(h),
 +            .IdrPicFlag                                  = IS_IDR(h),
 +            .IntraPicFlag                                = IS_IRAP(h),
 +        },
 +    };
 +
 +    fill_vaapi_pic(&pic->pic_param.CurrPic, h->ref, 0);
 +    fill_vaapi_reference_frames(h, &pic->pic_param);
 +
 +    if (pps->tiles_enabled_flag) {
 +        pic->pic_param.num_tile_columns_minus1 = pps->num_tile_columns - 1;
 +        pic->pic_param.num_tile_rows_minus1    = pps->num_tile_rows - 1;
 +
 +        for (i = 0; i < pps->num_tile_columns; i++)
 +            pic->pic_param.column_width_minus1[i] = pps->column_width[i] - 1;
 +
 +        for (i = 0; i < pps->num_tile_rows; i++)
 +            pic->pic_param.row_height_minus1[i] = pps->row_height[i] - 1;
 +    }
 +
 +    if (h->sh.short_term_ref_pic_set_sps_flag == 0 && h->sh.short_term_rps) {
 +        pic->pic_param.st_rps_bits = h->sh.short_term_ref_pic_set_size;
 +    } else {
 +        pic->pic_param.st_rps_bits = 0;
 +    }
 +
 +    err = ff_vaapi_decode_make_param_buffer(avctx, &pic->pic,
 +                                            VAPictureParameterBufferType,
 +                                            &pic->pic_param, sizeof(pic->pic_param));
 +    if (err < 0)
 +        goto fail;
 +
 +    if (pps->scaling_list_data_present_flag)
 +        scaling_list = &pps->scaling_list;
 +    else if (sps->scaling_list_enable_flag)
 +        scaling_list = &sps->scaling_list;
 +
 +    if (scaling_list) {
 +        VAIQMatrixBufferHEVC iq_matrix;
 +        int j;
 +
 +        for (i = 0; i < 6; i++) {
 +            for (j = 0; j < 16; j++)
 +                iq_matrix.ScalingList4x4[i][j] = scaling_list->sl[0][i][j];
 +            for (j = 0; j < 64; j++) {
 +                iq_matrix.ScalingList8x8[i][j]   = scaling_list->sl[1][i][j];
 +                iq_matrix.ScalingList16x16[i][j] = scaling_list->sl[2][i][j];
 +                if (i < 2)
 +                    iq_matrix.ScalingList32x32[i][j] = scaling_list->sl[3][i * 3][j];
 +            }
 +            iq_matrix.ScalingListDC16x16[i] = scaling_list->sl_dc[0][i];
 +            if (i < 2)
 +                iq_matrix.ScalingListDC32x32[i] = scaling_list->sl_dc[1][i * 3];
 +        }
 +
 +        err = ff_vaapi_decode_make_param_buffer(avctx, &pic->pic,
 +                                                VAIQMatrixBufferType,
 +                                                &iq_matrix, sizeof(iq_matrix));
 +        if (err < 0)
 +            goto fail;
 +    }
 +
 +    return 0;
 +
 +fail:
 +    ff_vaapi_decode_cancel(avctx, &pic->pic);
 +    return err;
 +}
 +
 +static int vaapi_hevc_end_frame(AVCodecContext *avctx)
 +{
 +    const HEVCContext        *h = avctx->priv_data;
 +    VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private;
 +    int ret;
 +
 +    if (pic->last_size) {
 +        pic->last_slice_param.LongSliceFlags.fields.LastSliceOfPic = 1;
 +        ret = ff_vaapi_decode_make_slice_buffer(avctx, &pic->pic,
 +                                                &pic->last_slice_param, sizeof(pic->last_slice_param),
 +                                                pic->last_buffer, pic->last_size);
 +        if (ret < 0)
 +            goto fail;
 +    }
 +
 +
 +    ret = ff_vaapi_decode_issue(avctx, &pic->pic);
 +    if (ret < 0)
 +        goto fail;
 +
 +    return 0;
 +fail:
 +    ff_vaapi_decode_cancel(avctx, &pic->pic);
 +    return ret;
 +}
 +
 +static void fill_pred_weight_table(const HEVCContext *h,
 +                                   const SliceHeader *sh,
 +                                   VASliceParameterBufferHEVC *slice_param)
 +{
 +    int i;
 +
