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 Libav.
11 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
23 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "libavutil/attributes.h"
27 #include "libavutil/common.h"
28 #include "libavutil/display.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"
36 #include "bytestream.h"
37 #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 };
45 static const uint8_t scan_1x1[1] = { 0 };
47 static const uint8_t horiz_scan2x2_x[4] = { 0, 1, 0, 1 };
49 static const uint8_t horiz_scan2x2_y[4] = { 0, 0, 1, 1 };
51 static const uint8_t horiz_scan4x4_x[16] = {
58 static const uint8_t horiz_scan4x4_y[16] = {
65 static const uint8_t horiz_scan8x8_inv[8][8] = {
66 { 0, 1, 2, 3, 16, 17, 18, 19, },
67 { 4, 5, 6, 7, 20, 21, 22, 23, },
68 { 8, 9, 10, 11, 24, 25, 26, 27, },
69 { 12, 13, 14, 15, 28, 29, 30, 31, },
70 { 32, 33, 34, 35, 48, 49, 50, 51, },
71 { 36, 37, 38, 39, 52, 53, 54, 55, },
72 { 40, 41, 42, 43, 56, 57, 58, 59, },
73 { 44, 45, 46, 47, 60, 61, 62, 63, },
76 static const uint8_t diag_scan2x2_x[4] = { 0, 0, 1, 1 };
78 static const uint8_t diag_scan2x2_y[4] = { 0, 1, 0, 1 };
80 static const uint8_t diag_scan2x2_inv[2][2] = {
85 const uint8_t ff_hevc_diag_scan4x4_x[16] = {
92 const uint8_t ff_hevc_diag_scan4x4_y[16] = {
99 static const uint8_t diag_scan4x4_inv[4][4] = {
106 const uint8_t ff_hevc_diag_scan8x8_x[64] = {
125 const uint8_t ff_hevc_diag_scan8x8_y[64] = {
144 static const uint8_t diag_scan8x8_inv[8][8] = {
145 { 0, 2, 5, 9, 14, 20, 27, 35, },
146 { 1, 4, 8, 13, 19, 26, 34, 42, },
147 { 3, 7, 12, 18, 25, 33, 41, 48, },
148 { 6, 11, 17, 24, 32, 40, 47, 53, },
149 { 10, 16, 23, 31, 39, 46, 52, 57, },
150 { 15, 22, 30, 38, 45, 51, 56, 60, },
151 { 21, 29, 37, 44, 50, 55, 59, 62, },
152 { 28, 36, 43, 49, 54, 58, 61, 63, },
156 * NOTE: Each function hls_foo correspond to the function foo in the
157 * specification (HLS stands for High Level Syntax).
164 /* free everything allocated by pic_arrays_init() */
165 static void pic_arrays_free(HEVCContext *s)
168 av_freep(&s->deblock);
170 av_freep(&s->skip_flag);
171 av_freep(&s->tab_ct_depth);
173 av_freep(&s->tab_ipm);
174 av_freep(&s->cbf_luma);
175 av_freep(&s->is_pcm);
177 av_freep(&s->qp_y_tab);
178 av_freep(&s->tab_slice_address);
179 av_freep(&s->filter_slice_edges);
181 av_freep(&s->horizontal_bs);
182 av_freep(&s->vertical_bs);
184 av_buffer_pool_uninit(&s->tab_mvf_pool);
185 av_buffer_pool_uninit(&s->rpl_tab_pool);
188 /* allocate arrays that depend on frame dimensions */
189 static int pic_arrays_init(HEVCContext *s, const HEVCSPS *sps)
191 int log2_min_cb_size = sps->log2_min_cb_size;
192 int width = sps->width;
193 int height = sps->height;
194 int pic_size_in_ctb = ((width >> log2_min_cb_size) + 1) *
195 ((height >> log2_min_cb_size) + 1);
196 int ctb_count = sps->ctb_width * sps->ctb_height;
197 int min_pu_size = sps->min_pu_width * sps->min_pu_height;
199 s->bs_width = width >> 3;
200 s->bs_height = height >> 3;
202 s->sao = av_mallocz_array(ctb_count, sizeof(*s->sao));
203 s->deblock = av_mallocz_array(ctb_count, sizeof(*s->deblock));
204 if (!s->sao || !s->deblock)
207 s->skip_flag = av_malloc(pic_size_in_ctb);
208 s->tab_ct_depth = av_malloc(sps->min_cb_height * sps->min_cb_width);
209 if (!s->skip_flag || !s->tab_ct_depth)
212 s->cbf_luma = av_malloc(sps->min_tb_width * sps->min_tb_height);
213 s->tab_ipm = av_mallocz(min_pu_size);
214 s->is_pcm = av_malloc(min_pu_size);
215 if (!s->tab_ipm || !s->cbf_luma || !s->is_pcm)
218 s->filter_slice_edges = av_malloc(ctb_count);
219 s->tab_slice_address = av_malloc(pic_size_in_ctb *
220 sizeof(*s->tab_slice_address));
221 s->qp_y_tab = av_malloc(pic_size_in_ctb *
222 sizeof(*s->qp_y_tab));
223 if (!s->qp_y_tab || !s->filter_slice_edges || !s->tab_slice_address)
226 s->horizontal_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
227 s->vertical_bs = av_mallocz(2 * s->bs_width * (s->bs_height + 1));
228 if (!s->horizontal_bs || !s->vertical_bs)
231 s->tab_mvf_pool = av_buffer_pool_init(min_pu_size * sizeof(MvField),
233 s->rpl_tab_pool = av_buffer_pool_init(ctb_count * sizeof(RefPicListTab),
235 if (!s->tab_mvf_pool || !s->rpl_tab_pool)
242 return AVERROR(ENOMEM);
245 static void pred_weight_table(HEVCContext *s, GetBitContext *gb)
249 uint8_t luma_weight_l0_flag[16];
250 uint8_t chroma_weight_l0_flag[16];
251 uint8_t luma_weight_l1_flag[16];
252 uint8_t chroma_weight_l1_flag[16];
254 s->sh.luma_log2_weight_denom = get_ue_golomb_long(gb);
255 if (s->sps->chroma_format_idc != 0) {
256 int delta = get_se_golomb(gb);
257 s->sh.chroma_log2_weight_denom = av_clip_c(s->sh.luma_log2_weight_denom + delta, 0, 7);
260 for (i = 0; i < s->sh.nb_refs[L0]; i++) {
261 luma_weight_l0_flag[i] = get_bits1(gb);
262 if (!luma_weight_l0_flag[i]) {
263 s->sh.luma_weight_l0[i] = 1 << s->sh.luma_log2_weight_denom;
264 s->sh.luma_offset_l0[i] = 0;
267 if (s->sps->chroma_format_idc != 0) { // FIXME: invert "if" and "for"
268 for (i = 0; i < s->sh.nb_refs[L0]; i++)
269 chroma_weight_l0_flag[i] = get_bits1(gb);
271 for (i = 0; i < s->sh.nb_refs[L0]; i++)
272 chroma_weight_l0_flag[i] = 0;
274 for (i = 0; i < s->sh.nb_refs[L0]; i++) {
275 if (luma_weight_l0_flag[i]) {
276 int delta_luma_weight_l0 = get_se_golomb(gb);
277 s->sh.luma_weight_l0[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l0;
278 s->sh.luma_offset_l0[i] = get_se_golomb(gb);
280 if (chroma_weight_l0_flag[i]) {
281 for (j = 0; j < 2; j++) {
282 int delta_chroma_weight_l0 = get_se_golomb(gb);
283 int delta_chroma_offset_l0 = get_se_golomb(gb);
284 s->sh.chroma_weight_l0[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l0;
285 s->sh.chroma_offset_l0[i][j] = av_clip_c((delta_chroma_offset_l0 - ((128 * s->sh.chroma_weight_l0[i][j])
286 >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
289 s->sh.chroma_weight_l0[i][0] = 1 << s->sh.chroma_log2_weight_denom;
290 s->sh.chroma_offset_l0[i][0] = 0;
291 s->sh.chroma_weight_l0[i][1] = 1 << s->sh.chroma_log2_weight_denom;
292 s->sh.chroma_offset_l0[i][1] = 0;
295 if (s->sh.slice_type == B_SLICE) {
296 for (i = 0; i < s->sh.nb_refs[L1]; i++) {
297 luma_weight_l1_flag[i] = get_bits1(gb);
298 if (!luma_weight_l1_flag[i]) {
