]> git.sesse.net Git - ffmpeg/blob - libavcodec/hevc.c
mpegvideo: fix overwriting hwaccel surface objects
[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     unsigned 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     if (!s->sh.slice_ctb_addr_rs && s->sh.dependent_slice_segment_flag) {
705         av_log(s->avctx, AV_LOG_ERROR, "Impossible slice segment.\n");
706         return AVERROR_INVALIDDATA;
707     }
708
709     s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
710
711     if (!s->pps->cu_qp_delta_enabled_flag)
712         s->HEVClc->qp_y = s->sh.slice_qp;
713
714     s->slice_initialized = 1;
715
716     return 0;
717 }
718
719 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
720
721 #define SET_SAO(elem, value)                            \
722 do {                                                    \
723     if (!sao_merge_up_flag && !sao_merge_left_flag)     \
724         sao->elem = value;                              \
725     else if (sao_merge_left_flag)                       \
726         sao->elem = CTB(s->sao, rx-1, ry).elem;         \
727     else if (sao_merge_up_flag)                         \
728         sao->elem = CTB(s->sao, rx, ry-1).elem;         \
729     else                                                \
730         sao->elem = 0;                                  \
731 } while (0)
732
733 static void hls_sao_param(HEVCContext *s, int rx, int ry)
734 {
735     HEVCLocalContext *lc    = s->HEVClc;
736     int sao_merge_left_flag = 0;
737     int sao_merge_up_flag   = 0;
738     int shift               = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
739     SAOParams *sao          = &CTB(s->sao, rx, ry);
740     int c_idx, i;
741
742     if (s->sh.slice_sample_adaptive_offset_flag[0] ||
743         s->sh.slice_sample_adaptive_offset_flag[1]) {
744         if (rx > 0) {
745             if (lc->ctb_left_flag)
746                 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
747         }
748         if (ry > 0 && !sao_merge_left_flag) {
749             if (lc->ctb_up_flag)
750                 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
751         }
752     }
753
754     for (c_idx = 0; c_idx < 3; c_idx++) {
755         if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
756             sao->type_idx[c_idx] = SAO_NOT_APPLIED;
757             continue;
758         }
759
760         if (c_idx == 2) {
761             sao->type_idx[2] = sao->type_idx[1];
762             sao->eo_class[2] = sao->eo_class[1];
763         } else {
764             SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
765         }
766
767         if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
768             continue;
769
770         for (i = 0; i < 4; i++)
771             SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
772
773         if (sao->type_idx[c_idx] == SAO_BAND) {
774             for (i = 0; i < 4; i++) {
775                 if (sao->offset_abs[c_idx][i]) {
776                     SET_SAO(offset_sign[c_idx][i],
777                             ff_hevc_sao_offset_sign_decode(s));
778                 } else {
779                     sao->offset_sign[c_idx][i] = 0;
780                 }
781             }
782             SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
783         } else if (c_idx != 2) {
784             SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
785         }
786
787         // Inferred parameters
788         sao->offset_val[c_idx][0] = 0;
789         for (i = 0; i < 4; i++) {
790             sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
791             if (sao->type_idx[c_idx] == SAO_EDGE) {
792                 if (i > 1)
793                     sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
794             } else if (sao->offset_sign[c_idx][i]) {
795                 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
796             }
797         }
798     }
799 }
800
801 #undef SET_SAO
802 #undef CTB
803
804 static int hls_transform_unit(HEVCContext *s, int x0, int y0,
805                               int xBase, int yBase, int cb_xBase, int cb_yBase,
806                               int log2_cb_size, int log2_trafo_size,
807                               int trafo_depth, int blk_idx)
808 {
809     HEVCLocalContext *lc = s->HEVClc;
810
811     if (lc->cu.pred_mode == MODE_INTRA) {
812         int trafo_size = 1 << log2_trafo_size;
813         ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
814
815         s->hpc.intra_pred(s, x0, y0, log2_trafo_size, 0);
816         if (log2_trafo_size > 2) {
817             trafo_size = trafo_size << (s->sps->hshift[1] - 1);
818             ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
819             s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 1);
820             s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 2);
821         } else if (blk_idx == 3) {
822             trafo_size = trafo_size << s->sps->hshift[1];
823             ff_hevc_set_neighbour_available(s, xBase, yBase,
824                                             trafo_size, trafo_size);
825             s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 1);
826             s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 2);
827         }
828     }
829
830     if (lc->tt.cbf_luma ||
831         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
832         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
833         int scan_idx   = SCAN_DIAG;
834         int scan_idx_c = SCAN_DIAG;
835
836         if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
837             lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
838             if (lc->tu.cu_qp_delta != 0)
839                 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
840                     lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
841             lc->tu.is_cu_qp_delta_coded = 1;
842
843             if (lc->tu.cu_qp_delta < -(26 + s->sps->qp_bd_offset / 2) ||
844                 lc->tu.cu_qp_delta >  (25 + s->sps->qp_bd_offset / 2)) {
845                 av_log(s->avctx, AV_LOG_ERROR,
846                        "The cu_qp_delta %d is outside the valid range "
847                        "[%d, %d].\n",
848                        lc->tu.cu_qp_delta,
849                        -(26 + s->sps->qp_bd_offset / 2),
850                         (25 + s->sps->qp_bd_offset / 2));
851                 return AVERROR_INVALIDDATA;
852             }
853
854             ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
855         }
856
857         if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
858             if (lc->tu.cur_intra_pred_mode >= 6 &&
859                 lc->tu.cur_intra_pred_mode <= 14) {
860                 scan_idx = SCAN_VERT;
861             } else if (lc->tu.cur_intra_pred_mode >= 22 &&
862                        lc->tu.cur_intra_pred_mode <= 30) {
863                 scan_idx = SCAN_HORIZ;
864             }
865
866             if (lc->pu.intra_pred_mode_c >=  6 &&
867                 lc->pu.intra_pred_mode_c <= 14) {
868                 scan_idx_c = SCAN_VERT;
869             } else if (lc->pu.intra_pred_mode_c >= 22 &&
870                        lc->pu.intra_pred_mode_c <= 30) {
871                 scan_idx_c = SCAN_HORIZ;
872             }
873         }
874
875         if (lc->tt.cbf_luma)
876             ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
877         if (log2_trafo_size > 2) {
878             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
879                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
880             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
881                 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
882         } else if (blk_idx == 3) {
883             if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
884                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
885             if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
886                 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
887         }
888     }
889     return 0;
890 }
891
892 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
893 {
894     int cb_size          = 1 << log2_cb_size;
895     int log2_min_pu_size = s->sps->log2_min_pu_size;
896
897     int min_pu_width     = s->sps->min_pu_width;
898     int x_end = FFMIN(x0 + cb_size, s->sps->width);
899     int y_end = FFMIN(y0 + cb_size, s->sps->height);
900     int i, j;
901
902     for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
903         for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
904             s->is_pcm[i + j * min_pu_width] = 2;
905 }
906
907 static int hls_transform_tree(HEVCContext *s, int x0, int y0,
908                               int xBase, int yBase, int cb_xBase, int cb_yBase,
909                               int log2_cb_size, int log2_trafo_size,
910                               int trafo_depth, int blk_idx)
911 {
912     HEVCLocalContext *lc = s->HEVClc;
913     uint8_t split_transform_flag;
914     int ret;
915
916     if (trafo_depth > 0 && log2_trafo_size == 2) {
917         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
918             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase);
919         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
920             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase);
921     } else {
922         SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
923         SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) = 0;
924     }
925
926     if (lc->cu.intra_split_flag) {
927         if (trafo_depth == 1)
928             lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
929     } else {
930         lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
931     }
932
933     lc->tt.cbf_luma = 1;
934
935     lc->tt.inter_split_flag = s->sps->max_transform_hierarchy_depth_inter == 0 &&
936                               lc->cu.pred_mode == MODE_INTER &&
937                               lc->cu.part_mode != PART_2Nx2N &&
938                               trafo_depth == 0;
939
940     if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
941         log2_trafo_size >  s->sps->log2_min_tb_size    &&
942         trafo_depth     < lc->cu.max_trafo_depth       &&
943         !(lc->cu.intra_split_flag && trafo_depth == 0)) {
944         split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
945     } else {
946         split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
947                                (lc->cu.intra_split_flag && trafo_depth == 0) ||
948                                lc->tt.inter_split_flag;
949     }
950
951     if (log2_trafo_size > 2) {
952         if (trafo_depth == 0 ||
953             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase)) {
954             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
955                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
956         }
957
958         if (trafo_depth == 0 ||
959             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase)) {
960             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
961                 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
962         }
963     }
964
965     if (split_transform_flag) {
966         int x1 = x0 + ((1 << log2_trafo_size) >> 1);
967         int y1 = y0 + ((1 << log2_trafo_size) >> 1);
968
969         ret = hls_transform_tree(s, x0, y0, x0, y0, cb_xBase, cb_yBase,
970                                  log2_cb_size, log2_trafo_size - 1,
971                                  trafo_depth + 1, 0);
972         if (ret < 0)
973             return ret;
974         ret = hls_transform_tree(s, x1, y0, x0, y0, cb_xBase, cb_yBase,
975                                  log2_cb_size, log2_trafo_size - 1,
976                                  trafo_depth + 1, 1);
977         if (ret < 0)
978             return ret;
979         ret = hls_transform_tree(s, x0, y1, x0, y0, cb_xBase, cb_yBase,
980                                  log2_cb_size, log2_trafo_size - 1,
981                                  trafo_depth + 1, 2);
982         if (ret < 0)
983             return ret;
984         ret = hls_transform_tree(s, x1, y1, x0, y0, cb_xBase, cb_yBase,
