hevc: skip invalid/ignored NALUs when splitting the packet
[ffmpeg.git] / libavcodec / hevc.c
index 08bed86..7080d00 100644 (file)
@@ -251,10 +251,10 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
     uint8_t luma_weight_l1_flag[16];
     uint8_t chroma_weight_l1_flag[16];
 
-    s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
-    if (s->sps->chroma_format_idc != 0) {
+    s->sh.luma_log2_weight_denom = av_clip(get_ue_golomb_long(gb), 0, 7);
+    if (s->ps.sps->chroma_format_idc != 0) {
         int delta = get_se_golomb(gb);
-        s->sh.chroma_log2_weight_denom = av_clip_c(s->sh.luma_log2_weight_denom + delta, 0, 7);
+        s->sh.chroma_log2_weight_denom = av_clip(s->sh.luma_log2_weight_denom + delta, 0, 7);
     }
 
     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
@@ -264,7 +264,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
             s->sh.luma_offset_l0[i] = 0;
         }
     }
-    if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
+    if (s->ps.sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
         for (i = 0; i < s->sh.nb_refs[L0]; i++)
             chroma_weight_l0_flag[i] = get_bits1(gb);
     } else {
@@ -282,7 +282,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                 int delta_chroma_weight_l0 = get_se_golomb(gb);
                 int delta_chroma_offset_l0 = get_se_golomb(gb);
                 s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
-                s->sh.chroma_offset_l0[i][j] = av_clip_c((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
+                s->sh.chroma_offset_l0[i][j] = av_clip((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
                                                                                     >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
             }
         } else {
@@ -300,7 +300,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                 s->sh.luma_offset_l1[i] = 0;
             }
         }
-        if (s->sps->chroma_format_idc != 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 {
@@ -318,7 +318,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
                     int delta_chroma_weight_l1 = get_se_golomb(gb);
                     int delta_chroma_offset_l1 = get_se_golomb(gb);
                     s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
-                    s->sh.chroma_offset_l1[i][j] = av_clip_c((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
+                    s->sh.chroma_offset_l1[i][j] = av_clip((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
                                                                                         >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
                 }
             } else {
@@ -333,7 +333,7 @@ static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
 
 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
 {
-    const HEVCSPS *sps = s->sps;
+    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;
@@ -383,46 +383,87 @@ static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
     return 0;
 }
 
-static int set_sps(HEVCContext *s, const HEVCSPS *sps)
+static void export_stream_params(AVCodecContext *avctx,
+                                 const HEVCContext *s, const HEVCSPS *sps)
 {
-    int ret;
+    const HEVCVPS *vps = (const HEVCVPS*)s->ps.vps_list[sps->vps_id]->data;
     unsigned int num = 0, den = 0;
 
-    pic_arrays_free(s);
-    ret = pic_arrays_init(s, sps);
-    if (ret < 0)
-        goto fail;
+    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;
 
-    s->avctx->coded_width         = sps->width;
-    s->avctx->coded_height        = sps->height;
-    s->avctx->width               = sps->output_width;
-    s->avctx->height              = sps->output_height;
-    s->avctx->pix_fmt             = sps->pix_fmt;
-    s->avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
-
-    ff_set_sar(s->avctx, sps->vui.sar);
+    ff_set_sar(avctx, sps->vui.sar);
 
     if (sps->vui.video_signal_type_present_flag)
-        s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
-                                                               : AVCOL_RANGE_MPEG;
+        avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
+                                                            : AVCOL_RANGE_MPEG;
     else
-        s->avctx->color_range = AVCOL_RANGE_MPEG;
+        avctx->color_range = AVCOL_RANGE_MPEG;
 
     if (sps->vui.colour_description_present_flag) {
-        s->avctx->color_primaries = sps->vui.colour_primaries;
-        s->avctx->color_trc       = sps->vui.transfer_characteristic;
-        s->avctx->colorspace      = sps->vui.matrix_coeffs;
+        avctx->color_primaries = sps->vui.colour_primaries;
+        avctx->color_trc       = sps->vui.transfer_characteristic;
+        avctx->colorspace      = sps->vui.matrix_coeffs;
     } else {
-        s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
-        s->avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
-        s->avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
+        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);
+}
+
+static int set_sps(HEVCContext *s, const HEVCSPS *sps)
+{
+    #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL)
+    enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
+    int ret;
+
+    export_stream_params(s->avctx, s, sps);
+
+    pic_arrays_free(s);
+    ret = pic_arrays_init(s, sps);
+    if (ret < 0)
+        goto fail;
+
+    if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P) {
+#if CONFIG_HEVC_DXVA2_HWACCEL
+        *fmt++ = AV_PIX_FMT_DXVA2_VLD;
+#endif
+#if CONFIG_HEVC_D3D11VA_HWACCEL
+        *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
+#endif
     }
 
+    *fmt++ = sps->pix_fmt;
+    *fmt = AV_PIX_FMT_NONE;
+
+    ret = ff_get_format(s->avctx, pix_fmts);
+    if (ret < 0)
+        goto fail;
+    s->avctx->pix_fmt = ret;
+
     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);
 