299 s->sh.luma_weight_l1[i] = 1 << s->sh.luma_log2_weight_denom;
300 s->sh.luma_offset_l1[i] = 0;
303 if (s->sps->chroma_format_idc != 0) {
304 for (i = 0; i < s->sh.nb_refs[L1]; i++)
305 chroma_weight_l1_flag[i] = get_bits1(gb);
307 for (i = 0; i < s->sh.nb_refs[L1]; i++)
308 chroma_weight_l1_flag[i] = 0;
310 for (i = 0; i < s->sh.nb_refs[L1]; i++) {
311 if (luma_weight_l1_flag[i]) {
312 int delta_luma_weight_l1 = get_se_golomb(gb);
313 s->sh.luma_weight_l1[i] = (1 << s->sh.luma_log2_weight_denom) + delta_luma_weight_l1;
314 s->sh.luma_offset_l1[i] = get_se_golomb(gb);
316 if (chroma_weight_l1_flag[i]) {
317 for (j = 0; j < 2; j++) {
318 int delta_chroma_weight_l1 = get_se_golomb(gb);
319 int delta_chroma_offset_l1 = get_se_golomb(gb);
320 s->sh.chroma_weight_l1[i][j] = (1 << s->sh.chroma_log2_weight_denom) + delta_chroma_weight_l1;
321 s->sh.chroma_offset_l1[i][j] = av_clip_c((delta_chroma_offset_l1 - ((128 * s->sh.chroma_weight_l1[i][j])
322 >> s->sh.chroma_log2_weight_denom) + 128), -128, 127);
325 s->sh.chroma_weight_l1[i][0] = 1 << s->sh.chroma_log2_weight_denom;
326 s->sh.chroma_offset_l1[i][0] = 0;
327 s->sh.chroma_weight_l1[i][1] = 1 << s->sh.chroma_log2_weight_denom;
328 s->sh.chroma_offset_l1[i][1] = 0;
334 static int decode_lt_rps(HEVCContext *s, LongTermRPS *rps, GetBitContext *gb)
336 const HEVCSPS *sps = s->sps;
337 int max_poc_lsb = 1 << sps->log2_max_poc_lsb;
338 int prev_delta_msb = 0;
339 unsigned int nb_sps = 0, nb_sh;
343 if (!sps->long_term_ref_pics_present_flag)
346 if (sps->num_long_term_ref_pics_sps > 0)
347 nb_sps = get_ue_golomb_long(gb);
348 nb_sh = get_ue_golomb_long(gb);
350 if (nb_sh + nb_sps > FF_ARRAY_ELEMS(rps->poc))
351 return AVERROR_INVALIDDATA;
353 rps->nb_refs = nb_sh + nb_sps;
355 for (i = 0; i < rps->nb_refs; i++) {
356 uint8_t delta_poc_msb_present;
359 uint8_t lt_idx_sps = 0;
361 if (sps->num_long_term_ref_pics_sps > 1)
362 lt_idx_sps = get_bits(gb, av_ceil_log2(sps->num_long_term_ref_pics_sps));
364 rps->poc[i] = sps->lt_ref_pic_poc_lsb_sps[lt_idx_sps];
365 rps->used[i] = sps->used_by_curr_pic_lt_sps_flag[lt_idx_sps];
367 rps->poc[i] = get_bits(gb, sps->log2_max_poc_lsb);
368 rps->used[i] = get_bits1(gb);
371 delta_poc_msb_present = get_bits1(gb);
372 if (delta_poc_msb_present) {
373 int delta = get_ue_golomb_long(gb);
375 if (i && i != nb_sps)
376 delta += prev_delta_msb;
378 rps->poc[i] += s->poc - delta * max_poc_lsb - s->sh.pic_order_cnt_lsb;
379 prev_delta_msb = delta;
386 static int set_sps(HEVCContext *s, const HEVCSPS *sps)
389 unsigned int num = 0, den = 0;
392 ret = pic_arrays_init(s, sps);
396 s->avctx->coded_width = sps->width;
397 s->avctx->coded_height = sps->height;
398 s->avctx->width = sps->output_width;
399 s->avctx->height = sps->output_height;
400 s->avctx->pix_fmt = sps->pix_fmt;
401 s->avctx->has_b_frames = sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics;
403 ff_set_sar(s->avctx, sps->vui.sar);
405 if (sps->vui.video_signal_type_present_flag)
406 s->avctx->color_range = sps->vui.video_full_range_flag ? AVCOL_RANGE_JPEG
409 s->avctx->color_range = AVCOL_RANGE_MPEG;
411 if (sps->vui.colour_description_present_flag) {
412 s->avctx->color_primaries = sps->vui.colour_primaries;
413 s->avctx->color_trc = sps->vui.transfer_characteristic;
414 s->avctx->colorspace = sps->vui.matrix_coeffs;
416 s->avctx->color_primaries = AVCOL_PRI_UNSPECIFIED;
417 s->avctx->color_trc = AVCOL_TRC_UNSPECIFIED;
418 s->avctx->colorspace = AVCOL_SPC_UNSPECIFIED;
421 ff_hevc_pred_init(&s->hpc, sps->bit_depth);
422 ff_hevc_dsp_init (&s->hevcdsp, sps->bit_depth);
423 ff_videodsp_init (&s->vdsp, sps->bit_depth);
425 if (sps->sao_enabled) {
426 av_frame_unref(s->tmp_frame);
427 ret = ff_get_buffer(s->avctx, s->tmp_frame, AV_GET_BUFFER_FLAG_REF);
430 s->frame = s->tmp_frame;
434 s->vps = (HEVCVPS*) s->vps_list[s->sps->vps_id]->data;
436 if (s->vps->vps_timing_info_present_flag) {
437 num = s->vps->vps_num_units_in_tick;
438 den = s->vps->vps_time_scale;
439 } else if (sps->vui.vui_timing_info_present_flag) {
440 num = sps->vui.vui_num_units_in_tick;
441 den = sps->vui.vui_time_scale;
444 if (num != 0 && den != 0)
445 av_reduce(&s->avctx->framerate.den, &s->avctx->framerate.num,
456 static int hls_slice_header(HEVCContext *s)
458 GetBitContext *gb = &s->HEVClc.gb;
459 SliceHeader *sh = &s->sh;
463 sh->first_slice_in_pic_flag = get_bits1(gb);
464 if ((IS_IDR(s) || IS_BLA(s)) && sh->first_slice_in_pic_flag) {
465 s->seq_decode = (s->seq_decode + 1) & 0xff;
468 ff_hevc_clear_refs(s);
471 sh->no_output_of_prior_pics_flag = get_bits1(gb);
473 sh->pps_id = get_ue_golomb_long(gb);
474 if (sh->pps_id >= MAX_PPS_COUNT || !s->pps_list[sh->pps_id]) {
475 av_log(s->avctx, AV_LOG_ERROR, "PPS id out of range: %d\n", sh->pps_id);
476 return AVERROR_INVALIDDATA;
478 if (!sh->first_slice_in_pic_flag &&
479 s->pps != (HEVCPPS*)s->pps_list[sh->pps_id]->data) {
480 av_log(s->avctx, AV_LOG_ERROR, "PPS changed between slices.\n");
481 return AVERROR_INVALIDDATA;
483 s->pps = (HEVCPPS*)s->pps_list[sh->pps_id]->data;
485 if (s->sps != (HEVCSPS*)s->sps_list[s->pps->sps_id]->data) {
486 s->sps = (HEVCSPS*)s->sps_list[s->pps->sps_id]->data;
488 ff_hevc_clear_refs(s);
489 ret = set_sps(s, s->sps);
493 s->seq_decode = (s->seq_decode + 1) & 0xff;
497 s->avctx->profile = s->sps->ptl.general_ptl.profile_idc;
498 s->avctx->level = s->sps->ptl.general_ptl.level_idc;
500 sh->dependent_slice_segment_flag = 0;
501 if (!sh->first_slice_in_pic_flag) {
502 int slice_address_length;
504 if (s->pps->dependent_slice_segments_enabled_flag)
505 sh->dependent_slice_segment_flag = get_bits1(gb);
507 slice_address_length = av_ceil_log2(s->sps->ctb_width *
509 sh->slice_segment_addr = get_bits(gb, slice_address_length);
510 if (sh->slice_segment_addr >= s->sps->ctb_width * s->sps->ctb_height) {
511 av_log(s->avctx, AV_LOG_ERROR,
512 "Invalid slice segment address: %u.\n",
513 sh->slice_segment_addr);
514 return AVERROR_INVALIDDATA;
517 if (!sh->dependent_slice_segment_flag) {
518 sh->slice_addr = sh->slice_segment_addr;
522 sh->slice_segment_addr = sh->slice_addr = 0;
524 s->slice_initialized = 0;
527 if (!sh->dependent_slice_segment_flag) {
528 s->slice_initialized = 0;
