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
9 * This file is part of FFmpeg.
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.
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.
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
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"
35 #include "bytestream.h"
36 #include "cabac_functions.h"
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 };
46 * NOTE: Each function hls_foo correspond to the function foo in the
47 * specification (HLS stands for High Level Syntax).
54 /* free everything allocated by pic_arrays_init() */
55 static void pic_arrays_free(HEVCContext *s)
58 av_freep(&s->deblock);
59 av_freep(&s->split_cu_flag);
61 av_freep(&s->skip_flag);
62 av_freep(&s->tab_ct_depth);
64 av_freep(&s->tab_ipm);
65 av_freep(&s->cbf_luma);
68 av_freep(&s->qp_y_tab);
69 av_freep(&s->tab_slice_address);
70 av_freep(&s->filter_slice_edges);
72 av_freep(&s->horizontal_bs);
73 av_freep(&s->vertical_bs);
75 av_freep(&s->sh.entry_point_offset);
76 av_freep(&s->sh.size);
77 av_freep(&s->sh.offset);
79 av_buffer_pool_uninit(&s->tab_mvf_pool);
80 av_buffer_pool_uninit(&s->rpl_tab_pool);
83 /* allocate arrays that depend on frame dimensions */
84 static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
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;
95 s->bs_width = width >> 3;
96 s->bs_height = height >> 3;
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)
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)
109 s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
110 s->tab_ipm = av_malloc(min_pu_size);
111 s->is_pcm = av_malloc(min_pu_size);
112 if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
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)
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)
128 s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
130 s->rpl_tab_pool = av_buffer_pool_init(ctb_count * sizeof(RefPicListTab),
132 if (!s->tab_mvf_pool || !s->rpl_tab_pool)
139 return AVERROR(ENOMEM);
142 static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
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];
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);
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;
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);
168 for (i = 0; i < s->sh.nb_refs[L0]; i++)
169 chroma_weight_l0_flag[i] = 0;
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);
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);
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;
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;
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);
204 for (i = 0; i < s->sh.nb_refs[L1]; i++)
205 chroma_weight_l1_flag[i] = 0;
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);
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);
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;
231 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
233 const HEVCSPS *sps = s->sps;
234 int max_poc_lsb = 1 << sps->log2_max_poc_lsb;
235 int prev_delta_msb = 0;
236 int nb_sps = 0, nb_sh;
240 if (!sps->long_term_ref_pics_present_flag)
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);
247 if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
248 return AVERROR_INVALIDDATA;
250 rps->nb_refs = nb_sh + nb_sps;
252 for (i = 0; i < rps->nb_refs; i++) {
253 uint8_t delta_poc_msb_present;
256 uint8_t lt_idx_sps = 0;
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));
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];
264 rps->poc[i] = get_bits(gb, sps->log2_max_poc_lsb);
265 rps->used[i] = get_bits1(gb);
268 delta_poc_msb_present = get_bits1(gb);
269 if (delta_poc_msb_present) {
270 int delta = get_ue_golomb_long(gb);
272 if (i && i != nb_sps)
273 delta += prev_delta_msb;
275 rps->poc[i] += s->poc - delta * max_poc_lsb - s->sh.pic_order_cnt_lsb;
276 prev_delta_msb = delta;
283 static int set_sps(HEVCContext *s, const HEVCSPS *sps)
286 int num = 0, den = 0;
289 ret = pic_arrays_init(s, sps);
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;
301 if (sps->vui.video_signal_type_present_flag)
302 s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
305 s->avctx->color_range = AVCOL_RANGE_MPEG;
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;
312 s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
313 s->avctx->color_trc = AVCOL_TRC_UNSPECIFIED;
314 s->avctx->colorspace = AVCOL_SPC_UNSPECIFIED;
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);
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);
326 s->frame = s->tmp_frame;
330 s->vps = (HEVCVPS*) s->vps_list[s->sps->vps_id]->data;
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;
340 if (num != 0 && den != 0)
341 av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
352 static int hls_slice_header(HEVCContext *s)
354 GetBitContext *gb = &s->HEVClc->gb;
355 SliceHeader *sh = &s->sh;
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;
364 ff_hevc_clear_refs(s);
366 if (s->nal_unit_type >= 16 && s->nal_unit_type <= 23)
367 sh->no_output_of_prior_pics_flag = get_bits1(gb);
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;
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;
379 s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
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);
388 s->seq_decode = (s->seq_decode + 1) & 0xff;
392 s->avctx->profile = s->sps->ptl.general_ptl.profile_idc;
393 s->avctx->level = s->sps->ptl.general_ptl.level_idc;
395 sh->dependent_slice_segment_flag = 0;
396 if (!sh->first_slice_in_pic_flag) {
397 int slice_address_length;
399 if (s->pps->dependent_slice_segments_enabled_flag)
400 sh->dependent_slice_segment_flag = get_bits1(gb);
402 slice_address_length = av_ceil_log2(s->sps->ctb_width *
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;
412 if (!sh->dependent_slice_segment_flag) {
413 sh->slice_addr = sh->slice_segment_addr;
417 sh->slice_segment_addr = sh->slice_addr = 0;
419 s->slice_initialized = 0;
422 if (!sh->dependent_slice_segment_flag) {
423 s->slice_initialized = 0;
425 for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
426 skip_bits(gb, 1); // slice_reserved_undetermined_flag[]
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",
434 return AVERROR_INVALIDDATA;
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;
441 if (s->pps->output_flag_present_flag)
442 sh->pic_output_flag = get_bits1(gb);
444 if (s->sps->separate_colour_plane_flag)
445 sh->colour_plane_id = get_bits(gb, 2);
448 int short_term_ref_pic_set_sps_flag, poc;
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;
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);
467 sh->short_term_rps = &sh->slice_rps;
469 int numbits, rps_idx;
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;
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];
481 ret = decode_lt_rps(s, &sh->long_term_rps, gb);
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;
488 if (s->sps->sps_temporal_mvp_enabled_flag)
489 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
491 sh->slice_temporal_mvp_enabled_flag = 0;
493 s->sh.short_term_rps = NULL;
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)
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);
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;
518 sh->nb_refs[L0] = sh->nb_refs[L1] = 0;
519 if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
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;
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;
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;
537 sh->rpl_modification_flag[0] = 0;
538 sh->rpl_modification_flag[1] = 0;
539 nb_refs = ff_hevc_frame_nb_refs(s);
541 av_log(s->avctx, AV_LOG_ERROR, "Zero refs for a frame with P or B slices.\n");
542 return AVERROR_INVALIDDATA;
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));
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));
560 if (sh->slice_type == B_SLICE)
561 sh->mvd_l1_zero_flag = get_bits1(gb);
563 if (s->pps->cabac_init_present_flag)
564 sh->cabac_init_flag = get_bits1(gb);
566 sh->cabac_init_flag = 0;
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);
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;
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);
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;
599 sh->slice_qp_delta = get_se_golomb(gb);
600 if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
601 sh->slice_cb_qp_offset = get_se_golomb(gb);
602 sh->slice_cr_qp_offset = get_se_golomb(gb);
604 sh->slice_cb_qp_offset = 0;
605 sh->slice_cr_qp_offset = 0;
608 if (s->pps->deblocking_filter_control_present_flag) {
609 int deblocking_filter_override_flag = 0;
611 if (s->pps->deblocking_filter_override_enabled_flag)
612 deblocking_filter_override_flag = get_bits1(gb);
614 if (deblocking_filter_override_flag) {
615 sh->disable_deblocking_filter_flag = get_bits1(gb);
616 if (!sh->disable_deblocking_filter_flag) {
617 sh->beta_offset = get_se_golomb(gb) * 2;
618 sh->tc_offset = get_se_golomb(gb) * 2;
621 sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
622 sh->beta_offset = s->pps->beta_offset;
623 sh->tc_offset = s->pps->tc_offset;
626 sh->disable_deblocking_filter_flag = 0;
631 if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
632 (sh->slice_sample_adaptive_offset_flag[0] ||
633 sh->slice_sample_adaptive_offset_flag[1] ||