 +    memset(slice_param->delta_luma_weight_l0,   0, sizeof(slice_param->delta_luma_weight_l0));
 +    memset(slice_param->delta_luma_weight_l1,   0, sizeof(slice_param->delta_luma_weight_l1));
 +    memset(slice_param->luma_offset_l0,         0, sizeof(slice_param->luma_offset_l0));
 +    memset(slice_param->luma_offset_l1,         0, sizeof(slice_param->luma_offset_l1));
 +    memset(slice_param->delta_chroma_weight_l0, 0, sizeof(slice_param->delta_chroma_weight_l0));
 +    memset(slice_param->delta_chroma_weight_l1, 0, sizeof(slice_param->delta_chroma_weight_l1));
 +    memset(slice_param->ChromaOffsetL0,         0, sizeof(slice_param->ChromaOffsetL0));
 +    memset(slice_param->ChromaOffsetL1,         0, sizeof(slice_param->ChromaOffsetL1));
 +
 +    slice_param->delta_chroma_log2_weight_denom = 0;
 +    slice_param->luma_log2_weight_denom         = 0;
 +
 +    if (sh->slice_type == I_SLICE ||
 +        (sh->slice_type == P_SLICE && !h->ps.pps->weighted_pred_flag) ||
 +        (sh->slice_type == B_SLICE && !h->ps.pps->weighted_bipred_flag))
 +        return;
 +
 +    slice_param->luma_log2_weight_denom = sh->luma_log2_weight_denom;
 +
 +    if (h->ps.sps->chroma_format_idc) {
 +        slice_param->delta_chroma_log2_weight_denom = sh->chroma_log2_weight_denom - sh->luma_log2_weight_denom;
 +    }
 +
 +    for (i = 0; i < 15 && i < sh->nb_refs[L0]; i++) {
 +        slice_param->delta_luma_weight_l0[i] = sh->luma_weight_l0[i] - (1 << sh->luma_log2_weight_denom);
 +        slice_param->luma_offset_l0[i] = sh->luma_offset_l0[i];
 +        slice_param->delta_chroma_weight_l0[i][0] = sh->chroma_weight_l0[i][0] - (1 << sh->chroma_log2_weight_denom);
 +        slice_param->delta_chroma_weight_l0[i][1] = sh->chroma_weight_l0[i][1] - (1 << sh->chroma_log2_weight_denom);
 +        slice_param->ChromaOffsetL0[i][0] = sh->chroma_offset_l0[i][0];
 +        slice_param->ChromaOffsetL0[i][1] = sh->chroma_offset_l0[i][1];
 +    }
 +
 +    if (sh->slice_type == B_SLICE) {
 +        for (i = 0; i < 15 && i < sh->nb_refs[L1]; i++) {
 +            slice_param->delta_luma_weight_l1[i] = sh->luma_weight_l1[i] - (1 << sh->luma_log2_weight_denom);
 +            slice_param->luma_offset_l1[i] = sh->luma_offset_l1[i];
 +            slice_param->delta_chroma_weight_l1[i][0] = sh->chroma_weight_l1[i][0] - (1 << sh->chroma_log2_weight_denom);
 +            slice_param->delta_chroma_weight_l1[i][1] = sh->chroma_weight_l1[i][1] - (1 << sh->chroma_log2_weight_denom);
 +            slice_param->ChromaOffsetL1[i][0] = sh->chroma_offset_l1[i][0];
 +            slice_param->ChromaOffsetL1[i][1] = sh->chroma_offset_l1[i][1];
 +        }
 +    }
 +}
 +
 +static uint8_t get_ref_pic_index(const HEVCContext *h, const HEVCFrame *frame)
 +{
 +    VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private;
 +    VAPictureParameterBufferHEVC *pp = &pic->pic_param;
 +    uint8_t i;
 +
 +    if (!frame)
 +        return 0xff;
 +
 +    for (i = 0; i < FF_ARRAY_ELEMS(pp->ReferenceFrames); i++) {
 +        VASurfaceID pid = pp->ReferenceFrames[i].picture_id;
 +        int poc = pp->ReferenceFrames[i].pic_order_cnt;
 +        if (pid != VA_INVALID_ID && pid == ff_vaapi_get_surface_id(frame->frame) && poc == frame->poc)
 +            return i;
 +    }
 +
 +    return 0xff;
 +}
 +
 +static int vaapi_hevc_decode_slice(AVCodecContext *avctx,
 +                                   const uint8_t  *buffer,
 +                                   uint32_t        size)
 +{
 +    const HEVCContext        *h = avctx->priv_data;
 +    const SliceHeader       *sh = &h->sh;
 +    VAAPIDecodePictureHEVC *pic = h->ref->hwaccel_picture_private;
 +
 +    int nb_list = (sh->slice_type == B_SLICE) ?