530 for (i = 0; i < s->pps->num_extra_slice_header_bits; i++)
531 skip_bits(gb, 1); // slice_reserved_undetermined_flag[]
533 sh->slice_type = get_ue_golomb_long(gb);
534 if (!(sh->slice_type == I_SLICE ||
535 sh->slice_type == P_SLICE ||
536 sh->slice_type == B_SLICE)) {
537 av_log(s->avctx, AV_LOG_ERROR, "Unknown slice type: %d.\n",
539 return AVERROR_INVALIDDATA;
541 if (IS_IRAP(s) && sh->slice_type != I_SLICE) {
542 av_log(s->avctx, AV_LOG_ERROR, "Inter slices in an IRAP frame.\n");
543 return AVERROR_INVALIDDATA;
546 // when flag is not present, picture is inferred to be output
547 sh->pic_output_flag = 1;
548 if (s->pps->output_flag_present_flag)
549 sh->pic_output_flag = get_bits1(gb);
551 if (s->sps->separate_colour_plane_flag)
552 sh->colour_plane_id = get_bits(gb, 2);
555 int short_term_ref_pic_set_sps_flag, poc;
557 sh->pic_order_cnt_lsb = get_bits(gb, s->sps->log2_max_poc_lsb);
558 poc = ff_hevc_compute_poc(s, sh->pic_order_cnt_lsb);
559 if (!sh->first_slice_in_pic_flag && poc != s->poc) {
560 av_log(s->avctx, AV_LOG_WARNING,
561 "Ignoring POC change between slices: %d -> %d\n", s->poc, poc);
562 if (s->avctx->err_recognition & AV_EF_EXPLODE)
563 return AVERROR_INVALIDDATA;
568 short_term_ref_pic_set_sps_flag = get_bits1(gb);
569 if (!short_term_ref_pic_set_sps_flag) {
570 ret = ff_hevc_decode_short_term_rps(s, &sh->slice_rps, s->sps, 1);
574 sh->short_term_rps = &sh->slice_rps;
576 int numbits, rps_idx;
578 if (!s->sps->nb_st_rps) {
579 av_log(s->avctx, AV_LOG_ERROR, "No ref lists in the SPS.\n");
580 return AVERROR_INVALIDDATA;
583 numbits = av_ceil_log2(s->sps->nb_st_rps);
584 rps_idx = numbits > 0 ? get_bits(gb, numbits) : 0;
585 sh->short_term_rps = &s->sps->st_rps[rps_idx];
588 ret = decode_lt_rps(s, &sh->long_term_rps, gb);
590 av_log(s->avctx, AV_LOG_WARNING, "Invalid long term RPS.\n");
591 if (s->avctx->err_recognition & AV_EF_EXPLODE)
592 return AVERROR_INVALIDDATA;
595 if (s->sps->sps_temporal_mvp_enabled_flag)
596 sh->slice_temporal_mvp_enabled_flag = get_bits1(gb);
598 sh->slice_temporal_mvp_enabled_flag = 0;
600 s->sh.short_term_rps = NULL;
605 if (s->temporal_id == 0 &&
606 s->nal_unit_type != NAL_TRAIL_N &&
607 s->nal_unit_type != NAL_TSA_N &&
608 s->nal_unit_type != NAL_STSA_N &&
609 s->nal_unit_type != NAL_RADL_N &&
610 s->nal_unit_type != NAL_RADL_R &&
611 s->nal_unit_type != NAL_RASL_N &&
612 s->nal_unit_type != NAL_RASL_R)
615 if (s->sps->sao_enabled) {
616 sh->slice_sample_adaptive_offset_flag[0] = get_bits1(gb);
617 sh->slice_sample_adaptive_offset_flag[1] =
618 sh->slice_sample_adaptive_offset_flag[2] = get_bits1(gb);
620 sh->slice_sample_adaptive_offset_flag[0] = 0;
621 sh->slice_sample_adaptive_offset_flag[1] = 0;
622 sh->slice_sample_adaptive_offset_flag[2] = 0;
625 sh->nb_refs[L0] = sh->nb_refs[L1] = 0;
626 if (sh->slice_type == P_SLICE || sh->slice_type == B_SLICE) {
629 sh->nb_refs[L0] = s->pps->num_ref_idx_l0_default_active;
630 if (sh->slice_type == B_SLICE)
631 sh->nb_refs[L1] = s->pps->num_ref_idx_l1_default_active;
633 if (get_bits1(gb)) { // num_ref_idx_active_override_flag
634 sh->nb_refs[L0] = get_ue_golomb_long(gb) + 1;
635 if (sh->slice_type == B_SLICE)
636 sh->nb_refs[L1] = get_ue_golomb_long(gb) + 1;
638 if (sh->nb_refs[L0] > MAX_REFS || sh->nb_refs[L1] > MAX_REFS) {
639 av_log(s->avctx, AV_LOG_ERROR, "Too many refs: %d/%d.\n",
640 sh->nb_refs[L0], sh->nb_refs[L1]);
641 return AVERROR_INVALIDDATA;
644 sh->rpl_modification_flag[0] = 0;
645 sh->rpl_modification_flag[1] = 0;
646 nb_refs = ff_hevc_frame_nb_refs(s);
648 av_log(s->avctx, AV_LOG_ERROR, "Zero refs for a frame with P or B slices.\n");
649 return AVERROR_INVALIDDATA;
652 if (s->pps->lists_modification_present_flag && nb_refs > 1) {
653 sh->rpl_modification_flag[0] = get_bits1(gb);
654 if (sh->rpl_modification_flag[0]) {
655 for (i = 0; i < sh->nb_refs[L0]; i++)
656 sh->list_entry_lx[0][i] = get_bits(gb, av_ceil_log2(nb_refs));
659 if (sh->slice_type == B_SLICE) {
660 sh->rpl_modification_flag[1] = get_bits1(gb);
661 if (sh->rpl_modification_flag[1] == 1)
662 for (i = 0; i < sh->nb_refs[L1]; i++)
663 sh->list_entry_lx[1][i] = get_bits(gb, av_ceil_log2(nb_refs));
667 if (sh->slice_type == B_SLICE)
668 sh->mvd_l1_zero_flag = get_bits1(gb);
670 if (s->pps->cabac_init_present_flag)
671 sh->cabac_init_flag = get_bits1(gb);
673 sh->cabac_init_flag = 0;
675 sh->collocated_ref_idx = 0;
676 if (sh->slice_temporal_mvp_enabled_flag) {
677 sh->collocated_list = L0;
678 if (sh->slice_type == B_SLICE)
679 sh->collocated_list = !get_bits1(gb);
681 if (sh->nb_refs[sh->collocated_list] > 1) {
682 sh->collocated_ref_idx = get_ue_golomb_long(gb);
683 if (sh->collocated_ref_idx >= sh->nb_refs[sh->collocated_list]) {
684 av_log(s->avctx, AV_LOG_ERROR,
685 "Invalid collocated_ref_idx: %d.\n",
686 sh->collocated_ref_idx);
687 return AVERROR_INVALIDDATA;
692 if ((s->pps->weighted_pred_flag && sh->slice_type == P_SLICE) ||
693 (s->pps->weighted_bipred_flag && sh->slice_type == B_SLICE)) {
694 pred_weight_table(s, gb);
697 sh->max_num_merge_cand = 5 - get_ue_golomb_long(gb);
698 if (sh->max_num_merge_cand < 1 || sh->max_num_merge_cand > 5) {
699 av_log(s->avctx, AV_LOG_ERROR,
700 "Invalid number of merging MVP candidates: %d.\n",
701 sh->max_num_merge_cand);
702 return AVERROR_INVALIDDATA;
706 sh->slice_qp_delta = get_se_golomb(gb);
708 if (s->pps->pic_slice_level_chroma_qp_offsets_present_flag) {
709 sh->slice_cb_qp_offset = get_se_golomb(gb);
710 sh->slice_cr_qp_offset = get_se_golomb(gb);
712 sh->slice_cb_qp_offset = 0;
713 sh->slice_cr_qp_offset = 0;
716 if (s->pps->deblocking_filter_control_present_flag) {
717 int deblocking_filter_override_flag = 0;
719 if (s->pps->deblocking_filter_override_enabled_flag)
720 deblocking_filter_override_flag = get_bits1(gb);
722 if (deblocking_filter_override_flag) {
723 sh->disable_deblocking_filter_flag = get_bits1(gb);
724 if (!sh->disable_deblocking_filter_flag) {
725 sh->beta_offset = get_se_golomb(gb) * 2;
726 sh->tc_offset = get_se_golomb(gb) * 2;
729 sh->disable_deblocking_filter_flag = s->pps->disable_dbf;
730 sh->beta_offset = s->pps->beta_offset;
731 sh->tc_offset = s->pps->tc_offset;
734 sh->disable_deblocking_filter_flag = 0;