634 !sh->disable_deblocking_filter_flag)) {
635 sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
637 sh->slice_loop_filter_across_slices_enabled_flag = s->pps->seq_loop_filter_across_slices_enabled_flag;
639 } else if (!s->slice_initialized) {
640 av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
641 return AVERROR_INVALIDDATA;
644 sh->num_entry_point_offsets = 0;
645 if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
646 sh->num_entry_point_offsets = get_ue_golomb_long(gb);
647 if (sh->num_entry_point_offsets > 0) {
648 int offset_len = get_ue_golomb_long(gb) + 1;
649 int segments = offset_len >> 4;
650 int rest = (offset_len & 15);
651 av_freep(&sh->entry_point_offset);
652 av_freep(&sh->offset);
654 sh->entry_point_offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
655 sh->offset = av_malloc(sh->num_entry_point_offsets * sizeof(int));
656 sh->size = av_malloc(sh->num_entry_point_offsets * sizeof(int));
657 for (i = 0; i < sh->num_entry_point_offsets; i++) {
659 for (j = 0; j < segments; j++) {
661 val += get_bits(gb, 16);
665 val += get_bits(gb, rest);
667 sh->entry_point_offset[i] = val + 1; // +1; // +1 to get the size
669 if (s->threads_number > 1 && (s->pps->num_tile_rows > 1 || s->pps->num_tile_columns > 1)) {
670 s->enable_parallel_tiles = 0; // TODO: you can enable tiles in parallel here
671 s->threads_number = 1;
673 s->enable_parallel_tiles = 0;
675 s->enable_parallel_tiles = 0;
678 if (s->pps->slice_header_extension_present_flag) {
679 int length = get_ue_golomb_long(gb);
680 for (i = 0; i < length; i++)
681 skip_bits(gb, 8); // slice_header_extension_data_byte
684 // Inferred parameters
685 sh->slice_qp = 26 + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
686 sh->slice_ctb_addr_rs = sh->slice_segment_addr;
688 s->HEVClc->first_qp_group = !s->sh.dependent_slice_segment_flag;
690 if (!s->pps->cu_qp_delta_enabled_flag)
691 s->HEVClc->qp_y = ((s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset) %
692 (52 + s->sps->qp_bd_offset)) - s->sps->qp_bd_offset;
694 s->slice_initialized = 1;
699 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
701 #define SET_SAO(elem, value) \
703 if (!sao_merge_up_flag && !sao_merge_left_flag) \
705 else if (sao_merge_left_flag) \
706 sao->elem = CTB(s->sao, rx-1, ry).elem; \
707 else if (sao_merge_up_flag) \
708 sao->elem = CTB(s->sao, rx, ry-1).elem; \
713 static void hls_sao_param(HEVCContext *s, int rx, int ry)
715 HEVCLocalContext *lc = s->HEVClc;
716 int sao_merge_left_flag = 0;
717 int sao_merge_up_flag = 0;
718 int shift = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
719 SAOParams *sao = &CTB(s->sao, rx, ry);
722 if (s->sh.slice_sample_adaptive_offset_flag[0] ||
723 s->sh.slice_sample_adaptive_offset_flag[1]) {
725 if (lc->ctb_left_flag)
726 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
728 if (ry > 0 && !sao_merge_left_flag) {
730 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
734 for (c_idx = 0; c_idx < 3; c_idx++) {
735 if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
736 sao->type_idx[c_idx] = SAO_NOT_APPLIED;
741 sao->type_idx[2] = sao->type_idx[1];
742 sao->eo_class[2] = sao->eo_class[1];
744 SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
747 if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
750 for (i = 0; i < 4; i++)
751 SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
753 if (sao->type_idx[c_idx] == SAO_BAND) {
754 for (i = 0; i < 4; i++) {
755 if (sao->offset_abs[c_idx][i]) {
756 SET_SAO(offset_sign[c_idx][i],
757 ff_hevc_sao_offset_sign_decode(s));
759 sao->offset_sign[c_idx][i] = 0;
762 SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
763 } else if (c_idx != 2) {
764 SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
767 // Inferred parameters
768 sao->offset_val[c_idx][0] = 0;
769 for (i = 0; i < 4; i++) {
770 sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
771 if (sao->type_idx[c_idx] == SAO_EDGE) {
773 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
774 } else if (sao->offset_sign[c_idx][i]) {
775 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
784 static void hls_transform_unit(HEVCContext *s, int x0, int y0,
785 int xBase, int yBase, int cb_xBase, int cb_yBase,
786 int log2_cb_size, int log2_trafo_size,
787 int trafo_depth, int blk_idx)
789 HEVCLocalContext *lc = s->HEVClc;
791 if (lc->cu.pred_mode == MODE_INTRA) {
792 int trafo_size = 1 << log2_trafo_size;
793 ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
795 s->hpc.intra_pred(s, x0, y0, log2_trafo_size, 0);
796 if (log2_trafo_size > 2) {
797 trafo_size = trafo_size << (s->sps->hshift[1] - 1);
798 ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
799 s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 1);
800 s->hpc.intra_pred(s, x0, y0, log2_trafo_size - 1, 2);
801 } else if (blk_idx == 3) {
802 trafo_size = trafo_size << s->sps->hshift[1];
803 ff_hevc_set_neighbour_available(s, xBase, yBase,
804 trafo_size, trafo_size);
805 s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 1);
806 s->hpc.intra_pred(s, xBase, yBase, log2_trafo_size, 2);
810 if (lc->tt.cbf_luma ||
811 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
812 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
813 int scan_idx = SCAN_DIAG;
814 int scan_idx_c = SCAN_DIAG;
816 if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
817 lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
818 if (lc->tu.cu_qp_delta != 0)
819 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
820 lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
821 lc->tu.is_cu_qp_delta_coded = 1;
822 ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
825 if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
826 if (lc->tu.cur_intra_pred_mode >= 6 &&
827 lc->tu.cur_intra_pred_mode <= 14) {
828 scan_idx = SCAN_VERT;
829 } else if (lc->tu.cur_intra_pred_mode >= 22 &&
830 lc->tu.cur_intra_pred_mode <= 30) {
831 scan_idx = SCAN_HORIZ;
834 if (lc->pu.intra_pred_mode_c >= 6 &&
835 lc->pu.intra_pred_mode_c <= 14) {
836 scan_idx_c = SCAN_VERT;
837 } else if (lc->pu.intra_pred_mode_c >= 22 &&
838 lc->pu.intra_pred_mode_c <= 30) {
839 scan_idx_c = SCAN_HORIZ;
844 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
845 if (log2_trafo_size > 2) {
846 if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0))
847 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
848 if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0))
849 ff_hevc_hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
850 } else if (blk_idx == 3) {
851 if (SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], xBase, yBase))
852 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
853 if (SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], xBase, yBase))
854 ff_hevc_hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
859 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
861 int cb_size = 1 << log2_cb_size;
862 int log2_min_pu_size = s->sps->log2_min_pu_size;
864 int min_pu_width = s->sps->min_pu_width;
865 int x_end = FFMIN(x0 + cb_size, s->sps->width);
866 int y_end = FFMIN(y0 + cb_size, s->sps->height);
869 for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
870 for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
871 s->is_pcm[i + j * min_pu_width] = 2;
874 static void hls_transform_tree(HEVCContext *s, int x0, int y0,
875 int xBase, int yBase, int cb_xBase, int cb_yBase,
876 int log2_cb_size, int log2_trafo_size,
877 int trafo_depth, int blk_idx)
879 HEVCLocalContext *lc = s->HEVClc;
880 uint8_t split_transform_flag;
882 if (trafo_depth > 0 && log2_trafo_size == 2) {
883 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
884 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase);
885 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
886 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase);
888 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
889 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) = 0;
892 if (lc->cu.intra_split_flag) {
893 if (trafo_depth == 1)
894 lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
896 lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
901 lc->tt.inter_split_flag = s->sps->max_transform_hierarchy_depth_inter == 0 &&
902 lc->cu.pred_mode == MODE_INTER &&
903 lc->cu.part_mode != PART_2Nx2N &&
906 if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
907 log2_trafo_size > s->sps->log2_min_tb_size &&
908 trafo_depth < lc->cu.max_trafo_depth &&
909 !(lc->cu.intra_split_flag && trafo_depth == 0)) {
910 split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
912 split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
913 (lc->cu.intra_split_flag && trafo_depth == 0) ||
914 lc->tt.inter_split_flag;
917 if (log2_trafo_size > 2) {
918 if (trafo_depth == 0 ||
919 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth - 1], xBase, yBase)) {
920 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) =
921 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
924 if (trafo_depth == 0 ||
925 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth - 1], xBase, yBase)) {
926 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0) =
927 ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
931 if (split_transform_flag) {
932 int x1 = x0 + ((1 << log2_trafo_size) >> 1);
933 int y1 = y0 + ((1 << log2_trafo_size) >> 1);
935 hls_transform_tree(s, x0, y0, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