-    if (sps->sao_enabled) {
+    if (sps->sao_enabled && !s->avctx->hwaccel) {
         av_frame_unref(s->tmp_frame);
         ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
         if (ret < 0)
@@ -430,26 +471,14 @@ static int set_sps(HEVCContext *s, const HEVCSPS *sps)
         s->frame = s->tmp_frame;
     }
 
-    s->sps = sps;
-    s->vps = (HEVCVPS*) s->vps_list[s->sps->vps_id]->data;
-
-    if (s->vps->vps_timing_info_present_flag) {
-        num = s->vps->vps_num_units_in_tick;
-        den = s->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(&s->avctx->framerate.den, &s->avctx->framerate.num,
-                  num, den, 1 << 30);
+    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->sps = NULL;
+    s->ps.sps = NULL;
     return ret;
 }
 
@@ -471,22 +500,22 @@ static int hls_slice_header(HEVCContext *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->pps_list[sh->pps_id]) {
+    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->pps != (HEVCPPS*)s->pps_list[sh->pps_id]->data) {
+        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->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
+    s->ps.pps = (HEVCPPS*)s->ps.pps_list[sh->pps_id]->data;
 
-    if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
-        s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
+    if (s->ps.sps != (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data) {
+        s->ps.sps = (HEVCSPS*)s->ps.sps_list[s->ps.pps->sps_id]->data;
 
         ff_hevc_clear_refs(s);
-        ret = set_sps(s, s->sps);
+        ret = set_sps(s, s->ps.sps);
         if (ret < 0)
             return ret;
 
@@ -494,20 +523,17 @@ static int hls_slice_header(HEVCContext *s)
         s->max_ra     = INT_MAX;
     }
 
-    s->avctx->profile = s->sps->ptl.general_ptl.profile_idc;
-    s->avctx->level   = s->sps->ptl.general_ptl.level_idc;
-
     sh->dependent_slice_segment_flag = 0;
     if (!sh->first_slice_in_pic_flag) {
         int slice_address_length;
 
-        if (s->pps->dependent_slice_segments_enabled_flag)
+        if (s->ps.pps->dependent_slice_segments_enabled_flag)
             sh->dependent_slice_segment_flag = get_bits1(gb);
 
-        slice_address_length = av_ceil_log2(s->sps->ctb_width *
-                                            s->sps->ctb_height);
-        sh->slice_segment_addr = get_bits(gb, slice_address_length);
-        if (sh->slice_segment_addr >= s->sps->ctb_width * s->sps->ctb_height) {
+        slice_address_length = av_ceil_log2(s->ps.sps->ctb_width *
+                                            s->ps.sps->ctb_height);
+        sh->slice_segment_addr = slice_address_length ? get_bits(gb, slice_address_length) : 0;
+        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);
@@ -527,7 +553,7 @@ static int hls_slice_header(HEVCContext *s)
     if (!sh->dependent_slice_segment_flag) {
         s->slice_initialized = 0;
 
-        for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
+        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);
@@ -545,16 +571,16 @@ static int hls_slice_header(HEVCContext *s)
 
         // when flag is not present, picture is inferred to be output
         sh->pic_output_flag = 1;
-        if (s->pps->output_flag_present_flag)
+        if (s->ps.pps->output_flag_present_flag)
             sh->pic_output_flag = get_bits1(gb);
 
-        if (s->sps->separate_colour_plane_flag)
+        if (s->ps.sps->separate_colour_plane_flag)
             sh->colour_plane_id = get_bits(gb, 2);
 
         if (!IS_IDR(s)) {
-            int short_term_ref_pic_set_sps_flag, poc;
+            int poc;
 
-            sh->pic_order_cnt_lsb = get_bits(gb, s->sps->log2_max_poc_lsb);
+            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,
@@ -565,24 +591,26 @@ static int hls_slice_header(HEVCContext *s)
             }
             s->poc = poc;
 
-            short_term_ref_pic_set_sps_flag = get_bits1(gb);
-            if (!short_term_ref_pic_set_sps_flag) {
-                ret = ff_hevc_decode_short_term_rps(s, &sh->slice_rps, s->sps, 1);
+            sh->short_term_ref_pic_set_sps_flag = get_bits1(gb);
+            if (!sh->short_term_ref_pic_set_sps_flag) {
+                int pos = get_bits_left(gb);
+                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_ref_pic_set_size = pos - get_bits_left(gb);
                 sh->short_term_rps = &sh->slice_rps;
             } else {
                 int numbits, rps_idx;
 
-                if (!s->sps->nb_st_rps) {
+                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->sps->nb_st_rps);
+                numbits = av_ceil_log2(s->ps.sps->nb_st_rps);
                 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
-                sh->short_term_rps = &s->sps->st_rps[rps_idx];
+                sh->short_term_rps = &s->ps.sps->st_rps[rps_idx];
             }
 
             ret = decode_lt_rps(s, &sh->long_term_rps, gb);
@@ -592,7 +620,7 @@ static int hls_slice_header(HEVCContext *s)
                     return AVERROR_INVALIDDATA;
             }
 
-            if (s->sps->sps_temporal_mvp_enabled_flag)
+            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;
@@ -612,7 +640,7 @@ static int hls_slice_header(HEVCContext *s)
             s->nal_unit_type != NAL_RASL_R)
             s->pocTid0 = s->poc;
 
