]> git.sesse.net Git - ffmpeg/blob - libavcodec/hevc.c
x86/vp9lpf: add ff_vp9_loop_filter_[vh]_88_16_sse2()
[ffmpeg] / libavcodec / hevc.c
1 /*
2  * HEVC video Decoder
3  *
4  * Copyright (C) 2012 - 2013 Guillaume Martres
5  * Copyright (C) 2012 - 2013 Mickael Raulet
6  * Copyright (C) 2012 - 2013 Gildas Cocherel
7  * Copyright (C) 2012 - 2013 Wassim Hamidouche
8  *
9  * This file is part of FFmpeg.
10  *
11  * FFmpeg is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU Lesser General Public
13  * License as published by the Free Software Foundation; either
14  * version 2.1 of the License, or (at your option) any later version.
15  *
16  * FFmpeg is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * Lesser General Public License for more details.
20  *
21  * You should have received a copy of the GNU Lesser General Public
22  * License along with FFmpeg; if not, write to the Free Software
23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24  */
25
26 #include "libavutil/atomic.h"
27 #include "libavutil/attributes.h"
28 #include "libavutil/common.h"
29 #include "libavutil/internal.h"
30 #include "libavutil/md5.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/pixdesc.h"
33 #include "libavutil/stereo3d.h"
34
35 #include "bytestream.h"
36 #include "cabac_functions.h"
37 #include "dsputil.h"
38 #include "golomb.h"
39 #include "hevc.h"
40
41 const uint8_t ff_hevc_qpel_extra_before[4] = { 0, 3, 3, 2 };
42 const uint8_t ff_hevc_qpel_extra_after[4]  = { 0, 3, 4, 4 };
43 const uint8_t ff_hevc_qpel_extra[4]        = { 0, 6, 7, 6 };
44
45 /**
46  * NOTE: Each function hls_foo correspond to the function foo in the
47  * specification (HLS stands for High Level Syntax).
48  */
49
50 /**
51  * Section 5.7
52  */
53
54 /* free everything allocated  by pic_arrays_init() */
55 static void pic_arrays_free(HEVCContext *s)
56 {
57     av_freep(&s->sao);
58     av_freep(&s->deblock);
59     av_freep(&s->split_cu_flag);
60
61     av_freep(&s->skip_flag);
62     av_freep(&s->tab_ct_depth);
63
64     av_freep(&s->tab_ipm);
65     av_freep(&s->cbf_luma);
66     av_freep(&s->is_pcm);
67
68     av_freep(&s->qp_y_tab);
69     av_freep(&s->tab_slice_address);
70     av_freep(&s->filter_slice_edges);
71
72     av_freep(&s->horizontal_bs);
73     av_freep(&s->vertical_bs);
74
75     av_freep(&s->sh.entry_point_offset);
76     av_freep(&s->sh.size);
77     av_freep(&s->sh.offset);
78
79     av_buffer_pool_uninit(&s->tab_mvf_pool);
80     av_buffer_pool_uninit(&s->rpl_tab_pool);
81 }
82
83 /* allocate arrays that depend on frame dimensions */
84 static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
85 {
86     int log2_min_cb_size = sps->log2_min_cb_size;
87     int width            = sps->width;
88     int height           = sps->height;
89     int pic_size         = width * height;
90     int pic_size_in_ctb  = ((width  >> log2_min_cb_size) + 1) *
91                            ((height >> log2_min_cb_size) + 1);
92     int ctb_count        = sps->ctb_width * sps->ctb_height;
93     int min_pu_size      = sps->min_pu_width * sps->min_pu_height;
94
95     s->bs_width  = width  >> 3;
96     s->bs_height = height >> 3;
97
98     s->sao           = av_mallocz_array(ctb_count, sizeof(*s->sao));
99     s->deblock       = av_mallocz_array(ctb_count, sizeof(*s->deblock));
100     s->split_cu_flag = av_malloc(pic_size);
101     if (!s->sao || !s->deblock || !s->split_cu_flag)
102         goto fail;
103
104     s->skip_flag    = av_malloc(pic_size_in_ctb);
105     s->tab_ct_depth = av_malloc(sps->min_cb_height * sps->min_cb_width);
106     if (!s->skip_flag || !s->tab_ct_depth)
107         goto fail;
108
109     s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
110     s->tab_ipm  = av_mallocz(min_pu_size);
111     s->is_pcm   = av_malloc(min_pu_size);
112     if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
113         goto fail;
114
115     s->filter_slice_edges = av_malloc(ctb_count);
116     s->tab_slice_address  = av_malloc(pic_size_in_ctb *
117                                       sizeof(*s->tab_slice_address));
118     s->qp_y_tab           = av_malloc(pic_size_in_ctb *
119                                       sizeof(*s->qp_y_tab));
120     if (!s->qp_y_tab || !s->filter_slice_edges || !s->tab_slice_address)
121         goto fail;
122
123     s->horizontal_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
124     s->vertical_bs   = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
125     if (!s->horizontal_bs || !s->vertical_bs)
126         goto fail;
127
128     s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
129                                           av_buffer_alloc);
130     s->rpl_tab_pool = av_buffer_pool_init(ctb_count * sizeof(RefPicListTab),
131                                           av_buffer_allocz);
132     if (!s->tab_mvf_pool || !s->rpl_tab_pool)
133         goto fail;
134
135     return 0;
136
137 fail:
138     pic_arrays_free(s);
139     return AVERROR(ENOMEM);
140 }
141
142 static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
143 {
144     int i = 0;
145     int j = 0;
146     uint8_t luma_weight_l0_flag[16];
147     uint8_t chroma_weight_l0_flag[16];
148     uint8_t luma_weight_l1_flag[16];
149     uint8_t chroma_weight_l1_flag[16];
150
151     s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
152     if (s->sps->chroma_format_idc != 0) {
153         int delta = get_se_golomb(gb);
154         s->sh.chroma_log2_weight_denom = av_clip_c(s->sh.luma_log2_weight_denom + delta, 0, 7);
155     }
156
157     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
158         luma_weight_l0_flag[i] = get_bits1(gb);
159         if (!luma_weight_l0_flag[i]) {
160             s->sh.luma_weight_l0[i] = 1 << s->sh.luma_log2_weight_denom;
161             s->sh.luma_offset_l0[i] = 0;
162         }
163     }
164     if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
165         for (i = 0; i < s->sh.nb_refs[L0]; i++)
166             chroma_weight_l0_flag[i] = get_bits1(gb);
167     } else {
168         for (i = 0; i < s->sh.nb_refs[L0]; i++)
169             chroma_weight_l0_flag[i] = 0;
170     }
171     for (i = 0; i < s->sh.nb_refs[L0]; i++) {
172         if (luma_weight_l0_flag[i]) {
173             int delta_luma_weight_l0 = get_se_golomb(gb);
174             s->sh.luma_weight_l0[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l0;
175             s->sh.luma_offset_l0[i] = get_se_golomb(gb);
176         }
177         if (chroma_weight_l0_flag[i]) {
178             for (j = 0; j < 2; j++) {
179                 int delta_chroma_weight_l0 = get_se_golomb(gb);
180                 int delta_chroma_offset_l0 = get_se_golomb(gb);
181                 s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
182                 s->sh.chroma_offset_l0[i][j] = av_clip_c((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
183                                                                                     >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
184             }
185         } else {
186             s->sh.chroma_weight_l0[i][0] = 1 << s->sh.chroma_log2_weight_denom;
187             s->sh.chroma_offset_l0[i][0] = 0;
188             s->sh.chroma_weight_l0[i][1] = 1 << s->sh.chroma_log2_weight_denom;
189             s->sh.chroma_offset_l0[i][1] = 0;
190         }
191     }
192     if (s->sh.slice_type == B_SLICE) {
193         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
194             luma_weight_l1_flag[i] = get_bits1(gb);
195             if (!luma_weight_l1_flag[i]) {
196                 s->sh.luma_weight_l1[i] = 1 << s->sh.luma_log2_weight_denom;
197                 s->sh.luma_offset_l1[i] = 0;
198             }
199         }
200         if (s->sps->chroma_format_idc != 0) {
201             for (i = 0; i < s->sh.nb_refs[L1]; i++)
202                 chroma_weight_l1_flag[i] = get_bits1(gb);
203         } else {
204             for (i = 0; i < s->sh.nb_refs[L1]; i++)
205                 chroma_weight_l1_flag[i] = 0;
206         }
207         for (i = 0; i < s->sh.nb_refs[L1]; i++) {
208             if (luma_weight_l1_flag[i]) {
209                 int delta_luma_weight_l1 = get_se_golomb(gb);
210                 s->sh.luma_weight_l1[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l1;
211                 s->sh.luma_offset_l1[i] = get_se_golomb(gb);
212             }
213             if (chroma_weight_l1_flag[i]) {
214                 for (j = 0; j < 2; j++) {
215                     int delta_chroma_weight_l1 = get_se_golomb(gb);
216                     int delta_chroma_offset_l1 = get_se_golomb(gb);
217                     s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
218                     s->sh.chroma_offset_l1[i][j] = av_clip_c((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
219                                                                                         >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
220                 }
221             } else {
222                 s->sh.chroma_weight_l1[i][0] = 1 << s->sh.chroma_log2_weight_denom;
223                 s->sh.chroma_offset_l1[i][0] = 0;
224                 s->sh.chroma_weight_l1[i][1] = 1 << s->sh.chroma_log2_weight_denom;
225                 s->sh.chroma_offset_l1[i][1] = 0;
226             }
227         }
228     }
229 }
230
231 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
232 {
233     const HEVCSPS *sps = s->sps;
234     int max_poc_lsb    = 1 << sps->log2_max_poc_lsb;
235     int prev_delta_msb = 0;
236     unsigned int nb_sps = 0, nb_sh;
237     int i;
238
239     rps->nb_refs = 0;
240     if (!sps->long_term_ref_pics_present_flag)
241         return 0;
242
243     if (sps->num_long_term_ref_pics_sps > 0)
244         nb_sps = get_ue_golomb_long(gb);
245     nb_sh = get_ue_golomb_long(gb);
246
247     if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
248         return AVERROR_INVALIDDATA;
249
250     rps->nb_refs = nb_sh + nb_sps;
251
252     for (i = 0; i < rps->nb_refs; i++) {
253         uint8_t delta_poc_msb_present;
254
255         if (i < nb_sps) {
256             uint8_t lt_idx_sps = 0;
257
258             if (sps->num_long_term_ref_pics_sps > 1)
259                 lt_idx_sps = get_bits(gb, av_ceil_log2(sps->num_long_term_ref_pics_sps));
260
261             rps->poc[i]  = sps->lt_ref_pic_poc_lsb_sps[lt_idx_sps];
262             rps->used[i] = sps->used_by_curr_pic_lt_sps_flag[lt_idx_sps];
263         } else {
264             rps->poc[i]  = get_bits(gb, sps->log2_max_poc_lsb);
265             rps->used[i] = get_bits1(gb);
266         }
267
268         delta_poc_msb_present = get_bits1(gb);
269         if (delta_poc_msb_present) {
270             int delta = get_ue_golomb_long(gb);
271
272             if (i && i != nb_sps)
273                 delta += prev_delta_msb;
274
275             rps->poc[i] += s->poc - delta * max_poc_lsb - s->sh.pic_order_cnt_lsb;
276             prev_delta_msb = delta;
277         }
278     }
279
280     return 0;
281 }
282
283 static int set_sps(HEVCContext *s, const HEVCSPS *sps)
284 {
285     int ret;
286     int num = 0, den = 0;
287
288     pic_arrays_free(s);
289     ret = pic_arrays_init(s, sps);
290     if (ret < 0)
291         goto fail;
292
293     s->avctx->coded_width         = sps->width;
294     s->avctx->coded_height        = sps->height;
295     s->avctx->width               = sps->output_width;
296     s->avctx->height              = sps->output_height;
297     s->avctx->pix_fmt             = sps->pix_fmt;
298     s->avctx->sample_aspect_ratio = sps->vui.sar;
299     s->avctx->has_b_frames        = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
300
301     if (sps->vui.video_signal_type_present_flag)
302         s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
303                                                                : AVCOL_RANGE_MPEG;
304     else
305         s->avctx->color_range = AVCOL_RANGE_MPEG;
306
307     if (sps->vui.colour_description_present_flag) {
308         s->avctx->color_primaries = sps->vui.colour_primaries;
309         s->avctx->color_trc       = sps->vui.transfer_characteristic;
310         s->avctx->colorspace      = sps->vui.matrix_coeffs;
311     } else {
312         s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
313         s->avctx->color_trc       = AVCOL_TRC_UNSPECIFIED;
314         s->avctx->colorspace      = AVCOL_SPC_UNSPECIFIED;
315     }
316
317     ff_hevc_pred_init(&s->hpc,     sps->bit_depth);
318     ff_hevc_dsp_init (&s->hevcdsp, sps->bit_depth);
319     ff_videodsp_init (&s->vdsp,    sps->bit_depth);
320
321     if (sps->sao_enabled) {
322         av_frame_unref(s->tmp_frame);
323         ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
324         if (ret < 0)
325             goto fail;
326         s->frame = s->tmp_frame;
327     }
328
329     s->sps = sps;
330     s->vps = (HEVCVPS*) s->vps_list[s->sps->vps_id]->data;
331
332     if (s->vps->vps_timing_info_present_flag) {
333         num = s->vps->vps_num_units_in_tick;
334         den = s->vps->vps_time_scale;
335     } else if (sps->vui.vui_timing_info_present_flag) {
336         num = sps->vui.vui_num_units_in_tick;
337         den = sps->vui.vui_time_scale;
338     }
339
340     if (num != 0 && den != 0)
341         av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
342                   num, den, 1 << 30);
343
344     return 0;
345
346 fail:
347     pic_arrays_free(s);
348     s->sps = NULL;
349     return ret;
350 }
351
352 static int hls_slice_header(HEVCContext *s)
353 {
354     GetBitContext *gb = &s->HEVClc->gb;
355     SliceHeader *sh   = &s->sh;
356     int i, j, ret;
357
358     // Coded parameters
359     sh->first_slice_in_pic_flag = get_bits1(gb);
360     if ((IS_IDR(s) || IS_BLA(s)) && sh->first_slice_in_pic_flag) {
361         s->seq_decode = (s->seq_decode + 1) & 0xff;
362         s->max_ra     = INT_MAX;
363         if (IS_IDR(s))
364             ff_hevc_clear_refs(s);
365     }
366     if (s->nal_unit_type >= 16 && s->nal_unit_type <= 23)
367         sh->no_output_of_prior_pics_flag = get_bits1(gb);
368
369     sh->pps_id = get_ue_golomb_long(gb);
370     if (sh->pps_id >= MAX_PPS_COUNT || !s->pps_list[sh->pps_id]) {
371         av_log(s->avctx, AV_LOG_ERROR, "PPS id out of range: %d\n", sh->pps_id);
372         return AVERROR_INVALIDDATA;
373     }
374     if (!sh->first_slice_in_pic_flag &&
375         s->pps != (HEVCPPS*)s->pps_list[sh->pps_id]->data) {
376         av_log(s->avctx, AV_LOG_ERROR, "PPS changed between slices.\n");
377         return AVERROR_INVALIDDATA;
378     }
379     s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
380
381     if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
382         s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
383         ff_hevc_clear_refs(s);
384         ret = set_sps(s, s->sps);
385         if (ret < 0)
386             return ret;
387
388         s->seq_decode = (s->seq_decode + 1) & 0xff;
389         s->max_ra     = INT_MAX;
390     }
391
392     s->avctx->profile = s->sps->ptl.general_ptl.profile_idc;
393     s->avctx->level   = s->sps->ptl.general_ptl.level_idc;
394
395     sh->dependent_slice_segment_flag = 0;
396     if (!sh->first_slice_in_pic_flag) {
397         int slice_address_length;
398
399         if (s->pps->dependent_slice_segments_enabled_flag)
400             sh->dependent_slice_segment_flag = get_bits1(gb);
401
402         slice_address_length = av_ceil_log2(s->sps->ctb_width *
403                                             s->sps->ctb_height);
404         sh->slice_segment_addr = get_bits(gb, slice_address_length);
405         if (sh->slice_segment_addr >= s->sps->ctb_width * s->sps->ctb_height) {
406             av_log(s->avctx, AV_LOG_ERROR,
407                    "Invalid slice segment address: %u.\n",
408                    sh->slice_segment_addr);
409             return AVERROR_INVALIDDATA;
410         }
411
412         if (!sh->dependent_slice_segment_flag) {
413             sh->slice_addr = sh->slice_segment_addr;
414             s->slice_idx++;
415         }
416     } else {
417         sh->slice_segment_addr = sh->slice_addr = 0;
418         s->slice_idx           = 0;
419         s->slice_initialized   = 0;
420     }
421
422     if (!sh->dependent_slice_segment_flag) {
423         s->slice_initialized = 0;
424
425         for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