985                                  log2_cb_size, log2_trafo_size - 1,
986                                  trafo_depth + 1, 3);
987         if (ret < 0)
988             return ret;
989     } else {
990         int min_tu_size      = 1 << s->sps->log2_min_tb_size;
991         int log2_min_tu_size = s->sps->log2_min_tb_size;
992         int min_tu_width     = s->sps->min_tb_width;
993
994         if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
995             SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
996             SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
997             lc->tt.cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
998         }
999
1000         ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
1001                                  log2_cb_size, log2_trafo_size, trafo_depth,
1002                                  blk_idx);
1003         if (ret < 0)
1004             return ret;
1005         // TODO: store cbf_luma somewhere else
1006         if (lc->tt.cbf_luma) {
1007             int i, j;
1008             for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
1009                 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
1010                     int x_tu = (x0 + j) >> log2_min_tu_size;
1011                     int y_tu = (y0 + i) >> log2_min_tu_size;
1012                     s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
1013                 }
1014         }
1015         if (!s->sh.disable_deblocking_filter_flag) {
1016             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size,
1017                                                   lc->slice_or_tiles_up_boundary,
1018                                                   lc->slice_or_tiles_left_boundary);
1019             if (s->pps->transquant_bypass_enable_flag &&
1020                 lc->cu.cu_transquant_bypass_flag)
1021                 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
1022         }
1023     }
1024     return 0;
1025 }
1026
1027 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
1028 {
1029     //TODO: non-4:2:0 support
1030     HEVCLocalContext *lc = s->HEVClc;
1031     GetBitContext gb;
1032     int cb_size   = 1 << log2_cb_size;
1033     int stride0   = s->frame->linesize[0];
1034     uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
1035     int   stride1 = s->frame->linesize[1];
1036     uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
1037     int   stride2 = s->frame->linesize[2];
1038     uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
1039
1040     int length         = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
1041     const uint8_t *pcm = skip_bytes(&s->HEVClc->cc, (length + 7) >> 3);
1042     int ret;
1043
1044     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1045                                           lc->slice_or_tiles_up_boundary,
1046                                           lc->slice_or_tiles_left_boundary);
1047
1048     ret = init_get_bits(&gb, pcm, length);
1049     if (ret < 0)
1050         return ret;
1051
1052     s->hevcdsp.put_pcm(dst0, stride0, cb_size,     &gb, s->sps->pcm.bit_depth);
1053     s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1054     s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1055     return 0;
1056 }
1057
1058 /**
1059  * 8.5.3.2.2.1 Luma sample interpolation process
1060  *
1061  * @param s HEVC decoding context
1062  * @param dst target buffer for block data at block position
1063  * @param dststride stride of the dst buffer
1064  * @param ref reference picture buffer at origin (0, 0)
1065  * @param mv motion vector (relative to block position) to get pixel data from
1066  * @param x_off horizontal position of block from origin (0, 0)
1067  * @param y_off vertical position of block from origin (0, 0)
1068  * @param block_w width of block
1069  * @param block_h height of block
1070  */
1071 static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
1072                     AVFrame *ref, const Mv *mv, int x_off, int y_off,
1073                     int block_w, int block_h)
1074 {
1075     HEVCLocalContext *lc = s->HEVClc;
1076     uint8_t *src         = ref->data[0];
1077     ptrdiff_t srcstride  = ref->linesize[0];
1078     int pic_width        = s->sps->width;
1079     int pic_height       = s->sps->height;
1080
1081     int mx         = mv->x & 3;
1082     int my         = mv->y & 3;
1083     int extra_left = ff_hevc_qpel_extra_before[mx];
1084     int extra_top  = ff_hevc_qpel_extra_before[my];
1085
1086     x_off += mv->x >> 2;
1087     y_off += mv->y >> 2;
1088     src   += y_off * srcstride + (x_off << s->sps->pixel_shift);
1089
1090     if (x_off < extra_left || y_off < extra_top ||
1091         x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
1092         y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
1093         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1094         int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
1095         int buf_offset = extra_top *
1096                          edge_emu_stride + (extra_left << s->sps->pixel_shift);
1097
1098         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1099                                  edge_emu_stride, srcstride,
1100                                  block_w + ff_hevc_qpel_extra[mx],
1101                                  block_h + ff_hevc_qpel_extra[my],
1102                                  x_off - extra_left, y_off - extra_top,
1103                                  pic_width, pic_height);
1104         src = lc->edge_emu_buffer + buf_offset;
1105         srcstride = edge_emu_stride;
1106     }
1107     s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
1108                                      block_h, lc->mc_buffer);
1109 }
1110
1111 /**
1112  * 8.5.3.2.2.2 Chroma sample interpolation process
1113  *
1114  * @param s HEVC decoding context
1115  * @param dst1 target buffer for block data at block position (U plane)
1116  * @param dst2 target buffer for block data at block position (V plane)
1117  * @param dststride stride of the dst1 and dst2 buffers
1118  * @param ref reference picture buffer at origin (0, 0)
1119  * @param mv motion vector (relative to block position) to get pixel data from
1120  * @param x_off horizontal position of block from origin (0, 0)
1121  * @param y_off vertical position of block from origin (0, 0)
1122  * @param block_w width of block
1123  * @param block_h height of block
1124  */
1125 static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
1126                       ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
1127                       int x_off, int y_off, int block_w, int block_h)
1128 {
1129     HEVCLocalContext *lc = s->HEVClc;
1130     uint8_t *src1        = ref->data[1];
1131     uint8_t *src2        = ref->data[2];
1132     ptrdiff_t src1stride = ref->linesize[1];
1133     ptrdiff_t src2stride = ref->linesize[2];
1134     int pic_width        = s->sps->width >> 1;
1135     int pic_height       = s->sps->height >> 1;
1136
1137     int mx = mv->x & 7;
1138     int my = mv->y & 7;
1139
1140     x_off += mv->x >> 3;
1141     y_off += mv->y >> 3;
1142     src1  += y_off * src1stride + (x_off << s->sps->pixel_shift);
1143     src2  += y_off * src2stride + (x_off << s->sps->pixel_shift);
1144
1145     if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1146         x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1147         y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1148         const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1149         int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1150         int buf_offset1 = EPEL_EXTRA_BEFORE *
1151                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1152         int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1153         int buf_offset2 = EPEL_EXTRA_BEFORE *
1154                           (edge_emu_stride + (1 << s->sps->pixel_shift));
1155
1156         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1157                                  edge_emu_stride, src1stride,
1158                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1159                                  x_off - EPEL_EXTRA_BEFORE,
1160                                  y_off - EPEL_EXTRA_BEFORE,
1161                                  pic_width, pic_height);
1162
1163         src1 = lc->edge_emu_buffer + buf_offset1;
1164         src1stride = edge_emu_stride;
1165         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1166                                              block_w, block_h, mx, my, lc->mc_buffer);
1167
1168         s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
1169                                  edge_emu_stride, src2stride,
1170                                  block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1171                                  x_off - EPEL_EXTRA_BEFORE,
1172                                  y_off - EPEL_EXTRA_BEFORE,
1173                                  pic_width, pic_height);
1174         src2 = lc->edge_emu_buffer + buf_offset2;
1175         src2stride = edge_emu_stride;
1176
1177         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1178                                              block_w, block_h, mx, my,
1179                                              lc->mc_buffer);
1180     } else {
1181         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1182                                              block_w, block_h, mx, my,
1183                                              lc->mc_buffer);
1184         s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1185                                              block_w, block_h, mx, my,
1186                                              lc->mc_buffer);
1187     }
1188 }
1189
1190 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1191                                 const Mv *mv, int y0, int height)
1192 {
1193     int y = (mv->y >> 2) + y0 + height + 9;
1194
1195     if (s->threads_type == FF_THREAD_FRAME )
1196         ff_thread_await_progress(&ref->tf, y, 0);
1197 }
1198
1199 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1200                                 int nPbW, int nPbH,
1201                                 int log2_cb_size, int partIdx)
1202 {
1203 #define POS(c_idx, x, y)                                                              \
1204     &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1205                            (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1206     HEVCLocalContext *lc = s->HEVClc;
1207     int merge_idx = 0;
1208     struct MvField current_mv = {{{ 0 }}};
1209
1210     int min_pu_width = s->sps->min_pu_width;
1211
1212     MvField *tab_mvf = s->ref->tab_mvf;
1213     RefPicList  *refPicList = s->ref->refPicList;
1214     HEVCFrame *ref0, *ref1;
1215
1216     int tmpstride = MAX_PB_SIZE;
1217
1218     uint8_t *dst0 = POS(0, x0, y0);
1219     uint8_t *dst1 = POS(1, x0, y0);
1220     uint8_t *dst2 = POS(2, x0, y0);
1221     int log2_min_cb_size = s->sps->log2_min_cb_size;
1222     int min_cb_width     = s->sps->min_cb_width;
1223     int x_cb             = x0 >> log2_min_cb_size;
1224     int y_cb             = y0 >> log2_min_cb_size;
1225     int ref_idx[2];
1226     int mvp_flag[2];
1227     int x_pu, y_pu;
1228     int i, j;
1229
1230     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1231         if (s->sh.max_num_merge_cand > 1)
1232             merge_idx = ff_hevc_merge_idx_decode(s);
1233         else
1234             merge_idx = 0;
1235
1236         ff_hevc_luma_mv_merge_mode(s, x0, y0,
1237                                    1 << log2_cb_size,
1238                                    1 << log2_cb_size,
1239                                    log2_cb_size, partIdx,
1240                                    merge_idx, &current_mv);
1241         x_pu = x0 >> s->sps->log2_min_pu_size;
1242         y_pu = y0 >> s->sps->log2_min_pu_size;
1243
1244         for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1245             for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1246                 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1247     } else { /* MODE_INTER */
1248         lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1249         if (lc->pu.merge_flag) {