936 log2_trafo_size - 1, trafo_depth + 1, 0);
937 hls_transform_tree(s, x1, y0, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
938 log2_trafo_size - 1, trafo_depth + 1, 1);
939 hls_transform_tree(s, x0, y1, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
940 log2_trafo_size - 1, trafo_depth + 1, 2);
941 hls_transform_tree(s, x1, y1, x0, y0, cb_xBase, cb_yBase, log2_cb_size,
942 log2_trafo_size - 1, trafo_depth + 1, 3);
944 int min_tu_size = 1 << s->sps->log2_min_tb_size;
945 int log2_min_tu_size = s->sps->log2_min_tb_size;
946 int min_tu_width = s->sps->min_tb_width;
948 if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
949 SAMPLE_CBF(lc->tt.cbf_cb[trafo_depth], x0, y0) ||
950 SAMPLE_CBF(lc->tt.cbf_cr[trafo_depth], x0, y0)) {
951 lc->tt.cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
954 hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
955 log2_cb_size, log2_trafo_size, trafo_depth, blk_idx);
957 // TODO: store cbf_luma somewhere else
958 if (lc->tt.cbf_luma) {
960 for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
961 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
962 int x_tu = (x0 + j) >> log2_min_tu_size;
963 int y_tu = (y0 + i) >> log2_min_tu_size;
964 s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
967 if (!s->sh.disable_deblocking_filter_flag) {
968 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size,
969 lc->slice_or_tiles_up_boundary,
970 lc->slice_or_tiles_left_boundary);
971 if (s->pps->transquant_bypass_enable_flag &&
972 lc->cu.cu_transquant_bypass_flag)
973 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
978 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
980 //TODO: non-4:2:0 support
981 HEVCLocalContext *lc = s->HEVClc;
983 int cb_size = 1 << log2_cb_size;
984 int stride0 = s->frame->linesize[0];
985 uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
986 int stride1 = s->frame->linesize[1];
987 uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
988 int stride2 = s->frame->linesize[2];
989 uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
991 int length = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
992 const uint8_t *pcm = skip_bytes(&s->HEVClc->cc, (length + 7) >> 3);
995 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
996 lc->slice_or_tiles_up_boundary,
997 lc->slice_or_tiles_left_boundary);
999 ret = init_get_bits(&gb, pcm, length);
1003 s->hevcdsp.put_pcm(dst0, stride0, cb_size, &gb, s->sps->pcm.bit_depth);
1004 s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1005 s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1010 * 8.5.3.2.2.1 Luma sample interpolation process
1012 * @param s HEVC decoding context
1013 * @param dst target buffer for block data at block position
1014 * @param dststride stride of the dst buffer
1015 * @param ref reference picture buffer at origin (0, 0)
1016 * @param mv motion vector (relative to block position) to get pixel data from
1017 * @param x_off horizontal position of block from origin (0, 0)
1018 * @param y_off vertical position of block from origin (0, 0)
1019 * @param block_w width of block
1020 * @param block_h height of block
1022 static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
1023 AVFrame *ref, const Mv *mv, int x_off, int y_off,
1024 int block_w, int block_h)
1026 HEVCLocalContext *lc = s->HEVClc;
1027 uint8_t *src = ref->data[0];
1028 ptrdiff_t srcstride = ref->linesize[0];
1029 int pic_width = s->sps->width;
1030 int pic_height = s->sps->height;
1034 int extra_left = ff_hevc_qpel_extra_before[mx];
1035 int extra_top = ff_hevc_qpel_extra_before[my];
1037 x_off += mv->x >> 2;
1038 y_off += mv->y >> 2;
1039 src += y_off * srcstride + (x_off << s->sps->pixel_shift);
1041 if (x_off < extra_left || y_off < extra_top ||
1042 x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
1043 y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
1044 int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
1046 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1047 srcstride, srcstride,
1048 block_w + ff_hevc_qpel_extra[mx],
1049 block_h + ff_hevc_qpel_extra[my],
1050 x_off - extra_left, y_off - extra_top,
1051 pic_width, pic_height);
1052 src = lc->edge_emu_buffer + offset;
1054 s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
1055 block_h, lc->mc_buffer);
1059 * 8.5.3.2.2.2 Chroma sample interpolation process
1061 * @param s HEVC decoding context
1062 * @param dst1 target buffer for block data at block position (U plane)
1063 * @param dst2 target buffer for block data at block position (V plane)
1064 * @param dststride stride of the dst1 and dst2 buffers
1065 * @param ref reference picture buffer at origin (0, 0)
1066 * @param mv motion vector (relative to block position) to get pixel data from
1067 * @param x_off horizontal position of block from origin (0, 0)
1068 * @param y_off vertical position of block from origin (0, 0)
1069 * @param block_w width of block
1070 * @param block_h height of block
1072 static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
1073 ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
1074 int x_off, int y_off, int block_w, int block_h)
1076 HEVCLocalContext *lc = s->HEVClc;
1077 uint8_t *src1 = ref->data[1];
1078 uint8_t *src2 = ref->data[2];
1079 ptrdiff_t src1stride = ref->linesize[1];
1080 ptrdiff_t src2stride = ref->linesize[2];
1081 int pic_width = s->sps->width >> 1;
1082 int pic_height = s->sps->height >> 1;
1087 x_off += mv->x >> 3;
1088 y_off += mv->y >> 3;
1089 src1 += y_off * src1stride + (x_off << s->sps->pixel_shift);
1090 src2 += y_off * src2stride + (x_off << s->sps->pixel_shift);
1092 if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1093 x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1094 y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1095 int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1096 int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1098 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1099 src1stride, src1stride,
1100 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1101 x_off - EPEL_EXTRA_BEFORE,
1102 y_off - EPEL_EXTRA_BEFORE,
1103 pic_width, pic_height);
1105 src1 = lc->edge_emu_buffer + offset1;
1106 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1107 block_w, block_h, mx, my, lc->mc_buffer);
1109 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
1110 src2stride, src2stride,
1111 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1112 x_off - EPEL_EXTRA_BEFORE,
1113 y_off - EPEL_EXTRA_BEFORE,
1114 pic_width, pic_height);
1115 src2 = lc->edge_emu_buffer + offset2;
1116 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1117 block_w, block_h, mx, my,
1120 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1121 block_w, block_h, mx, my,
1123 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1124 block_w, block_h, mx, my,
1129 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1130 const Mv *mv, int y0, int height)
1132 int y = (mv->y >> 2) + y0 + height + 9;
1134 if (s->threads_type == FF_THREAD_FRAME )
1135 ff_thread_await_progress(&ref->tf, y, 0);
1138 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1140 int log2_cb_size, int partIdx)
1142 #define POS(c_idx, x, y) \
1143 &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1144 (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1145 HEVCLocalContext *lc = s->HEVClc;
1147 struct MvField current_mv = {{{ 0 }}};
1149 int min_pu_width = s->sps->min_pu_width;
1151 MvField *tab_mvf = s->ref->tab_mvf;
1152 RefPicList *refPicList = s->ref->refPicList;
1153 HEVCFrame *ref0, *ref1;
1155 int tmpstride = MAX_PB_SIZE;
1157 uint8_t *dst0 = POS(0, x0, y0);
1158 uint8_t *dst1 = POS(1, x0, y0);
1159 uint8_t *dst2 = POS(2, x0, y0);
1160 int log2_min_cb_size = s->sps->log2_min_cb_size;
1161 int min_cb_width = s->sps->min_cb_width;
1162 int x_cb = x0 >> log2_min_cb_size;
1163 int y_cb = y0 >> log2_min_cb_size;
1169 if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1170 if (s->sh.max_num_merge_cand > 1)
1171 merge_idx = ff_hevc_merge_idx_decode(s);
1175 ff_hevc_luma_mv_merge_mode(s, x0, y0,
1178 log2_cb_size, partIdx,
1179 merge_idx, ¤t_mv);
1180 x_pu = x0 >> s->sps->log2_min_pu_size;
1181 y_pu = y0 >> s->sps->log2_min_pu_size;
1183 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1184 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1185 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1186 } else { /* MODE_INTER */
1187 lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1188 if (lc->pu.merge_flag) {
1189 if (s->sh.max_num_merge_cand > 1)
1190 merge_idx = ff_hevc_merge_idx_decode(s);
1194 ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1195 partIdx, merge_idx, ¤t_mv);
1196 x_pu = x0 >> s->sps->log2_min_pu_size;
1197 y_pu = y0 >> s->sps->log2_min_pu_size;
1199 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1200 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1201 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1203 enum InterPredIdc inter_pred_idc = PRED_L0;
1204 ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1205 if (s->sh.slice_type == B_SLICE)
1206 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1208 if (inter_pred_idc != PRED_L1) {
1209 if (s->sh.nb_refs[L0]) {
1210 ref_idx[0] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1211 current_mv.ref_idx[0] = ref_idx[0];