426             skip_bits(gb, 1);  // slice_reserved_undetermined_flag[]
427
428         sh->slice_type = get_ue_golomb_long(gb);
429         if (!(sh->slice_type == I_SLICE ||
430               sh->slice_type == P_SLICE ||
431               sh->slice_type == B_SLICE)) {
432             av_log(s->avctx, AV_LOG_ERROR, "Unknown slice type: %d.\n",
433                    sh->slice_type);
434             return AVERROR_INVALIDDATA;
435         }
436         if (IS_IRAP(s) && sh->slice_type != I_SLICE) {
437             av_log(s->avctx, AV_LOG_ERROR, "Inter slices in an IRAP frame.\n");
438             return AVERROR_INVALIDDATA;
439         }
440
441         if (s->pps->output_flag_present_flag)
442             sh->pic_output_flag = get_bits1(gb);
443
444         if (s->sps->separate_colour_plane_flag)
445             sh->colour_plane_id = get_bits(gb, 2);
446
447         if (!IS_IDR(s)) {
448             int short_term_ref_pic_set_sps_flag, poc;
449
450             sh->pic_order_cnt_lsb = get_bits(gb, s->sps->log2_max_poc_lsb);
451             poc = ff_hevc_compute_poc(s, sh->pic_order_cnt_lsb);
452             if (!sh->first_slice_in_pic_flag && poc != s->poc) {
453                 av_log(s->avctx, AV_LOG_WARNING,
454                        "Ignoring POC change between slices: %d -> %d\n", s->poc, poc);
455                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
456                     return AVERROR_INVALIDDATA;
457                 poc = s->poc;
458             }
459             s->poc = poc;
460
461             short_term_ref_pic_set_sps_flag = get_bits1(gb);
462             if (!short_term_ref_pic_set_sps_flag) {
463                 ret = ff_hevc_decode_short_term_rps(s, &sh->slice_rps, s->sps, 1);
464                 if (ret < 0)
465                     return ret;
466
467                 sh->short_term_rps = &sh->slice_rps;
468             } else {
469                 int numbits, rps_idx;
470
471                 if (!s->sps->nb_st_rps) {
472                     av_log(s->avctx, AV_LOG_ERROR, "No ref lists in the SPS.\n");
473                     return AVERROR_INVALIDDATA;
474                 }
475
476                 numbits = av_ceil_log2(s->sps->nb_st_rps);
477                 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
478                 sh->short_term_rps = &s->sps->st_rps[rps_idx];
479             }
480
481             ret = decode_lt_rps(s, &sh->long_term_rps, gb);
482             if (ret < 0) {
483                 av_log(s->avctx, AV_LOG_WARNING, "Invalid long term RPS.\n");
484                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
485                     return AVERROR_INVALIDDATA;
486             }
487
488             if (s->sps->sps_temporal_mvp_enabled_flag)
489                 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
490             else
491                 sh->slice_temporal_mvp_enabled_flag = 0;
492         } else {
493             s->sh.short_term_rps = NULL;
494             s->poc               = 0;
495         }
496
497         /* 8.3.1 */
498         if (s->temporal_id == 0 &&
499             s->nal_unit_type != NAL_TRAIL_N &&
500             s->nal_unit_type != NAL_TSA_N   &&
501             s->nal_unit_type != NAL_STSA_N  &&
502             s->nal_unit_type != NAL_RADL_N  &&
503             s->nal_unit_type != NAL_RADL_R  &&
504             s->nal_unit_type != NAL_RASL_N  &&
505             s->nal_unit_type != NAL_RASL_R)
506             s->pocTid0 = s->poc;
507
508         if (s->sps->sao_enabled) {
509             sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
510             sh->slice_sample_adaptive_offset_flag[1] =
511             sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
512         } else {
513             sh->slice_sample_adaptive_offset_flag[0] = 0;
514             sh->slice_sample_adaptive_offset_flag[1] = 0;
515             sh->slice_sample_adaptive_offset_flag[2] = 0;
516         }
517
518         sh->nb_refs[L0] = sh->nb_refs[L1] = 0;
519         if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
520             int nb_refs;
521
522             sh->nb_refs[L0] = s->pps->num_ref_idx_l0_default_active;
523             if (sh->slice_type == B_SLICE)
524                 sh->nb_refs[L1] = s->pps->num_ref_idx_l1_default_active;
525
526             if (get_bits1(gb)) { // num_ref_idx_active_override_flag
527                 sh->nb_refs[L0] = get_ue_golomb_long(gb) + 1;
528                 if (sh->slice_type == B_SLICE)
529                     sh->nb_refs[L1] = get_ue_golomb_long(gb) + 1;
530             }
531             if (sh->nb_refs[L0] > MAX_REFS || sh->nb_refs[L1] > MAX_REFS) {
532                 av_log(s->avctx, AV_LOG_ERROR, "Too many refs: %d/%d.\n",
533                        sh->nb_refs[L0], sh->nb_refs[L1]);
534                 return AVERROR_INVALIDDATA;
535             }
536
537             sh->rpl_modification_flag[0] = 0;
538             sh->rpl_modification_flag[1] = 0;
539             nb_refs = ff_hevc_frame_nb_refs(s);
540             if (!nb_refs) {
541                 av_log(s->avctx, AV_LOG_ERROR, "Zero refs for a frame with P or B slices.\n");
542                 return AVERROR_INVALIDDATA;
543             }
544
545             if (s->pps->lists_modification_present_flag && nb_refs > 1) {
546                 sh->rpl_modification_flag[0] = get_bits1(gb);
547                 if (sh->rpl_modification_flag[0]) {
548                     for (i = 0; i < sh->nb_refs[L0]; i++)
549                         sh->list_entry_lx[0][i] = get_bits(gb, av_ceil_log2(nb_refs));
550                 }
551
552                 if (sh->slice_type == B_SLICE) {
553                     sh->rpl_modification_flag[1] = get_bits1(gb);
554                     if (sh->rpl_modification_flag[1] == 1)
555                         for (i = 0; i < sh->nb_refs[L1]; i++)
556                             sh->list_entry_lx[1][i] = get_bits(gb, av_ceil_log2(nb_refs));
557                 }
558             }
559
560             if (sh->slice_type == B_SLICE)
561                 sh->mvd_l1_zero_flag = get_bits1(gb);
562
563             if (s->pps->cabac_init_present_flag)
564                 sh->cabac_init_flag = get_bits1(gb);
565             else
566                 sh->cabac_init_flag = 0;
567
568             sh->collocated_ref_idx = 0;
569             if (sh->slice_temporal_mvp_enabled_flag) {
570                 sh->collocated_list = L0;
571                 if (sh->slice_type == B_SLICE)
572                     sh->collocated_list = !get_bits1(gb);
573
574                 if (sh->nb_refs[sh->collocated_list] > 1) {
575                     sh->collocated_ref_idx = get_ue_golomb_long(gb);
576                     if (sh->collocated_ref_idx >= sh->nb_refs[sh->collocated_list]) {
577                         av_log(s->avctx, AV_LOG_ERROR,
578                                "Invalid collocated_ref_idx: %d.\n",
579                                sh->collocated_ref_idx);
580                         return AVERROR_INVALIDDATA;
581                     }
582                 }
583             }
584
585             if ((s->pps->weighted_pred_flag   && sh->slice_type == P_SLICE) ||
586                 (s->pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
587                 pred_weight_table(s, gb);
588             }
589
590             sh->max_num_merge_cand = 5 - get_ue_golomb_long(gb);
591             if (sh->max_num_merge_cand < 1 || sh->max_num_merge_cand > 5) {
592                 av_log(s->avctx, AV_LOG_ERROR,
593                        "Invalid number of merging MVP candidates: %d.\n",
594                        sh->max_num_merge_cand);
595                 return AVERROR_INVALIDDATA;
596             }
597         }
598
599         sh->slice_qp_delta = get_se_golomb(gb);
600
601         if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
602             sh->slice_cb_qp_offset = get_se_golomb(gb);
603             sh->slice_cr_qp_offset = get_se_golomb(gb);
604         } else {
605             sh->slice_cb_qp_offset = 0;
606             sh->slice_cr_qp_offset = 0;
607         }
608
609         if (s->pps->deblocking_filter_control_present_flag) {
610             int deblocking_filter_override_flag = 0;
611
612             if (s->pps->deblocking_filter_override_enabled_flag)
613                 deblocking_filter_override_flag = get_bits1(gb);
614
615             if (deblocking_filter_override_flag) {
616                 sh->disable_deblocking_filter_flag = get_bits1(gb);
617                 if (!sh->disable_deblocking_filter_flag) {
618                     sh->beta_offset = get_se_golomb(gb) * 2;
619                     sh->tc_offset   = get_se_golomb(gb) * 2;
620                 }
621             } else {
622                 sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
623                 sh->beta_offset                    = s->pps->beta_offset;
624                 sh->tc_offset                      = s->pps->tc_offset;
625             }
626         } else {
627             sh->disable_deblocking_filter_flag = 0;
628             sh->beta_offset                    = 0;
629             sh->tc_offset                      = 0;
630         }
631
632         if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
633             (sh->slice_sample_adaptive_offset_flag[0] ||
634              sh->slice_sample_adaptive_offset_flag[1] ||
635              !sh->disable_deblocking_filter_flag)) {
636             sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
637         } else {
638             sh->slice_loop_filter_across_slices_enabled_flag = s->pps->seq_loop_filter_across_slices_enabled_flag;
639         }
640     } else if (!s->slice_initialized) {
641         av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
642         return AVERROR_INVALIDDATA;
643     }
644
645     sh->num_entry_point_offsets = 0;
646     if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
647         sh->num_entry_point_offsets = get_ue_golomb_long(gb);
648         if (sh->num_entry_point_offsets > 0) {
649             int offset_len = get_ue_golomb_long(gb) + 1;
650             int segments = offset_len >> 4;
651             int rest = (offset_len & 15);
652             av_freep(&sh->entry_point_offset);
653             av_freep(&sh->offset);
654             av_freep(&sh->size);
655             sh->entry_point_offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
656             sh->offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
657             sh->size = av_malloc(sh->num_entry_point_offsets * sizeof(int));
658             if (!sh->entry_point_offset || !sh->offset || !sh->size) {
659                 sh->num_entry_point_offsets = 0;
660                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate memory\n");
661                 return AVERROR(ENOMEM);
662             }
663             for (i = 0; i < sh->num_entry_point_offsets; i++) {
664                 int val = 0;
665                 for (j = 0; j < segments; j++) {
666                     val <<= 16;
667                     val += get_bits(gb, 16);
668                 }
669                 if (rest) {
670                     val <<= rest;
671                     val += get_bits(gb, rest);
672                 }
673                 sh->entry_point_offset[i] = val + 1; // +1; // +1 to get the size
674             }
675             if (s->threads_number > 1 && (s->pps->num_tile_rows > 1 || s->pps->num_tile_columns > 1)) {
676                 s->enable_parallel_tiles = 0; // TODO: you can enable tiles in parallel here
677                 s->threads_number = 1;
678             } else
679                 s->enable_parallel_tiles = 0;
680         } else
681             s->enable_parallel_tiles = 0;
682     }
683
684     if (s->pps->slice_header_extension_present_flag) {
685         unsigned int length = get_ue_golomb_long(gb);
686         for (i = 0; i < length; i++)
687             skip_bits(gb, 8);  // slice_header_extension_data_byte
688     }
689
690     // Inferred parameters
691     sh->slice_qp = 26U + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
692     if (sh->slice_qp > 51 ||
693         sh->slice_qp < -s->sps->qp_bd_offset) {
694         av_log(s->avctx, AV_LOG_ERROR,
695                "The slice_qp %d is outside the valid range "
696                "[%d, 51].\n",
697                sh->slice_qp,
698                -s->sps->qp_bd_offset);
699         return AVERROR_INVALIDDATA;
700     }
701
702     sh->slice_ctb_addr_rs = sh->slice_segment_addr;
703
704     s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
705
706     if (!s->pps->cu_qp_delta_enabled_flag)
707         s->HEVClc->qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset,
708                                  52 + s->sps->qp_bd_offset) - s->sps->qp_bd_offset;
709
710     s->slice_initialized = 1;
711
712     return 0;
713 }
714
715 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
716
717 #define SET_SAO(elem, value)                            \
718 do {                                                    \
719     if (!sao_merge_up_flag && !sao_merge_left_flag)     \
720         sao->elem = value;                              \
721     else if (sao_merge_left_flag)                       \
722         sao->elem = CTB(s->sao, rx-1, ry).elem;         \
723     else if (sao_merge_up_flag)                         \
724         sao->elem = CTB(s->sao, rx, ry-1).elem;         \
725     else                                                \
726         sao->elem = 0;                                  \
727 } while (0)
728
729 static void hls_sao_param(HEVCContext *s, int rx, int ry)
730 {
731     HEVCLocalContext *lc    = s->HEVClc;
732     int sao_merge_left_flag = 0;
733     int sao_merge_up_flag   = 0;
734     int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
735     SAOParams *sao          = &CTB(s->sao, rx, ry);
736     int c_idx, i;
737
738     if (s->sh.slice_sample_adaptive_offset_flag[0] ||
739         s->sh.slice_sample_adaptive_offset_flag[1]) {
740         if (rx > 0) {
741             if (lc->ctb_left_flag)
742                 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
743         }
744         if (ry > 0 && !sao_merge_left_flag) {
745             if (lc->ctb_up_flag)
746                 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
747         }
748     }
749
750     for (c_idx = 0; c_idx < 3; c_idx++) {
751         if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
752             sao->type_idx[c_idx] = SAO_NOT_APPLIED;
753             continue;
754         }
755
756         if (c_idx == 2) {
757             sao->type_idx[2] = sao->type_idx[1];
758             sao->eo_class[2] = sao->eo_class[1];
759         } else {
760             SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
761         }
762
763         if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
764             continue;
765
766         for (i = 0; i < 4; i++)
767             SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
768
769         if (sao->type_idx[c_idx] == SAO_BAND) {
770             for (i = 0; i < 4; i++) {
771                 if (sao->offset_abs[c_idx][i]) {
772                     SET_SAO(offset_sign[c_idx][i],
773                             ff_hevc_sao_offset_sign_decode(s));
774                 } else {
775                     sao->offset_sign[c_idx][i] = 0;
776                 }
777             }
778             SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
779         } else if (c_idx != 2) {
780             SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
781         }
782
783         // Inferred parameters
784         sao->offset_val[c_idx][0] = 0;
785         for (i = 0; i < 4; i++) {
786             sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
787             if (sao->type_idx[c_idx] == SAO_EDGE) {
788                 if (i > 1)
789                     sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
790             } else if (sao->offset_sign[c_idx][i]) {
791                 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
792             }
793         }
794     }
795 }
796
797 #undef SET_SAO
798 #undef CTB
799
800 static int hls_transform_unit(HEVCContext *s, int x0, int y0,
801                               int xBase, int yBase, int cb_xBase, int cb_yBase,
802                               int log2_cb_size, int log2_trafo_size,