1250             if (s->sh.max_num_merge_cand > 1)
1251                 merge_idx = ff_hevc_merge_idx_decode(s);
1252             else
1253                 merge_idx = 0;
1254
1255             ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1256                                        partIdx, merge_idx, &current_mv);
1257             x_pu = x0 >> s->sps->log2_min_pu_size;
1258             y_pu = y0 >> s->sps->log2_min_pu_size;
1259
1260             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1261                 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1262                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1263         } else {
1264             enum InterPredIdc inter_pred_idc = PRED_L0;
1265             ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1266             if (s->sh.slice_type == B_SLICE)
1267                 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1268
1269             if (inter_pred_idc != PRED_L1) {
1270                 if (s->sh.nb_refs[L0]) {
1271                     ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1272                     current_mv.ref_idx[0] = ref_idx[0];
1273                 }
1274                 current_mv.pred_flag[0] = 1;
1275                 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
1276                 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
1277                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1278                                          partIdx, merge_idx, &current_mv,
1279                                          mvp_flag[0], 0);
1280                 current_mv.mv[0].x += lc->pu.mvd.x;
1281                 current_mv.mv[0].y += lc->pu.mvd.y;
1282             }
1283
1284             if (inter_pred_idc != PRED_L0) {
1285                 if (s->sh.nb_refs[L1]) {
1286                     ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1287                     current_mv.ref_idx[1] = ref_idx[1];
1288                 }
1289
1290                 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1291                     lc->pu.mvd.x = 0;
1292                     lc->pu.mvd.y = 0;
1293                 } else {
1294                     ff_hevc_hls_mvd_coding(s, x0, y0, 1);
1295                 }
1296
1297                 current_mv.pred_flag[1] = 1;
1298                 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
1299                 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1300                                          partIdx, merge_idx, &current_mv,
1301                                          mvp_flag[1], 1);
1302                 current_mv.mv[1].x += lc->pu.mvd.x;
1303                 current_mv.mv[1].y += lc->pu.mvd.y;
1304             }
1305
1306             x_pu = x0 >> s->sps->log2_min_pu_size;
1307             y_pu = y0 >> s->sps->log2_min_pu_size;
1308
1309             for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1310                 for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1311                     tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1312         }
1313     }
1314
1315     if (current_mv.pred_flag[0]) {
1316         ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1317         if (!ref0)
1318             return;
1319         hevc_await_progress(s, ref0, &current_mv.mv[0], y0, nPbH);
1320     }
1321     if (current_mv.pred_flag[1]) {
1322         ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1323         if (!ref1)
1324             return;
1325         hevc_await_progress(s, ref1, &current_mv.mv[1], y0, nPbH);
1326     }
1327
1328     if (current_mv.pred_flag[0] && !current_mv.pred_flag[1]) {
1329         DECLARE_ALIGNED(16, int16_t,  tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1330         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1331
1332         luma_mc(s, tmp, tmpstride, ref0->frame,
1333                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
1334
1335         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1336             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1337             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1338                                      s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1339                                      s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1340                                      dst0, s->frame->linesize[0], tmp,
1341                                      tmpstride, nPbW, nPbH);
1342         } else {
1343             s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1344         }
1345         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1346                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1347
1348         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1349             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1350             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1351                                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1352                                      s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1353                                      dst1, s->frame->linesize[1], tmp, tmpstride,
1354                                      nPbW / 2, nPbH / 2);
1355             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1356                                      s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1357                                      s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1358                                      dst2, s->frame->linesize[2], tmp2, tmpstride,
1359                                      nPbW / 2, nPbH / 2);
1360         } else {
1361             s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1362             s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1363         }
1364     } else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1365         DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1366         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1367
1368         if (!ref1)
1369             return;
1370
1371         luma_mc(s, tmp, tmpstride, ref1->frame,
1372                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
1373
1374         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1375             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1376             s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1377                                       s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1378                                       s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1379                                       dst0, s->frame->linesize[0], tmp, tmpstride,
1380                                       nPbW, nPbH);
1381         } else {
1382             s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1383         }
1384
1385         chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
1386                   &current_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
1387
1388         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1389             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1390             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1391                                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1392                                      s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1393                                      dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1394             s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1395                                      s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1396                                      s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1397                                      dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1398         } else {
1399             s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1400             s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1401         }
1402     } else if (current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1403         DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1404         DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1405         DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
1406         DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
1407         HEVCFrame *ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1408         HEVCFrame *ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1409
1410         if (!ref0 || !ref1)
1411             return;
1412
1413         luma_mc(s, tmp, tmpstride, ref0->frame,
1414                 &current_mv.mv[0], x0, y0, nPbW, nPbH);
1415         luma_mc(s, tmp2, tmpstride, ref1->frame,
1416                 &current_mv.mv[1], x0, y0, nPbW, nPbH);
1417
1418         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1419             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1420             s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
1421                                          s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1422                                          s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1423                                          s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1424                                          s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1425                                          dst0, s->frame->linesize[0],
1426                                          tmp, tmp2, tmpstride, nPbW, nPbH);
1427         } else {
1428             s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
1429                                              tmp, tmp2, tmpstride, nPbW, nPbH);
1430         }
1431
1432         chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1433                   &current_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1434         chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
1435                   &current_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1436
1437         if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1438             (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1439             s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1440                                          s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1441                                          s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1442                                          s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1443                                          s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1444                                          dst1, s->frame->linesize[1], tmp, tmp3,
1445                                          tmpstride, nPbW / 2, nPbH / 2);
1446             s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1447                                          s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1448                                          s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1449                                          s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1450                                          s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1451                                          dst2, s->frame->linesize[2], tmp2, tmp4,
1452                                          tmpstride, nPbW / 2, nPbH / 2);
1453         } else {
1454             s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
1455             s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
1456         }
1457     }
1458 }
1459
1460 /**
1461  * 8.4.1
1462  */
1463 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1464                                 int prev_intra_luma_pred_flag)
1465 {
1466     HEVCLocalContext *lc = s->HEVClc;
1467     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1468     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1469     int min_pu_width     = s->sps->min_pu_width;
1470     int size_in_pus      = pu_size >> s->sps->log2_min_pu_size;