739 if (s->pps->seq_loop_filter_across_slices_enabled_flag &&
740 (sh->slice_sample_adaptive_offset_flag[0] ||
741 sh->slice_sample_adaptive_offset_flag[1] ||
742 !sh->disable_deblocking_filter_flag)) {
743 sh->slice_loop_filter_across_slices_enabled_flag = get_bits1(gb);
745 sh->slice_loop_filter_across_slices_enabled_flag = s->pps->seq_loop_filter_across_slices_enabled_flag;
747 } else if (!s->slice_initialized) {
748 av_log(s->avctx, AV_LOG_ERROR, "Independent slice segment missing.\n");
749 return AVERROR_INVALIDDATA;
752 sh->num_entry_point_offsets = 0;
753 if (s->pps->tiles_enabled_flag || s->pps->entropy_coding_sync_enabled_flag) {
754 sh->num_entry_point_offsets = get_ue_golomb_long(gb);
755 if (sh->num_entry_point_offsets > 0) {
756 int offset_len = get_ue_golomb_long(gb) + 1;
758 for (i = 0; i < sh->num_entry_point_offsets; i++)
759 skip_bits(gb, offset_len);
763 if (s->pps->slice_header_extension_present_flag) {
764 unsigned int length = get_ue_golomb_long(gb);
765 for (i = 0; i < length; i++)
766 skip_bits(gb, 8); // slice_header_extension_data_byte
769 // Inferred parameters
770 sh->slice_qp = 26 + s->pps->pic_init_qp_minus26 + sh->slice_qp_delta;
771 if (sh->slice_qp > 51 ||
772 sh->slice_qp < -s->sps->qp_bd_offset) {
773 av_log(s->avctx, AV_LOG_ERROR,
774 "The slice_qp %d is outside the valid range "
777 -s->sps->qp_bd_offset);
778 return AVERROR_INVALIDDATA;
781 sh->slice_ctb_addr_rs = sh->slice_segment_addr;
783 if (!s->sh.slice_ctb_addr_rs && s->sh.dependent_slice_segment_flag) {
784 av_log(s->avctx, AV_LOG_ERROR, "Impossible slice segment.\n");
785 return AVERROR_INVALIDDATA;
788 s->HEVClc.first_qp_group = !s->sh.dependent_slice_segment_flag;
790 if (!s->pps->cu_qp_delta_enabled_flag)
791 s->HEVClc.qp_y = FFUMOD(s->sh.slice_qp + 52 + 2 * s->sps->qp_bd_offset,
792 52 + s->sps->qp_bd_offset) - s->sps->qp_bd_offset;
794 s->slice_initialized = 1;
799 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
801 #define SET_SAO(elem, value) \
803 if (!sao_merge_up_flag && !sao_merge_left_flag) \
805 else if (sao_merge_left_flag) \
806 sao->elem = CTB(s->sao, rx-1, ry).elem; \
807 else if (sao_merge_up_flag) \
808 sao->elem = CTB(s->sao, rx, ry-1).elem; \
813 static void hls_sao_param(HEVCContext *s, int rx, int ry)
815 HEVCLocalContext *lc = &s->HEVClc;
816 int sao_merge_left_flag = 0;
817 int sao_merge_up_flag = 0;
818 int shift = s->sps->bit_depth - FFMIN(s->sps->bit_depth, 10);
819 SAOParams *sao = &CTB(s->sao, rx, ry);
822 if (s->sh.slice_sample_adaptive_offset_flag[0] ||
823 s->sh.slice_sample_adaptive_offset_flag[1]) {
825 if (lc->ctb_left_flag)
826 sao_merge_left_flag = ff_hevc_sao_merge_flag_decode(s);
828 if (ry > 0 && !sao_merge_left_flag) {
830 sao_merge_up_flag = ff_hevc_sao_merge_flag_decode(s);
834 for (c_idx = 0; c_idx < 3; c_idx++) {
835 if (!s->sh.slice_sample_adaptive_offset_flag[c_idx]) {
836 sao->type_idx[c_idx] = SAO_NOT_APPLIED;
841 sao->type_idx[2] = sao->type_idx[1];
842 sao->eo_class[2] = sao->eo_class[1];
844 SET_SAO(type_idx[c_idx], ff_hevc_sao_type_idx_decode(s));
847 if (sao->type_idx[c_idx] == SAO_NOT_APPLIED)
850 for (i = 0; i < 4; i++)
851 SET_SAO(offset_abs[c_idx][i], ff_hevc_sao_offset_abs_decode(s));
853 if (sao->type_idx[c_idx] == SAO_BAND) {
854 for (i = 0; i < 4; i++) {
855 if (sao->offset_abs[c_idx][i]) {
856 SET_SAO(offset_sign[c_idx][i],
857 ff_hevc_sao_offset_sign_decode(s));
859 sao->offset_sign[c_idx][i] = 0;
862 SET_SAO(band_position[c_idx], ff_hevc_sao_band_position_decode(s));
863 } else if (c_idx != 2) {
864 SET_SAO(eo_class[c_idx], ff_hevc_sao_eo_class_decode(s));
867 // Inferred parameters
868 sao->offset_val[c_idx][0] = 0;
869 for (i = 0; i < 4; i++) {
870 sao->offset_val[c_idx][i + 1] = sao->offset_abs[c_idx][i] << shift;
871 if (sao->type_idx[c_idx] == SAO_EDGE) {
873 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
874 } else if (sao->offset_sign[c_idx][i]) {
875 sao->offset_val[c_idx][i + 1] = -sao->offset_val[c_idx][i + 1];
884 static void hls_residual_coding(HEVCContext *s, int x0, int y0,
885 int log2_trafo_size, enum ScanType scan_idx,
888 #define GET_COORD(offset, n) \
890 x_c = (scan_x_cg[offset >> 4] << 2) + scan_x_off[n]; \
891 y_c = (scan_y_cg[offset >> 4] << 2) + scan_y_off[n]; \
893 HEVCLocalContext *lc = &s->HEVClc;
894 int transform_skip_flag = 0;
896 int last_significant_coeff_x, last_significant_coeff_y;
900 int greater1_ctx = 1;
903 int x_cg_last_sig, y_cg_last_sig;
905 const uint8_t *scan_x_cg, *scan_y_cg, *scan_x_off, *scan_y_off;
907 ptrdiff_t stride = s->frame->linesize[c_idx];
908 int hshift = s->sps->hshift[c_idx];
909 int vshift = s->sps->vshift[c_idx];
910 uint8_t *dst = &s->frame->data[c_idx][(y0 >> vshift) * stride +
911 ((x0 >> hshift) << s->sps->pixel_shift)];
912 DECLARE_ALIGNED(16, int16_t, coeffs[MAX_TB_SIZE * MAX_TB_SIZE]) = { 0 };
913 DECLARE_ALIGNED(8, uint8_t, significant_coeff_group_flag[8][8]) = { { 0 } };
915 int trafo_size = 1 << log2_trafo_size;
916 int i, qp, shift, add, scale, scale_m;
917 const uint8_t level_scale[] = { 40, 45, 51, 57, 64, 72 };
918 const uint8_t *scale_matrix;
921 // Derive QP for dequant
922 if (!lc->cu.cu_transquant_bypass_flag) {
923 static const int qp_c[] = {
924 29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
927 static const uint8_t rem6[51 + 2 * 6 + 1] = {
928 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
929 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
930 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
933 static const uint8_t div6[51 + 2 * 6 + 1] = {
934 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
935 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
936 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
941 qp = qp_y + s->sps->qp_bd_offset;
946 offset = s->pps->cb_qp_offset + s->sh.slice_cb_qp_offset;
948 offset = s->pps->cr_qp_offset + s->sh.slice_cr_qp_offset;
950 qp_i = av_clip_c(qp_y + offset, -s->sps->qp_bd_offset, 57);
956 qp = qp_c[qp_i - 30];
958 qp += s->sps->qp_bd_offset;
961 shift = s->sps->bit_depth + log2_trafo_size - 5;
962 add = 1 << (shift - 1);
963 scale = level_scale[rem6[qp]] << (div6[qp]);
964 scale_m = 16; // default when no custom scaling lists.