803                               int trafo_depth, int blk_idx)
804 {
805     HEVCLocalContext *lc = s->HEVClc;
806
807     if (lc->cu.pred_mode == MODE_INTRA) {
808         int trafo_size = 1 << log2_trafo_size;
809         ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
810
811         s->hpc.intra_pred(s, x0, y0, log2_trafo_size, 0);
812         if (log2_trafo_size > 2) {
813             trafo_size = trafo_size << (s->sps->hshift[1] - 1);
814             ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
815             s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 1);
816             s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 2);
817         } else if (blk_idx == 3) {
818             trafo_size = trafo_size << s->sps->hshift[1];
819             ff_hevc_set_neighbour_available(s, xBase, yBase,
820                                             trafo_size, trafo_size);
821             s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 1);
822             s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 2);
823         }
824     }
825
826     if (lc->tt.cbf_luma ||
827         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
828         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
829         int scan_idx   = SCAN_DIAG;
830         int scan_idx_c = SCAN_DIAG;
831
832         if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
833             lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
834             if (lc->tu.cu_qp_delta != 0)
835                 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
836                     lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
837             lc->tu.is_cu_qp_delta_coded = 1;
838
839             if (lc->tu.cu_qp_delta < -(26 + s->sps->qp_bd_offset / 2) ||
840                 lc->tu.cu_qp_delta >  (25 + s->sps->qp_bd_offset / 2)) {
841                 av_log(s->avctx, AV_LOG_ERROR,
842                        "The cu_qp_delta %d is outside the valid range "
843                        "[%d, %d].\n",
844                        lc->tu.cu_qp_delta,
845                        -(26 + s->sps->qp_bd_offset / 2),
846                         (25 + s->sps->qp_bd_offset / 2));
847                 return AVERROR_INVALIDDATA;
848             }
849
850             ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
851         }
852
853         if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
854             if (lc->tu.cur_intra_pred_mode >= 6 &&
855                 lc->tu.cur_intra_pred_mode <= 14) {
856                 scan_idx = SCAN_VERT;
857             } else if (lc->tu.cur_intra_pred_mode >= 22 &&
858                        lc->tu.cur_intra_pred_mode <= 30) {
859                 scan_idx = SCAN_HORIZ;
860             }
861
862             if (lc->pu.intra_pred_mode_c >=  6 &&
863                 lc->pu.intra_pred_mode_c <= 14) {
864                 scan_idx_c = SCAN_VERT;
865             } else if (lc->pu.intra_pred_mode_c >= 22 &&
866                        lc->pu.intra_pred_mode_c <= 30) {
867                 scan_idx_c = SCAN_HORIZ;
868             }
869         }
870
871         if (lc->tt.cbf_luma)
872             ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
873         if (log2_trafo_size > 2) {
874             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
875                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
876             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
877                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
878         } else if (blk_idx == 3) {
879             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
880                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
881             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
882                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
883         }
884     }
885     return 0;
886 }
887
888 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
889 {
890     int cb_size          = 1 << log2_cb_size;
891     int log2_min_pu_size = s->sps->log2_min_pu_size;
892
893     int min_pu_width     = s->sps->min_pu_width;
894     int x_end = FFMIN(x0 + cb_size, s->sps->width);
895     int y_end = FFMIN(y0 + cb_size, s->sps->height);
896     int i, j;
897
898     for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
899         for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
900             s->is_pcm[i + j * min_pu_width] = 2;
901 }
902
903 static int hls_transform_tree(HEVCContext *s, int x0, int y0,
904                               int xBase, int yBase, int cb_xBase, int cb_yBase,
905                               int log2_cb_size, int log2_trafo_size,
906                               int trafo_depth, int blk_idx)
907 {
908     HEVCLocalContext *lc = s->HEVClc;
909     uint8_t split_transform_flag;
910     int ret;
911
912     if (trafo_depth > 0 && log2_trafo_size == 2) {
913         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
914             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase);
915         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
916             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase);
917     } else {
918         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
919         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) = 0;
920     }
921
922     if (lc->cu.intra_split_flag) {
923         if (trafo_depth == 1)
924             lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
925     } else {
926         lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
927     }
928
929     lc->tt.cbf_luma = 1;
930
931     lc->tt.inter_split_flag = s->sps->max_transform_hierarchy_depth_inter == 0 &&
932                               lc->cu.pred_mode == MODE_INTER &&
933                               lc->cu.part_mode != PART_2Nx2N &&
934                               trafo_depth == 0;
935
936     if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
937         log2_trafo_size >  s->sps->log2_min_tb_size    &&
938         trafo_depth     < lc->cu.max_trafo_depth       &&
939         !(lc->cu.intra_split_flag && trafo_depth == 0)) {
940         split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
941     } else {
942         split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
943                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
944                                lc->tt.inter_split_flag;
945     }
946
947     if (log2_trafo_size > 2) {
948         if (trafo_depth == 0 ||
949             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase)) {
950             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
951                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
952         }
953
954         if (trafo_depth == 0 ||
955             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase)) {
956             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
957                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
958         }
959     }
960
961     if (split_transform_flag) {
962         int x1 = x0 + ((1 << log2_trafo_size) >> 1);
963         int y1 = y0 + ((1 << log2_trafo_size) >> 1);
964
965         ret = hls_transform_tree(s, x0, y0, x0, y0, cb_xBase, cb_yBase,
966                                  log2_cb_size, log2_trafo_size - 1,
967                                  trafo_depth + 1, 0);
968         if (ret < 0)
969             return ret;
970         ret = hls_transform_tree(s, x1, y0, x0, y0, cb_xBase, cb_yBase,
971                                  log2_cb_size, log2_trafo_size - 1,
972                                  trafo_depth + 1, 1);
973         if (ret < 0)
974             return ret;
975         ret = hls_transform_tree(s, x0, y1, x0, y0, cb_xBase, cb_yBase,
976                                  log2_cb_size, log2_trafo_size - 1,
977                                  trafo_depth + 1, 2);
978         if (ret < 0)
979             return ret;
980         ret = hls_transform_tree(s, x1, y1, x0, y0, cb_xBase, cb_yBase,
981                                  log2_cb_size, log2_trafo_size - 1,
982                                  trafo_depth + 1, 3);
983         if (ret < 0)
984             return ret;
985     } else {
986         int min_tu_size      = 1 << s->sps->log2_min_tb_size;
987         int log2_min_tu_size = s->sps->log2_min_tb_size;
988         int min_tu_width     = s->sps->min_tb_width;
989
990         if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
991             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
992             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
993             lc->tt.cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
994         }
995
996         ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
997                                  log2_cb_size, log2_trafo_size, trafo_depth,
998                                  blk_idx);
999         if (ret < 0)
1000             return ret;
1001         // TODO: store cbf_luma somewhere else
1002         if (lc->tt.cbf_luma) {
1003             int i, j;
1004             for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
1005                 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
1006                     int x_tu = (x0 + j) >> log2_min_tu_size;
1007                     int y_tu = (y0 + i) >> log2_min_tu_size;
1008                     s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
1009                 }
1010         }
1011         if (!s->sh.disable_deblocking_filter_flag) {
1012             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size,
1013                                                   lc->slice_or_tiles_up_boundary,
1014                                                   lc->slice_or_tiles_left_boundary);
1015             if (s->pps->transquant_bypass_enable_flag &&
1016                 lc->cu.cu_transquant_bypass_flag)
1017                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
1018         }
1019     }
1020     return 0;
1021 }
1022
1023 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
1024 {
1025     //TODO: non-4:2:0 support
1026     HEVCLocalContext *lc = s->HEVClc;
1027     GetBitContext gb;
1028     int cb_size   = 1 << log2_cb_size;
1029     int stride0   = s->frame->linesize[0];
1030     uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
1031     int   stride1 = s->frame->linesize[1];
1032     uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
1033     int   stride2 = s->frame->linesize[2];
1034     uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
1035
1036     int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
1037     const uint8_t *pcm = skip_bytes(&s->HEVClc->cc, (length + 7) >> 3);
1038     int ret;
1039
1040     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1041                                           lc->slice_or_tiles_up_boundary,
1042                                           lc->slice_or_tiles_left_boundary);
1043
1044     ret = init_get_bits(&gb, pcm, length);
1045     if (ret < 0)
1046         return ret;
1047
1048     s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
1049     s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1050     s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1051     return 0;
1052 }
1053
1054 /**
1055  * 8.5.3.2.2.1 Luma sample interpolation process
1056  *
1057  * @param s HEVC decoding context
1058  * @param dst target buffer for block data at block position
1059  * @param dststride stride of the dst buffer
1060  * @param ref reference picture buffer at origin (0, 0)
1061  * @param mv motion vector (relative to block position) to get pixel data from
1062  * @param x_off horizontal position of block from origin (0, 0)
1063  * @param y_off vertical position of block from origin (0, 0)
1064  * @param block_w width of block
1065  * @param block_h height of block
1066  */
1067 static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
1068                     AVFrame *ref, const Mv *mv, int x_off, int y_off,
1069                     int block_w, int block_h)
1070 {
1071     HEVCLocalContext *lc = s->HEVClc;
1072     uint8_t *src         = ref->data[0];
1073     ptrdiff_t srcstride  = ref->linesize[0];
1074     int pic_width        = s->sps->width;
1075     int pic_height       = s->sps->height;
1076
1077     int mx         = mv->x & 3;
1078     int my         = mv->y & 3;
1079     int extra_left = ff_hevc_qpel_extra_before[mx];
1080     int extra_top  = ff_hevc_qpel_extra_before[my];
1081
1082     x_off += mv->x >> 2;
1083     y_off += mv->y >> 2;
1084     src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
1085
1086     if (x_off < extra_left || y_off < extra_top ||
1087         x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
1088         y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
1089         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1090         int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
1091         int buf_offset = extra_top *
1092                          edge_emu_stride + (extra_left << s->sps->pixel_shift);
1093
1094         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1095                                  edge_emu_stride, srcstride,
1096                                  block_w + ff_hevc_qpel_extra[mx],
1097                                  block_h + ff_hevc_qpel_extra[my],
1098                                  x_off - extra_left, y_off - extra_top,
1099                                  pic_width, pic_height);
1100         src = lc->edge_emu_buffer + buf_offset;
1101         srcstride = edge_emu_stride;
1102     }
1103     s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
1104                                      block_h, lc->mc_buffer);
1105 }
1106
1107 /**
1108  * 8.5.3.2.2.2 Chroma sample interpolation process
1109  *
1110  * @param s HEVC decoding context
1111  * @param dst1 target buffer for block data at block position (U plane)
1112  * @param dst2 target buffer for block data at block position (V plane)
1113  * @param dststride stride of the dst1 and dst2 buffers
1114  * @param ref reference picture buffer at origin (0, 0)
1115  * @param mv motion vector (relative to block position) to get pixel data from
1116  * @param x_off horizontal position of block from origin (0, 0)
1117  * @param y_off vertical position of block from origin (0, 0)
1118  * @param block_w width of block
1119  * @param block_h height of block
1120  */
1121 static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
1122                       ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
1123                       int x_off, int y_off, int block_w, int block_h)
1124 {
1125     HEVCLocalContext *lc = s->HEVClc;
1126     uint8_t *src1        = ref->data[1];
1127     uint8_t *src2        = ref->data[2];
1128     ptrdiff_t src1stride = ref->linesize[1];
1129     ptrdiff_t src2stride = ref->linesize[2];
1130     int pic_width        = s->sps->width >> 1;
1131     int pic_height       = s->sps->height >> 1;
1132
1133     int mx = mv->x & 7;
1134     int my = mv->y & 7;
1135
1136     x_off += mv->x >> 3;
1137     y_off += mv->y >> 3;
1138     src1  += y_off * src1stride + (x_off << s->sps->pixel_shift);
1139     src2  += y_off * src2stride + (x_off << s->sps->pixel_shift);
1140
1141     if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1142         x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1143         y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1144         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1145         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1146         int buf_offset1 = EPEL_EXTRA_BEFORE *
1147                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1148         int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1149         int buf_offset2 = EPEL_EXTRA_BEFORE *
1150                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1151
1152         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1153                                  edge_emu_stride, src1stride,
1154                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1155                                  x_off - EPEL_EXTRA_BEFORE,
1156                                  y_off - EPEL_EXTRA_BEFORE,