1471     int x0b              = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1472     int y0b              = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1473
1474     int cand_up   = (lc->ctb_up_flag || y0b) ?
1475                     s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1476     int cand_left = (lc->ctb_left_flag || x0b) ?
1477                     s->tab_ipm[y_pu * min_pu_width + x_pu - 1]   : INTRA_DC;
1478
1479     int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1480
1481     MvField *tab_mvf = s->ref->tab_mvf;
1482     int intra_pred_mode;
1483     int candidate[3];
1484     int i, j;
1485
1486     // intra_pred_mode prediction does not cross vertical CTB boundaries
1487     if ((y0 - 1) < y_ctb)
1488         cand_up = INTRA_DC;
1489
1490     if (cand_left == cand_up) {
1491         if (cand_left < 2) {
1492             candidate[0] = INTRA_PLANAR;
1493             candidate[1] = INTRA_DC;
1494             candidate[2] = INTRA_ANGULAR_26;
1495         } else {
1496             candidate[0] = cand_left;
1497             candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1498             candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1499         }
1500     } else {
1501         candidate[0] = cand_left;
1502         candidate[1] = cand_up;
1503         if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1504             candidate[2] = INTRA_PLANAR;
1505         } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1506             candidate[2] = INTRA_DC;
1507         } else {
1508             candidate[2] = INTRA_ANGULAR_26;
1509         }
1510     }
1511
1512     if (prev_intra_luma_pred_flag) {
1513         intra_pred_mode = candidate[lc->pu.mpm_idx];
1514     } else {
1515         if (candidate[0] > candidate[1])
1516             FFSWAP(uint8_t, candidate[0], candidate[1]);
1517         if (candidate[0] > candidate[2])
1518             FFSWAP(uint8_t, candidate[0], candidate[2]);
1519         if (candidate[1] > candidate[2])
1520             FFSWAP(uint8_t, candidate[1], candidate[2]);
1521
1522         intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1523         for (i = 0; i < 3; i++)
1524             if (intra_pred_mode >= candidate[i])
1525                 intra_pred_mode++;
1526     }
1527
1528     /* write the intra prediction units into the mv array */
1529     if (!size_in_pus)
1530         size_in_pus = 1;
1531     for (i = 0; i < size_in_pus; i++) {
1532         memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1533                intra_pred_mode, size_in_pus);
1534
1535         for (j = 0; j < size_in_pus; j++) {
1536             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].is_intra     = 1;
1537             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[0] = 0;
1538             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[1] = 0;
1539             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[0]   = 0;
1540             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[1]   = 0;
1541             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].x      = 0;
1542             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].y      = 0;
1543             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].x      = 0;
1544             tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].y      = 0;
1545         }
1546     }
1547
1548     return intra_pred_mode;
1549 }
1550
1551 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1552                                           int log2_cb_size, int ct_depth)
1553 {
1554     int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1555     int x_cb   = x0 >> s->sps->log2_min_cb_size;
1556     int y_cb   = y0 >> s->sps->log2_min_cb_size;
1557     int y;
1558
1559     for (y = 0; y < length; y++)
1560         memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1561                ct_depth, length);
1562 }
1563
1564 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1565                                   int log2_cb_size)
1566 {
1567     HEVCLocalContext *lc = s->HEVClc;
1568     static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1569     uint8_t prev_intra_luma_pred_flag[4];
1570     int split   = lc->cu.part_mode == PART_NxN;
1571     int pb_size = (1 << log2_cb_size) >> split;
1572     int side    = split + 1;
1573     int chroma_mode;
1574     int i, j;
1575
1576     for (i = 0; i < side; i++)
1577         for (j = 0; j < side; j++)
1578             prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1579
1580     for (i = 0; i < side; i++) {
1581         for (j = 0; j < side; j++) {
1582             if (prev_intra_luma_pred_flag[2 * i + j])
1583                 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1584             else
1585                 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1586
1587             lc->pu.intra_pred_mode[2 * i + j] =
1588                 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1589                                      prev_intra_luma_pred_flag[2 * i + j]);
1590         }
1591     }
1592
1593     chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1594     if (chroma_mode != 4) {
1595         if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1596             lc->pu.intra_pred_mode_c = 34;
1597         else
1598             lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1599     } else {
1600         lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1601     }
1602 }
1603
1604 static void intra_prediction_unit_default_value(HEVCContext *s,
1605                                                 int x0, int y0,
1606                                                 int log2_cb_size)
1607 {
1608     HEVCLocalContext *lc = s->HEVClc;
1609     int pb_size          = 1 << log2_cb_size;
1610     int size_in_pus      = pb_size >> s->sps->log2_min_pu_size;
1611     int min_pu_width     = s->sps->min_pu_width;
1612     MvField *tab_mvf     = s->ref->tab_mvf;
1613     int x_pu             = x0 >> s->sps->log2_min_pu_size;
1614     int y_pu             = y0 >> s->sps->log2_min_pu_size;
1615     int j, k;
1616
1617     if (size_in_pus == 0)
1618         size_in_pus = 1;
1619     for (j = 0; j < size_in_pus; j++) {
1620         memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
1621         for (k = 0; k < size_in_pus; k++)
1622             tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].is_intra = lc->cu.pred_mode == MODE_INTRA;
1623     }
1624 }
1625
1626 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
1627 {
1628     int cb_size          = 1 << log2_cb_size;
1629     HEVCLocalContext *lc = s->HEVClc;
1630     int log2_min_cb_size = s->sps->log2_min_cb_size;
1631     int length           = cb_size >> log2_min_cb_size;
1632     int min_cb_width     = s->sps->min_cb_width;
1633     int x_cb             = x0 >> log2_min_cb_size;
1634     int y_cb             = y0 >> log2_min_cb_size;
1635     int x, y, ret;
1636     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1637
1638     lc->cu.x                = x0;
1639     lc->cu.y                = y0;
1640     lc->cu.rqt_root_cbf     = 1;
1641     lc->cu.pred_mode        = MODE_INTRA;
1642     lc->cu.part_mode        = PART_2Nx2N;
1643     lc->cu.intra_split_flag = 0;
1644     lc->cu.pcm_flag         = 0;
1645
1646     SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
1647     for (x = 0; x < 4; x++)
1648         lc->pu.intra_pred_mode[x] = 1;
1649     if (s->pps->transquant_bypass_enable_flag) {
1650         lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
1651         if (lc->cu.cu_transquant_bypass_flag)
1652             set_deblocking_bypass(s, x0, y0, log2_cb_size);
1653     } else
1654         lc->cu.cu_transquant_bypass_flag = 0;
1655
1656     if (s->sh.slice_type != I_SLICE) {
1657         uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
1658
1659         lc->cu.pred_mode = MODE_SKIP;
1660         x = y_cb * min_cb_width + x_cb;
1661         for (y = 0; y < length; y++) {
1662             memset(&s->skip_flag[x], skip_flag, length);
1663             x += min_cb_width;
1664         }
1665         lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
1666     }
1667
1668     if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1669         hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1670         intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1671
1672         if (!s->sh.disable_deblocking_filter_flag)
1673             ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1674                                                   lc->slice_or_tiles_up_boundary,
1675                                                   lc->slice_or_tiles_left_boundary);
1676     } else {
1677         if (s->sh.slice_type != I_SLICE)
1678             lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
1679         if (lc->cu.pred_mode != MODE_INTRA ||
1680             log2_cb_size == s->sps->log2_min_cb_size) {
1681             lc->cu.part_mode        = ff_hevc_part_mode_decode(s, log2_cb_size);
1682             lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
1683                                       lc->cu.pred_mode == MODE_INTRA;
1684         }
1685
1686         if (lc->cu.pred_mode == MODE_INTRA) {
1687             if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
1688                 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
1689                 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
1690                 lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
1691             }
1692             if (lc->cu.pcm_flag) {
1693                 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1694                 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
1695                 if (s->sps->pcm.loop_filter_disable_flag)
1696                     set_deblocking_bypass(s, x0, y0, log2_cb_size);
1697
1698                 if (ret < 0)
1699                     return ret;
1700             } else {
1701                 intra_prediction_unit(s, x0, y0, log2_cb_size);
1702             }
1703         } else {
1704             intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1705             switch (lc->cu.part_mode) {
1706             case PART_2Nx2N:
1707                 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1708                 break;
1709             case PART_2NxN:
1710                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size / 2, log2_cb_size, 0);
1711                 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
1712                 break;
1713             case PART_Nx2N:
1714                 hls_prediction_unit(s, x0,               y0, cb_size / 2, cb_size, log2_cb_size, 0);
1715                 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
1716                 break;
1717             case PART_2NxnU:
1718                 hls_prediction_unit(s, x0, y0,               cb_size, cb_size     / 4, log2_cb_size, 0);
1719                 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
1720                 break;
1721             case PART_2NxnD:
1722                 hls_prediction_unit(s, x0, y0,                   cb_size, cb_size * 3 / 4, log2_cb_size, 0);
1723                 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size     / 4, log2_cb_size, 1);
1724                 break;
1725             case PART_nLx2N:
1726                 hls_prediction_unit(s, x0,               y0, cb_size     / 4, cb_size, log2_cb_size, 0);
1727                 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
1728                 break;
1729             case PART_nRx2N:
1730                 hls_prediction_unit(s, x0,                   y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
1731                 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size     / 4, cb_size, log2_cb_size, 1);
1732                 break;
1733             case PART_NxN:
1734                 hls_prediction_unit(s, x0,               y0,               cb_size / 2, cb_size / 2, log2_cb_size, 0);
1735                 hls_prediction_unit(s, x0 + cb_size / 2, y0,               cb_size / 2, cb_size / 2, log2_cb_size, 1);
1736                 hls_prediction_unit(s, x0,               y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