-        if (s->sps->sao_enabled) {
+        if (s->ps.sps->sao_enabled) {
             sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
             sh->slice_sample_adaptive_offset_flag[1] =
             sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
@@ -626,9 +654,9 @@ static int hls_slice_header(HEVCContext *s)
         if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
             int nb_refs;
 
-            sh->nb_refs[L0] = s->pps->num_ref_idx_l0_default_active;
+            sh->nb_refs[L0] = s->ps.pps->num_ref_idx_l0_default_active;
             if (sh->slice_type == B_SLICE)
-                sh->nb_refs[L1] = s->pps->num_ref_idx_l1_default_active;
+                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;
@@ -649,7 +677,7 @@ static int hls_slice_header(HEVCContext *s)
                 return AVERROR_INVALIDDATA;
             }
 
-            if (s->pps->lists_modification_present_flag && nb_refs > 1) {
+            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++)
@@ -667,7 +695,7 @@ static int hls_slice_header(HEVCContext *s)
             if (sh->slice_type == B_SLICE)
                 sh->mvd_l1_zero_flag = get_bits1(gb);
 
-            if (s->pps->cabac_init_present_flag)
+            if (s->ps.pps->cabac_init_present_flag)
                 sh->cabac_init_flag = get_bits1(gb);
             else
                 sh->cabac_init_flag = 0;
@@ -689,8 +717,8 @@ static int hls_slice_header(HEVCContext *s)
                 }
             }
 
-            if ((s->pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
-                (s->pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
+            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);
             }
 
@@ -705,7 +733,7 @@ static int hls_slice_header(HEVCContext *s)
 
         sh->slice_qp_delta = get_se_golomb(gb);
 
-        if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
+        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 {
@@ -713,10 +741,10 @@ static int hls_slice_header(HEVCContext *s)
             sh->slice_cr_qp_offset = 0;
         }
 
-        if (s->pps->deblocking_filter_control_present_flag) {
+        if (s->ps.pps->deblocking_filter_control_present_flag) {
             int deblocking_filter_override_flag = 0;
 
-            if (s->pps->deblocking_filter_override_enabled_flag)
+            if (s->ps.pps->deblocking_filter_override_enabled_flag)
                 deblocking_filter_override_flag = get_bits1(gb);
 
             if (deblocking_filter_override_flag) {
@@ -726,9 +754,9 @@ static int hls_slice_header(HEVCContext *s)
                     sh->tc_offset   = get_se_golomb(gb) * 2;
                 }
             } else {
-                sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
-                sh->beta_offset                    = s->pps->beta_offset;
-                sh->tc_offset                      = s->pps->tc_offset;
+                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;
@@ -736,13 +764,13 @@ static int hls_slice_header(HEVCContext *s)
             sh->tc_offset                      = 0;
         }
 
-        if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
+        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->pps->seq_loop_filter_across_slices_enabled_flag;
+            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");
@@ -750,7 +778,7 @@ static int hls_slice_header(HEVCContext *s)
     }
 
     sh->num_entry_point_offsets = 0;
-    if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
+    if (s->ps.pps->tiles_enabled_flag || s->ps.pps->entropy_coding_sync_enabled_flag) {
         sh->num_entry_point_offsets = get_ue_golomb_long(gb);
         if (sh->num_entry_point_offsets > 0) {
             int offset_len = get_ue_golomb_long(gb) + 1;
@@ -760,21 +788,21 @@ static int hls_slice_header(HEVCContext *s)
         }
     }
 
-    if (s->pps->slice_header_extension_present_flag) {
+    if (s->ps.pps->slice_header_extension_present_flag) {
         unsigned int length = get_ue_golomb_long(gb);
         for (i = 0; i < length; i++)
             skip_bits(gb, 8);  // slice_header_extension_data_byte
     }
 
     // Inferred parameters
-    sh->slice_qp = 26 + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
+    sh->slice_qp = 26 + s->ps.pps->pic_init_qp_minus26 + sh->slice_qp_delta;
     if (sh->slice_qp > 51 ||
-        sh->slice_qp < -s->sps->qp_bd_offset) {
+        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->sps->qp_bd_offset);
+               -s->ps.sps->qp_bd_offset);
         return AVERROR_INVALIDDATA;
     }
 
@@ -787,16 +815,16 @@ static int hls_slice_header(HEVCContext *s)
 
     s->HEVClc.first_qp_group = !s->sh.dependent_slice_segment_flag;
 
-    if (!s->pps->cu_qp_delta_enabled_flag)
-        s->HEVClc.qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset,
-                                52 + s->sps->qp_bd_offset) - s->sps->qp_bd_offset;
+    if (!s->ps.pps->cu_qp_delta_enabled_flag)
+        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;
 
     s->slice_initialized = 1;
 
     return 0;
 }
 
-#define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
+#define CTB(tab, x, y) ((tab)[(y) * s->ps.sps->ctb_width + (x)])
 