1157                                  pic_width, pic_height);
1158
1159         src1 = lc->edge_emu_buffer + buf_offset1;
1160         src1stride = edge_emu_stride;
1161         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1162                                              block_w, block_h, mx, my, lc->mc_buffer);
1163
1164         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
1165                                  edge_emu_stride, src2stride,
1166                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1167                                  x_off - EPEL_EXTRA_BEFORE,
1168                                  y_off - EPEL_EXTRA_BEFORE,
1169                                  pic_width, pic_height);
1170         src2 = lc->edge_emu_buffer + buf_offset2;
1171         src2stride = edge_emu_stride;
1172
1173         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1174                                              block_w, block_h, mx, my,
1175                                              lc->mc_buffer);
1176     } else {
1177         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1178                                              block_w, block_h, mx, my,
1179                                              lc->mc_buffer);
1180         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1181                                              block_w, block_h, mx, my,
1182                                              lc->mc_buffer);
1183     }
1184 }
1185
1186 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1187                                 const Mv *mv, int y0, int height)
1188 {
1189     int y = (mv->y >> 2) + y0 + height + 9;
1190
1191     if (s->threads_type == FF_THREAD_FRAME )
1192         ff_thread_await_progress(&ref->tf, y, 0);
1193 }
1194
1195 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1196                                 int nPbW, int nPbH,
1197                                 int log2_cb_size, int partIdx)
1198 {
1199 #define POS(c_idx, x, y)                                                              \
1200     &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1201                            (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1202     HEVCLocalContext *lc = s->HEVClc;
1203     int merge_idx = 0;
1204     struct MvField current_mv = {{{ 0 }}};
1205
1206     int min_pu_width = s->sps->min_pu_width;
1207
1208     MvField *tab_mvf = s->ref->tab_mvf;
1209     RefPicList  *refPicList = s->ref->refPicList;
1210     HEVCFrame *ref0, *ref1;
1211
1212     int tmpstride = MAX_PB_SIZE;
1213
1214     uint8_t *dst0 = POS(0, x0, y0);
1215     uint8_t *dst1 = POS(1, x0, y0);
1216     uint8_t *dst2 = POS(2, x0, y0);
1217     int log2_min_cb_size = s->sps->log2_min_cb_size;
1218     int min_cb_width     = s->sps->min_cb_width;
1219     int x_cb             = x0 >> log2_min_cb_size;
1220     int y_cb             = y0 >> log2_min_cb_size;
1221     int ref_idx[2];
1222     int mvp_flag[2];
1223     int x_pu, y_pu;
1224     int i, j;
1225
1226     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1227         if (s->sh.max_num_merge_cand > 1)
1228             merge_idx = ff_hevc_merge_idx_decode(s);
1229         else
1230             merge_idx = 0;
1231
1232         ff_hevc_luma_mv_merge_mode(s, x0, y0,
1233                                    1 << log2_cb_size,
1234                                    1 << log2_cb_size,
1235                                    log2_cb_size, partIdx,
1236                                    merge_idx, &current_mv);
1237         x_pu = x0 >> s->sps->log2_min_pu_size;
1238         y_pu = y0 >> s->sps->log2_min_pu_size;
1239
1240         for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1241             for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1242                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1243     } else { /* MODE_INTER */
1244         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1245         if (lc->pu.merge_flag) {
1246             if (s->sh.max_num_merge_cand > 1)
1247                 merge_idx = ff_hevc_merge_idx_decode(s);
1248             else
1249                 merge_idx = 0;
1250
1251             ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1252                                        partIdx, merge_idx, &current_mv);
1253             x_pu = x0 >> s->sps->log2_min_pu_size;
1254             y_pu = y0 >> s->sps->log2_min_pu_size;
1255
1256             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1257                 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1258                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1259         } else {
1260             enum InterPredIdc inter_pred_idc = PRED_L0;
1261             ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1262             if (s->sh.slice_type == B_SLICE)
1263                 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1264
1265             if (inter_pred_idc != PRED_L1) {
1266                 if (s->sh.nb_refs[L0]) {
1267                     ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1268                     current_mv.ref_idx[0] = ref_idx[0];
1269                 }
1270                 current_mv.pred_flag[0] = 1;
1271                 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
1272                 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
1273                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1274                                          partIdx, merge_idx, &current_mv,
1275                                          mvp_flag[0], 0);
1276                 current_mv.mv[0].x += lc->pu.mvd.x;
1277                 current_mv.mv[0].y += lc->pu.mvd.y;
1278             }
1279
1280             if (inter_pred_idc != PRED_L0) {
1281                 if (s->sh.nb_refs[L1]) {
1282                     ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1283                     current_mv.ref_idx[1] = ref_idx[1];
1284                 }
1285
1286                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1287                     lc->pu.mvd.x = 0;
1288                     lc->pu.mvd.y = 0;
1289                 } else {
1290                     ff_hevc_hls_mvd_coding(s, x0, y0, 1);
1291                 }
1292
1293                 current_mv.pred_flag[1] = 1;
1294                 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
1295                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1296                                          partIdx, merge_idx, &current_mv,
1297                                          mvp_flag[1], 1);
1298                 current_mv.mv[1].x += lc->pu.mvd.x;
1299                 current_mv.mv[1].y += lc->pu.mvd.y;
1300             }
1301
1302             x_pu = x0 >> s->sps->log2_min_pu_size;
1303             y_pu = y0 >> s->sps->log2_min_pu_size;
1304
1305             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1306                 for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1307                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1308         }
1309     }
1310
1311     if (current_mv.pred_flag[0]) {
1312         ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1313         if (!ref0)
1314             return;
1315         hevc_await_progress(s, ref0, &current_mv.mv[0], y0, nPbH);
1316     }
1317     if (current_mv.pred_flag[1]) {
1318         ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1319         if (!ref1)
1320             return;
1321         hevc_await_progress(s, ref1, &current_mv.mv[1], y0, nPbH);
1322     }
1323
1324     if (current_mv.pred_flag[0] && !current_mv.pred_flag[1]) {
1325         DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1326         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1327
1328         luma_mc(s, tmp, tmpstride, ref0->frame,
1329                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
1330
1331         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1332             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1333             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1334                                      s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1335                                      s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1336                                      dst0, s->frame->linesize[0], tmp,
1337                                      tmpstride, nPbW, nPbH);
1338         } else {
1339             s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1340         }
1341         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1342                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1343
1344         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1345             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1346             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1347                                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1348                                      s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1349                                      dst1, s->frame->linesize[1], tmp, tmpstride,
1350                                      nPbW / 2, nPbH / 2);
1351             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1352                                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1353                                      s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1354                                      dst2, s->frame->linesize[2], tmp2, tmpstride,
1355                                      nPbW / 2, nPbH / 2);
1356         } else {
1357             s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1358             s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1359         }
1360     } else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1361         DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1362         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1363
1364         if (!ref1)
1365             return;
1366
1367         luma_mc(s, tmp, tmpstride, ref1->frame,
1368                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
1369
1370         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1371             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1372             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1373                                       s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1374                                       s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1375                                       dst0, s->frame->linesize[0], tmp, tmpstride,
1376                                       nPbW, nPbH);
1377         } else {
1378             s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1379         }
1380
1381         chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
1382                   &current_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
1383
1384         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1385             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1386             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1387                                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1388                                      s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1389                                      dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1390             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1391                                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1392                                      s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1393                                      dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1394         } else {
1395             s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1396             s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1397         }
1398     } else if (current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1399         DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1400         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1401         DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
1402         DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
1403         HEVCFrame *ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1404         HEVCFrame *ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1405
1406         if (!ref0 || !ref1)
1407             return;
1408
1409         luma_mc(s, tmp, tmpstride, ref0->frame,
1410                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
1411         luma_mc(s, tmp2, tmpstride, ref1->frame,
1412                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
1413
1414         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1415             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1416             s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
1417                                          s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1418                                          s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1419                                          s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1420                                          s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1421                                          dst0, s->frame->linesize[0],
1422                                          tmp, tmp2, tmpstride, nPbW, nPbH);
1423         } else {
1424             s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
1425                                              tmp, tmp2, tmpstride, nPbW, nPbH);
1426         }
1427
1428         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1429                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1430         chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
1431                   &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1432
1433         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1434             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1435             s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1436                                          s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1437                                          s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1438                                          s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1439                                          s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1440                                          dst1, s->frame->linesize[1], tmp, tmp3,
1441                                          tmpstride, nPbW / 2, nPbH / 2);
1442             s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1443                                          s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1444                                          s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1445                                          s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1446                                          s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1447                                          dst2, s->frame->linesize[2], tmp2, tmp4,
1448                                          tmpstride, nPbW / 2, nPbH / 2);
1449         } else {
1450             s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
1451             s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
1452         }
1453     }
1454 }
1455
1456 /**
1457  * 8.4.1
1458  */
1459 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1460                                 int prev_intra_luma_pred_flag)
1461 {
1462     HEVCLocalContext *lc = s->HEVClc;
1463     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1464     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1465     int min_pu_width     = s->sps->min_pu_width;
1466     int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
1467     int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1468     int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1469
1470     int cand_up   = (lc->ctb_up_flag || y0b) ?