1213 current_mv.pred_flag[0] = 1;
1214 ff_hevc_hls_mvd_coding(s, x0, y0, 0);
1215 mvp_flag[0] = ff_hevc_mvp_lx_flag_decode(s);
1216 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1217 partIdx, merge_idx, ¤t_mv,
1219 current_mv.mv[0].x += lc->pu.mvd.x;
1220 current_mv.mv[0].y += lc->pu.mvd.y;
1223 if (inter_pred_idc != PRED_L0) {
1224 if (s->sh.nb_refs[L1]) {
1225 ref_idx[1] = ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1226 current_mv.ref_idx[1] = ref_idx[1];
1229 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1233 ff_hevc_hls_mvd_coding(s, x0, y0, 1);
1236 current_mv.pred_flag[1] = 1;
1237 mvp_flag[1] = ff_hevc_mvp_lx_flag_decode(s);
1238 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1239 partIdx, merge_idx, ¤t_mv,
1241 current_mv.mv[1].x += lc->pu.mvd.x;
1242 current_mv.mv[1].y += lc->pu.mvd.y;
1245 x_pu = x0 >> s->sps->log2_min_pu_size;
1246 y_pu = y0 >> s->sps->log2_min_pu_size;
1248 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1249 for(j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1250 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1254 if (current_mv.pred_flag[0]) {
1255 ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1258 hevc_await_progress(s, ref0, ¤t_mv.mv[0], y0, nPbH);
1260 if (current_mv.pred_flag[1]) {
1261 ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1264 hevc_await_progress(s, ref1, ¤t_mv.mv[1], y0, nPbH);
1267 if (current_mv.pred_flag[0] && !current_mv.pred_flag[1]) {
1268 DECLARE_ALIGNED(16, int16_t, tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1269 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1271 luma_mc(s, tmp, tmpstride, ref0->frame,
1272 ¤t_mv.mv[0], x0, y0, nPbW, nPbH);
1274 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1275 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1276 s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1277 s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1278 s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1279 dst0, s->frame->linesize[0], tmp,
1280 tmpstride, nPbW, nPbH);
1282 s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1284 chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1285 ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1287 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1288 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1289 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1290 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1291 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1292 dst1, s->frame->linesize[1], tmp, tmpstride,
1293 nPbW / 2, nPbH / 2);
1294 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1295 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1296 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1297 dst2, s->frame->linesize[2], tmp2, tmpstride,
1298 nPbW / 2, nPbH / 2);
1300 s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1301 s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1303 } else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1304 DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1305 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1310 luma_mc(s, tmp, tmpstride, ref1->frame,
1311 ¤t_mv.mv[1], x0, y0, nPbW, nPbH);
1313 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1314 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1315 s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1316 s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1317 s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1318 dst0, s->frame->linesize[0], tmp, tmpstride,
1321 s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1324 chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
1325 ¤t_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
1327 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1328 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1329 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1330 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1331 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1332 dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1333 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1334 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1335 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1336 dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1338 s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1339 s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1341 } else if (current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1342 DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1343 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1344 DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
1345 DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
1346 HEVCFrame *ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1347 HEVCFrame *ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1352 luma_mc(s, tmp, tmpstride, ref0->frame,
1353 ¤t_mv.mv[0], x0, y0, nPbW, nPbH);
1354 luma_mc(s, tmp2, tmpstride, ref1->frame,
1355 ¤t_mv.mv[1], x0, y0, nPbW, nPbH);
1357 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1358 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1359 s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
1360 s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1361 s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1362 s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1363 s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1364 dst0, s->frame->linesize[0],
1365 tmp, tmp2, tmpstride, nPbW, nPbH);
1367 s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
1368 tmp, tmp2, tmpstride, nPbW, nPbH);
1371 chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1372 ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1373 chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
1374 ¤t_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1376 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1377 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1378 s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1379 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1380 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1381 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1382 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1383 dst1, s->frame->linesize[1], tmp, tmp3,
1384 tmpstride, nPbW / 2, nPbH / 2);
1385 s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1386 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1387 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1388 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1389 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1390 dst2, s->frame->linesize[2], tmp2, tmp4,
1391 tmpstride, nPbW / 2, nPbH / 2);
1393 s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
1394 s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
1402 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1403 int prev_intra_luma_pred_flag)
1405 HEVCLocalContext *lc = s->HEVClc;
1406 int x_pu = x0 >> s->sps->log2_min_pu_size;
1407 int y_pu = y0 >> s->sps->log2_min_pu_size;
1408 int min_pu_width = s->sps->min_pu_width;
1409 int size_in_pus = pu_size >> s->sps->log2_min_pu_size;
1410 int x0b = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1411 int y0b = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1413 int cand_up = (lc->ctb_up_flag || y0b) ?
1414 s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1415 int cand_left = (lc->ctb_left_flag || x0b) ?
1416 s->tab_ipm[y_pu * min_pu_width + x_pu - 1] : INTRA_DC;
1418 int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1420 MvField *tab_mvf = s->ref->tab_mvf;
1421 int intra_pred_mode;
1425 // intra_pred_mode prediction does not cross vertical CTB boundaries
1426 if ((y0 - 1) < y_ctb)
1429 if (cand_left == cand_up) {
1430 if (cand_left < 2) {
1431 candidate[0] = INTRA_PLANAR;
1432 candidate[1] = INTRA_DC;
1433 candidate[2] = INTRA_ANGULAR_26;
1435 candidate[0] = cand_left;
1436 candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1437 candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1440 candidate[0] = cand_left;
1441 candidate[1] = cand_up;
1442 if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1443 candidate[2] = INTRA_PLANAR;
1444 } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1445 candidate[2] = INTRA_DC;
1447 candidate[2] = INTRA_ANGULAR_26;
1451 if (prev_intra_luma_pred_flag) {
1452 intra_pred_mode = candidate[lc->pu.mpm_idx];
1454 if (candidate[0] > candidate[1])
1455 FFSWAP(uint8_t, candidate[0], candidate[1]);
1456 if (candidate[0] > candidate[2])
1457 FFSWAP(uint8_t, candidate[0], candidate[2]);
1458 if (candidate[1] > candidate[2])
1459 FFSWAP(uint8_t, candidate[1], candidate[2]);
1461 intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1462 for (i = 0; i < 3; i++)
1463 if (intra_pred_mode >= candidate[i])
1467 /* write the intra prediction units into the mv array */
1470 for (i = 0; i < size_in_pus; i++) {
1471 memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1472 intra_pred_mode, size_in_pus);
1474 for (j = 0; j < size_in_pus; j++) {
1475 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].is_intra = 1;
1476 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[0] = 0;
1477 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[1] = 0;
1478 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[0] = 0;