 #define SET_SAO(elem, value)                            \
 do {                                                    \
@@ -815,7 +843,7 @@ static void hls_sao_param(HEVCContext *s, int rx, int ry)
     HEVCLocalContext *lc    = &s->HEVClc;
     int sao_merge_left_flag = 0;
     int sao_merge_up_flag   = 0;
-    int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
+    int shift               = s->ps.sps->bit_depth - FFMIN(s->ps.sps->bit_depth, 10);
     SAOParams *sao          = &CTB(s->sao, rx, ry);
     int c_idx, i;
 
@@ -905,10 +933,10 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
     const uint8_t *scan_x_cg, *scan_y_cg, *scan_x_off, *scan_y_off;
 
     ptrdiff_t stride = s->frame->linesize[c_idx];
-    int hshift       = s->sps->hshift[c_idx];
-    int vshift       = s->sps->vshift[c_idx];
+    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->sps->pixel_shift)];
+                                              ((x0 >> hshift) << s->ps.sps->pixel_shift)];
     DECLARE_ALIGNED(16, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
     DECLARE_ALIGNED(8, uint8_t, significant_coeff_group_flag[8][8]) = { { 0 } };
 
@@ -938,16 +966,16 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
         int qp_y = lc->qp_y;
 
         if (c_idx == 0) {
-            qp = qp_y + s->sps->qp_bd_offset;
+            qp = qp_y + s->ps.sps->qp_bd_offset;
         } else {
             int qp_i, offset;
 
             if (c_idx == 1)
-                offset = s->pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
+                offset = s->ps.pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
             else
-                offset = s->pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
+                offset = s->ps.pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
 
-            qp_i = av_clip_c(qp_y + offset, -s->sps->qp_bd_offset, 57);
+            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)
@@ -955,18 +983,18 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
             else
                 qp = qp_c[qp_i - 30];
 
-            qp += s->sps->qp_bd_offset;
+            qp += s->ps.sps->qp_bd_offset;
         }
 
-        shift    = s->sps->bit_depth + log2_trafo_size - 5;
+        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->sps->scaling_list_enable_flag) {
-            const ScalingList *sl = s->pps->scaling_list_data_present_flag ?
-                                    &s->pps->scaling_list : &s->sps->scaling_list;
+        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)
@@ -978,7 +1006,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
         }
     }
 
-    if (s->pps->transform_skip_enabled_flag &&
+    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);
@@ -1162,7 +1190,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
             if (first_greater1_coeff_idx != -1) {
                 coeff_abs_level_greater1_flag[first_greater1_coeff_idx] += ff_hevc_coeff_abs_level_greater2_flag_decode(s, c_idx, ctx_set);
             }
-            if (!s->pps->sign_data_hiding_flag || !sign_hidden) {
+            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));
@@ -1180,7 +1208,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
                     if ((trans_coeff_level) > (3 * (1 << c_rice_param)))
                         c_rice_param = FFMIN(c_rice_param + 1, 4);
                 }
-                if (s->pps->sign_data_hiding_flag && sign_hidden) {
+                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;
@@ -1189,7 +1217,7 @@ static void hls_residual_coding(HEVCContext *s, int x0, int y0,
                     trans_coeff_level = -trans_coeff_level;
                 coeff_sign_flag <<= 1;
                 if (!lc->cu.cu_transquant_bypass_flag) {
-                    if (s->sps->scaling_list_enable_flag) {
+                    if (s->ps.sps->scaling_list_enable_flag) {
                         if (y_c || x_c || log2_trafo_size < 4) {
                             int pos;
                             switch (log2_trafo_size) {
@@ -1243,12 +1271,12 @@ static int hls_transform_unit(HEVCContext *s, int x0, int y0,
 
         s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
         if (log2_trafo_size > 2) {
-            trafo_size = trafo_size << (s->sps->hshift[1] - 1);
+            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->sps->hshift[1];
+            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);
@@ -1260,21 +1288,21 @@ static int hls_transform_unit(HEVCContext *s, int x0, int y0,
         int scan_idx   = SCAN_DIAG;
         int scan_idx_c = SCAN_DIAG;
 
-        if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
+        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->sps->qp_bd_offset / 2) ||
-                lc->tu.cu_qp_delta >  (25 + s->sps->qp_bd_offset / 2)) {
+            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->sps->qp_bd_offset / 2),
-                        (25 + s->sps->qp_bd_offset / 2));
+                       -(26 + s->ps.sps->qp_bd_offset / 2),
+                        (25 + s->ps.sps->qp_bd_offset / 2));
                 return AVERROR_INVALIDDATA;
             }
 
@@ -1319,11 +1347,11 @@ static int hls_transform_unit(HEVCContext *s, int x0, int y0,
 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->sps->log2_min_pu_size;
+    int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
 
-    int min_pu_width     = s->sps->min_pu_width;
-    int x_end = FFMIN(x0 + cb_size, s->sps->width);
-    int y_end = FFMIN(y0 + cb_size, s->sps->height);
+    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++)
@@ -1348,18 +1376,18 @@ static int hls_transform_tree(HEVCContext *s, int x0, int y0,
         lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
     }
 
-    if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
-        log2_trafo_size >  s->sps->log2_min_tb_size    &&
+    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->sps->max_transform_hierarchy_depth_inter == 0 &&
+        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->sps->log2_max_trafo_size ||
+        split_transform_flag = log2_trafo_size > s->ps.sps->log2_max_trafo_size ||
                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
                                inter_split;
     }
@@ -1394,9 +1422,9 @@ do {
 