1471                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1472     int cand_left = (lc->ctb_left_flag || x0b) ?
1473                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
1474
1475     int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1476
1477     MvField *tab_mvf = s->ref->tab_mvf;
1478     int intra_pred_mode;
1479     int candidate[3];
1480     int i, j;
1481
1482     // intra_pred_mode prediction does not cross vertical CTB boundaries
1483     if ((y0 - 1) < y_ctb)
1484         cand_up = INTRA_DC;
1485
1486     if (cand_left == cand_up) {
1487         if (cand_left < 2) {
1488             candidate[0] = INTRA_PLANAR;
1489             candidate[1] = INTRA_DC;
1490             candidate[2] = INTRA_ANGULAR_26;
1491         } else {
1492             candidate[0] = cand_left;
1493             candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1494             candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1495         }
1496     } else {
1497         candidate[0] = cand_left;
1498         candidate[1] = cand_up;
1499         if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1500             candidate[2] = INTRA_PLANAR;
1501         } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1502             candidate[2] = INTRA_DC;
1503         } else {
1504             candidate[2] = INTRA_ANGULAR_26;
1505         }
1506     }
1507
1508     if (prev_intra_luma_pred_flag) {
1509         intra_pred_mode = candidate[lc->pu.mpm_idx];
1510     } else {
1511         if (candidate[0] > candidate[1])
1512             FFSWAP(uint8_t, candidate[0], candidate[1]);
1513         if (candidate[0] > candidate[2])
1514             FFSWAP(uint8_t, candidate[0], candidate[2]);
1515         if (candidate[1] > candidate[2])
1516             FFSWAP(uint8_t, candidate[1], candidate[2]);
1517
1518         intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1519         for (i = 0; i < 3; i++)
1520             if (intra_pred_mode >= candidate[i])
1521                 intra_pred_mode++;
1522     }
1523
1524     /* write the intra prediction units into the mv array */
1525     if (!size_in_pus)
1526         size_in_pus = 1;
1527     for (i = 0; i < size_in_pus; i++) {
1528         memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1529                intra_pred_mode, size_in_pus);
1530
1531         for (j = 0; j < size_in_pus; j++) {
1532             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].is_intra     = 1;
1533             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[0] = 0;
1534             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[1] = 0;
1535             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[0]   = 0;
1536             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[1]   = 0;
1537             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].x      = 0;
1538             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].y      = 0;
1539             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].x      = 0;
1540             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].y      = 0;
1541         }
1542     }
1543
1544     return intra_pred_mode;
1545 }
1546
1547 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1548                                           int log2_cb_size, int ct_depth)
1549 {
1550     int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1551     int x_cb   = x0 >> s->sps->log2_min_cb_size;
1552     int y_cb   = y0 >> s->sps->log2_min_cb_size;
1553     int y;
1554
1555     for (y = 0; y < length; y++)
1556         memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1557                ct_depth, length);
1558 }
1559
1560 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1561                                   int log2_cb_size)
1562 {
1563     HEVCLocalContext *lc = s->HEVClc;
1564     static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1565     uint8_t prev_intra_luma_pred_flag[4];
1566     int split   = lc->cu.part_mode == PART_NxN;
1567     int pb_size = (1 << log2_cb_size) >> split;
1568     int side    = split + 1;
1569     int chroma_mode;
1570     int i, j;
1571
1572     for (i = 0; i < side; i++)
1573         for (j = 0; j < side; j++)
1574             prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1575
1576     for (i = 0; i < side; i++) {
1577         for (j = 0; j < side; j++) {
1578             if (prev_intra_luma_pred_flag[2 * i + j])
1579                 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1580             else
1581                 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1582
1583             lc->pu.intra_pred_mode[2 * i + j] =
1584                 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1585                                      prev_intra_luma_pred_flag[2 * i + j]);
1586         }
1587     }
1588
1589     chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1590     if (chroma_mode != 4) {
1591         if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1592             lc->pu.intra_pred_mode_c = 34;
1593         else
1594             lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1595     } else {
1596         lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1597     }
1598 }
1599
1600 static void intra_prediction_unit_default_value(HEVCContext *s,
1601                                                 int x0, int y0,
1602                                                 int log2_cb_size)
1603 {
1604     HEVCLocalContext *lc = s->HEVClc;
1605     int pb_size          = 1 << log2_cb_size;
1606     int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
1607     int min_pu_width     = s->sps->min_pu_width;
1608     MvField *tab_mvf     = s->ref->tab_mvf;
1609     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1610     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1611     int j, k;
1612
1613     if (size_in_pus == 0)
1614         size_in_pus = 1;
1615     for (j = 0; j < size_in_pus; j++) {
1616         memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
1617         for (k = 0; k < size_in_pus; k++)
1618             tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].is_intra = lc->cu.pred_mode == MODE_INTRA;
1619     }
1620 }
1621
1622 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
1623 {
1624     int cb_size          = 1 << log2_cb_size;
1625     HEVCLocalContext *lc = s->HEVClc;
1626     int log2_min_cb_size = s->sps->log2_min_cb_size;
1627     int length           = cb_size >> log2_min_cb_size;
1628     int min_cb_width     = s->sps->min_cb_width;
1629     int x_cb             = x0 >> log2_min_cb_size;
1630     int y_cb             = y0 >> log2_min_cb_size;
1631     int x, y, ret;
1632
1633     lc->cu.x                = x0;
1634     lc->cu.y                = y0;
1635     lc->cu.rqt_root_cbf     = 1;
1636     lc->cu.pred_mode        = MODE_INTRA;
1637     lc->cu.part_mode        = PART_2Nx2N;
1638     lc->cu.intra_split_flag = 0;
1639     lc->cu.pcm_flag         = 0;
1640
1641     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
1642     for (x = 0; x < 4; x++)
1643         lc->pu.intra_pred_mode[x] = 1;
1644     if (s->pps->transquant_bypass_enable_flag) {
1645         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
1646         if (lc->cu.cu_transquant_bypass_flag)
1647             set_deblocking_bypass(s, x0, y0, log2_cb_size);
1648     } else
1649         lc->cu.cu_transquant_bypass_flag = 0;
1650
1651     if (s->sh.slice_type != I_SLICE) {
1652         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
1653
1654         lc->cu.pred_mode = MODE_SKIP;
1655         x = y_cb * min_cb_width + x_cb;
1656         for (y = 0; y < length; y++) {
1657             memset(&s->skip_flag[x], skip_flag, length);
1658             x += min_cb_width;
1659         }
1660         lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
1661     }
1662
1663     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1664         hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1665         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1666
1667         if (!s->sh.disable_deblocking_filter_flag)
1668             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1669                                                   lc->slice_or_tiles_up_boundary,
1670                                                   lc->slice_or_tiles_left_boundary);
1671     } else {
1672         if (s->sh.slice_type != I_SLICE)
1673             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
1674         if (lc->cu.pred_mode != MODE_INTRA ||
1675             log2_cb_size == s->sps->log2_min_cb_size) {
1676             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
1677             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
1678                                       lc->cu.pred_mode == MODE_INTRA;
1679         }
1680
1681         if (lc->cu.pred_mode == MODE_INTRA) {
1682             if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
1683                 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
1684                 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
1685                 lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
1686             }
1687             if (lc->cu.pcm_flag) {
1688                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1689                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
1690                 if (s->sps->pcm.loop_filter_disable_flag)
1691                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
1692
1693                 if (ret < 0)
1694                     return ret;
1695             } else {
1696                 intra_prediction_unit(s, x0, y0, log2_cb_size);
1697             }
1698         } else {
1699             intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1700             switch (lc->cu.part_mode) {
1701             case PART_2Nx2N:
1702                 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1703                 break;
1704             case PART_2NxN:
1705                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
1706                 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
1707                 break;
1708             case PART_Nx2N:
1709                 hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
1710                 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
1711                 break;
1712             case PART_2NxnU:
1713                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
1714                 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
1715                 break;
1716             case PART_2NxnD:
1717                 hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
1718                 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
1719                 break;
1720             case PART_nLx2N:
1721                 hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
1722                 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
1723                 break;
1724             case PART_nRx2N:
1725                 hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
1726                 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
1727                 break;
1728             case PART_NxN:
1729                 hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
1730                 hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
1731                 hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
1732                 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
1733                 break;
1734             }
1735         }
1736
1737         if (!lc->cu.pcm_flag) {
1738             if (lc->cu.pred_mode != MODE_INTRA &&
1739                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
1740                 lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
1741             }
1742             if (lc->cu.rqt_root_cbf) {
1743                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
1744                                          s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
1745                                          s->sps->max_transform_hierarchy_depth_inter;
1746                 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
1747                                          log2_cb_size,
1748                                          log2_cb_size, 0, 0);
1749                 if (ret < 0)
1750                     return ret;
1751             } else {
1752                 if (!s->sh.disable_deblocking_filter_flag)
1753                     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1754                                                           lc->slice_or_tiles_up_boundary,
1755                                                           lc->slice_or_tiles_left_boundary);
1756             }
1757         }
1758     }
1759
1760     if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
1761         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
1762
1763     x = y_cb * min_cb_width + x_cb;
1764     for (y = 0; y < length; y++) {
1765         memset(&s->qp_y_tab[x], lc->qp_y, length);
1766         x += min_cb_width;
1767     }
1768
1769     set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
1770
1771     return 0;
1772 }
1773
1774 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
1775                                int log2_cb_size, int cb_depth)
1776 {
1777     HEVCLocalContext *lc = s->HEVClc;
1778     const int cb_size    = 1 << log2_cb_size;
1779     int ret;
1780
1781     lc->ct.depth = cb_depth;
1782     if (x0 + cb_size <= s->sps->width  &&
1783         y0 + cb_size <= s->sps->height &&
1784         log2_cb_size > s->sps->log2_min_cb_size) {
1785         SAMPLE(s->split_cu_flag, x0, y0) =
1786             ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
1787     } else {
1788         SAMPLE(s->split_cu_flag, x0, y0) =
1789             (log2_cb_size > s->sps->log2_min_cb_size);
1790     }
1791     if (s->pps->cu_qp_delta_enabled_flag &&
1792         log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
1793         lc->tu.is_cu_qp_delta_coded = 0;
1794         lc->tu.cu_qp_delta          = 0;
1795     }
1796
1797     if (SAMPLE(s->split_cu_flag, x0, y0)) {
1798         const int cb_size_split = cb_size >> 1;
1799         const int x1 = x0 + cb_size_split;
1800         const int y1 = y0 + cb_size_split;
1801
1802         int more_data = 0;
1803
1804         more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
1805         if (more_data < 0)
1806             return more_data;
1807
1808         if (more_data && x1 < s->sps->width)