 +                  2 : (sh->slice_type == I_SLICE ? 0 : 1);
 +
 +    int err, i, list_idx;
 +
 +    if (!sh->first_slice_in_pic_flag) {
 +        err = ff_vaapi_decode_make_slice_buffer(avctx, &pic->pic,
 +                                                &pic->last_slice_param, sizeof(pic->last_slice_param),
 +                                                pic->last_buffer, pic->last_size);
 +        pic->last_buffer = NULL;
 +        pic->last_size   = 0;
 +        if (err) {
 +            ff_vaapi_decode_cancel(avctx, &pic->pic);
 +            return err;
 +        }
 +    }
 +
 +    pic->last_slice_param = (VASliceParameterBufferHEVC) {
 +        .slice_data_size               = size,
 +        .slice_data_offset             = 0,
 +        .slice_data_flag               = VA_SLICE_DATA_FLAG_ALL,
 +        /* Add 1 to the bits count here to account for the byte_alignment bit, which
 +         * always is at least one bit and not accounted for otherwise. */
 +        .slice_data_byte_offset        = (get_bits_count(&h->HEVClc->gb) + 1 + 7) / 8,
 +        .slice_segment_address         = sh->slice_segment_addr,
 +        .slice_qp_delta                = sh->slice_qp_delta,
 +        .slice_cb_qp_offset            = sh->slice_cb_qp_offset,
 +        .slice_cr_qp_offset            = sh->slice_cr_qp_offset,
 +        .slice_beta_offset_div2        = sh->beta_offset / 2,
 +        .slice_tc_offset_div2          = sh->tc_offset / 2,
 +        .collocated_ref_idx            = sh->slice_temporal_mvp_enabled_flag ? sh->collocated_ref_idx : 0xFF,
 +        .five_minus_max_num_merge_cand = sh->slice_type == I_SLICE ? 0 : 5 - sh->max_num_merge_cand,
 +        .num_ref_idx_l0_active_minus1  = sh->nb_refs[L0] ? sh->nb_refs[L0] - 1 : 0,
 +        .num_ref_idx_l1_active_minus1  = sh->nb_refs[L1] ? sh->nb_refs[L1] - 1 : 0,
 +
 +        .LongSliceFlags.fields = {
 +            .dependent_slice_segment_flag                 = sh->dependent_slice_segment_flag,
 +            .slice_type                                   = sh->slice_type,
 +            .color_plane_id                               = sh->colour_plane_id,
 +            .mvd_l1_zero_flag                             = sh->mvd_l1_zero_flag,
 +            .cabac_init_flag                              = sh->cabac_init_flag,
 +            .slice_temporal_mvp_enabled_flag              = sh->slice_temporal_mvp_enabled_flag,
 +            .slice_deblocking_filter_disabled_flag        = sh->disable_deblocking_filter_flag,
 +            .collocated_from_l0_flag                      = sh->collocated_list == L0 ? 1 : 0,
 +            .slice_loop_filter_across_slices_enabled_flag = sh->slice_loop_filter_across_slices_enabled_flag,
 +            .slice_sao_luma_flag                          = sh->slice_sample_adaptive_offset_flag[0],
 +            .slice_sao_chroma_flag                        = sh->slice_sample_adaptive_offset_flag[1],
 +        },
 +    };
 +
 +    memset(pic->last_slice_param.RefPicList, 0xFF, sizeof(pic->last_slice_param.RefPicList));
 +
 +    for (list_idx = 0; list_idx < nb_list; list_idx++) {
 +        RefPicList *rpl = &h->ref->refPicList[list_idx];
 +
 +        for (i = 0; i < rpl->nb_refs; i++)
 +            pic->last_slice_param.RefPicList[list_idx][i] = get_ref_pic_index(h, rpl->ref[i]);
 +    }
 +
 +    fill_pred_weight_table(h, sh, &pic->last_slice_param);
 +
 +    pic->last_buffer = buffer;
 +    pic->last_size   = size;
 +
 +    return 0;
 +}
 +
 +AVHWAccel ff_hevc_vaapi_hwaccel = {
 +    .name                 = "hevc_vaapi",
 +    .type                 = AVMEDIA_TYPE_VIDEO,
 +    .id                   = AV_CODEC_ID_HEVC,
 +    .pix_fmt              = AV_PIX_FMT_VAAPI,
 +    .start_frame          = vaapi_hevc_start_frame,
 +    .end_frame            = vaapi_hevc_end_frame,
 +    .decode_slice         = vaapi_hevc_decode_slice,
 +    .frame_priv_data_size = sizeof(VAAPIDecodePictureHEVC),
 +    .init                 = ff_vaapi_decode_init,
 +    .uninit               = ff_vaapi_decode_uninit,
 +    .priv_data_size       = sizeof(VAAPIDecodeContext),
 +    .caps_internal        = HWACCEL_CAP_ASYNC_SAFE,
 +};
@@@ -24,8 -24,7 +24,8 @@@
  
  #include "avcodec.h"
  #include "internal.h"
- #include "hevc.h"
+ #include "hevcdec.h"
 +#include "hwaccel.h"
  #include "vdpau.h"
  #include "vdpau_internal.h"
  
Simple merge
Simple merge