1479 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[1] = 0;
1480 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].x = 0;
1481 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].y = 0;
1482 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].x = 0;
1483 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].y = 0;
1487 return intra_pred_mode;
1490 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1491 int log2_cb_size, int ct_depth)
1493 int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1494 int x_cb = x0 >> s->sps->log2_min_cb_size;
1495 int y_cb = y0 >> s->sps->log2_min_cb_size;
1498 for (y = 0; y < length; y++)
1499 memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1503 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1506 HEVCLocalContext *lc = s->HEVClc;
1507 static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1508 uint8_t prev_intra_luma_pred_flag[4];
1509 int split = lc->cu.part_mode == PART_NxN;
1510 int pb_size = (1 << log2_cb_size) >> split;
1511 int side = split + 1;
1515 for (i = 0; i < side; i++)
1516 for (j = 0; j < side; j++)
1517 prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1519 for (i = 0; i < side; i++) {
1520 for (j = 0; j < side; j++) {
1521 if (prev_intra_luma_pred_flag[2 * i + j])
1522 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1524 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1526 lc->pu.intra_pred_mode[2 * i + j] =
1527 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1528 prev_intra_luma_pred_flag[2 * i + j]);
1532 chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1533 if (chroma_mode != 4) {
1534 if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1535 lc->pu.intra_pred_mode_c = 34;
1537 lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1539 lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1543 static void intra_prediction_unit_default_value(HEVCContext *s,
1547 HEVCLocalContext *lc = s->HEVClc;
1548 int pb_size = 1 << log2_cb_size;
1549 int size_in_pus = pb_size >> s->sps->log2_min_pu_size;
1550 int min_pu_width = s->sps->min_pu_width;
1551 MvField *tab_mvf = s->ref->tab_mvf;
1552 int x_pu = x0 >> s->sps->log2_min_pu_size;
1553 int y_pu = y0 >> s->sps->log2_min_pu_size;
1556 if (size_in_pus == 0)
1558 for (j = 0; j < size_in_pus; j++) {
1559 memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
1560 for (k = 0; k < size_in_pus; k++)
1561 tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].is_intra = lc->cu.pred_mode == MODE_INTRA;
1565 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
1567 int cb_size = 1 << log2_cb_size;
1568 HEVCLocalContext *lc = s->HEVClc;
1569 int log2_min_cb_size = s->sps->log2_min_cb_size;
1570 int length = cb_size >> log2_min_cb_size;
1571 int min_cb_width = s->sps->min_cb_width;
1572 int x_cb = x0 >> log2_min_cb_size;
1573 int y_cb = y0 >> log2_min_cb_size;
1578 lc->cu.rqt_root_cbf = 1;
1579 lc->cu.pred_mode = MODE_INTRA;
1580 lc->cu.part_mode = PART_2Nx2N;
1581 lc->cu.intra_split_flag = 0;
1582 lc->cu.pcm_flag = 0;
1584 SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
1585 for (x = 0; x < 4; x++)
1586 lc->pu.intra_pred_mode[x] = 1;
1587 if (s->pps->transquant_bypass_enable_flag) {
1588 lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
1589 if (lc->cu.cu_transquant_bypass_flag)
1590 set_deblocking_bypass(s, x0, y0, log2_cb_size);
1592 lc->cu.cu_transquant_bypass_flag = 0;
1594 if (s->sh.slice_type != I_SLICE) {
1595 uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
1597 lc->cu.pred_mode = MODE_SKIP;
1598 x = y_cb * min_cb_width + x_cb;
1599 for (y = 0; y < length; y++) {
1600 memset(&s->skip_flag[x], skip_flag, length);
1603 lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
1606 if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
1607 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1608 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1610 if (!s->sh.disable_deblocking_filter_flag)
1611 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1612 lc->slice_or_tiles_up_boundary,
1613 lc->slice_or_tiles_left_boundary);
1615 if (s->sh.slice_type != I_SLICE)
1616 lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
1617 if (lc->cu.pred_mode != MODE_INTRA ||
1618 log2_cb_size == s->sps->log2_min_cb_size) {
1619 lc->cu.part_mode = ff_hevc_part_mode_decode(s, log2_cb_size);
1620 lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
1621 lc->cu.pred_mode == MODE_INTRA;
1624 if (lc->cu.pred_mode == MODE_INTRA) {
1625 if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
1626 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
1627 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
1628 lc->cu.pcm_flag = ff_hevc_pcm_flag_decode(s);
1630 if (lc->cu.pcm_flag) {
1632 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1633 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
1634 if (s->sps->pcm.loop_filter_disable_flag)
1635 set_deblocking_bypass(s, x0, y0, log2_cb_size);
1640 intra_prediction_unit(s, x0, y0, log2_cb_size);
1643 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
1644 switch (lc->cu.part_mode) {
1646 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
1649 hls_prediction_unit(s, x0, y0, cb_size, cb_size / 2, log2_cb_size, 0);
1650 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
1653 hls_prediction_unit(s, x0, y0, cb_size / 2, cb_size, log2_cb_size, 0);
1654 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
1657 hls_prediction_unit(s, x0, y0, cb_size, cb_size / 4, log2_cb_size, 0);
1658 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
1661 hls_prediction_unit(s, x0, y0, cb_size, cb_size * 3 / 4, log2_cb_size, 0);
1662 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size / 4, log2_cb_size, 1);
1665 hls_prediction_unit(s, x0, y0, cb_size / 4, cb_size, log2_cb_size, 0);
1666 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
1669 hls_prediction_unit(s, x0, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
1670 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size / 4, cb_size, log2_cb_size, 1);
1673 hls_prediction_unit(s, x0, y0, cb_size / 2, cb_size / 2, log2_cb_size, 0);
1674 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size / 2, log2_cb_size, 1);
1675 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
1676 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
1681 if (!lc->cu.pcm_flag) {
1682 if (lc->cu.pred_mode != MODE_INTRA &&
1683 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
1684 lc->cu.rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
1686 if (lc->cu.rqt_root_cbf) {
1687 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
1688 s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
1689 s->sps->max_transform_hierarchy_depth_inter;
1690 hls_transform_tree(s, x0, y0, x0, y0, x0, y0, log2_cb_size,
1691 log2_cb_size, 0, 0);
1693 if (!s->sh.disable_deblocking_filter_flag)
1694 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size,
1695 lc->slice_or_tiles_up_boundary,
1696 lc->slice_or_tiles_left_boundary);
1701 if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
1702 ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
1704 x = y_cb * min_cb_width + x_cb;
1705 for (y = 0; y < length; y++) {
1706 memset(&s->qp_y_tab[x], lc->qp_y, length);
1710 set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
1715 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
1716 int log2_cb_size, int cb_depth)
1718 HEVCLocalContext *lc = s->HEVClc;
1719 const int cb_size = 1 << log2_cb_size;
1722 lc->ct.depth = cb_depth;
1723 if (x0 + cb_size <= s->sps->width &&
1724 y0 + cb_size <= s->sps->height &&
1725 log2_cb_size > s->sps->log2_min_cb_size) {
1726 SAMPLE(s->split_cu_flag, x0, y0) =
1727 ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
1729 SAMPLE(s->split_cu_flag, x0, y0) =
1730 (log2_cb_size > s->sps->log2_min_cb_size);
1732 if (s->pps->cu_qp_delta_enabled_flag &&
1733 log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
1734 lc->tu.is_cu_qp_delta_coded = 0;
1735 lc->tu.cu_qp_delta = 0;
1738 if (SAMPLE(s->split_cu_flag, x0, y0)) {
1739 const int cb_size_split = cb_size >> 1;
1740 const int x1 = x0 + cb_size_split;
1741 const int y1 = y0 + cb_size_split;
1745 more_data = hls_coding_quadtree(s, x0, y0, log2_cb_size - 1, cb_depth + 1);
1749 if (more_data && x1 < s->sps->width)
1750 more_data = hls_coding_quadtree(s, x1, y0, log2_cb_size - 1, cb_depth + 1);
1751 if (more_data && y1 < s->sps->height)
1752 more_data = hls_coding_quadtree(s, x0, y1, log2_cb_size - 1, cb_depth + 1);
1753 if (more_data && x1 < s->sps->width &&
1754 y1 < s->sps->height) {
1755 return hls_coding_quadtree(s, x1, y1, log2_cb_size - 1, cb_depth + 1);
1758 return ((x1 + cb_size_split) < s->sps->width ||
1759 (y1 + cb_size_split) < s->sps->height);
1763 ret = hls_coding_unit(s, x0, y0, log2_cb_size);
1766 if ((!((x0 + cb_size) %
1767 (1 << (s->sps->log2_ctb_size))) ||
1768 (x0 + cb_size >= s->sps->width)) &&
1770 (1 << (s->sps->log2_ctb_size))) ||
1771 (y0 + cb_size >= s->sps->height))) {
1772 int end_of_slice_flag = ff_hevc_end_of_slice_flag_decode(s);
1773 return !end_of_slice_flag;
1782 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
1785 HEVCLocalContext *lc = s->HEVClc;
1786 int ctb_size = 1 << s->sps->log2_ctb_size;
1787 int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1788 int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