1809             more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
1810         if (more_data && y1 < s->sps->height)
1811             more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
1812         if (more_data && x1 < s->sps->width &&
1813             y1 < s->sps->height) {
1814             return hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
1815         }
1816         if (more_data)
1817             return ((x1 + cb_size_split) < s->sps->width ||
1818                     (y1 + cb_size_split) < s->sps->height);
1819         else
1820             return 0;
1821     } else {
1822         ret = hls_coding_unit(s, x0, y0, log2_cb_size);
1823         if (ret < 0)
1824             return ret;
1825         if ((!((x0 + cb_size) %
1826                (1 << (s->sps->log2_ctb_size))) ||
1827              (x0 + cb_size >= s->sps->width)) &&
1828             (!((y0 + cb_size) %
1829                (1 << (s->sps->log2_ctb_size))) ||
1830              (y0 + cb_size >= s->sps->height))) {
1831             int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
1832             return !end_of_slice_flag;
1833         } else {
1834             return 1;
1835         }
1836     }
1837
1838     return 0;
1839 }
1840
1841 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
1842                                  int ctb_addr_ts)
1843 {
1844     HEVCLocalContext *lc  = s->HEVClc;
1845     int ctb_size          = 1 << s->sps->log2_ctb_size;
1846     int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1847     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
1848
1849     int tile_left_boundary, tile_up_boundary;
1850     int slice_left_boundary, slice_up_boundary;
1851
1852     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
1853
1854     if (s->pps->entropy_coding_sync_enabled_flag) {
1855         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
1856             lc->first_qp_group = 1;
1857         lc->end_of_tiles_x = s->sps->width;
1858     } else if (s->pps->tiles_enabled_flag) {
1859         if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
1860             int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
1861             lc->start_of_tiles_x = x_ctb;
1862             lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
1863             lc->first_qp_group   = 1;
1864         }
1865     } else {
1866         lc->end_of_tiles_x = s->sps->width;
1867     }
1868
1869     lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
1870
1871     if (s->pps->tiles_enabled_flag) {
1872         tile_left_boundary = x_ctb > 0 &&
1873                              s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
1874         slice_left_boundary = x_ctb > 0 &&
1875                               s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
1876         tile_up_boundary  = y_ctb > 0 &&
1877                             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]];
1878         slice_up_boundary = y_ctb > 0 &&
1879                             s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
1880     } else {
1881         tile_left_boundary =
1882         tile_up_boundary   = 0;
1883         slice_left_boundary = ctb_addr_in_slice <= 0;
1884         slice_up_boundary   = ctb_addr_in_slice < s->sps->ctb_width;
1885     }
1886     lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
1887     lc->slice_or_tiles_up_boundary   = slice_up_boundary   + (tile_up_boundary   << 1);
1888     lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0)                  && !tile_left_boundary);
1889     lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
1890     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]]));
1891     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]]));
1892 }
1893
1894 static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
1895 {
1896     HEVCContext *s  = avctxt->priv_data;
1897     int ctb_size    = 1 << s->sps->log2_ctb_size;
1898     int more_data   = 1;
1899     int x_ctb       = 0;
1900     int y_ctb       = 0;
1901     int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
1902
1903     if (!ctb_addr_ts && s->sh.dependent_slice_segment_flag) {
1904         av_log(s->avctx, AV_LOG_ERROR, "Impossible initial tile.\n");
1905         return AVERROR_INVALIDDATA;
1906     }
1907
1908     while (more_data && ctb_addr_ts < s->sps->ctb_size) {
1909         int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1910
1911         x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1912         y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1913         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
1914
1915         ff_hevc_cabac_init(s, ctb_addr_ts);
1916
1917         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
1918
1919         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
1920         s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
1921         s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
1922
1923         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
1924         if (more_data < 0)
1925             return more_data;
1926
1927         ctb_addr_ts++;
1928         ff_hevc_save_states(s, ctb_addr_ts);
1929         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
1930     }
1931
1932     if (x_ctb + ctb_size >= s->sps->width &&
1933         y_ctb + ctb_size >= s->sps->height)
1934         ff_hevc_hls_filter(s, x_ctb, y_ctb);
1935
1936     return ctb_addr_ts;
1937 }
1938
1939 static int hls_slice_data(HEVCContext *s)
1940 {
1941     int arg[2];
1942     int ret[2];
1943
1944     arg[0] = 0;
1945     arg[1] = 1;
1946
1947     s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
1948     return ret[0];
1949 }
1950 static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
1951 {
1952     HEVCContext *s1  = avctxt->priv_data, *s;
1953     HEVCLocalContext *lc;
1954     int ctb_size    = 1<< s1->sps->log2_ctb_size;
1955     int more_data   = 1;
1956     int *ctb_row_p    = input_ctb_row;
1957     int ctb_row = ctb_row_p[job];
1958     int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
1959     int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
1960     int thread = ctb_row % s1->threads_number;
1961     int ret;
1962
1963     s = s1->sList[self_id];
1964     lc = s->HEVClc;
1965
1966     if(ctb_row) {
1967         ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
1968
1969         if (ret < 0)
1970             return ret;
1971         ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
1972     }
1973
1974     while(more_data && ctb_addr_ts < s->sps->ctb_size) {
1975         int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
1976         int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
1977
1978         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
1979
1980         ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
1981
1982         if (avpriv_atomic_int_get(&s1->wpp_err)){
1983             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
1984             return 0;
1985         }
1986
1987         ff_hevc_cabac_init(s, ctb_addr_ts);
1988         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
1989         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
1990
1991         if (more_data < 0)
1992             return more_data;
1993
1994         ctb_addr_ts++;
1995
1996         ff_hevc_save_states(s, ctb_addr_ts);
1997         ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
1998         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
1999
2000         if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
2001             avpriv_atomic_int_set(&s1->wpp_err,  1);
2002             ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2003             return 0;
2004         }
2005
2006         if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
2007             ff_hevc_hls_filter(s, x_ctb, y_ctb);
2008             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2009             return ctb_addr_ts;
2010         }
2011         ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2012         x_ctb+=ctb_size;
2013
2014         if(x_ctb >= s->sps->width) {
2015             break;
2016         }
2017     }
2018     ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2019
2020     return 0;
2021 }
2022
2023 static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
2024 {
2025     HEVCLocalContext *lc = s->HEVClc;
2026     int *ret = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2027     int *arg = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2028     int offset;
2029     int startheader, cmpt = 0;
2030     int i, j, res = 0;
2031
2032
2033     if (!s->sList[1]) {
2034         ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
2035
2036
2037         for (i = 1; i < s->threads_number; i++) {
2038             s->sList[i] = av_malloc(sizeof(HEVCContext));
2039             memcpy(s->sList[i], s, sizeof(HEVCContext));
2040             s->HEVClcList[i] = av_malloc(sizeof(HEVCLocalContext));
2041             s->sList[i]->HEVClc = s->HEVClcList[i];
2042         }
2043     }
2044
2045     offset = (lc->gb.index >> 3);
2046
2047     for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
2048         if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2049             startheader--;
2050             cmpt++;
2051         }
2052     }
2053
2054     for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
2055         offset += (s->sh.entry_point_offset[i - 1] - cmpt);
2056         for (j = 0, cmpt = 0, startheader = offset
2057              + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
2058             if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2059                 startheader--;
2060                 cmpt++;
2061             }
2062         }
2063         s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
2064         s->sh.offset[i - 1] = offset;
2065
2066     }
2067     if (s->sh.num_entry_point_offsets != 0) {
2068         offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
2069         s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
2070         s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
2071
2072     }
2073     s->data = nal;
2074
2075     for (i = 1; i < s->threads_number; i++) {
2076         s->sList[i]->HEVClc->first_qp_group = 1;
2077         s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
2078         memcpy(s->sList[i], s, sizeof(HEVCContext));
2079         s->sList[i]->HEVClc = s->HEVClcList[i];
2080     }
2081
2082     avpriv_atomic_int_set(&s->wpp_err, 0);
2083     ff_reset_entries(s->avctx);
2084
2085     for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
2086         arg[i] = i;
2087         ret[i] = 0;
2088     }
2089
2090     if (s->pps->entropy_coding_sync_enabled_flag)
2091         s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
2092
2093     for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
2094         res += ret[i];
2095     av_free(ret);
2096     av_free(arg);
2097     return res;
2098 }
2099
2100 /**
2101  * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2102  * 0 if the unit should be skipped, 1 otherwise
2103  */
2104 static int hls_nal_unit(HEVCContext *s)
2105 {
2106     GetBitContext *gb = &s->HEVClc->gb;
2107     int nuh_layer_id;
2108
2109     if (get_bits1(gb) != 0)
2110         return AVERROR_INVALIDDATA;
2111
2112     s->nal_unit_type = get_bits(gb, 6);
2113
2114     nuh_layer_id   = get_bits(gb, 6);
2115     s->temporal_id = get_bits(gb, 3) - 1;
2116     if (s->temporal_id < 0)
2117         return AVERROR_INVALIDDATA;
2118
2119     av_log(s->avctx, AV_LOG_DEBUG,
2120            "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2121            s->nal_unit_type, nuh_layer_id, s->temporal_id);
2122
2123     return nuh_layer_id == 0;
2124 }
2125
2126 static void restore_tqb_pixels(HEVCContext *s)
2127 {
2128     int min_pu_size = 1 << s->sps->log2_min_pu_size;
2129     int x, y, c_idx;
2130
2131     for (c_idx = 0; c_idx < 3; c_idx++) {
2132         ptrdiff_t stride = s->frame->linesize[c_idx];
2133         int hshift       = s->sps->hshift[c_idx];
2134         int vshift       = s->sps->vshift[c_idx];
2135         for (y = 0; y < s->sps->min_pu_height; y++) {
2136             for (x = 0; x < s->sps->min_pu_width; x++) {
2137                 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2138                     int n;
2139                     int len      = min_pu_size >> hshift;
2140                     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)];
2141                     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)];
2142                     for (n = 0; n < (min_pu_size >> vshift); n++) {
2143                         memcpy(dst, src, len);
2144                         src += stride;
2145                         dst += stride;
2146                     }
2147                 }
2148             }
2149         }
2150     }
2151 }
2152
2153 static int set_side_data(HEVCContext *s)
2154 {
2155     AVFrame *out = s->ref->frame;
2156
2157     if (s->sei_frame_packing_present &&
2158         s->frame_packing_arrangement_type >= 3 &&
2159         s->frame_packing_arrangement_type <= 5 &&
2160         s->content_interpretation_type > 0 &&
2161         s->content_interpretation_type < 3) {
2162         AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2163         if (!stereo)
2164             return AVERROR(ENOMEM);
2165
2166         switch (s->frame_packing_arrangement_type) {
2167         case 3:
2168             if (s->quincunx_subsampling)
2169                 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2170             else
2171                 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2172             break;
2173         case 4:
2174             stereo->type = AV_STEREO3D_TOPBOTTOM;
2175             break;
2176         case 5:
2177             stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2178             break;
2179         }
2180
2181         if (s->content_interpretation_type == 2)
2182             stereo->flags = AV_STEREO3D_FLAG_INVERT;
2183     }
2184
2185     return 0;
2186 }
2187
2188 static int hevc_frame_start(HEVCContext *s)
2189 {
2190     HEVCLocalContext *lc = s->HEVClc;
2191     int ret;
2192
2193     memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2194     memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
2195     memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
2196     memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
2197
2198     lc->start_of_tiles_x = 0;
2199     s->is_decoded        = 0;
2200
2201     if (s->pps->tiles_enabled_flag)
2202         lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2203