 #undef SUBDIVIDE
     } else {
-        int min_tu_size      = 1 << s->sps->log2_min_tb_size;
-        int log2_min_tu_size = s->sps->log2_min_tb_size;
-        int min_tu_width     = s->sps->min_tb_width;
+        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 ||
@@ -1420,7 +1448,7 @@ do {
         }
         if (!s->sh.disable_deblocking_filter_flag) {
             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
-            if (s->pps->transquant_bypass_enable_flag &&
+            if (s->ps.pps->transquant_bypass_enable_flag &&
                 lc->cu.cu_transquant_bypass_flag)
                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
         }
@@ -1435,13 +1463,13 @@ static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
     GetBitContext gb;
     int cb_size   = 1 << log2_cb_size;
     int stride0   = s->frame->linesize[0];
-    uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
+    uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->ps.sps->pixel_shift)];
     int   stride1 = s->frame->linesize[1];
-    uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->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)];
     int   stride2 = s->frame->linesize[2];
-    uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
+    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)];
 
-    int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
+    int length         = cb_size * cb_size * s->ps.sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->ps.sps->pcm.bit_depth_chroma;
     const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
     int ret;
 
@@ -1452,9 +1480,9 @@ static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
     if (ret < 0)
         return ret;
 
-    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
-    s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
-    s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
+    s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &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;
 }
 
@@ -1502,8 +1530,8 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
     HEVCLocalContext *lc = &s->HEVClc;
     uint8_t *src         = ref->data[0];
     ptrdiff_t srcstride  = ref->linesize[0];
-    int pic_width        = s->sps->width;
-    int pic_height       = s->sps->height;
+    int pic_width        = s->ps.sps->width;
+    int pic_height       = s->ps.sps->height;
 
     int mx         = mv->x & 3;
     int my         = mv->y & 3;
@@ -1512,15 +1540,15 @@ static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
 
     x_off += mv->x >> 2;
     y_off += mv->y >> 2;
-    src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
+    src   += y_off * srcstride + (x_off << s->ps.sps->pixel_shift);
 
     if (x_off < extra_left || y_off < extra_top ||
         x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
         y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
-        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
-        int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
+        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << 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->sps->pixel_shift);
+                         edge_emu_stride + (extra_left << s->ps.sps->pixel_shift);
 
         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
                                  edge_emu_stride, srcstride,
@@ -1558,27 +1586,27 @@ static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
     uint8_t *src2        = ref->data[2];
     ptrdiff_t src1stride = ref->linesize[1];
     ptrdiff_t src2stride = ref->linesize[2];
-    int pic_width        = s->sps->width >> 1;
-    int pic_height       = s->sps->height >> 1;
+    int 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 << s->sps->pixel_shift);
-    src2  += y_off * src2stride + (x_off << s->sps->pixel_shift);
+    src1  += y_off * src1stride + (x_off << s->ps.sps->pixel_shift);
+    src2  += y_off * src2stride + (x_off << 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) {
-        const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
-        int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
+        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->sps->pixel_shift));
-        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
+                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
+        int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->ps.sps->pixel_shift));
         int buf_offset2 = EPEL_EXTRA_BEFORE *
-                          (edge_emu_stride + (1 << s->sps->pixel_shift));
+                          (edge_emu_stride + (1 << s->ps.sps->pixel_shift));
 
         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
                                  edge_emu_stride, src1stride,
@@ -1670,13 +1698,13 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                                 int log2_cb_size, int partIdx)
 {
 #define POS(c_idx, x, y)                                                              \
-    &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
-                           (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
+    &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)]
     HEVCLocalContext *lc = &s->HEVClc;
     int merge_idx = 0;
     struct MvField current_mv = {{{ 0 }}};
 
-    int min_pu_width = s->sps->min_pu_width;
+    int min_pu_width = s->ps.sps->min_pu_width;
 
     MvField *tab_mvf = s->ref->tab_mvf;
     RefPicList  *refPicList = s->ref->refPicList;
@@ -1687,8 +1715,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
     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->sps->log2_min_cb_size;
-    int min_cb_width     = s->sps->min_cb_width;
+    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;
@@ -1712,11 +1740,11 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
                               partIdx, merge_idx, &current_mv);
     }
 
-    x_pu = x0 >> s->sps->log2_min_pu_size;
-    y_pu = y0 >> s->sps->log2_min_pu_size;
+    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->sps->log2_min_pu_size; j++)
-        for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
+    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[0]) {
@@ -1739,8 +1767,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         luma_mc(s, tmp, tmpstride, ref0->frame,
                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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]],
@@ -1752,8 +1780,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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],
@@ -1775,8 +1803,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         luma_mc(s, tmp, tmpstride, ref1->frame,
                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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]],
@@ -1789,8 +1817,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
                   &current_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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],
@@ -1814,8 +1842,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         luma_mc(s, tmp2, tmpstride, ref1->frame,
                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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]],
@@ -1833,8 +1861,8 @@ static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
         chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
                   &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
 