967 if (s->sps->scaling_list_enable_flag) {
968 const ScalingList *sl = s->pps->scaling_list_data_present_flag ?
969 &s->pps->scaling_list : &s->sps->scaling_list;
970 int matrix_id = lc->cu.pred_mode != MODE_INTRA;
972 if (log2_trafo_size != 5)
973 matrix_id = 3 * matrix_id + c_idx;
975 scale_matrix = sl->sl[log2_trafo_size - 2][matrix_id];
976 if (log2_trafo_size >= 4)
977 dc_scale = sl->sl_dc[log2_trafo_size - 4][matrix_id];
981 if (s->pps->transform_skip_enabled_flag &&
982 !lc->cu.cu_transquant_bypass_flag &&
983 log2_trafo_size == 2) {
984 transform_skip_flag = ff_hevc_transform_skip_flag_decode(s, c_idx);
987 last_significant_coeff_x =
988 ff_hevc_last_significant_coeff_x_prefix_decode(s, c_idx, log2_trafo_size);
989 last_significant_coeff_y =
990 ff_hevc_last_significant_coeff_y_prefix_decode(s, c_idx, log2_trafo_size);
992 if (last_significant_coeff_x > 3) {
993 int suffix = ff_hevc_last_significant_coeff_suffix_decode(s, last_significant_coeff_x);
994 last_significant_coeff_x = (1 << ((last_significant_coeff_x >> 1) - 1)) *
995 (2 + (last_significant_coeff_x & 1)) +
999 if (last_significant_coeff_y > 3) {
1000 int suffix = ff_hevc_last_significant_coeff_suffix_decode(s, last_significant_coeff_y);
1001 last_significant_coeff_y = (1 << ((last_significant_coeff_y >> 1) - 1)) *
1002 (2 + (last_significant_coeff_y & 1)) +
1006 if (scan_idx == SCAN_VERT)
1007 FFSWAP(int, last_significant_coeff_x, last_significant_coeff_y);
1009 x_cg_last_sig = last_significant_coeff_x >> 2;
1010 y_cg_last_sig = last_significant_coeff_y >> 2;
1014 int last_x_c = last_significant_coeff_x & 3;
1015 int last_y_c = last_significant_coeff_y & 3;
1017 scan_x_off = ff_hevc_diag_scan4x4_x;
1018 scan_y_off = ff_hevc_diag_scan4x4_y;
1019 num_coeff = diag_scan4x4_inv[last_y_c][last_x_c];
1020 if (trafo_size == 4) {
1021 scan_x_cg = scan_1x1;
1022 scan_y_cg = scan_1x1;
1023 } else if (trafo_size == 8) {
1024 num_coeff += diag_scan2x2_inv[y_cg_last_sig][x_cg_last_sig] << 4;
1025 scan_x_cg = diag_scan2x2_x;
1026 scan_y_cg = diag_scan2x2_y;
1027 } else if (trafo_size == 16) {
1028 num_coeff += diag_scan4x4_inv[y_cg_last_sig][x_cg_last_sig] << 4;
1029 scan_x_cg = ff_hevc_diag_scan4x4_x;
1030 scan_y_cg = ff_hevc_diag_scan4x4_y;
1031 } else { // trafo_size == 32
1032 num_coeff += diag_scan8x8_inv[y_cg_last_sig][x_cg_last_sig] << 4;
1033 scan_x_cg = ff_hevc_diag_scan8x8_x;
1034 scan_y_cg = ff_hevc_diag_scan8x8_y;
1039 scan_x_cg = horiz_scan2x2_x;
1040 scan_y_cg = horiz_scan2x2_y;
1041 scan_x_off = horiz_scan4x4_x;
1042 scan_y_off = horiz_scan4x4_y;
1043 num_coeff = horiz_scan8x8_inv[last_significant_coeff_y][last_significant_coeff_x];
1045 default: //SCAN_VERT
1046 scan_x_cg = horiz_scan2x2_y;
1047 scan_y_cg = horiz_scan2x2_x;
1048 scan_x_off = horiz_scan4x4_y;
1049 scan_y_off = horiz_scan4x4_x;
1050 num_coeff = horiz_scan8x8_inv[last_significant_coeff_x][last_significant_coeff_y];
1054 num_last_subset = (num_coeff - 1) >> 4;
1056 for (i = num_last_subset; i >= 0; i--) {
1058 int x_cg, y_cg, x_c, y_c;
1059 int implicit_non_zero_coeff = 0;
1060 int64_t trans_coeff_level;
1062 int offset = i << 4;
1064 uint8_t significant_coeff_flag_idx[16];
1065 uint8_t nb_significant_coeff_flag = 0;
1067 x_cg = scan_x_cg[i];
1068 y_cg = scan_y_cg[i];
1070 if (i < num_last_subset && i > 0) {
1072 if (x_cg < (1 << (log2_trafo_size - 2)) - 1)
1073 ctx_cg += significant_coeff_group_flag[x_cg + 1][y_cg];
1074 if (y_cg < (1 << (log2_trafo_size - 2)) - 1)
1075 ctx_cg += significant_coeff_group_flag[x_cg][y_cg + 1];
1077 significant_coeff_group_flag[x_cg][y_cg] =
1078 ff_hevc_significant_coeff_group_flag_decode(s, c_idx, ctx_cg);
1079 implicit_non_zero_coeff = 1;
1081 significant_coeff_group_flag[x_cg][y_cg] =
1082 ((x_cg == x_cg_last_sig && y_cg == y_cg_last_sig) ||
1083 (x_cg == 0 && y_cg == 0));
1086 last_scan_pos = num_coeff - offset - 1;
1088 if (i == num_last_subset) {
1089 n_end = last_scan_pos - 1;
1090 significant_coeff_flag_idx[0] = last_scan_pos;
1091 nb_significant_coeff_flag = 1;
1096 if (x_cg < ((1 << log2_trafo_size) - 1) >> 2)
1097 prev_sig = significant_coeff_group_flag[x_cg + 1][y_cg];
1098 if (y_cg < ((1 << log2_trafo_size) - 1) >> 2)
1099 prev_sig += significant_coeff_group_flag[x_cg][y_cg + 1] << 1;
1101 for (n = n_end; n >= 0; n--) {
1102 GET_COORD(offset, n);
1104 if (significant_coeff_group_flag[x_cg][y_cg] &&
1105 (n > 0 || implicit_non_zero_coeff == 0)) {
1106 if (ff_hevc_significant_coeff_flag_decode(s, c_idx, x_c, y_c,
1110 significant_coeff_flag_idx[nb_significant_coeff_flag] = n;
1111 nb_significant_coeff_flag++;
1112 implicit_non_zero_coeff = 0;
1115 int last_cg = (x_c == (x_cg << 2) && y_c == (y_cg << 2));
1116 if (last_cg && implicit_non_zero_coeff && significant_coeff_group_flag[x_cg][y_cg]) {
1117 significant_coeff_flag_idx[nb_significant_coeff_flag] = n;
1118 nb_significant_coeff_flag++;
1123 n_end = nb_significant_coeff_flag;
1126 int first_nz_pos_in_cg = 16;
1127 int last_nz_pos_in_cg = -1;
1128 int c_rice_param = 0;
1129 int first_greater1_coeff_idx = -1;
1130 uint8_t coeff_abs_level_greater1_flag[16] = { 0 };
1131 uint16_t coeff_sign_flag;
1133 int sign_hidden = 0;
1135 // initialize first elem of coeff_bas_level_greater1_flag
1136 int ctx_set = (i > 0 && c_idx == 0) ? 2 : 0;
1138 if (!(i == num_last_subset) && greater1_ctx == 0)
1141 last_nz_pos_in_cg = significant_coeff_flag_idx[0];
1143 for (m = 0; m < (n_end > 8 ? 8 : n_end); m++) {
1144 int n_idx = significant_coeff_flag_idx[m];
1145 int inc = (ctx_set << 2) + greater1_ctx;
1146 coeff_abs_level_greater1_flag[n_idx] =