2204     ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2205                               s->poc);
2206     if (ret < 0)
2207         goto fail;
2208
2209     ret = ff_hevc_frame_rps(s);
2210     if (ret < 0) {
2211         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2212         goto fail;
2213     }
2214
2215     ret = set_side_data(s);
2216     if (ret < 0)
2217         goto fail;
2218
2219     av_frame_unref(s->output_frame);
2220     ret = ff_hevc_output_frame(s, s->output_frame, 0);
2221     if (ret < 0)
2222         goto fail;
2223
2224     ff_thread_finish_setup(s->avctx);
2225
2226     return 0;
2227
2228 fail:
2229     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2230         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2231     s->ref = NULL;
2232     return ret;
2233 }
2234
2235 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2236 {
2237     HEVCLocalContext *lc = s->HEVClc;
2238     GetBitContext *gb    = &lc->gb;
2239     int ctb_addr_ts, ret;
2240
2241     ret = init_get_bits8(gb, nal, length);
2242     if (ret < 0)
2243         return ret;
2244
2245     ret = hls_nal_unit(s);
2246     if (ret < 0) {
2247         av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2248                s->nal_unit_type);
2249         if (s->avctx->err_recognition & AV_EF_EXPLODE)
2250             return ret;
2251         return 0;
2252     } else if (!ret)
2253         return 0;
2254
2255     switch (s->nal_unit_type) {
2256     case NAL_VPS:
2257         ret = ff_hevc_decode_nal_vps(s);
2258         if (ret < 0)
2259             return ret;
2260         break;
2261     case NAL_SPS:
2262         ret = ff_hevc_decode_nal_sps(s);
2263         if (ret < 0)
2264             return ret;
2265         break;
2266     case NAL_PPS:
2267         ret = ff_hevc_decode_nal_pps(s);
2268         if (ret < 0)
2269             return ret;
2270         break;
2271     case NAL_SEI_PREFIX:
2272     case NAL_SEI_SUFFIX:
2273         ret = ff_hevc_decode_nal_sei(s);
2274         if (ret < 0)
2275             return ret;
2276         break;
2277     case NAL_TRAIL_R:
2278     case NAL_TRAIL_N:
2279     case NAL_TSA_N:
2280     case NAL_TSA_R:
2281     case NAL_STSA_N:
2282     case NAL_STSA_R:
2283     case NAL_BLA_W_LP:
2284     case NAL_BLA_W_RADL:
2285     case NAL_BLA_N_LP:
2286     case NAL_IDR_W_RADL:
2287     case NAL_IDR_N_LP:
2288     case NAL_CRA_NUT:
2289     case NAL_RADL_N:
2290     case NAL_RADL_R:
2291     case NAL_RASL_N:
2292     case NAL_RASL_R:
2293         ret = hls_slice_header(s);
2294         if (ret < 0)
2295             return ret;
2296
2297         if (s->max_ra == INT_MAX) {
2298             if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2299                 s->max_ra = s->poc;
2300             } else {
2301                 if (IS_IDR(s))
2302                     s->max_ra = INT_MIN;
2303             }
2304         }
2305
2306         if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2307             s->poc <= s->max_ra) {
2308             s->is_decoded = 0;
2309             break;
2310         } else {
2311             if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2312                 s->max_ra = INT_MIN;
2313         }
2314
2315         if (s->sh.first_slice_in_pic_flag) {
2316             ret = hevc_frame_start(s);
2317             if (ret < 0)
2318                 return ret;
2319         } else if (!s->ref) {
2320             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2321             return AVERROR_INVALIDDATA;
2322         }
2323
2324         if (!s->sh.dependent_slice_segment_flag &&
2325             s->sh.slice_type != I_SLICE) {
2326             ret = ff_hevc_slice_rpl(s);
2327             if (ret < 0) {
2328                 av_log(s->avctx, AV_LOG_WARNING,
2329                        "Error constructing the reference lists for the current slice.\n");
2330                 if (s->avctx->err_recognition & AV_EF_EXPLODE)
2331                     return ret;
2332             }
2333         }
2334
2335         if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
2336             ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
2337         else
2338             ctb_addr_ts = hls_slice_data(s);
2339         if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2340             s->is_decoded = 1;
2341             if ((s->pps->transquant_bypass_enable_flag ||
2342                  (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2343                 s->sps->sao_enabled)
2344                 restore_tqb_pixels(s);
2345         }
2346
2347         if (ctb_addr_ts < 0)
2348             return ctb_addr_ts;
2349         break;
2350     case NAL_EOS_NUT:
2351     case NAL_EOB_NUT:
2352         s->seq_decode = (s->seq_decode + 1) & 0xff;
2353         s->max_ra     = INT_MAX;
2354         break;
2355     case NAL_AUD:
2356     case NAL_FD_NUT:
2357         break;
2358     default:
2359         av_log(s->avctx, AV_LOG_INFO,
2360                "Skipping NAL unit %d\n", s->nal_unit_type);
2361     }
2362
2363     return 0;
2364 }
2365
2366 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2367  * between these functions would be nice. */
2368 int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
2369                          HEVCNAL *nal)
2370 {
2371     int i, si, di;
2372     uint8_t *dst;
2373
2374     s->skipped_bytes = 0;
2375 #define STARTCODE_TEST                                                  \
2376         if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
2377             if (src[i + 2] != 3) {                                      \
2378                 /* startcode, so we must be past the end */             \
2379                 length = i;                                             \
2380             }                                                           \
2381             break;                                                      \
2382         }
2383 #if HAVE_FAST_UNALIGNED
2384 #define FIND_FIRST_ZERO                                                 \
2385         if (i > 0 && !src[i])                                           \
2386             i--;                                                        \
2387         while (src[i])                                                  \
2388             i++
2389 #if HAVE_FAST_64BIT
2390     for (i = 0; i + 1 < length; i += 9) {
2391         if (!((~AV_RN64A(src + i) &
2392                (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2393               0x8000800080008080ULL))
2394             continue;
2395         FIND_FIRST_ZERO;
2396         STARTCODE_TEST;
2397         i -= 7;
2398     }
2399 #else
2400     for (i = 0; i + 1 < length; i += 5) {
2401         if (!((~AV_RN32A(src + i) &
2402                (AV_RN32A(src + i) - 0x01000101U)) &
2403               0x80008080U))
2404             continue;
2405         FIND_FIRST_ZERO;
2406         STARTCODE_TEST;
2407         i -= 3;
2408     }
2409 #endif /* HAVE_FAST_64BIT */
2410 #else
2411     for (i = 0; i + 1 < length; i += 2) {
2412         if (src[i])
2413             continue;
2414         if (i > 0 && src[i - 1] == 0)
2415             i--;
2416         STARTCODE_TEST;
2417     }
2418 #endif /* HAVE_FAST_UNALIGNED */
2419
2420     if (i >= length - 1) { // no escaped 0
2421         nal->data = src;
2422         nal->size = length;
2423         return length;
2424     }
2425
2426     av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2427                    length + FF_INPUT_BUFFER_PADDING_SIZE);
2428     if (!nal->rbsp_buffer)
2429         return AVERROR(ENOMEM);
2430
2431     dst = nal->rbsp_buffer;
2432
2433     memcpy(dst, src, i);
2434     si = di = i;
2435     while (si + 2 < length) {
2436         // remove escapes (very rare 1:2^22)
2437         if (src[si + 2] > 3) {
2438             dst[di++] = src[si++];
2439             dst[di++] = src[si++];
2440         } else if (src[si] == 0 && src[si + 1] == 0) {
2441             if (src[si + 2] == 3) { // escape
2442                 dst[di++] = 0;
2443                 dst[di++] = 0;
2444                 si       += 3;
2445
2446                 s->skipped_bytes++;
2447                 if (s->skipped_bytes_pos_size < s->skipped_bytes) {
2448                     s->skipped_bytes_pos_size *= 2;
2449                     av_reallocp_array(&s->skipped_bytes_pos,
2450                             s->skipped_bytes_pos_size,
2451                             sizeof(*s->skipped_bytes_pos));
2452                     if (!s->skipped_bytes_pos)
2453                         return AVERROR(ENOMEM);
2454                 }
2455                 if (s->skipped_bytes_pos)
2456                     s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
2457                 continue;
2458             } else // next start code
2459                 goto nsc;
2460         }
2461
2462         dst[di++] = src[si++];
2463     }
2464     while (si < length)
2465         dst[di++] = src[si++];
2466
2467 nsc:
2468     memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2469
2470     nal->data = dst;
2471     nal->size = di;
2472     return si;
2473 }
2474
2475 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2476 {
2477     int i, consumed, ret = 0;
2478
2479     s->ref = NULL;
2480     s->eos = 0;
2481
2482     /* split the input packet into NAL units, so we know the upper bound on the
2483      * number of slices in the frame */
2484     s->nb_nals = 0;
2485     while (length >= 4) {
2486         HEVCNAL *nal;
2487         int extract_length = 0;
2488
2489         if (s->is_nalff) {
2490             int i;
2491             for (i = 0; i < s->nal_length_size; i++)
2492                 extract_length = (extract_length << 8) | buf[i];
2493             buf    += s->nal_length_size;
2494             length -= s->nal_length_size;
2495
2496             if (extract_length > length) {
2497                 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2498                 ret = AVERROR_INVALIDDATA;
2499                 goto fail;
2500             }
2501         } else {
2502             /* search start code */
2503             while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2504                 ++buf;
2505                 --length;
2506                 if (length < 4) {
2507                     av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
2508                     ret = AVERROR_INVALIDDATA;
2509                     goto fail;
2510                 }
2511             }
2512
2513             buf           += 3;
2514             length        -= 3;
2515         }
2516
2517         if (!s->is_nalff)
2518             extract_length = length;
2519
2520         if (s->nals_allocated < s->nb_nals + 1) {
2521             int new_size = s->nals_allocated + 1;
2522             HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2523             if (!tmp) {
2524                 ret = AVERROR(ENOMEM);
2525                 goto fail;
2526             }
2527             s->nals = tmp;
2528             memset(s->nals + s->nals_allocated, 0,
2529                    (new_size - s->nals_allocated) * sizeof(*tmp));
2530             av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
2531             av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
2532             av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
2533             s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
2534             s->skipped_bytes_pos_nal[s->nals_allocated] = av_malloc_array(s->skipped_bytes_pos_size_nal[s->nals_allocated], sizeof(*s->skipped_bytes_pos));
2535             s->nals_allocated = new_size;
2536         }
2537         s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
2538         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
2539         nal = &s->nals[s->nb_nals];
2540
2541         consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
2542
2543         s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
2544         s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
2545         s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
2546
2547
2548         if (consumed < 0) {
2549             ret = consumed;
2550             goto fail;
2551         }
2552
2553         ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
2554         if (ret < 0)
2555             goto fail;
2556         hls_nal_unit(s);
2557
2558         if (s->nal_unit_type == NAL_EOB_NUT ||
2559             s->nal_unit_type == NAL_EOS_NUT)
2560             s->eos = 1;
2561
2562         buf    += consumed;
2563         length -= consumed;
2564     }
2565
2566     /* parse the NAL units */
2567     for (i = 0; i < s->nb_nals; i++) {
2568         int ret;
2569         s->skipped_bytes = s->skipped_bytes_nal[i];
2570         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
2571
2572         ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2573         if (ret < 0) {
2574             av_log(s->avctx, AV_LOG_WARNING,
2575                    "Error parsing NAL unit #%d.\n", i);
2576             if (s->avctx->err_recognition & AV_EF_EXPLODE)
2577                 goto fail;
2578         }
2579     }
2580
2581 fail:
2582     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2583         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2584
2585     return ret;
2586 }
2587
2588 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2589 {
2590     int i;
2591     for (i = 0; i < 16; i++)
2592         av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2593 }
2594
2595 static int verify_md5(HEVCContext *s, AVFrame *frame)
2596 {
2597     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2598     int pixel_shift;
2599     int i, j;
2600
2601     if (!desc)
2602         return AVERROR(EINVAL);
2603
2604     pixel_shift = desc->comp[0].depth_minus1 > 7;
2605
2606     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2607            s->poc);
2608
2609     /* the checksums are LE, so we have to byteswap for >8bpp formats
2610      * on BE arches */
2611 #if HAVE_BIGENDIAN
2612     if (pixel_shift && !s->checksum_buf) {
2613         av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2614                        FFMAX3(frame->linesize[0], frame->linesize[1],
2615                               frame->linesize[2]));
2616         if (!s->checksum_buf)
2617             return AVERROR(ENOMEM);
2618     }
2619 #endif
2620
2621     for (i = 0; frame->data[i]; i++) {
2622         int width  = s->avctx->coded_width;
2623         int height = s->avctx->coded_height;
2624         int w = (i == 1 || i == 2) ? (width  >> desc->log2_chroma_w) : width;
2625         int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2626         uint8_t md5[16];
2627
2628         av_md5_init(s->md5_ctx);