-        if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
-            (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
+        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(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],
@@ -1863,19 +1891,19 @@ static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
                                 int prev_intra_luma_pred_flag)
 {
     HEVCLocalContext *lc = &s->HEVClc;
-    int x_pu             = x0 >> s->sps->log2_min_pu_size;
-    int y_pu             = y0 >> s->sps->log2_min_pu_size;
-    int min_pu_width     = s->sps->min_pu_width;
-    int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
-    int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
-    int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
+    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;
+    int x0b              = x0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
+    int y0b              = y0 & ((1 << s->ps.sps->log2_ctb_size) - 1);
 
     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->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
+    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;
@@ -1950,13 +1978,13 @@ static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
 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->sps->log2_min_cb_size;
-    int x_cb   = x0 >> s->sps->log2_min_cb_size;
-    int y_cb   = y0 >> s->sps->log2_min_cb_size;
+    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->sps->min_cb_width + x_cb],
+        memset(&s->tab_ct_depth[(y_cb + y) * s->ps.sps->min_cb_width + x_cb],
                ct_depth, length);
 }
 
@@ -2006,11 +2034,11 @@ static void intra_prediction_unit_default_value(HEVCContext *s,
 {
     HEVCLocalContext *lc = &s->HEVClc;
     int pb_size          = 1 << log2_cb_size;
-    int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
-    int min_pu_width     = s->sps->min_pu_width;
+    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->sps->log2_min_pu_size;
-    int y_pu             = y0 >> s->sps->log2_min_pu_size;
+    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)
@@ -2026,9 +2054,9 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
 {
     int cb_size          = 1 << log2_cb_size;
     HEVCLocalContext *lc = &s->HEVClc;
-    int log2_min_cb_size = s->sps->log2_min_cb_size;
+    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->sps->min_cb_width;
+    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, y, ret;
@@ -2042,7 +2070,7 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
     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->pps->transquant_bypass_enable_flag) {
+    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);
@@ -2072,22 +2100,22 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
         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->sps->log2_min_cb_size) {
+            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->sps->pcm_enabled_flag &&
-                log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
-                log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
+            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->sps->pcm.loop_filter_disable_flag)
+                if (s->ps.sps->pcm.loop_filter_disable_flag)
                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
 
                 if (ret < 0)
@@ -2143,8 +2171,8 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
             }
             if (rqt_root_cbf) {
                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
-                                         s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
-                                         s->sps->max_transform_hierarchy_depth_inter;
+                                         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,
                                          log2_cb_size, 0, 0, 0, 0);
@@ -2157,7 +2185,7 @@ static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
         }
     }
 
-    if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
+    if (s->ps.pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
 
     x = y_cb * min_cb_width + x_cb;
@@ -2179,15 +2207,15 @@ static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
     int split_cu;
 
     lc->ct.depth = cb_depth;
-    if (x0 + cb_size <= s->sps->width  &&
-        y0 + cb_size <= s->sps->height &&
-        log2_cb_size > s->sps->log2_min_cb_size) {
+    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->sps->log2_min_cb_size);
+        split_cu = (log2_cb_size > s->ps.sps->log2_min_cb_size);
     }
-    if (s->pps->cu_qp_delta_enabled_flag &&
-        log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
+    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;
     }
@@ -2202,7 +2230,7 @@ static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
 
 #define SUBDIVIDE(x, y)                                                \
 do {                                                                   \
-    if (x < s->sps->width && y < s->sps->height) {                     \
+    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;                                                \
@@ -2226,77 +2254,77 @@ static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
                                  int ctb_addr_ts)
 {
     HEVCLocalContext *lc  = &s->HEVClc;
-    int ctb_size          = 1 << s->sps->log2_ctb_size;
-    int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+    int ctb_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->pps->entropy_coding_sync_enabled_flag) {
+    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->sps->width;
-    } else if (s->pps->tiles_enabled_flag) {
-        if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
-            int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
+        lc->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];
             lc->start_of_tiles_x = x_ctb;
-            lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
+            lc->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->sps->width;
+        lc->end_of_tiles_x = s->ps.sps->width;
     }
 
-    lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
+    lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->ps.sps->height);
 
     lc->boundary_flags = 0;
-    if (s->pps->tiles_enabled_flag) {
-        if (x_ctb > 0 && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]])
+    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->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]])
+        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->sps->ctb_width])
+        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 {
         if (!ctb_addr_in_slice > 0)
             lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
-        if (ctb_addr_in_slice < s->sps->ctb_width)
+        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->sps->ctb_width) && !(lc->boundary_flags & BOUNDARY_UPPER_TILE));
-    lc->ctb_up_right_flag = ((y_ctb > 0)  && (ctb_addr_in_slice+1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1 - s->sps->ctb_width]]));
-    lc->ctb_up_left_flag = ((x_ctb > 0) && (y_ctb > 0)  && (ctb_addr_in_slice-1 >= s->sps->ctb_width) && (s->pps->tile_id[ctb_addr_ts] == s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1 - s->sps->ctb_width]]));
+    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]]));
 }
 
 static int hls_slice_data(HEVCContext *s)
 {
-    int ctb_size    = 1 << s->sps->log2_ctb_size;
+    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->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
+    int ctb_addr_ts = s->ps.pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
     int ret;
 
-    while (more_data && ctb_addr_ts < s->sps->ctb_size) {
-        int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
+    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->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
-        y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
+        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->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
+        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;
 