1790 int tile_left_boundary, tile_up_boundary;
1791 int slice_left_boundary, slice_up_boundary;
1793 s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
1795 if (s->pps->entropy_coding_sync_enabled_flag) {
1796 if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
1797 lc->first_qp_group = 1;
1798 lc->end_of_tiles_x = s->sps->width;
1799 } else if (s->pps->tiles_enabled_flag) {
1800 if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
1801 int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
1802 lc->start_of_tiles_x = x_ctb;
1803 lc->end_of_tiles_x = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
1804 lc->first_qp_group = 1;
1807 lc->end_of_tiles_x = s->sps->width;
1810 lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
1812 if (s->pps->tiles_enabled_flag) {
1813 tile_left_boundary = x_ctb > 0 &&
1814 s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
1815 slice_left_boundary = x_ctb > 0 &&
1816 s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1];
1817 tile_up_boundary = y_ctb > 0 &&
1818 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]];
1819 slice_up_boundary = y_ctb > 0 &&
1820 s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width];
1822 tile_left_boundary =
1823 tile_up_boundary = 0;
1824 slice_left_boundary = ctb_addr_in_slice <= 0;
1825 slice_up_boundary = ctb_addr_in_slice < s->sps->ctb_width;
1827 lc->slice_or_tiles_left_boundary = slice_left_boundary + (tile_left_boundary << 1);
1828 lc->slice_or_tiles_up_boundary = slice_up_boundary + (tile_up_boundary << 1);
1829 lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && !tile_left_boundary);
1830 lc->ctb_up_flag = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !tile_up_boundary);
1831 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]]));
1832 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]]));
1835 static int hls_decode_entry(AVCodecContext *avctxt, void *isFilterThread)
1837 HEVCContext *s = avctxt->priv_data;
1838 int ctb_size = 1 << s->sps->log2_ctb_size;
1842 int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
1844 while (more_data && ctb_addr_ts < s->sps->ctb_size) {
1845 int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1847 x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1848 y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
1849 hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
1851 ff_hevc_cabac_init(s, ctb_addr_ts);
1853 hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
1855 s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
1856 s->deblock[ctb_addr_rs].tc_offset = s->sh.tc_offset;
1857 s->filter_slice_edges[ctb_addr_rs] = s->sh.slice_loop_filter_across_slices_enabled_flag;
1859 more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
1864 ff_hevc_save_states(s, ctb_addr_ts);
1865 ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
1868 if (x_ctb + ctb_size >= s->sps->width &&
1869 y_ctb + ctb_size >= s->sps->height)
1870 ff_hevc_hls_filter(s, x_ctb, y_ctb);
1875 static int hls_slice_data(HEVCContext *s)
1883 s->avctx->execute(s->avctx, hls_decode_entry, arg, ret , 1, sizeof(int));
1886 static int hls_decode_entry_wpp(AVCodecContext *avctxt, void *input_ctb_row, int job, int self_id)
1888 HEVCContext *s1 = avctxt->priv_data, *s;
1889 HEVCLocalContext *lc;
1890 int ctb_size = 1<< s1->sps->log2_ctb_size;
1892 int *ctb_row_p = input_ctb_row;
1893 int ctb_row = ctb_row_p[job];
1894 int ctb_addr_rs = s1->sh.slice_ctb_addr_rs + ctb_row * ((s1->sps->width + ctb_size - 1) >> s1->sps->log2_ctb_size);
1895 int ctb_addr_ts = s1->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
1896 int thread = ctb_row % s1->threads_number;
1899 s = s1->sList[self_id];
1903 ret = init_get_bits8(&lc->gb, s->data + s->sh.offset[ctb_row - 1], s->sh.size[ctb_row - 1]);
1907 ff_init_cabac_decoder(&lc->cc, s->data + s->sh.offset[(ctb_row)-1], s->sh.size[ctb_row - 1]);
1910 while(more_data && ctb_addr_ts < s->sps->ctb_size) {
1911 int x_ctb = (ctb_addr_rs % s->sps->ctb_width) << s->sps->log2_ctb_size;
1912 int y_ctb = (ctb_addr_rs / s->sps->ctb_width) << s->sps->log2_ctb_size;
1914 hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
1916 ff_thread_await_progress2(s->avctx, ctb_row, thread, SHIFT_CTB_WPP);
1918 if (avpriv_atomic_int_get(&s1->wpp_err)){
1919 ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
1923 ff_hevc_cabac_init(s, ctb_addr_ts);
1924 hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
1925 more_data = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
1932 ff_hevc_save_states(s, ctb_addr_ts);
1933 ff_thread_report_progress2(s->avctx, ctb_row, thread, 1);
1934 ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
1936 if (!more_data && (x_ctb+ctb_size) < s->sps->width && ctb_row != s->sh.num_entry_point_offsets) {
1937 avpriv_atomic_int_set(&s1->wpp_err, 1);
1938 ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
1942 if ((x_ctb+ctb_size) >= s->sps->width && (y_ctb+ctb_size) >= s->sps->height ) {
1943 ff_hevc_hls_filter(s, x_ctb, y_ctb);
1944 ff_thread_report_progress2(s->avctx, ctb_row , thread, SHIFT_CTB_WPP);
1947 ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
1950 if(x_ctb >= s->sps->width) {
1954 ff_thread_report_progress2(s->avctx, ctb_row ,thread, SHIFT_CTB_WPP);
1959 static int hls_slice_data_wpp(HEVCContext *s, const uint8_t *nal, int length)
1961 HEVCLocalContext *lc = s->HEVClc;
1962 int *ret = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
1963 int *arg = av_malloc((s->sh.num_entry_point_offsets + 1) * sizeof(int));
1965 int startheader, cmpt = 0;
1970 ff_alloc_entries(s->avctx, s->sh.num_entry_point_offsets + 1);
1973 for (i = 1; i < s->threads_number; i++) {
1974 s->sList[i] = av_malloc(sizeof(HEVCContext));
1975 memcpy(s->sList[i], s, sizeof(HEVCContext));
1976 s->HEVClcList[i] = av_malloc(sizeof(HEVCLocalContext));
1977 s->HEVClcList[i]->edge_emu_buffer = av_malloc((MAX_PB_SIZE + 7) * s->frame->linesize[0]);
1978 s->sList[i]->HEVClc = s->HEVClcList[i];
1982 offset = (lc->gb.index >> 3);
1984 for (j = 0, cmpt = 0, startheader = offset + s->sh.entry_point_offset[0]; j < s->skipped_bytes; j++) {
1985 if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
1991 for (i = 1; i < s->sh.num_entry_point_offsets; i++) {
1992 offset += (s->sh.entry_point_offset[i - 1] - cmpt);
1993 for (j = 0, cmpt = 0, startheader = offset
1994 + s->sh.entry_point_offset[i]; j < s->skipped_bytes; j++) {
1995 if (s->skipped_bytes_pos[j] >= offset && s->skipped_bytes_pos[j] < startheader) {
2000 s->sh.size[i - 1] = s->sh.entry_point_offset[i] - cmpt;
2001 s->sh.offset[i - 1] = offset;
2004 if (s->sh.num_entry_point_offsets != 0) {
2005 offset += s->sh.entry_point_offset[s->sh.num_entry_point_offsets - 1] - cmpt;
2006 s->sh.size[s->sh.num_entry_point_offsets - 1] = length - offset;
2007 s->sh.offset[s->sh.num_entry_point_offsets - 1] = offset;
2012 for (i = 1; i < s->threads_number; i++) {
2013 s->sList[i]->HEVClc->first_qp_group = 1;
2014 s->sList[i]->HEVClc->qp_y = s->sList[0]->HEVClc->qp_y;
2015 memcpy(s->sList[i], s, sizeof(HEVCContext));
2016 s->sList[i]->HEVClc = s->HEVClcList[i];
2019 avpriv_atomic_int_set(&s->wpp_err, 0);
2020 ff_reset_entries(s->avctx);
2022 for (i = 0; i <= s->sh.num_entry_point_offsets; i++) {
2027 if (s->pps->entropy_coding_sync_enabled_flag)
2028 s->avctx->execute2(s->avctx, (void *) hls_decode_entry_wpp, arg, ret, s->sh.num_entry_point_offsets + 1);
2030 for (i = 0; i <= s->sh.num_entry_point_offsets; i++)
2038 * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2039 * 0 if the unit should be skipped, 1 otherwise
2041 static int hls_nal_unit(HEVCContext *s)
2043 GetBitContext *gb = &s->HEVClc->gb;
2046 if (get_bits1(gb) != 0)
2047 return AVERROR_INVALIDDATA;
2049 s->nal_unit_type = get_bits(gb, 6);
2051 nuh_layer_id = get_bits(gb, 6);
2052 s->temporal_id = get_bits(gb, 3) - 1;
2053 if (s->temporal_id < 0)
2054 return AVERROR_INVALIDDATA;
2056 av_log(s->avctx, AV_LOG_DEBUG,
2057 "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2058 s->nal_unit_type, nuh_layer_id, s->temporal_id);
2060 return nuh_layer_id == 0;
2063 static void restore_tqb_pixels(HEVCContext *s)
2065 int min_pu_size = 1 << s->sps->log2_min_pu_size;
2068 for (c_idx = 0; c_idx < 3; c_idx++) {
2069 ptrdiff_t stride = s->frame->linesize[c_idx];
2070 int hshift = s->sps->hshift[c_idx];
2071 int vshift = s->sps->vshift[c_idx];
2072 for (y = 0; y < s->sps->min_pu_height; y++) {
2073 for (x = 0; x < s->sps->min_pu_width; x++) {
2074 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2076 int len = min_pu_size >> hshift;
2077 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)];
2078 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)];
2079 for (n = 0; n < (min_pu_size >> vshift); n++) {
2080 memcpy(dst, src, len);
2090 static int set_side_data(HEVCContext *s)
2092 AVFrame *out = s->ref->frame;
2094 if (s->sei_frame_packing_present &&
2095 s->frame_packing_arrangement_type >= 3 &&
2096 s->frame_packing_arrangement_type <= 5 &&
2097 s->content_interpretation_type > 0 &&
2098 s->content_interpretation_type < 3) {
2099 AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2101 return AVERROR(ENOMEM);
2103 switch (s->frame_packing_arrangement_type) {
2105 if (s->quincunx_subsampling)
2106 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2108 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2111 stereo->type = AV_STEREO3D_TOPBOTTOM;