1147 ff_hevc_coeff_abs_level_greater1_flag_decode(s, c_idx, inc);
1148 if (coeff_abs_level_greater1_flag[n_idx]) {
1150 } else if (greater1_ctx > 0 && greater1_ctx < 3) {
1154 if (coeff_abs_level_greater1_flag[n_idx] &&
1155 first_greater1_coeff_idx == -1)
1156 first_greater1_coeff_idx = n_idx;
1158 first_nz_pos_in_cg = significant_coeff_flag_idx[n_end - 1];
1159 sign_hidden = last_nz_pos_in_cg - first_nz_pos_in_cg >= 4 &&
1160 !lc->cu.cu_transquant_bypass_flag;
1162 if (first_greater1_coeff_idx != -1) {
1163 coeff_abs_level_greater1_flag[first_greater1_coeff_idx] += ff_hevc_coeff_abs_level_greater2_flag_decode(s, c_idx, ctx_set);
1165 if (!s->pps->sign_data_hiding_flag || !sign_hidden) {
1166 coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag) << (16 - nb_significant_coeff_flag);
1168 coeff_sign_flag = ff_hevc_coeff_sign_flag(s, nb_significant_coeff_flag - 1) << (16 - (nb_significant_coeff_flag - 1));
1171 for (m = 0; m < n_end; m++) {
1172 n = significant_coeff_flag_idx[m];
1173 GET_COORD(offset, n);
1174 trans_coeff_level = 1 + coeff_abs_level_greater1_flag[n];
1175 if (trans_coeff_level == ((m < 8) ?
1176 ((n == first_greater1_coeff_idx) ? 3 : 2) : 1)) {
1177 int last_coeff_abs_level_remaining = ff_hevc_coeff_abs_level_remaining(s, trans_coeff_level, c_rice_param);
1179 trans_coeff_level += last_coeff_abs_level_remaining;
1180 if ((trans_coeff_level) > (3 * (1 << c_rice_param)))
1181 c_rice_param = FFMIN(c_rice_param + 1, 4);
1183 if (s->pps->sign_data_hiding_flag && sign_hidden) {
1184 sum_abs += trans_coeff_level;
1185 if (n == first_nz_pos_in_cg && ((sum_abs & 1) == 1))
1186 trans_coeff_level = -trans_coeff_level;
1188 if (coeff_sign_flag >> 15)
1189 trans_coeff_level = -trans_coeff_level;
1190 coeff_sign_flag <<= 1;
1191 if (!lc->cu.cu_transquant_bypass_flag) {
1192 if (s->sps->scaling_list_enable_flag) {
1193 if (y_c || x_c || log2_trafo_size < 4) {
1195 switch (log2_trafo_size) {
1196 case 3: pos = (y_c << 3) + x_c; break;
1197 case 4: pos = ((y_c >> 1) << 3) + (x_c >> 1); break;
1198 case 5: pos = ((y_c >> 2) << 3) + (x_c >> 2); break;
1199 default: pos = (y_c << 2) + x_c;
1201 scale_m = scale_matrix[pos];
1206 trans_coeff_level = (trans_coeff_level * (int64_t)scale * (int64_t)scale_m + add) >> shift;
1207 if(trans_coeff_level < 0) {
1208 if((~trans_coeff_level) & 0xFffffffffff8000)
1209 trans_coeff_level = -32768;
1211 if (trans_coeff_level & 0xffffffffffff8000)
1212 trans_coeff_level = 32767;
1215 coeffs[y_c * trafo_size + x_c] = trans_coeff_level;
1220 if (lc->cu.cu_transquant_bypass_flag) {
1221 s->hevcdsp.transquant_bypass[log2_trafo_size - 2](dst, coeffs, stride);
1223 if (transform_skip_flag)
1224 s->hevcdsp.transform_skip(dst, coeffs, stride);
1225 else if (lc->cu.pred_mode == MODE_INTRA && c_idx == 0 &&
1226 log2_trafo_size == 2)
1227 s->hevcdsp.transform_4x4_luma_add(dst, coeffs, stride);
1229 s->hevcdsp.transform_add[log2_trafo_size - 2](dst, coeffs, stride);
1233 static int hls_transform_unit(HEVCContext *s, int x0, int y0,
1234 int xBase, int yBase, int cb_xBase, int cb_yBase,
1235 int log2_cb_size, int log2_trafo_size,
1236 int blk_idx, int cbf_luma, int cbf_cb, int cbf_cr)
1238 HEVCLocalContext *lc = &s->HEVClc;
1240 if (lc->cu.pred_mode == MODE_INTRA) {
1241 int trafo_size = 1 << log2_trafo_size;
1242 ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
1244 s->hpc.intra_pred[log2_trafo_size - 2](s, x0, y0, 0);
1245 if (log2_trafo_size > 2) {
1246 trafo_size = trafo_size << (s->sps->hshift[1] - 1);
1247 ff_hevc_set_neighbour_available(s, x0, y0, trafo_size, trafo_size);
1248 s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 1);
1249 s->hpc.intra_pred[log2_trafo_size - 3](s, x0, y0, 2);
1250 } else if (blk_idx == 3) {
1251 trafo_size = trafo_size << s->sps->hshift[1];
1252 ff_hevc_set_neighbour_available(s, xBase, yBase,
1253 trafo_size, trafo_size);
1254 s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 1);
1255 s->hpc.intra_pred[log2_trafo_size - 2](s, xBase, yBase, 2);
1259 if (cbf_luma || cbf_cb || cbf_cr) {
1260 int scan_idx = SCAN_DIAG;
1261 int scan_idx_c = SCAN_DIAG;
1263 if (s->pps->cu_qp_delta_enabled_flag && !lc->tu.is_cu_qp_delta_coded) {
1264 lc->tu.cu_qp_delta = ff_hevc_cu_qp_delta_abs(s);
1265 if (lc->tu.cu_qp_delta != 0)
1266 if (ff_hevc_cu_qp_delta_sign_flag(s) == 1)
1267 lc->tu.cu_qp_delta = -lc->tu.cu_qp_delta;
1268 lc->tu.is_cu_qp_delta_coded = 1;
1270 if (lc->tu.cu_qp_delta < -(26 + s->sps->qp_bd_offset / 2) ||
1271 lc->tu.cu_qp_delta > (25 + s->sps->qp_bd_offset / 2)) {
1272 av_log(s->avctx, AV_LOG_ERROR,
1273 "The cu_qp_delta %d is outside the valid range "
1276 -(26 + s->sps->qp_bd_offset / 2),
1277 (25 + s->sps->qp_bd_offset / 2));
1278 return AVERROR_INVALIDDATA;
1281 ff_hevc_set_qPy(s, x0, y0, cb_xBase, cb_yBase, log2_cb_size);
1284 if (lc->cu.pred_mode == MODE_INTRA && log2_trafo_size < 4) {
1285 if (lc->tu.cur_intra_pred_mode >= 6 &&
1286 lc->tu.cur_intra_pred_mode <= 14) {
1287 scan_idx = SCAN_VERT;
1288 } else if (lc->tu.cur_intra_pred_mode >= 22 &&
1289 lc->tu.cur_intra_pred_mode <= 30) {
1290 scan_idx = SCAN_HORIZ;
1293 if (lc->pu.intra_pred_mode_c >= 6 &&
1294 lc->pu.intra_pred_mode_c <= 14) {
1295 scan_idx_c = SCAN_VERT;
1296 } else if (lc->pu.intra_pred_mode_c >= 22 &&
1297 lc->pu.intra_pred_mode_c <= 30) {
1298 scan_idx_c = SCAN_HORIZ;
1303 hls_residual_coding(s, x0, y0, log2_trafo_size, scan_idx, 0);
1304 if (log2_trafo_size > 2) {
1306 hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 1);
1308 hls_residual_coding(s, x0, y0, log2_trafo_size - 1, scan_idx_c, 2);
1309 } else if (blk_idx == 3) {