-        ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
+        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);
@@ -2306,8 +2334,8 @@ static int hls_slice_data(HEVCContext *s)
         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
     }
 
-    if (x_ctb + ctb_size >= s->sps->width &&
-        y_ctb + ctb_size >= s->sps->height)
+    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);
 
     return ctb_addr_ts;
@@ -2341,20 +2369,20 @@ static int hls_nal_unit(HEVCContext *s)
 
 static void restore_tqb_pixels(HEVCContext *s)
 {
-    int min_pu_size = 1 << s->sps->log2_min_pu_size;
+    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->sps->hshift[c_idx];
-        int vshift       = s->sps->vshift[c_idx];
-        for (y = 0; y < s->sps->min_pu_height; y++) {
-            for (x = 0; x < s->sps->min_pu_width; x++) {
-                if (s->is_pcm[y * s->sps->min_pu_width + x]) {
+        int 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->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
-                    uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
+                    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;
@@ -2422,17 +2450,17 @@ 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));
-    memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
-    memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
+    memset(s->cbf_luma,      0, s->ps.sps->min_tb_width * s->ps.sps->min_tb_height);
+    memset(s->is_pcm,        0, s->ps.sps->min_pu_width * s->ps.sps->min_pu_height);
 
     lc->start_of_tiles_x = 0;
     s->is_decoded        = 0;
     s->first_nal_type    = s->nal_unit_type;
 
-    if (s->pps->tiles_enabled_flag)
-        lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
+    if (s->ps.pps->tiles_enabled_flag)
+        lc->end_of_tiles_x = s->ps.pps->column_width[0] << s->ps.sps->log2_ctb_size;
 
-    ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
+    ret = ff_hevc_set_new_ref(s, s->ps.sps->sao_enabled ? &s->sao_frame : &s->frame,
                               s->poc);
     if (ret < 0)
         goto fail;
@@ -2460,42 +2488,37 @@ static int hevc_frame_start(HEVCContext *s)
 
 fail:
     if (s->ref)
-        ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
+        ff_hevc_unref_frame(s, s->ref, ~0);
     s->ref = NULL;
     return ret;
 }
 
-static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
+static int decode_nal_unit(HEVCContext *s, const HEVCNAL *nal)
 {
     HEVCLocalContext *lc = &s->HEVClc;
     GetBitContext *gb    = &lc->gb;
     int ctb_addr_ts, ret;
 
-    ret = init_get_bits8(gb, nal, length);
+    ret = init_get_bits8(gb, nal->data, nal->size);
     if (ret < 0)
         return ret;
 
-    ret = hls_nal_unit(s);
-    if (ret < 0) {
-        av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
-               s->nal_unit_type);
-        goto fail;
-    } else if (!ret)
-        return 0;
+    hls_nal_unit(s);
 
     switch (s->nal_unit_type) {
     case NAL_VPS:
-        ret = ff_hevc_decode_nal_vps(s);
+        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(s);
+        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(s);
+        ret = ff_hevc_decode_nal_pps(gb, s->avctx, &s->ps);
         if (ret < 0)
             goto fail;
         break;
@@ -2569,18 +2592,30 @@ static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
             }
         }
 
-        ctb_addr_ts = hls_slice_data(s);
-        if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
-            s->is_decoded = 1;
-            if ((s->pps->transquant_bypass_enable_flag ||
-                 (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
-                s->sps->sao_enabled)
-                restore_tqb_pixels(s);
+        if (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 (ctb_addr_ts < 0) {
-            ret = ctb_addr_ts;
-            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 {
+            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;
+                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 (ctb_addr_ts < 0) {
+                ret = ctb_addr_ts;
+                goto fail;
+            }
         }
         break;
     case NAL_EOS_NUT:
@@ -2603,103 +2638,6 @@ fail:
     return 0;
 }
 