2114 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2118 if (s->content_interpretation_type == 2)
2119 stereo->flags = AV_STEREO3D_FLAG_INVERT;
2125 static int hevc_frame_start(HEVCContext *s)
2127 HEVCLocalContext *lc = s->HEVClc;
2130 memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2131 memset(s->vertical_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2132 memset(s->cbf_luma, 0, s->sps->min_tb_width * s->sps->min_tb_height);
2133 memset(s->is_pcm, 0, s->sps->min_pu_width * s->sps->min_pu_height);
2135 lc->start_of_tiles_x = 0;
2138 if (s->pps->tiles_enabled_flag)
2139 lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2141 ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2146 av_fast_malloc(&lc->edge_emu_buffer, &lc->edge_emu_buffer_size,
2147 (MAX_PB_SIZE + 7) * s->ref->frame->linesize[0]);
2148 if (!lc->edge_emu_buffer) {
2149 ret = AVERROR(ENOMEM);
2153 ret = ff_hevc_frame_rps(s);
2155 av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2159 ret = set_side_data(s);
2163 av_frame_unref(s->output_frame);
2164 ret = ff_hevc_output_frame(s, s->output_frame, 0);
2168 ff_thread_finish_setup(s->avctx);
2173 if (s->ref && s->threads_type == FF_THREAD_FRAME)
2174 ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2179 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2181 HEVCLocalContext *lc = s->HEVClc;
2182 GetBitContext *gb = &lc->gb;
2183 int ctb_addr_ts, ret;
2185 ret = init_get_bits8(gb, nal, length);
2189 ret = hls_nal_unit(s);
2191 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2193 if (s->avctx->err_recognition & AV_EF_EXPLODE)
2199 switch (s->nal_unit_type) {
2201 ret = ff_hevc_decode_nal_vps(s);
2206 ret = ff_hevc_decode_nal_sps(s);
2211 ret = ff_hevc_decode_nal_pps(s);
2215 case NAL_SEI_PREFIX:
2216 case NAL_SEI_SUFFIX:
2217 ret = ff_hevc_decode_nal_sei(s);
2228 case NAL_BLA_W_RADL:
2230 case NAL_IDR_W_RADL:
2237 ret = hls_slice_header(s);
2241 if (s->max_ra == INT_MAX) {
2242 if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2246 s->max_ra = INT_MIN;
2250 if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2251 s->poc <= s->max_ra) {
2255 if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2256 s->max_ra = INT_MIN;
2259 if (s->sh.first_slice_in_pic_flag) {
2260 ret = hevc_frame_start(s);
2263 } else if (!s->ref) {
2264 av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2265 return AVERROR_INVALIDDATA;
2268 if (!s->sh.dependent_slice_segment_flag &&
2269 s->sh.slice_type != I_SLICE) {
2270 ret = ff_hevc_slice_rpl(s);
2272 av_log(s->avctx, AV_LOG_WARNING,
2273 "Error constructing the reference lists for the current slice.\n");
2274 if (s->avctx->err_recognition & AV_EF_EXPLODE)
2279 if (s->threads_number > 1 && s->sh.num_entry_point_offsets > 0)
2280 ctb_addr_ts = hls_slice_data_wpp(s, nal, length);
2282 ctb_addr_ts = hls_slice_data(s);
2283 if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2285 if ((s->pps->transquant_bypass_enable_flag ||
2286 (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2287 s->sps->sao_enabled)
2288 restore_tqb_pixels(s);
2291 if (ctb_addr_ts < 0)
2296 s->seq_decode = (s->seq_decode + 1) & 0xff;
2297 s->max_ra = INT_MAX;
2303 av_log(s->avctx, AV_LOG_INFO,
2304 "Skipping NAL unit %d\n", s->nal_unit_type);
2310 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2311 * between these functions would be nice. */
2312 int ff_hevc_extract_rbsp(HEVCContext *s, const uint8_t *src, int length,
2318 s->skipped_bytes = 0;
2319 #define STARTCODE_TEST \
2320 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
2321 if (src[i + 2] != 3) { \
2322 /* startcode, so we must be past the end */ \
2327 #if HAVE_FAST_UNALIGNED
2328 #define FIND_FIRST_ZERO \
2329 if (i > 0 && !src[i]) \
2334 for (i = 0; i + 1 < length; i += 9) {
2335 if (!((~AV_RN64A(src + i) &
2336 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2337 0x8000800080008080ULL))
2344 for (i = 0; i + 1 < length; i += 5) {
2345 if (!((~AV_RN32A(src + i) &
2346 (AV_RN32A(src + i) - 0x01000101U)) &
2353 #endif /* HAVE_FAST_64BIT */
2355 for (i = 0; i + 1 < length; i += 2) {
2358 if (i > 0 && src[i - 1] == 0)
2362 #endif /* HAVE_FAST_UNALIGNED */
2364 if (i >= length - 1) { // no escaped 0
2370 av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2371 length + FF_INPUT_BUFFER_PADDING_SIZE);
2372 if (!nal->rbsp_buffer)
2373 return AVERROR(ENOMEM);
2375 dst = nal->rbsp_buffer;
2377 memcpy(dst, src, i);
2379 while (si + 2 < length) {
2380 // remove escapes (very rare 1:2^22)
2381 if (src[si + 2] > 3) {
2382 dst[di++] = src[si++];
2383 dst[di++] = src[si++];
2384 } else if (src[si] == 0 && src[si + 1] == 0) {
2385 if (src[si + 2] == 3) { // escape
2391 if (s->skipped_bytes_pos_size < s->skipped_bytes) {
2392 s->skipped_bytes_pos_size *= 2;
2393 av_reallocp_array(&s->skipped_bytes_pos,
2394 s->skipped_bytes_pos_size,
2395 sizeof(*s->skipped_bytes_pos));
2396 if (!s->skipped_bytes_pos)
2397 return AVERROR(ENOMEM);
2399 if (s->skipped_bytes_pos)
2400 s->skipped_bytes_pos[s->skipped_bytes-1] = di - 1;
2402 } else // next start code
2406 dst[di++] = src[si++];
2409 dst[di++] = src[si++];
2412 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2419 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2421 int i, consumed, ret = 0;
2426 /* split the input packet into NAL units, so we know the upper bound on the
2427 * number of slices in the frame */
2429 while (length >= 4) {
2431 int extract_length = 0;
2435 for (i = 0; i < s->nal_length_size; i++)
2436 extract_length = (extract_length << 8) | buf[i];
2437 buf += s->nal_length_size;
2438 length -= s->nal_length_size;
2440 if (extract_length > length) {
2441 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2442 ret = AVERROR_INVALIDDATA;
2446 /* search start code */
2447 while (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2451 av_log(s->avctx, AV_LOG_ERROR, "No start code is found.\n");
2452 ret = AVERROR_INVALIDDATA;
2462 extract_length = length;
2464 if (s->nals_allocated < s->nb_nals + 1) {
2465 int new_size = s->nals_allocated + 1;
2466 HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2468 ret = AVERROR(ENOMEM);
2472 memset(s->nals + s->nals_allocated, 0,
2473 (new_size - s->nals_allocated) * sizeof(*tmp));
2474 av_reallocp_array(&s->skipped_bytes_nal, new_size, sizeof(*s->skipped_bytes_nal));
2475 av_reallocp_array(&s->skipped_bytes_pos_size_nal, new_size, sizeof(*s->skipped_bytes_pos_size_nal));
2476 av_reallocp_array(&s->skipped_bytes_pos_nal, new_size, sizeof(*s->skipped_bytes_pos_nal));
2477 s->skipped_bytes_pos_size_nal[s->nals_allocated] = 1024; // initial buffer size
2478 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));
2479 s->nals_allocated = new_size;
2481 s->skipped_bytes_pos_size = s->skipped_bytes_pos_size_nal[s->nb_nals];
2482 s->skipped_bytes_pos = s->skipped_bytes_pos_nal[s->nb_nals];
2483 nal = &s->nals[s->nb_nals];
2485 consumed = ff_hevc_extract_rbsp(s, buf, extract_length, nal);
2487 s->skipped_bytes_nal[s->nb_nals] = s->skipped_bytes;
2488 s->skipped_bytes_pos_size_nal[s->nb_nals] = s->skipped_bytes_pos_size;
2489 s->skipped_bytes_pos_nal[s->nb_nals++] = s->skipped_bytes_pos;
2497 ret = init_get_bits8(&s->HEVClc->gb, nal->data, nal->size);
2502 if (s->nal_unit_type == NAL_EOB_NUT ||
2503 s->nal_unit_type == NAL_EOS_NUT)
2510 /* parse the NAL units */
2511 for (i = 0; i < s->nb_nals; i++) {
2513 s->skipped_bytes = s->skipped_bytes_nal[i];
2514 s->skipped_bytes_pos = s->skipped_bytes_pos_nal[i];
2516 ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2518 av_log(s->avctx, AV_LOG_WARNING,
2519 "Error parsing NAL unit #%d.\n", i);
2520 if (s->avctx->err_recognition & AV_EF_EXPLODE)
2526 if (s->ref && s->threads_type == FF_THREAD_FRAME)
2527 ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2532 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2535 for (i = 0; i < 16; i++)
2536 av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2539 static int verify_md5(HEVCContext *s, AVFrame *frame)
2541 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2546 return AVERROR(EINVAL);
2548 pixel_shift = desc->comp[0].depth_minus1 > 7;
2550 av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2553 /* the checksums are LE, so we have to byteswap for >8bpp formats
2556 if (pixel_shift && !s->checksum_buf) {
2557 av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2558 FFMAX3(frame->linesize[0], frame->linesize[1],
2559 frame->linesize[2]));
2560 if (!s->checksum_buf)
2561 return AVERROR(ENOMEM);
2565 for (i = 0; frame->data[i]; i++) {
2566 int width = s->avctx->coded_width;
2567 int height = s->avctx->coded_height;
2568 int w = (i == 1 || i == 2) ? (width >> desc->log2_chroma_w) : width;
2569 int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2572 av_md5_init(s->md5_ctx);
2573 for (j = 0; j < h; j++) {
2574 const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2577 s->dsp.bswap16_buf((uint16_t*)s->checksum_buf,
2578 (const uint16_t*)src, w);
2579 src = s->checksum_buf;
2582 av_md5_update(s->md5_ctx, src, w << pixel_shift);
2584 av_md5_final(s->md5_ctx, md5);
2586 if (!memcmp(md5, s->md5[i], 16)) {
2587 av_log (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2588 print_md5(s->avctx, AV_LOG_DEBUG, md5);
2589 av_log (s->avctx, AV_LOG_DEBUG, "; ");
2591 av_log (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2592 print_md5(s->avctx, AV_LOG_ERROR, md5);
2593 av_log (s->avctx, AV_LOG_ERROR, " != ");