1311 hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 1);
1313 hls_residual_coding(s, xBase, yBase, log2_trafo_size, scan_idx_c, 2);
1319 static void set_deblocking_bypass(HEVCContext *s, int x0, int y0, int log2_cb_size)
1321 int cb_size = 1 << log2_cb_size;
1322 int log2_min_pu_size = s->sps->log2_min_pu_size;
1324 int min_pu_width = s->sps->min_pu_width;
1325 int x_end = FFMIN(x0 + cb_size, s->sps->width);
1326 int y_end = FFMIN(y0 + cb_size, s->sps->height);
1329 for (j = (y0 >> log2_min_pu_size); j < (y_end >> log2_min_pu_size); j++)
1330 for (i = (x0 >> log2_min_pu_size); i < (x_end >> log2_min_pu_size); i++)
1331 s->is_pcm[i + j * min_pu_width] = 2;
1334 static int hls_transform_tree(HEVCContext *s, int x0, int y0,
1335 int xBase, int yBase, int cb_xBase, int cb_yBase,
1336 int log2_cb_size, int log2_trafo_size,
1337 int trafo_depth, int blk_idx,
1338 int cbf_cb, int cbf_cr)
1340 HEVCLocalContext *lc = &s->HEVClc;
1341 uint8_t split_transform_flag;
1344 if (lc->cu.intra_split_flag) {
1345 if (trafo_depth == 1)
1346 lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[blk_idx];
1348 lc->tu.cur_intra_pred_mode = lc->pu.intra_pred_mode[0];
1351 if (log2_trafo_size <= s->sps->log2_max_trafo_size &&
1352 log2_trafo_size > s->sps->log2_min_tb_size &&
1353 trafo_depth < lc->cu.max_trafo_depth &&
1354 !(lc->cu.intra_split_flag && trafo_depth == 0)) {
1355 split_transform_flag = ff_hevc_split_transform_flag_decode(s, log2_trafo_size);
1357 int inter_split = s->sps->max_transform_hierarchy_depth_inter == 0 &&
1358 lc->cu.pred_mode == MODE_INTER &&
1359 lc->cu.part_mode != PART_2Nx2N &&
1362 split_transform_flag = log2_trafo_size > s->sps->log2_max_trafo_size ||
1363 (lc->cu.intra_split_flag && trafo_depth == 0) ||
1367 if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cb))
1368 cbf_cb = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
1369 else if (log2_trafo_size > 2 || trafo_depth == 0)
1371 if (log2_trafo_size > 2 && (trafo_depth == 0 || cbf_cr))
1372 cbf_cr = ff_hevc_cbf_cb_cr_decode(s, trafo_depth);
1373 else if (log2_trafo_size > 2 || trafo_depth == 0)
1376 if (split_transform_flag) {
1377 const int trafo_size_split = 1 << (log2_trafo_size - 1);
1378 const int x1 = x0 + trafo_size_split;
1379 const int y1 = y0 + trafo_size_split;
1381 #define SUBDIVIDE(x, y, idx) \
1383 ret = hls_transform_tree(s, x, y, x0, y0, cb_xBase, cb_yBase, log2_cb_size, \
1384 log2_trafo_size - 1, trafo_depth + 1, idx, \
1390 SUBDIVIDE(x0, y0, 0);
1391 SUBDIVIDE(x1, y0, 1);
1392 SUBDIVIDE(x0, y1, 2);
1393 SUBDIVIDE(x1, y1, 3);
1397 int min_tu_size = 1 << s->sps->log2_min_tb_size;
1398 int log2_min_tu_size = s->sps->log2_min_tb_size;
1399 int min_tu_width = s->sps->min_tb_width;
1402 if (lc->cu.pred_mode == MODE_INTRA || trafo_depth != 0 ||
1404 cbf_luma = ff_hevc_cbf_luma_decode(s, trafo_depth);
1406 ret = hls_transform_unit(s, x0, y0, xBase, yBase, cb_xBase, cb_yBase,
1407 log2_cb_size, log2_trafo_size,
1408 blk_idx, cbf_luma, cbf_cb, cbf_cr);
1411 // TODO: store cbf_luma somewhere else
1414 for (i = 0; i < (1 << log2_trafo_size); i += min_tu_size)
1415 for (j = 0; j < (1 << log2_trafo_size); j += min_tu_size) {
1416 int x_tu = (x0 + j) >> log2_min_tu_size;
1417 int y_tu = (y0 + i) >> log2_min_tu_size;
1418 s->cbf_luma[y_tu * min_tu_width + x_tu] = 1;
1421 if (!s->sh.disable_deblocking_filter_flag) {
1422 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_trafo_size);
1423 if (s->pps->transquant_bypass_enable_flag &&
1424 lc->cu.cu_transquant_bypass_flag)
1425 set_deblocking_bypass(s, x0, y0, log2_trafo_size);
1431 static int hls_pcm_sample(HEVCContext *s, int x0, int y0, int log2_cb_size)
1433 //TODO: non-4:2:0 support
1434 HEVCLocalContext *lc = &s->HEVClc;
1436 int cb_size = 1 << log2_cb_size;
1437 int stride0 = s->frame->linesize[0];
1438 uint8_t *dst0 = &s->frame->data[0][y0 * stride0 + (x0 << s->sps->pixel_shift)];
1439 int stride1 = s->frame->linesize[1];
1440 uint8_t *dst1 = &s->frame->data[1][(y0 >> s->sps->vshift[1]) * stride1 + ((x0 >> s->sps->hshift[1]) << s->sps->pixel_shift)];
1441 int stride2 = s->frame->linesize[2];
1442 uint8_t *dst2 = &s->frame->data[2][(y0 >> s->sps->vshift[2]) * stride2 + ((x0 >> s->sps->hshift[2]) << s->sps->pixel_shift)];
1444 int length = cb_size * cb_size * s->sps->pcm.bit_depth + ((cb_size * cb_size) >> 1) * s->sps->pcm.bit_depth_chroma;
1445 const uint8_t *pcm = skip_bytes(&lc->cc, (length + 7) >> 3);
1448 if (!s->sh.disable_deblocking_filter_flag)
1449 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
1451 ret = init_get_bits(&gb, pcm, length);
1455 s->hevcdsp.put_pcm(dst0, stride0, cb_size, &gb, s->sps->pcm.bit_depth);
1456 s->hevcdsp.put_pcm(dst1, stride1, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1457 s->hevcdsp.put_pcm(dst2, stride2, cb_size / 2, &gb, s->sps->pcm.bit_depth_chroma);
1461 static void hls_mvd_coding(HEVCContext *s, int x0, int y0, int log2_cb_size)
1463 HEVCLocalContext *lc = &s->HEVClc;
1464 int x = ff_hevc_abs_mvd_greater0_flag_decode(s);
1465 int y = ff_hevc_abs_mvd_greater0_flag_decode(s);
1468 x += ff_hevc_abs_mvd_greater1_flag_decode(s);
1470 y += ff_hevc_abs_mvd_greater1_flag_decode(s);
1473 case 2: lc->pu.mvd.x = ff_hevc_mvd_decode(s); break;
1474 case 1: lc->pu.mvd.x = ff_hevc_mvd_sign_flag_decode(s); break;
1475 case 0: lc->pu.mvd.x = 0; break;
1479 case 2: lc->pu.mvd.y = ff_hevc_mvd_decode(s); break;
1480 case 1: lc->pu.mvd.y = ff_hevc_mvd_sign_flag_decode(s); break;
1481 case 0: lc->pu.mvd.y = 0; break;
1486 * 8.5.3.2.2.1 Luma sample interpolation process
1488 * @param s HEVC decoding context