1737                 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
1738                 break;
1739             }
1740         }
1741
1742         if (!lc->cu.pcm_flag) {
1743             if (lc->cu.pred_mode != MODE_INTRA &&
1744                 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
1745                 lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
1746             }
1747             if (lc->cu.rqt_root_cbf) {
1748                 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
1749                                          s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
1750                                          s->sps->max_transform_hierarchy_depth_inter;
1751                 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
1752                                          log2_cb_size,
1753                                          log2_cb_size, 0, 0);
1754                 if (ret < 0)
1755                     return ret;
1756             } else {
1757                 if (!s->sh.disable_deblocking_filter_flag)
1758                     ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1759                                                           lc->slice_or_tiles_up_boundary,
1760                                                           lc->slice_or_tiles_left_boundary);
1761             }
1762         }
1763     }
1764
1765     if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
1766         ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
1767
1768     x = y_cb * min_cb_width + x_cb;
1769     for (y = 0; y < length; y++) {
1770         memset(&s->qp_y_tab[x], lc->qp_y, length);
1771         x += min_cb_width;
1772     }
1773
1774     if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1775        ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0) {
1776         lc->qPy_pred = lc->qp_y;
1777     }
1778
1779     set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
1780
1781     return 0;
1782 }
1783
1784 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
1785                                int log2_cb_size, int cb_depth)
1786 {
1787     HEVCLocalContext *lc = s->HEVClc;
1788     const int cb_size    = 1 << log2_cb_size;
1789     int ret;
1790     int qp_block_mask = (1<<(s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
1791
1792     lc->ct.depth = cb_depth;
1793     if (x0 + cb_size <= s->sps->width  &&
1794         y0 + cb_size <= s->sps->height &&
1795         log2_cb_size > s->sps->log2_min_cb_size) {
1796         SAMPLE(s->split_cu_flag, x0, y0) =
1797             ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
1798     } else {
1799         SAMPLE(s->split_cu_flag, x0, y0) =
1800             (log2_cb_size > s->sps->log2_min_cb_size);
1801     }
1802     if (s->pps->cu_qp_delta_enabled_flag &&
1803         log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
1804         lc->tu.is_cu_qp_delta_coded = 0;
1805         lc->tu.cu_qp_delta          = 0;
1806     }
1807
1808     if (SAMPLE(s->split_cu_flag, x0, y0)) {
1809         const int cb_size_split = cb_size >> 1;
1810         const int x1 = x0 + cb_size_split;
1811         const int y1 = y0 + cb_size_split;
1812
1813         int more_data = 0;
1814
1815         more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
1816         if (more_data < 0)
1817             return more_data;
1818
1819         if (more_data && x1 < s->sps->width) {
1820             more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
1821             if (more_data < 0)
1822                 return more_data;
1823         }
1824         if (more_data && y1 < s->sps->height) {
1825             more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
1826             if (more_data < 0)
1827                 return more_data;
1828         }
1829         if (more_data && x1 < s->sps->width &&
1830             y1 < s->sps->height) {
1831             more_data = hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
1832             if (more_data < 0)
1833                 return more_data;
1834         }
1835
1836         if(((x0 + (1<<log2_cb_size)) & qp_block_mask) == 0 &&
1837             ((y0 + (1<<log2_cb_size)) & qp_block_mask) == 0)
1838             lc->qPy_pred = lc->qp_y;
1839
1840         if (more_data)
1841             return ((x1 + cb_size_split) < s->sps->width ||
1842                     (y1 + cb_size_split) < s->sps->height);
1843         else
1844             return 0;
1845     } else {
1846         ret = hls_coding_unit(s, x0, y0, log2_cb_size);
1847         if (ret < 0)
1848             return ret;
1849         if ((!((x0 + cb_size) %
1850                (1 << (s->sps->log2_ctb_size))) ||
1851              (x0 + cb_size >= s->sps->width)) &&
1852             (!((y0 + cb_size) %
1853                (1 << (s->sps->log2_ctb_size))) ||
1854              (y0 + cb_size >= s->sps->height))) {
1855             int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
1856             return !end_of_slice_flag;
1857         } else {
1858             return 1;
1859         }
1860     }
1861
1862     return 0;
1863 }
1864
1865 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
1866                                  int ctb_addr_ts)
1867 {
1868     HEVCLocalContext *lc  = s->HEVClc;
1869     int ctb_size          = 1 << s->sps->log2_ctb_size;
1870     int ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1871     int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
1872
1873     int tile_left_boundary, tile_up_boundary;
1874     int slice_left_boundary, slice_up_boundary;
1875
1876     s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
1877
1878     if (s->pps->entropy_coding_sync_enabled_flag) {
1879         if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
1880             lc->first_qp_group = 1;
1881         lc->end_of_tiles_x = s->sps->width;
1882     } else if (s->pps->tiles_enabled_flag) {
1883         if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
1884             int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
1885             lc->end_of_tiles_x   = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
1886             lc->first_qp_group   = 1;
1887         }
1888     } else {
1889         lc->end_of_tiles_x = s->sps->width;
1890     }
1891
1892     lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
1893
1894     if (s->pps->tiles_enabled_flag) {
1895         tile_left_boundary = x_ctb > 0 &&
1896                              s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
1897         slice_left_boundary = x_ctb > 0 &&
1898                               s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
1899         tile_up_boundary  = y_ctb > 0 &&
1900                             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]];
1901         slice_up_boundary = y_ctb > 0 &&
1902                             s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
1903     } else {
1904         tile_left_boundary =
1905         tile_up_boundary   = 0;
1906         slice_left_boundary = ctb_addr_in_slice <= 0;
1907         slice_up_boundary   = ctb_addr_in_slice < s->sps->ctb_width;
1908     }
1909     lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
1910     lc->slice_or_tiles_up_boundary   = slice_up_boundary   + (tile_up_boundary   << 1);
1911     lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0)                  && !tile_left_boundary);
1912     lc->ctb_up_flag   = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
1913     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]]));
1914     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]]));
1915 }
1916
1917 static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
1918 {
1919     HEVCContext *s  = avctxt->priv_data;
1920     int ctb_size    = 1 << s->sps->log2_ctb_size;
1921     int more_data   = 1;
1922     int x_ctb       = 0;
1923     int y_ctb       = 0;
1924     int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
1925
1926     if (!ctb_addr_ts && s->sh.dependent_slice_segment_flag) {
1927         av_log(s->avctx, AV_LOG_ERROR, "Impossible initial tile.\n");
1928         return AVERROR_INVALIDDATA;
1929     }
1930
1931     if (s->sh.dependent_slice_segment_flag) {
1932         int prev_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts - 1];
1933         if (s->tab_slice_address[prev_rs] != s->sh.slice_addr) {
1934             av_log(s->avctx, AV_LOG_ERROR, "Previous slice segment missing\n");
1935             return AVERROR_INVALIDDATA;
1936         }
1937     }
1938
1939     while (more_data && ctb_addr_ts < s->sps->ctb_size) {
1940         int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1941
1942         x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1943         y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1944         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
1945
1946         ff_hevc_cabac_init(s, ctb_addr_ts);
1947
1948         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
1949
1950         s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
1951         s->deblock[ctb_addr_rs].tc_offset   = s->sh.tc_offset;
1952         s->filter_slice_edges[ctb_addr_rs]  = s->sh.slice_loop_filter_across_slices_enabled_flag;
1953
1954         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
1955         if (more_data < 0) {
1956             s->tab_slice_address[ctb_addr_rs] = -1;
1957             return more_data;
1958         }
1959
1960
1961         ctb_addr_ts++;
1962         ff_hevc_save_states(s, ctb_addr_ts);
1963         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
1964     }
1965
1966     if (x_ctb + ctb_size >= s->sps->width &&
1967         y_ctb + ctb_size >= s->sps->height)
1968         ff_hevc_hls_filter(s, x_ctb, y_ctb);
1969
1970     return ctb_addr_ts;
1971 }
1972
1973 static int hls_slice_data(HEVCContext *s)
1974 {
1975     int arg[2];
1976     int ret[2];
1977
1978     arg[0] = 0;
1979     arg[1] = 1;
1980
1981     s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
1982     return ret[0];
1983 }
1984 static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
1985 {
1986     HEVCContext *s1  = avctxt->priv_data, *s;
1987     HEVCLocalContext *lc;
1988     int ctb_size    = 1<< s1->sps->log2_ctb_size;
1989     int more_data   = 1;
1990     int *ctb_row_p    = input_ctb_row;
1991     int ctb_row = ctb_row_p[job];
1992     int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
1993     int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
1994     int thread = ctb_row % s1->threads_number;
1995     int ret;
1996
1997     s = s1->sList[self_id];
1998     lc = s->HEVClc;
1999
2000     if(ctb_row) {
2001         ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
2002
2003         if (ret < 0)
2004             return ret;
2005         ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
2006     }
2007
2008     while(more_data && ctb_addr_ts < s->sps->ctb_size) {
2009         int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
2010         int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
2011
2012         hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2013
2014         ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
2015
2016         if (avpriv_atomic_int_get(&s1->wpp_err)){
2017             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2018             return 0;
2019         }
2020
2021         ff_hevc_cabac_init(s, ctb_addr_ts);
2022         hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2023         more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2024
2025         if (more_data < 0) {
2026             s->tab_slice_address[ctb_addr_rs] = -1;
2027             return more_data;
2028         }
2029
2030         ctb_addr_ts++;
2031
2032         ff_hevc_save_states(s, ctb_addr_ts);
2033         ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
2034         ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2035
2036         if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
2037             avpriv_atomic_int_set(&s1->wpp_err,  1);
2038             ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2039             return 0;
2040         }
2041
2042         if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