2594 print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2595 av_log (s->avctx, AV_LOG_ERROR, "\n");
2596 return AVERROR_INVALIDDATA;
2600 av_log(s->avctx, AV_LOG_DEBUG, "\n");
2605 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2609 HEVCContext *s = avctx->priv_data;
2612 ret = ff_hevc_output_frame(s, data, 1);
2621 ret = decode_nal_units(s, avpkt->data, avpkt->size);
2625 /* verify the SEI checksum */
2626 if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2628 ret = verify_md5(s, s->ref->frame);
2629 if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2630 ff_hevc_unref_frame(s, s->ref, ~0);
2636 if (s->is_decoded) {
2637 av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2641 if (s->output_frame->buf[0]) {
2642 av_frame_move_ref(data, s->output_frame);
2649 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2653 ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2657 dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2658 if (!dst->tab_mvf_buf)
2660 dst->tab_mvf = src->tab_mvf;
2662 dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2663 if (!dst->rpl_tab_buf)
2665 dst->rpl_tab = src->rpl_tab;
2667 dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2671 dst->poc = src->poc;
2672 dst->ctb_count = src->ctb_count;
2673 dst->window = src->window;
2674 dst->flags = src->flags;
2675 dst->sequence = src->sequence;
2679 ff_hevc_unref_frame(s, dst, ~0);
2680 return AVERROR(ENOMEM);
2683 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2685 HEVCContext *s = avctx->priv_data;
2686 HEVCLocalContext *lc = s->HEVClc;
2692 av_freep(&lc->edge_emu_buffer);
2693 av_freep(&s->md5_ctx);
2695 for(i=0; i < s->nals_allocated; i++) {
2696 av_freep(&s->skipped_bytes_pos_nal[i]);
2698 av_freep(&s->skipped_bytes_pos_size_nal);
2699 av_freep(&s->skipped_bytes_nal);
2700 av_freep(&s->skipped_bytes_pos_nal);
2702 av_freep(&s->cabac_state);
2704 av_frame_free(&s->tmp_frame);
2705 av_frame_free(&s->output_frame);
2707 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2708 ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2709 av_frame_free(&s->DPB[i].frame);
2712 for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2713 av_buffer_unref(&s->vps_list[i]);
2714 for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2715 av_buffer_unref(&s->sps_list[i]);
2716 for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2717 av_buffer_unref(&s->pps_list[i]);
2719 av_freep(&s->sh.entry_point_offset);
2720 av_freep(&s->sh.offset);
2721 av_freep(&s->sh.size);
2723 for (i = 1; i < s->threads_number; i++) {
2724 lc = s->HEVClcList[i];
2726 av_freep(&lc->edge_emu_buffer);
2727 av_freep(&s->HEVClcList[i]);
2728 av_freep(&s->sList[i]);
2731 if (s->HEVClc == s->HEVClcList[0])
2733 av_freep(&s->HEVClcList[0]);
2735 for (i = 0; i < s->nals_allocated; i++)
2736 av_freep(&s->nals[i].rbsp_buffer);
2738 s->nals_allocated = 0;
2743 static av_cold int hevc_init_context(AVCodecContext *avctx)
2745 HEVCContext *s = avctx->priv_data;
2750 s->HEVClc = av_mallocz(sizeof(HEVCLocalContext));
2753 s->HEVClcList[0] = s->HEVClc;
2756 s->cabac_state = av_malloc(HEVC_CONTEXTS);
2757 if (!s->cabac_state)
2760 s->tmp_frame = av_frame_alloc();
2764 s->output_frame = av_frame_alloc();
2765 if (!s->output_frame)
2768 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2769 s->DPB[i].frame = av_frame_alloc();
2770 if (!s->DPB[i].frame)
2772 s->DPB[i].tf.f = s->DPB[i].frame;
2775 s->max_ra = INT_MAX;
2777 s->md5_ctx = av_md5_alloc();
2781 ff_dsputil_init(&s->dsp, avctx);
2783 s->context_initialized = 1;
2788 hevc_decode_free(avctx);
2789 return AVERROR(ENOMEM);
2792 static int hevc_update_thread_context(AVCodecContext *dst,
2793 const AVCodecContext *src)
2795 HEVCContext *s = dst->priv_data;
2796 HEVCContext *s0 = src->priv_data;
2799 if (!s->context_initialized) {
2800 ret = hevc_init_context(dst);
2805 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2806 ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2807 if (s0->DPB[i].frame->buf[0]) {
2808 ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
2814 for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
2815 av_buffer_unref(&s->vps_list[i]);
2816 if (s0->vps_list[i]) {
2817 s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
2818 if (!s->vps_list[i])
2819 return AVERROR(ENOMEM);
2823 for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
2824 av_buffer_unref(&s->sps_list[i]);
2825 if (s0->sps_list[i]) {
2826 s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
2827 if (!s->sps_list[i])
2828 return AVERROR(ENOMEM);
2832 for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
2833 av_buffer_unref(&s->pps_list[i]);
2834 if (s0->pps_list[i]) {
2835 s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
2836 if (!s->pps_list[i])
2837 return AVERROR(ENOMEM);
2841 if (s->sps != s0->sps)
2842 ret = set_sps(s, s0->sps);
2844 s->seq_decode = s0->seq_decode;
2845 s->seq_output = s0->seq_output;
2846 s->pocTid0 = s0->pocTid0;
2847 s->max_ra = s0->max_ra;
2849 s->is_nalff = s0->is_nalff;
2850 s->nal_length_size = s0->nal_length_size;
2852 s->threads_number = s0->threads_number;
2853 s->threads_type = s0->threads_type;
2856 s->seq_decode = (s->seq_decode + 1) & 0xff;
2857 s->max_ra = INT_MAX;
2863 static int hevc_decode_extradata(HEVCContext *s)
2865 AVCodecContext *avctx = s->avctx;
2869 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
2871 if (avctx->extradata_size > 3 &&
2872 (avctx->extradata[0] || avctx->extradata[1] ||
2873 avctx->extradata[2] > 1)) {
2874 /* It seems the extradata is encoded as hvcC format.
2875 * Temporarily, we support configurationVersion==0 until 14496-15 3rd
2876 * is finalized. When finalized, configurationVersion will be 1 and we
2877 * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
2878 int i, j, num_arrays, nal_len_size;
2882 bytestream2_skip(&gb, 21);
2883 nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
2884 num_arrays = bytestream2_get_byte(&gb);
2886 /* nal units in the hvcC always have length coded with 2 bytes,
2887 * so put a fake nal_length_size = 2 while parsing them */
2888 s->nal_length_size = 2;
2890 /* Decode nal units from hvcC. */
2891 for (i = 0; i < num_arrays; i++) {
2892 int type = bytestream2_get_byte(&gb) & 0x3f;
2893 int cnt = bytestream2_get_be16(&gb);
2895 for (j = 0; j < cnt; j++) {
2896 // +2 for the nal size field
2897 int nalsize = bytestream2_peek_be16(&gb) + 2;
2898 if (bytestream2_get_bytes_left(&gb) < nalsize) {
2899 av_log(s->avctx, AV_LOG_ERROR,
2900 "Invalid NAL unit size in extradata.\n");
2901 return AVERROR_INVALIDDATA;
2904 ret = decode_nal_units(s, gb.buffer, nalsize);
2906 av_log(avctx, AV_LOG_ERROR,
2907 "Decoding nal unit %d %d from hvcC failed\n",
2911 bytestream2_skip(&gb, nalsize);
2915 /* Now store right nal length size, that will be used to parse
2917 s->nal_length_size = nal_len_size;
2920 ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
2927 static av_cold int hevc_decode_init(AVCodecContext *avctx)
2929 HEVCContext *s = avctx->priv_data;
2932 ff_init_cabac_states();
2934 avctx->internal->allocate_progress = 1;
2936 ret = hevc_init_context(avctx);
2940 s->enable_parallel_tiles = 0;
2941 s->picture_struct = 0;
2943 if(avctx->active_thread_type & FF_THREAD_SLICE)
2944 s->threads_number = avctx->thread_count;
2946 s->threads_number = 1;
2948 if (avctx->extradata_size > 0 && avctx->extradata) {
2949 ret = hevc_decode_extradata(s);
2951 hevc_decode_free(avctx);
2956 if((avctx->active_thread_type & FF_THREAD_FRAME) && avctx->thread_count > 1)
2957 s->threads_type = FF_THREAD_FRAME;
2959 s->threads_type = FF_THREAD_SLICE;
2964 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
2966 HEVCContext *s = avctx->priv_data;
2969 memset(s, 0, sizeof(*s));
2971 ret = hevc_init_context(avctx);
2978 static void hevc_decode_flush(AVCodecContext *avctx)
2980 HEVCContext *s = avctx->priv_data;
2981 ff_hevc_flush_dpb(s);
2982 s->max_ra = INT_MAX;
2985 #define OFFSET(x) offsetof(HEVCContext, x)
2986 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
2988 static const AVProfile profiles[] = {
2989 { FF_PROFILE_HEVC_MAIN, "Main" },
2990 { FF_PROFILE_HEVC_MAIN_10, "Main 10" },
2991 { FF_PROFILE_HEVC_MAIN_STILL_PICTURE, "Main Still Picture" },
2992 { FF_PROFILE_UNKNOWN },
2995 static const AVOption options[] = {
2996 { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
2997 AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
2998 { "strict-displaywin", "stricly apply default display window size", OFFSET(apply_defdispwin),
2999 AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3003 static const AVClass hevc_decoder_class = {
3004 .class_name = "HEVC decoder",
3005 .item_name = av_default_item_name,
3007 .version = LIBAVUTIL_VERSION_INT,
3010 AVCodec ff_hevc_decoder = {
3012 .long_name = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3013 .type = AVMEDIA_TYPE_VIDEO,
3014 .id = AV_CODEC_ID_HEVC,
3015 .priv_data_size = sizeof(HEVCContext),
3016 .priv_class = &hevc_decoder_class,
3017 .init = hevc_decode_init,
3018 .close = hevc_decode_free,
3019 .decode = hevc_decode_frame,
3020 .flush = hevc_decode_flush,
3021 .update_thread_context = hevc_update_thread_context,
3022 .init_thread_copy = hevc_init_thread_copy,
3023 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3024 CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
3025 .profiles = NULL_IF_CONFIG_SMALL(profiles),