-/* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
- * between these functions would be nice. */
-static int extract_rbsp(const uint8_t *src, int length,
-                        HEVCNAL *nal)
-{
-    int i, si, di;
-    uint8_t *dst;
-
-#define STARTCODE_TEST                                                  \
-        if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
-            if (src[i + 2] != 3) {                                      \
-                /* startcode, so we must be past the end */             \
-                length = i;                                             \
-            }                                                           \
-            break;                                                      \
-        }
-#if HAVE_FAST_UNALIGNED
-#define FIND_FIRST_ZERO                                                 \
-        if (i > 0 && !src[i])                                           \
-            i--;                                                        \
-        while (src[i])                                                  \
-            i++
-#if HAVE_FAST_64BIT
-    for (i = 0; i + 1 < length; i += 9) {
-        if (!((~AV_RN64A(src + i) &
-               (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
-              0x8000800080008080ULL))
-            continue;
-        FIND_FIRST_ZERO;
-        STARTCODE_TEST;
-        i -= 7;
-    }
-#else
-    for (i = 0; i + 1 < length; i += 5) {
-        if (!((~AV_RN32A(src + i) &
-               (AV_RN32A(src + i) - 0x01000101U)) &
-              0x80008080U))
-            continue;
-        FIND_FIRST_ZERO;
-        STARTCODE_TEST;
-        i -= 3;
-    }
-#endif /* HAVE_FAST_64BIT */
-#else
-    for (i = 0; i + 1 < length; i += 2) {
-        if (src[i])
-            continue;
-        if (i > 0 && src[i - 1] == 0)
-            i--;
-        STARTCODE_TEST;
-    }
-#endif /* HAVE_FAST_UNALIGNED */
-
-    if (i >= length - 1) { // no escaped 0
-        nal->data = src;
-        nal->size = length;
-        return length;
-    }
-
-    av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
-                   length + FF_INPUT_BUFFER_PADDING_SIZE);
-    if (!nal->rbsp_buffer)
-        return AVERROR(ENOMEM);
-
-    dst = nal->rbsp_buffer;
-
-    memcpy(dst, src, i);
-    si = di = i;
-    while (si + 2 < length) {
-        // remove escapes (very rare 1:2^22)
-        if (src[si + 2] > 3) {
-            dst[di++] = src[si++];
-            dst[di++] = src[si++];
-        } else if (src[si] == 0 && src[si + 1] == 0) {
-            if (src[si + 2] == 3) { // escape
-                dst[di++] = 0;
-                dst[di++] = 0;
-                si       += 3;
-
-                continue;
-            } else // next start code
-                goto nsc;
-        }
-
-        dst[di++] = src[si++];
-    }
-    while (si < length)
-        dst[di++] = src[si++];
-
-nsc:
-    memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
-
-    nal->data = dst;
-    nal->size = di;
-    return si;
-}
-
 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
 {
     int i, consumed, ret = 0;
@@ -2756,7 +2694,7 @@ static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
         }
         nal = &s->nals[s->nb_nals++];
 
-        consumed = extract_rbsp(buf, extract_length, nal);
+        consumed = ff_hevc_extract_rbsp(buf, extract_length, nal);
         if (consumed < 0) {
             ret = consumed;
             goto fail;
@@ -2765,19 +2703,29 @@ static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
         ret = init_get_bits8(&s->HEVClc.gb, nal->data, nal->size);
         if (ret < 0)
             goto fail;
-        hls_nal_unit(s);
+
+        ret = hls_nal_unit(s);
+        if (ret <= 0) {
+            if (ret < 0) {
+                av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
+                       s->nal_unit_type);
+            }
+            s->nb_nals--;
+            goto skip_nal;
+        }
 
         if (s->nal_unit_type == NAL_EOB_NUT ||
             s->nal_unit_type == NAL_EOS_NUT)
             s->eos = 1;
 
+skip_nal:
         buf    += consumed;
         length -= consumed;
     }
 
     /* parse the NAL units */
     for (i = 0; i < s->nb_nals; i++) {
-        int ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
+        ret = decode_nal_unit(s, &s->nals[i]);
         if (ret < 0) {
             av_log(s->avctx, AV_LOG_WARNING,
                    "Error parsing NAL unit #%d.\n", i);
@@ -2885,13 +2833,19 @@ static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
     if (ret < 0)
         return ret;
 
-    /* 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;
+    if (avctx->hwaccel) {
+        if (s->ref && avctx->hwaccel->end_frame(avctx) < 0)
+            av_log(avctx, AV_LOG_ERROR,
+                   "hardware accelerator failed to decode picture\n");
+    } 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;
@@ -2935,6 +2889,13 @@ static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
     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);
@@ -2958,12 +2919,12 @@ static av_cold int hevc_decode_free(AVCodecContext *avctx)
         av_frame_free(&s->DPB[i].frame);
     }
 
-    for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
-        av_buffer_unref(&s->vps_list[i]);
-    for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
-        av_buffer_unref(&s->sps_list[i]);
-    for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
-        av_buffer_unref(&s->pps_list[i]);
+    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]);
 
     for (i = 0; i < s->nals_allocated; i++)
         av_freep(&s->nals[i].rbsp_buffer);
@@ -3034,35 +2995,35 @@ static int hevc_update_thread_context(AVCodecContext *dst,
         }
     }
 
-    for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
-        av_buffer_unref(&s->vps_list[i]);
-        if (s0->vps_list[i]) {
-            s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
-            if (!s->vps_list[i])
+    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->sps_list); i++) {
-        av_buffer_unref(&s->sps_list[i]);
-        if (s0->sps_list[i]) {
-            s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
-            if (!s->sps_list[i])
+    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->pps_list); i++) {
-        av_buffer_unref(&s->pps_list[i]);
-        if (s0->pps_list[i]) {
-            s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
-            if (!s->pps_list[i])
+    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->sps != s0->sps)
-        ret = set_sps(s, s0->sps);
+    if (s->ps.sps != s0->ps.sps)
+        ret = set_sps(s, s0->ps.sps);
 
     s->seq_decode = s0->seq_decode;
     s->seq_output = s0->seq_output;
@@ -3084,7 +3045,7 @@ static int hevc_decode_extradata(HEVCContext *s)
 {
     AVCodecContext *avctx = s->avctx;
     GetByteContext gb;
-    int ret;
+    int ret, i;
 
     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
 
@@ -3141,6 +3102,16 @@ static int hevc_decode_extradata(HEVCContext *s)
         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, sps);
+            break;
+        }
+    }
+
     return 0;
 }