2043             ff_hevc_hls_filter(s, x_ctb, y_ctb);
2044             ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
2045             return ctb_addr_ts;
2046         }
2047         ctb_addr_rs       = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2048         x_ctb+=ctb_size;
2049
2050         if(x_ctb >= s->sps->width) {
2051             break;
2052         }
2053     }
2054     ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
2055
2056     return 0;
2057 }
2058
2059 static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
2060 {
2061     HEVCLocalContext *lc = s->HEVClc;
2062     int *ret = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2063     int *arg = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
2064     int offset;
2065     int startheader, cmpt = 0;
2066     int i, j, res = 0;
2067
2068
2069     if (!s->sList[1]) {
2070         ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
2071
2072
2073         for (i = 1; i < s->threads_number; i++) {
2074             s->sList[i] = av_malloc(sizeof(HEVCContext));
2075             memcpy(s->sList[i], s, sizeof(HEVCContext));
2076             s->HEVClcList[i] = av_malloc(sizeof(HEVCLocalContext));
2077             s->sList[i]->HEVClc = s->HEVClcList[i];
2078         }
2079     }
2080
2081     offset = (lc->gb.index >> 3);
2082
2083     for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
2084         if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2085             startheader--;
2086             cmpt++;
2087         }
2088     }
2089
2090     for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
2091         offset += (s->sh.entry_point_offset[i - 1] - cmpt);
2092         for (j = 0, cmpt = 0, startheader = offset
2093              + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
2094             if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2095                 startheader--;
2096                 cmpt++;
2097             }
2098         }
2099         s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
2100         s->sh.offset[i - 1] = offset;
2101
2102     }
2103     if (s->sh.num_entry_point_offsets != 0) {
2104         offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
2105         s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
2106         s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
2107
2108     }
2109     s->data = nal;
2110
2111     for (i = 1; i < s->threads_number; i++) {
2112         s->sList[i]->HEVClc->first_qp_group = 1;
2113         s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
2114         memcpy(s->sList[i], s, sizeof(HEVCContext));
2115         s->sList[i]->HEVClc = s->HEVClcList[i];
2116     }
2117
2118     avpriv_atomic_int_set(&s->wpp_err, 0);
2119     ff_reset_entries(s->avctx);
2120
2121     for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
2122         arg[i] = i;
2123         ret[i] = 0;
2124     }
2125
2126     if (s->pps->entropy_coding_sync_enabled_flag)
2127         s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
2128
2129     for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
2130         res += ret[i];
2131     av_free(ret);
2132     av_free(arg);
2133     return res;
2134 }
2135
2136 /**
2137  * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2138  * 0 if the unit should be skipped, 1 otherwise
2139  */
2140 static int hls_nal_unit(HEVCContext *s)
2141 {
2142     GetBitContext *gb = &s->HEVClc->gb;
2143     int nuh_layer_id;
2144
2145     if (get_bits1(gb) != 0)
2146         return AVERROR_INVALIDDATA;
2147
2148     s->nal_unit_type = get_bits(gb, 6);
2149
2150     nuh_layer_id   = get_bits(gb, 6);
2151     s->temporal_id = get_bits(gb, 3) - 1;
2152     if (s->temporal_id < 0)
2153         return AVERROR_INVALIDDATA;
2154
2155     av_log(s->avctx, AV_LOG_DEBUG,
2156            "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2157            s->nal_unit_type, nuh_layer_id, s->temporal_id);
2158
2159     return nuh_layer_id == 0;
2160 }
2161
2162 static void restore_tqb_pixels(HEVCContext *s)
2163 {
2164     int min_pu_size = 1 << s->sps->log2_min_pu_size;
2165     int x, y, c_idx;
2166
2167     for (c_idx = 0; c_idx < 3; c_idx++) {
2168         ptrdiff_t stride = s->frame->linesize[c_idx];
2169         int hshift       = s->sps->hshift[c_idx];
2170         int vshift       = s->sps->vshift[c_idx];
2171         for (y = 0; y < s->sps->min_pu_height; y++) {
2172             for (x = 0; x < s->sps->min_pu_width; x++) {
2173                 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2174                     int n;
2175                     int len      = min_pu_size >> hshift;
2176                     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)];
2177                     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)];
2178                     for (n = 0; n < (min_pu_size >> vshift); n++) {
2179                         memcpy(dst, src, len);
2180                         src += stride;
2181                         dst += stride;
2182                     }
2183                 }
2184             }
2185         }
2186     }
2187 }
2188
2189 static int set_side_data(HEVCContext *s)
2190 {
2191     AVFrame *out = s->ref->frame;
2192
2193     if (s->sei_frame_packing_present &&
2194         s->frame_packing_arrangement_type >= 3 &&
2195         s->frame_packing_arrangement_type <= 5 &&
2196         s->content_interpretation_type > 0 &&
2197         s->content_interpretation_type < 3) {
2198         AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2199         if (!stereo)
2200             return AVERROR(ENOMEM);
2201
2202         switch (s->frame_packing_arrangement_type) {
2203         case 3:
2204             if (s->quincunx_subsampling)
2205                 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2206             else
2207                 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2208             break;
2209         case 4:
2210             stereo->type = AV_STEREO3D_TOPBOTTOM;
2211             break;
2212         case 5:
2213             stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2214             break;
2215         }
2216
2217         if (s->content_interpretation_type == 2)
2218             stereo->flags = AV_STEREO3D_FLAG_INVERT;
2219     }
2220
2221     return 0;
2222 }
2223
2224 static int hevc_frame_start(HEVCContext *s)
2225 {
2226     HEVCLocalContext *lc = s->HEVClc;
2227     int pic_size_in_ctb  = ((s->sps->width  >> s->sps->log2_min_cb_size) + 1) *
2228                            ((s->sps->height >> s->sps->log2_min_cb_size) + 1);
2229     int ret;
2230
2231     memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2232     memset(s->vertical_bs,   0, 2 * s->bs_width * (s->bs_height + 1));
2233     memset(s->cbf_luma,      0, s->sps->min_tb_width * s->sps->min_tb_height);
2234     memset(s->is_pcm,        0, s->sps->min_pu_width * s->sps->min_pu_height);
2235     memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
2236
2237     s->is_decoded        = 0;
2238     s->first_nal_type    = s->nal_unit_type;
2239
2240     if (s->pps->tiles_enabled_flag)
2241         lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2242
2243     ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2244                               s->poc);
2245     if (ret < 0)
2246         goto fail;
2247
2248     ret = ff_hevc_frame_rps(s);
2249     if (ret < 0) {
2250         av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2251         goto fail;
2252     }
2253
2254     ret = set_side_data(s);
2255     if (ret < 0)
2256         goto fail;
2257
2258     av_frame_unref(s->output_frame);
2259     ret = ff_hevc_output_frame(s, s->output_frame, 0);
2260     if (ret < 0)
2261         goto fail;
2262
2263     ff_thread_finish_setup(s->avctx);
2264
2265     return 0;
2266
2267 fail:
2268     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2269         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2270     s->ref = NULL;
2271     return ret;
2272 }
2273
2274 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2275 {
2276     HEVCLocalContext *lc = s->HEVClc;
2277     GetBitContext *gb    = &lc->gb;
2278     int ctb_addr_ts, ret;
2279
2280     ret = init_get_bits8(gb, nal, length);
2281     if (ret < 0)
2282         return ret;
2283
2284     ret = hls_nal_unit(s);
2285     if (ret < 0) {
2286         av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2287                s->nal_unit_type);
2288         goto fail;
2289     } else if (!ret)
2290         return 0;
2291
2292     switch (s->nal_unit_type) {
2293     case NAL_VPS:
2294         ret = ff_hevc_decode_nal_vps(s);
2295         if (ret < 0)
2296             goto fail;
2297         break;
2298     case NAL_SPS:
2299         ret = ff_hevc_decode_nal_sps(s);
2300         if (ret < 0)
2301             goto fail;
2302         break;
2303     case NAL_PPS:
2304         ret = ff_hevc_decode_nal_pps(s);
2305         if (ret < 0)
2306             goto fail;
2307         break;
2308     case NAL_SEI_PREFIX:
2309     case NAL_SEI_SUFFIX:
2310         ret = ff_hevc_decode_nal_sei(s);
2311         if (ret < 0)
2312             goto fail;
2313         break;
2314     case NAL_TRAIL_R:
2315     case NAL_TRAIL_N:
2316     case NAL_TSA_N:
2317     case NAL_TSA_R:
2318     case NAL_STSA_N:
2319     case NAL_STSA_R:
2320     case NAL_BLA_W_LP:
2321     case NAL_BLA_W_RADL:
2322     case NAL_BLA_N_LP:
2323     case NAL_IDR_W_RADL:
2324     case NAL_IDR_N_LP:
2325     case NAL_CRA_NUT:
2326     case NAL_RADL_N:
2327     case NAL_RADL_R:
2328     case NAL_RASL_N:
2329     case NAL_RASL_R:
2330         ret = hls_slice_header(s);
2331         if (ret < 0)
2332             return ret;
2333
2334         if (s->max_ra == INT_MAX) {
2335             if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2336                 s->max_ra = s->poc;
2337             } else {
2338                 if (IS_IDR(s))
2339                     s->max_ra = INT_MIN;
2340             }
2341         }
2342
2343         if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2344             s->poc <= s->max_ra) {
2345             s->is_decoded = 0;
2346             break;
2347         } else {
2348             if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2349                 s->max_ra = INT_MIN;
2350         }
2351
2352         if (s->sh.first_slice_in_pic_flag) {
2353             ret = hevc_frame_start(s);
2354             if (ret < 0)
2355                 return ret;
2356         } else if (!s->ref) {
2357             av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2358             goto fail;
2359         }
2360
2361         if (s->nal_unit_type != s->first_nal_type) {
2362             av_log(s->avctx, AV_LOG_ERROR,
2363                    "Non-matching NAL types of the VCL NALUs: %d %d\n",
2364                    s->first_nal_type, s->nal_unit_type);
2365             return AVERROR_INVALIDDATA;
2366         }
2367
2368         if (!s->sh.dependent_slice_segment_flag &&
2369             s->sh.slice_type != I_SLICE) {
2370             ret = ff_hevc_slice_rpl(s);
2371             if (ret < 0) {
2372                 av_log(s->avctx, AV_LOG_WARNING,
2373                        "Error constructing the reference lists for the current slice.\n");
2374                 goto fail;
2375             }
2376         }
2377
2378         if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
2379             ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
2380         else
2381             ctb_addr_ts = hls_slice_data(s);
2382         if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2383             s->is_decoded = 1;
2384             if ((s->pps->transquant_bypass_enable_flag ||
2385                  (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2386                 s->sps->sao_enabled)
2387                 restore_tqb_pixels(s);
2388         }
2389
2390         if (ctb_addr_ts < 0) {
2391             ret = ctb_addr_ts;
2392             goto fail;
2393         }
2394         break;
2395     case NAL_EOS_NUT:
2396     case NAL_EOB_NUT:
2397         s->seq_decode = (s->seq_decode + 1) & 0xff;
2398         s->max_ra     = INT_MAX;
2399         break;
2400     case NAL_AUD:
2401     case NAL_FD_NUT:
2402         break;
2403     default:
2404         av_log(s->avctx, AV_LOG_INFO,
2405                "Skipping NAL unit %d\n", s->nal_unit_type);
2406     }
2407
2408     return 0;
2409 fail:
2410     if (s->avctx->err_recognition & AV_EF_EXPLODE)
2411         return ret;
2412     return 0;
2413 }
2414
2415 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2416  * between these functions would be nice. */
2417 int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
2418                          HEVCNAL *nal)
2419 {
2420     int i, si, di;
2421     uint8_t *dst;
2422
2423     s->skipped_bytes = 0;
2424 #define STARTCODE_TEST                                                  \
2425         if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
2426             if (src[i + 2] != 3) {                                      \
2427                 /* startcode, so we must be past the end */             \
2428                 length = i;                                             \
2429             }                                                           \
2430             break;                                                      \
2431         }
2432 #if HAVE_FAST_UNALIGNED
2433 #define FIND_FIRST_ZERO                                                 \
2434         if (i > 0 && !src[i])                                           \
2435             i--;                                                        \
2436         while (src[i])                                                  \
2437             i++
2438 #if HAVE_FAST_64BIT
2439     for (i = 0; i + 1 < length; i += 9) {
2440         if (!((~AV_RN64A(src + i) &
2441                (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2442               0x8000800080008080ULL))
2443             continue;
2444         FIND_FIRST_ZERO;
2445         STARTCODE_TEST;
2446         i -= 7;
2447     }
2448 #else
2449     for (i = 0; i + 1 < length; i += 5) {
2450         if (!((~AV_RN32A(src + i) &
2451                (AV_RN32A(src + i) - 0x01000101U)) &
2452               0x80008080U))
2453             continue;
2454         FIND_FIRST_ZERO;
2455         STARTCODE_TEST;
2456         i -= 3;
2457     }
2458 #endif /* HAVE_FAST_64BIT */
2459 #else
2460     for (i = 0; i + 1 < length; i += 2) {
2461         if (src[i])
2462             continue;
2463         if (i > 0 && src[i - 1] == 0)
2464             i--;
2465         STARTCODE_TEST;
2466     }
2467 #endif /* HAVE_FAST_UNALIGNED */
2468
2469     if (i >= length - 1) { // no escaped 0
2470         nal->data = src;
2471         nal->size = length;
2472         return length;
2473     }
2474
2475     av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2476                    length + FF_INPUT_BUFFER_PADDING_SIZE);
2477     if (!nal->rbsp_buffer)
2478         return AVERROR(ENOMEM);
2479
2480     dst = nal->rbsp_buffer;
2481
2482     memcpy(dst, src, i);
2483     si = di = i;
2484     while (si + 2 < length) {
2485         // remove escapes (very rare 1:2^22)
2486         if (src[si + 2] > 3) {
2487             dst[di++] = src[si++];
2488             dst[di++] = src[si++];
2489         } else if (src[si] == 0 && src[si + 1] == 0) {
2490             if (src[si + 2] == 3) { // escape
2491                 dst[di++] = 0;
2492                 dst[di++] = 0;
2493                 si       += 3;
2494
2495                 s->skipped_bytes++;
2496                 if (s->skipped_bytes_pos_size < s->skipped_bytes) {
2497                     s->skipped_bytes_pos_size *= 2;
2498                     av_reallocp_array(&s->skipped_bytes_pos,
2499                             s->skipped_bytes_pos_size,
2500                             sizeof(*s->skipped_bytes_pos));
2501                     if (!s->skipped_bytes_pos)
2502                         return AVERROR(ENOMEM);
2503                 }
2504                 if (s->skipped_bytes_pos)
2505                     s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
2506                 continue;
2507             } else // next start code
2508                 goto nsc;
2509         }
2510
2511         dst[di++] = src[si++];
2512     }
2513     while (si < length)
2514         dst[di++] = src[si++];
2515
2516 nsc:
2517     memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2518
2519     nal->data = dst;
2520     nal->size = di;
2521     return si;
2522 }
2523
2524 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2525 {
2526     int i, consumed, ret = 0;
2527
2528     s->ref = NULL;
2529     s->eos = 0;
2530
2531     /* split the input packet into NAL units, so we know the upper bound on the
2532      * number of slices in the frame */
2533     s->nb_nals = 0;
2534     while (length >= 4) {
2535         HEVCNAL *nal;
2536         int extract_length = 0;
2537
2538         if (s->is_nalff) {
2539             int i;
2540             for (i = 0; i < s->nal_length_size; i++)
2541                 extract_length = (extract_length << 8) | buf[i];
2542             buf    += s->nal_length_size;
2543             length -= s->nal_length_size;
2544
2545             if (extract_length > length) {
2546                 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2547                 ret = AVERROR_INVALIDDATA;
2548                 goto fail;
2549             }
2550         } else {
2551             /* search start code */
2552             while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2553                 ++buf;
2554                 --length;
2555                 if (length < 4) {
2556                     av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
2557                     ret = AVERROR_INVALIDDATA;
2558                     goto fail;
2559                 }
2560             }
2561
2562             buf           += 3;
2563             length        -= 3;
2564         }
2565
2566         if (!s->is_nalff)
2567             extract_length = length;
2568
2569         if (s->nals_allocated < s->nb_nals + 1) {
2570             int new_size = s->nals_allocated + 1;
2571             HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2572             if (!tmp) {
2573                 ret = AVERROR(ENOMEM);
2574                 goto fail;
2575             }
2576             s->nals = tmp;
2577             memset(s->nals + s->nals_allocated, 0,
2578                    (new_size - s->nals_allocated) * sizeof(*tmp));
2579             av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
2580             av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
2581             av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
2582             s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
2583             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));
2584             s->nals_allocated = new_size;
2585         }
2586         s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
2587         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
2588         nal = &s->nals[s->nb_nals];
2589
2590         consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
2591
2592         s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
2593         s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
2594         s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
2595
2596
2597         if (consumed < 0) {
2598             ret = consumed;
2599             goto fail;
2600         }
2601
2602         ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
2603         if (ret < 0)
2604             goto fail;
2605         hls_nal_unit(s);
2606
2607         if (s->nal_unit_type == NAL_EOB_NUT ||
2608             s->nal_unit_type == NAL_EOS_NUT)
2609             s->eos = 1;
2610
2611         buf    += consumed;
2612         length -= consumed;
2613     }
2614
2615     /* parse the NAL units */
2616     for (i = 0; i < s->nb_nals; i++) {
2617         int ret;
2618         s->skipped_bytes = s->skipped_bytes_nal[i];
2619         s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
2620
2621         ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2622         if (ret < 0) {
2623             av_log(s->avctx, AV_LOG_WARNING,
2624                    "Error parsing NAL unit #%d.\n", i);
2625             goto fail;
2626         }
2627     }
2628
2629 fail:
2630     if (s->ref && s->threads_type == FF_THREAD_FRAME)
2631         ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2632
2633     return ret;
2634 }
2635
2636 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2637 {
2638     int i;
2639     for (i = 0; i < 16; i++)
2640         av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2641 }
2642
2643 static int verify_md5(HEVCContext *s, AVFrame *frame)
2644 {
2645     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2646     int pixel_shift;
2647     int i, j;
2648
2649     if (!desc)
2650         return AVERROR(EINVAL);
2651
2652     pixel_shift = desc->comp[0].depth_minus1 > 7;
2653
2654     av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2655            s->poc);
2656
2657     /* the checksums are LE, so we have to byteswap for >8bpp formats
2658      * on BE arches */
2659 #if HAVE_BIGENDIAN
2660     if (pixel_shift && !s->checksum_buf) {
2661         av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2662                        FFMAX3(frame->linesize[0], frame->linesize[1],
2663                               frame->linesize[2]));
2664         if (!s->checksum_buf)
2665             return AVERROR(ENOMEM);
2666     }
2667 #endif
2668
2669     for (i = 0; frame->data[i]; i++) {
2670         int width  = s->avctx->coded_width;
2671         int height = s->avctx->coded_height;
2672         int w = (i == 1 || i == 2) ? (width  >> desc->log2_chroma_w) : width;
2673         int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2674         uint8_t md5[16];
2675
2676         av_md5_init(s->md5_ctx);
2677         for (j = 0; j < h; j++) {
2678             const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2679 #if HAVE_BIGENDIAN
2680             if (pixel_shift) {
2681                 s->dsp.bswap16_buf((uint16_t*)s->checksum_buf,
2682                                    (const uint16_t*)src, w);
2683                 src = s->checksum_buf;
2684             }
2685 #endif
2686             av_md5_update(s->md5_ctx, src, w << pixel_shift);
2687         }
2688         av_md5_final(s->md5_ctx, md5);
2689
2690         if (!memcmp(md5, s->md5[i], 16)) {
2691             av_log   (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2692             print_md5(s->avctx, AV_LOG_DEBUG, md5);
2693             av_log   (s->avctx, AV_LOG_DEBUG, "; ");
2694         } else {
2695             av_log   (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2696             print_md5(s->avctx, AV_LOG_ERROR, md5);
2697             av_log   (s->avctx, AV_LOG_ERROR, " != ");
2698             print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2699             av_log   (s->avctx, AV_LOG_ERROR, "\n");
2700             return AVERROR_INVALIDDATA;
2701         }
2702     }
2703
2704     av_log(s->avctx, AV_LOG_DEBUG, "\n");
2705
2706     return 0;
2707 }
2708
2709 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2710                              AVPacket *avpkt)
2711 {
2712     int ret;
2713     HEVCContext *s = avctx->priv_data;
2714
2715     if (!avpkt->size) {
2716         ret = ff_hevc_output_frame(s, data, 1);
2717         if (ret < 0)
2718             return ret;
2719
2720         *got_output = ret;
2721         return 0;
2722     }
2723
2724     s->ref = NULL;
2725     ret    = decode_nal_units(s, avpkt->data, avpkt->size);
2726     if (ret < 0)
2727         return ret;
2728
2729     /* verify the SEI checksum */
2730     if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2731         s->is_md5) {
2732         ret = verify_md5(s, s->ref->frame);
2733         if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2734             ff_hevc_unref_frame(s, s->ref, ~0);
2735             return ret;
2736         }
2737     }
2738     s->is_md5 = 0;
2739
2740     if (s->is_decoded) {
2741         av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2742         s->is_decoded = 0;
2743     }
2744
2745     if (s->output_frame->buf[0]) {
2746         av_frame_move_ref(data, s->output_frame);
2747         *got_output = 1;
2748     }
2749
2750     return avpkt->size;
2751 }
2752