2629         for (j = 0; j < h; j++) {
2630             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2631 #if HAVE_BIGENDIAN
2632             if (pixel_shift) {
2633                 s->dsp.bswap16_buf((uint16_t*)s->checksum_buf,
2634                                    (const uint16_t*)src, w);
2635                 src = s->checksum_buf;
2636             }
2637 #endif
2638             av_md5_update(s->md5_ctx, src, w << pixel_shift);
2639         }
2640         av_md5_final(s->md5_ctx, md5);
2641
2642         if (!memcmp(md5, s->md5[i], 16)) {
2643             av_log   (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2644             print_md5(s->avctx, AV_LOG_DEBUG, md5);
2645             av_log   (s->avctx, AV_LOG_DEBUG, "; ");
2646         } else {
2647             av_log   (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2648             print_md5(s->avctx, AV_LOG_ERROR, md5);
2649             av_log   (s->avctx, AV_LOG_ERROR, " != ");
2650             print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2651             av_log   (s->avctx, AV_LOG_ERROR, "\n");
2652             return AVERROR_INVALIDDATA;
2653         }
2654     }
2655
2656     av_log(s->avctx, AV_LOG_DEBUG, "\n");
2657
2658     return 0;
2659 }
2660
2661 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2662                              AVPacket *avpkt)
2663 {
2664     int ret;
2665     HEVCContext *s = avctx->priv_data;
2666
2667     if (!avpkt->size) {
2668         ret = ff_hevc_output_frame(s, data, 1);
2669         if (ret < 0)
2670             return ret;
2671
2672         *got_output = ret;
2673         return 0;
2674     }
2675
2676     s->ref = NULL;
2677     ret    = decode_nal_units(s, avpkt->data, avpkt->size);
2678     if (ret < 0)
2679         return ret;
2680
2681     /* verify the SEI checksum */
2682     if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2683         s->is_md5) {
2684         ret = verify_md5(s, s->ref->frame);
2685         if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2686             ff_hevc_unref_frame(s, s->ref, ~0);
2687             return ret;
2688         }
2689     }
2690     s->is_md5 = 0;
2691
2692     if (s->is_decoded) {
2693         av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2694         s->is_decoded = 0;
2695     }
2696
2697     if (s->output_frame->buf[0]) {
2698         av_frame_move_ref(data, s->output_frame);
2699         *got_output = 1;
2700     }
2701
2702     return avpkt->size;
2703 }
2704
2705 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2706 {
2707     int ret;
2708
2709     ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2710     if (ret < 0)
2711         return ret;
2712
2713     dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2714     if (!dst->tab_mvf_buf)
2715         goto fail;
2716     dst->tab_mvf = src->tab_mvf;
2717
2718     dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2719     if (!dst->rpl_tab_buf)
2720         goto fail;
2721     dst->rpl_tab = src->rpl_tab;
2722
2723     dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2724     if (!dst->rpl_buf)
2725         goto fail;
2726
2727     dst->poc        = src->poc;
2728     dst->ctb_count  = src->ctb_count;
2729     dst->window     = src->window;
2730     dst->flags      = src->flags;
2731     dst->sequence   = src->sequence;
2732
2733     return 0;
2734 fail:
2735     ff_hevc_unref_frame(s, dst, ~0);
2736     return AVERROR(ENOMEM);
2737 }
2738
2739 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2740 {
2741     HEVCContext       *s = avctx->priv_data;
2742     HEVCLocalContext *lc = s->HEVClc;
2743     int i;
2744
2745     pic_arrays_free(s);
2746
2747     av_freep(&s->md5_ctx);
2748
2749     for(i=0; i < s->nals_allocated; i++) {
2750         av_freep(&s->skipped_bytes_pos_nal[i]);
2751     }
2752     av_freep(&s->skipped_bytes_pos_size_nal);
2753     av_freep(&s->skipped_bytes_nal);
2754     av_freep(&s->skipped_bytes_pos_nal);
2755
2756     av_freep(&s->cabac_state);
2757
2758     av_frame_free(&s->tmp_frame);
2759     av_frame_free(&s->output_frame);
2760
2761     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2762         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2763         av_frame_free(&s->DPB[i].frame);
2764     }
2765
2766     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2767         av_buffer_unref(&s->vps_list[i]);
2768     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2769         av_buffer_unref(&s->sps_list[i]);
2770     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2771         av_buffer_unref(&s->pps_list[i]);
2772
2773     av_freep(&s->sh.entry_point_offset);
2774     av_freep(&s->sh.offset);
2775     av_freep(&s->sh.size);
2776
2777     for (i = 1; i < s->threads_number; i++) {
2778         lc = s->HEVClcList[i];
2779         if (lc) {
2780             av_freep(&s->HEVClcList[i]);
2781             av_freep(&s->sList[i]);
2782         }
2783     }
2784     if (s->HEVClc == s->HEVClcList[0])
2785         s->HEVClc = NULL;
2786     av_freep(&s->HEVClcList[0]);
2787
2788     for (i = 0; i < s->nals_allocated; i++)
2789         av_freep(&s->nals[i].rbsp_buffer);
2790     av_freep(&s->nals);
2791     s->nals_allocated = 0;
2792
2793     return 0;
2794 }
2795
2796 static av_cold int hevc_init_context(AVCodecContext *avctx)
2797 {
2798     HEVCContext *s = avctx->priv_data;
2799     int i;
2800
2801     s->avctx = avctx;
2802
2803     s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
2804     if (!s->HEVClc)
2805         goto fail;
2806     s->HEVClcList[0] = s->HEVClc;
2807     s->sList[0] = s;
2808
2809     s->cabac_state = av_malloc(HEVC_CONTEXTS);
2810     if (!s->cabac_state)
2811         goto fail;
2812
2813     s->tmp_frame = av_frame_alloc();
2814     if (!s->tmp_frame)
2815         goto fail;
2816
2817     s->output_frame = av_frame_alloc();
2818     if (!s->output_frame)
2819         goto fail;
2820
2821     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2822         s->DPB[i].frame = av_frame_alloc();
2823         if (!s->DPB[i].frame)
2824             goto fail;
2825         s->DPB[i].tf.f = s->DPB[i].frame;
2826     }
2827
2828     s->max_ra = INT_MAX;
2829
2830     s->md5_ctx = av_md5_alloc();
2831     if (!s->md5_ctx)
2832         goto fail;
2833
2834     ff_dsputil_init(&s->dsp, avctx);
2835
2836     s->context_initialized = 1;
2837
2838     return 0;
2839
2840 fail:
2841     hevc_decode_free(avctx);
2842     return AVERROR(ENOMEM);
2843 }
2844
2845 static int hevc_update_thread_context(AVCodecContext *dst,
2846                                       const AVCodecContext *src)
2847 {
2848     HEVCContext *s  = dst->priv_data;
2849     HEVCContext *s0 = src->priv_data;
2850     int i, ret;
2851
2852     if (!s->context_initialized) {
2853         ret = hevc_init_context(dst);
2854         if (ret < 0)
2855             return ret;
2856     }
2857
2858     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2859         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2860         if (s0->DPB[i].frame->buf[0]) {
2861             ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
2862             if (ret < 0)
2863                 return ret;
2864         }
2865     }
2866
2867     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
2868         av_buffer_unref(&s->vps_list[i]);
2869         if (s0->vps_list[i]) {
2870             s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
2871             if (!s->vps_list[i])
2872                 return AVERROR(ENOMEM);
2873         }
2874     }
2875
2876     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
2877         av_buffer_unref(&s->sps_list[i]);
2878         if (s0->sps_list[i]) {
2879             s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
2880             if (!s->sps_list[i])
2881                 return AVERROR(ENOMEM);
2882         }
2883     }
2884
2885     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
2886         av_buffer_unref(&s->pps_list[i]);
2887         if (s0->pps_list[i]) {
2888             s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
2889             if (!s->pps_list[i])
2890                 return AVERROR(ENOMEM);
2891         }
2892     }
2893
2894     if (s->sps != s0->sps)
2895         ret = set_sps(s, s0->sps);
2896
2897     s->seq_decode = s0->seq_decode;
2898     s->seq_output = s0->seq_output;
2899     s->pocTid0    = s0->pocTid0;
2900     s->max_ra     = s0->max_ra;
2901
2902     s->is_nalff        = s0->is_nalff;
2903     s->nal_length_size = s0->nal_length_size;
2904
2905     s->threads_number      = s0->threads_number;
2906     s->threads_type        = s0->threads_type;
2907
2908     if (s0->eos) {
2909         s->seq_decode = (s->seq_decode + 1) & 0xff;
2910         s->max_ra = INT_MAX;
2911     }
2912
2913     return 0;
2914 }
2915
2916 static int hevc_decode_extradata(HEVCContext *s)
2917 {
2918     AVCodecContext *avctx = s->avctx;
2919     GetByteContext gb;
2920     int ret;
2921
2922     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
2923
2924     if (avctx->extradata_size > 3 &&
2925         (avctx->extradata[0] || avctx->extradata[1] ||
2926          avctx->extradata[2] > 1)) {
2927         /* It seems the extradata is encoded as hvcC format.
2928          * Temporarily, we support configurationVersion==0 until 14496-15 3rd
2929          * is finalized. When finalized, configurationVersion will be 1 and we
2930          * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
2931         int i, j, num_arrays, nal_len_size;
2932
2933         s->is_nalff = 1;
2934
2935         bytestream2_skip(&gb, 21);
2936         nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
2937         num_arrays   = bytestream2_get_byte(&gb);
2938
2939         /* nal units in the hvcC always have length coded with 2 bytes,
2940          * so put a fake nal_length_size = 2 while parsing them */
2941         s->nal_length_size = 2;
2942
2943         /* Decode nal units from hvcC. */
2944         for (i = 0; i < num_arrays; i++) {
2945             int type = bytestream2_get_byte(&gb) & 0x3f;
2946             int cnt  = bytestream2_get_be16(&gb);
2947
2948             for (j = 0; j < cnt; j++) {
2949                 // +2 for the nal size field
2950                 int nalsize = bytestream2_peek_be16(&gb) + 2;
2951                 if (bytestream2_get_bytes_left(&gb) < nalsize) {
2952                     av_log(s->avctx, AV_LOG_ERROR,
2953                            "Invalid NAL unit size in extradata.\n");
2954                     return AVERROR_INVALIDDATA;
2955                 }
2956
2957                 ret = decode_nal_units(s, gb.buffer, nalsize);
2958                 if (ret < 0) {
2959                     av_log(avctx, AV_LOG_ERROR,
2960                            "Decoding nal unit %d %d from hvcC failed\n",
2961                            type, i);
2962                     return ret;
2963                 }
2964                 bytestream2_skip(&gb, nalsize);
2965             }
2966         }
2967
2968         /* Now store right nal length size, that will be used to parse
2969          * all other nals */
2970         s->nal_length_size = nal_len_size;
2971     } else {
2972         s->is_nalff = 0;
2973         ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
2974         if (ret < 0)
2975             return ret;
2976     }
2977     return 0;
2978 }
2979
2980 static av_cold int hevc_decode_init(AVCodecContext *avctx)
2981 {
2982     HEVCContext *s = avctx->priv_data;
2983     int ret;
2984
2985     ff_init_cabac_states();
2986
2987     avctx->internal->allocate_progress = 1;
2988
2989     ret = hevc_init_context(avctx);
2990     if (ret < 0)
2991         return ret;
2992
2993     s->enable_parallel_tiles = 0;
2994     s->picture_struct = 0;
2995
2996     if(avctx->active_thread_type & FF_THREAD_SLICE)
2997         s->threads_number = avctx->thread_count;
2998     else
2999         s->threads_number = 1;
3000
3001     if (avctx->extradata_size > 0 && avctx->extradata) {
3002         ret = hevc_decode_extradata(s);
3003         if (ret < 0) {
3004             hevc_decode_free(avctx);
3005             return ret;
3006         }
3007     }
3008
3009     if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
3010             s->threads_type = FF_THREAD_FRAME;
3011         else
3012             s->threads_type = FF_THREAD_SLICE;
3013
3014     return 0;
3015 }
3016
3017 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
3018 {
3019     HEVCContext *s = avctx->priv_data;
3020     int ret;
3021
3022     memset(s, 0, sizeof(*s));
3023
3024     ret = hevc_init_context(avctx);
3025     if (ret < 0)
3026         return ret;
3027
3028     return 0;
3029 }
3030
3031 static void hevc_decode_flush(AVCodecContext *avctx)
3032 {
3033     HEVCContext *s = avctx->priv_data;
3034     ff_hevc_flush_dpb(s);
3035     s->max_ra = INT_MAX;
3036 }
3037
3038 #define OFFSET(x) offsetof(HEVCContext, x)
3039 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
3040
3041 static const AVProfile profiles[] = {
3042     { FF_PROFILE_HEVC_MAIN,                 "Main"                },
3043     { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
3044     { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
3045     { FF_PROFILE_UNKNOWN },
3046 };
3047
3048 static const AVOption options[] = {
3049     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
3050         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3051     { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
3052         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3053     { NULL },
3054 };
3055
3056 static const AVClass hevc_decoder_class = {
3057     .class_name = "HEVC decoder",
3058     .item_name  = av_default_item_name,
3059     .option     = options,
3060     .version    = LIBAVUTIL_VERSION_INT,
3061 };
3062
3063 AVCodec ff_hevc_decoder = {
3064     .name                  = "hevc",
3065     .long_name             = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3066     .type                  = AVMEDIA_TYPE_VIDEO,
3067     .id                    = AV_CODEC_ID_HEVC,
3068     .priv_data_size        = sizeof(HEVCContext),
3069     .priv_class            = &hevc_decoder_class,
3070     .init                  = hevc_decode_init,
3071     .close                 = hevc_decode_free,
3072     .decode                = hevc_decode_frame,
3073     .flush                 = hevc_decode_flush,
3074     .update_thread_context = hevc_update_thread_context,
3075     .init_thread_copy      = hevc_init_thread_copy,
3076     .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3077                              CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
3078     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
3079 };