2753 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2754 {
2755     int ret;
2756
2757     ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2758     if (ret < 0)
2759         return ret;
2760
2761     dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2762     if (!dst->tab_mvf_buf)
2763         goto fail;
2764     dst->tab_mvf = src->tab_mvf;
2765
2766     dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2767     if (!dst->rpl_tab_buf)
2768         goto fail;
2769     dst->rpl_tab = src->rpl_tab;
2770
2771     dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2772     if (!dst->rpl_buf)
2773         goto fail;
2774
2775     dst->poc        = src->poc;
2776     dst->ctb_count  = src->ctb_count;
2777     dst->window     = src->window;
2778     dst->flags      = src->flags;
2779     dst->sequence   = src->sequence;
2780
2781     return 0;
2782 fail:
2783     ff_hevc_unref_frame(s, dst, ~0);
2784     return AVERROR(ENOMEM);
2785 }
2786
2787 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2788 {
2789     HEVCContext       *s = avctx->priv_data;
2790     HEVCLocalContext *lc = s->HEVClc;
2791     int i;
2792
2793     pic_arrays_free(s);
2794
2795     av_freep(&s->md5_ctx);
2796
2797     for(i=0; i < s->nals_allocated; i++) {
2798         av_freep(&s->skipped_bytes_pos_nal[i]);
2799     }
2800     av_freep(&s->skipped_bytes_pos_size_nal);
2801     av_freep(&s->skipped_bytes_nal);
2802     av_freep(&s->skipped_bytes_pos_nal);
2803
2804     av_freep(&s->cabac_state);
2805
2806     av_frame_free(&s->tmp_frame);
2807     av_frame_free(&s->output_frame);
2808
2809     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2810         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2811         av_frame_free(&s->DPB[i].frame);
2812     }
2813
2814     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2815         av_buffer_unref(&s->vps_list[i]);
2816     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2817         av_buffer_unref(&s->sps_list[i]);
2818     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2819         av_buffer_unref(&s->pps_list[i]);
2820
2821     av_freep(&s->sh.entry_point_offset);
2822     av_freep(&s->sh.offset);
2823     av_freep(&s->sh.size);
2824
2825     for (i = 1; i < s->threads_number; i++) {
2826         lc = s->HEVClcList[i];
2827         if (lc) {
2828             av_freep(&s->HEVClcList[i]);
2829             av_freep(&s->sList[i]);
2830         }
2831     }
2832     if (s->HEVClc == s->HEVClcList[0])
2833         s->HEVClc = NULL;
2834     av_freep(&s->HEVClcList[0]);
2835
2836     for (i = 0; i < s->nals_allocated; i++)
2837         av_freep(&s->nals[i].rbsp_buffer);
2838     av_freep(&s->nals);
2839     s->nals_allocated = 0;
2840
2841     return 0;
2842 }
2843
2844 static av_cold int hevc_init_context(AVCodecContext *avctx)
2845 {
2846     HEVCContext *s = avctx->priv_data;
2847     int i;
2848
2849     s->avctx = avctx;
2850
2851     s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
2852     if (!s->HEVClc)
2853         goto fail;
2854     s->HEVClcList[0] = s->HEVClc;
2855     s->sList[0] = s;
2856
2857     s->cabac_state = av_malloc(HEVC_CONTEXTS);
2858     if (!s->cabac_state)
2859         goto fail;
2860
2861     s->tmp_frame = av_frame_alloc();
2862     if (!s->tmp_frame)
2863         goto fail;
2864
2865     s->output_frame = av_frame_alloc();
2866     if (!s->output_frame)
2867         goto fail;
2868
2869     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2870         s->DPB[i].frame = av_frame_alloc();
2871         if (!s->DPB[i].frame)
2872             goto fail;
2873         s->DPB[i].tf.f = s->DPB[i].frame;
2874     }
2875
2876     s->max_ra = INT_MAX;
2877
2878     s->md5_ctx = av_md5_alloc();
2879     if (!s->md5_ctx)
2880         goto fail;
2881
2882     ff_dsputil_init(&s->dsp, avctx);
2883
2884     s->context_initialized = 1;
2885
2886     return 0;
2887
2888 fail:
2889     hevc_decode_free(avctx);
2890     return AVERROR(ENOMEM);
2891 }
2892
2893 static int hevc_update_thread_context(AVCodecContext *dst,
2894                                       const AVCodecContext *src)
2895 {
2896     HEVCContext *s  = dst->priv_data;
2897     HEVCContext *s0 = src->priv_data;
2898     int i, ret;
2899
2900     if (!s->context_initialized) {
2901         ret = hevc_init_context(dst);
2902         if (ret < 0)
2903             return ret;
2904     }
2905
2906     for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2907         ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2908         if (s0->DPB[i].frame->buf[0]) {
2909             ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
2910             if (ret < 0)
2911                 return ret;
2912         }
2913     }
2914
2915     for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
2916         av_buffer_unref(&s->vps_list[i]);
2917         if (s0->vps_list[i]) {
2918             s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
2919             if (!s->vps_list[i])
2920                 return AVERROR(ENOMEM);
2921         }
2922     }
2923
2924     for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
2925         av_buffer_unref(&s->sps_list[i]);
2926         if (s0->sps_list[i]) {
2927             s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
2928             if (!s->sps_list[i])
2929                 return AVERROR(ENOMEM);
2930         }
2931     }
2932
2933     for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
2934         av_buffer_unref(&s->pps_list[i]);
2935         if (s0->pps_list[i]) {
2936             s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
2937             if (!s->pps_list[i])
2938                 return AVERROR(ENOMEM);
2939         }
2940     }
2941
2942     if (s->sps != s0->sps)
2943         ret = set_sps(s, s0->sps);
2944
2945     s->seq_decode = s0->seq_decode;
2946     s->seq_output = s0->seq_output;
2947     s->pocTid0    = s0->pocTid0;
2948     s->max_ra     = s0->max_ra;
2949
2950     s->is_nalff        = s0->is_nalff;
2951     s->nal_length_size = s0->nal_length_size;
2952
2953     s->threads_number      = s0->threads_number;
2954     s->threads_type        = s0->threads_type;
2955
2956     if (s0->eos) {
2957         s->seq_decode = (s->seq_decode + 1) & 0xff;
2958         s->max_ra = INT_MAX;
2959     }
2960
2961     return 0;
2962 }
2963
2964 static int hevc_decode_extradata(HEVCContext *s)
2965 {
2966     AVCodecContext *avctx = s->avctx;
2967     GetByteContext gb;
2968     int ret;
2969
2970     bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
2971
2972     if (avctx->extradata_size > 3 &&
2973         (avctx->extradata[0] || avctx->extradata[1] ||
2974          avctx->extradata[2] > 1)) {
2975         /* It seems the extradata is encoded as hvcC format.
2976          * Temporarily, we support configurationVersion==0 until 14496-15 3rd
2977          * is finalized. When finalized, configurationVersion will be 1 and we
2978          * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
2979         int i, j, num_arrays, nal_len_size;
2980
2981         s->is_nalff = 1;
2982
2983         bytestream2_skip(&gb, 21);
2984         nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
2985         num_arrays   = bytestream2_get_byte(&gb);
2986
2987         /* nal units in the hvcC always have length coded with 2 bytes,
2988          * so put a fake nal_length_size = 2 while parsing them */
2989         s->nal_length_size = 2;
2990
2991         /* Decode nal units from hvcC. */
2992         for (i = 0; i < num_arrays; i++) {
2993             int type = bytestream2_get_byte(&gb) & 0x3f;
2994             int cnt  = bytestream2_get_be16(&gb);
2995
2996             for (j = 0; j < cnt; j++) {
2997                 // +2 for the nal size field
2998                 int nalsize = bytestream2_peek_be16(&gb) + 2;
2999                 if (bytestream2_get_bytes_left(&gb) < nalsize) {
3000                     av_log(s->avctx, AV_LOG_ERROR,
3001                            "Invalid NAL unit size in extradata.\n");
3002                     return AVERROR_INVALIDDATA;
3003                 }
3004
3005                 ret = decode_nal_units(s, gb.buffer, nalsize);
3006                 if (ret < 0) {
3007                     av_log(avctx, AV_LOG_ERROR,
3008                            "Decoding nal unit %d %d from hvcC failed\n",
3009                            type, i);
3010                     return ret;
3011                 }
3012                 bytestream2_skip(&gb, nalsize);
3013             }
3014         }
3015
3016         /* Now store right nal length size, that will be used to parse
3017          * all other nals */
3018         s->nal_length_size = nal_len_size;
3019     } else {
3020         s->is_nalff = 0;
3021         ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
3022         if (ret < 0)
3023             return ret;
3024     }
3025     return 0;
3026 }
3027
3028 static av_cold int hevc_decode_init(AVCodecContext *avctx)
3029 {
3030     HEVCContext *s = avctx->priv_data;
3031     int ret;
3032
3033     ff_init_cabac_states();
3034
3035     avctx->internal->allocate_progress = 1;
3036
3037     ret = hevc_init_context(avctx);
3038     if (ret < 0)
3039         return ret;
3040
3041     s->enable_parallel_tiles = 0;
3042     s->picture_struct = 0;
3043
3044     if(avctx->active_thread_type & FF_THREAD_SLICE)
3045         s->threads_number = avctx->thread_count;
3046     else
3047         s->threads_number = 1;
3048
3049     if (avctx->extradata_size > 0 && avctx->extradata) {
3050         ret = hevc_decode_extradata(s);
3051         if (ret < 0) {
3052             hevc_decode_free(avctx);
3053             return ret;
3054         }
3055     }
3056
3057     if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
3058             s->threads_type = FF_THREAD_FRAME;
3059         else
3060             s->threads_type = FF_THREAD_SLICE;
3061
3062     return 0;
3063 }
3064
3065 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
3066 {
3067     HEVCContext *s = avctx->priv_data;
3068     int ret;
3069
3070     memset(s, 0, sizeof(*s));
3071
3072     ret = hevc_init_context(avctx);
3073     if (ret < 0)
3074         return ret;
3075
3076     return 0;
3077 }
3078
3079 static void hevc_decode_flush(AVCodecContext *avctx)
3080 {
3081     HEVCContext *s = avctx->priv_data;
3082     ff_hevc_flush_dpb(s);
3083     s->max_ra = INT_MAX;
3084 }
3085
3086 #define OFFSET(x) offsetof(HEVCContext, x)
3087 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
3088
3089 static const AVProfile profiles[] = {
3090     { FF_PROFILE_HEVC_MAIN,                 "Main"                },
3091     { FF_PROFILE_HEVC_MAIN_10,              "Main 10"             },
3092     { FF_PROFILE_HEVC_MAIN_STILL_PICTURE,   "Main Still Picture"  },
3093     { FF_PROFILE_UNKNOWN },
3094 };
3095
3096 static const AVOption options[] = {
3097     { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
3098         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3099     { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
3100         AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3101     { NULL },
3102 };
3103
3104 static const AVClass hevc_decoder_class = {
3105     .class_name = "HEVC decoder",
3106     .item_name  = av_default_item_name,
3107     .option     = options,
3108     .version    = LIBAVUTIL_VERSION_INT,
3109 };
3110
3111 AVCodec ff_hevc_decoder = {
3112     .name                  = "hevc",
3113     .long_name             = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3114     .type                  = AVMEDIA_TYPE_VIDEO,
3115     .id                    = AV_CODEC_ID_HEVC,
3116     .priv_data_size        = sizeof(HEVCContext),
3117     .priv_class            = &hevc_decoder_class,
3118     .init                  = hevc_decode_init,
3119     .close                 = hevc_decode_free,
3120     .decode                = hevc_decode_frame,
3121     .flush                 = hevc_decode_flush,
3122     .update_thread_context = hevc_update_thread_context,
3123     .init_thread_copy      = hevc_init_thread_copy,
3124     .capabilities          = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3125                              CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
3126     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
3127 };