1489 * @param dst target buffer for block data at block position
1490 * @param dststride stride of the dst buffer
1491 * @param ref reference picture buffer at origin (0, 0)
1492 * @param mv motion vector (relative to block position) to get pixel data from
1493 * @param x_off horizontal position of block from origin (0, 0)
1494 * @param y_off vertical position of block from origin (0, 0)
1495 * @param block_w width of block
1496 * @param block_h height of block
1498 static void luma_mc(HEVCContext *s, int16_t *dst, ptrdiff_t dststride,
1499 AVFrame *ref, const Mv *mv, int x_off, int y_off,
1500 int block_w, int block_h)
1502 HEVCLocalContext *lc = &s->HEVClc;
1503 uint8_t *src = ref->data[0];
1504 ptrdiff_t srcstride = ref->linesize[0];
1505 int pic_width = s->sps->width;
1506 int pic_height = s->sps->height;
1510 int extra_left = ff_hevc_qpel_extra_before[mx];
1511 int extra_top = ff_hevc_qpel_extra_before[my];
1513 x_off += mv->x >> 2;
1514 y_off += mv->y >> 2;
1515 src += y_off * srcstride + (x_off << s->sps->pixel_shift);
1517 if (x_off < extra_left || y_off < extra_top ||
1518 x_off >= pic_width - block_w - ff_hevc_qpel_extra_after[mx] ||
1519 y_off >= pic_height - block_h - ff_hevc_qpel_extra_after[my]) {
1520 const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1521 int offset = extra_top * srcstride + (extra_left << s->sps->pixel_shift);
1522 int buf_offset = extra_top *
1523 edge_emu_stride + (extra_left << s->sps->pixel_shift);
1525 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src - offset,
1526 edge_emu_stride, srcstride,
1527 block_w + ff_hevc_qpel_extra[mx],
1528 block_h + ff_hevc_qpel_extra[my],
1529 x_off - extra_left, y_off - extra_top,
1530 pic_width, pic_height);
1531 src = lc->edge_emu_buffer + buf_offset;
1532 srcstride = edge_emu_stride;
1534 s->hevcdsp.put_hevc_qpel[my][mx](dst, dststride, src, srcstride, block_w,
1535 block_h, lc->mc_buffer);
1539 * 8.5.3.2.2.2 Chroma sample interpolation process
1541 * @param s HEVC decoding context
1542 * @param dst1 target buffer for block data at block position (U plane)
1543 * @param dst2 target buffer for block data at block position (V plane)
1544 * @param dststride stride of the dst1 and dst2 buffers
1545 * @param ref reference picture buffer at origin (0, 0)
1546 * @param mv motion vector (relative to block position) to get pixel data from
1547 * @param x_off horizontal position of block from origin (0, 0)
1548 * @param y_off vertical position of block from origin (0, 0)
1549 * @param block_w width of block
1550 * @param block_h height of block
1552 static void chroma_mc(HEVCContext *s, int16_t *dst1, int16_t *dst2,
1553 ptrdiff_t dststride, AVFrame *ref, const Mv *mv,
1554 int x_off, int y_off, int block_w, int block_h)
1556 HEVCLocalContext *lc = &s->HEVClc;
1557 uint8_t *src1 = ref->data[1];
1558 uint8_t *src2 = ref->data[2];
1559 ptrdiff_t src1stride = ref->linesize[1];
1560 ptrdiff_t src2stride = ref->linesize[2];
1561 int pic_width = s->sps->width >> 1;
1562 int pic_height = s->sps->height >> 1;
1567 x_off += mv->x >> 3;
1568 y_off += mv->y >> 3;
1569 src1 += y_off * src1stride + (x_off << s->sps->pixel_shift);
1570 src2 += y_off * src2stride + (x_off << s->sps->pixel_shift);
1572 if (x_off < EPEL_EXTRA_BEFORE || y_off < EPEL_EXTRA_AFTER ||
1573 x_off >= pic_width - block_w - EPEL_EXTRA_AFTER ||
1574 y_off >= pic_height - block_h - EPEL_EXTRA_AFTER) {
1575 const int edge_emu_stride = EDGE_EMU_BUFFER_STRIDE << s->sps->pixel_shift;
1576 int offset1 = EPEL_EXTRA_BEFORE * (src1stride + (1 << s->sps->pixel_shift));
1577 int buf_offset1 = EPEL_EXTRA_BEFORE *
1578 (edge_emu_stride + (1 << s->sps->pixel_shift));
1579 int offset2 = EPEL_EXTRA_BEFORE * (src2stride + (1 << s->sps->pixel_shift));
1580 int buf_offset2 = EPEL_EXTRA_BEFORE *
1581 (edge_emu_stride + (1 << s->sps->pixel_shift));
1583 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src1 - offset1,
1584 edge_emu_stride, src1stride,
1585 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1586 x_off - EPEL_EXTRA_BEFORE,
1587 y_off - EPEL_EXTRA_BEFORE,
1588 pic_width, pic_height);
1590 src1 = lc->edge_emu_buffer + buf_offset1;
1591 src1stride = edge_emu_stride;
1592 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1593 block_w, block_h, mx, my, lc->mc_buffer);
1595 s->vdsp.emulated_edge_mc(lc->edge_emu_buffer, src2 - offset2,
1596 edge_emu_stride, src2stride,
1597 block_w + EPEL_EXTRA, block_h + EPEL_EXTRA,
1598 x_off - EPEL_EXTRA_BEFORE,
1599 y_off - EPEL_EXTRA_BEFORE,
1600 pic_width, pic_height);
1601 src2 = lc->edge_emu_buffer + buf_offset2;
1602 src2stride = edge_emu_stride;
1604 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1605 block_w, block_h, mx, my,
1608 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst1, dststride, src1, src1stride,
1609 block_w, block_h, mx, my,
1611 s->hevcdsp.put_hevc_epel[!!my][!!mx](dst2, dststride, src2, src2stride,
1612 block_w, block_h, mx, my,
1617 static void hevc_await_progress(HEVCContext *s, HEVCFrame *ref,
1618 const Mv *mv, int y0, int height)
1620 int y = (mv->y >> 2) + y0 + height + 9;
1621 ff_thread_await_progress(&ref->tf, y, 0);
1624 static void hls_prediction_unit(HEVCContext *s, int x0, int y0,
1626 int log2_cb_size, int partIdx)
1628 #define POS(c_idx, x, y) \
1629 &s->frame->data[c_idx][((y) >> s->sps->vshift[c_idx]) * s->frame->linesize[c_idx] + \
1630 (((x) >> s->sps->hshift[c_idx]) << s->sps->pixel_shift)]
1631 HEVCLocalContext *lc = &s->HEVClc;
1633 struct MvField current_mv = {{{ 0 }}};
1635 int min_pu_width = s->sps->min_pu_width;
1637 MvField *tab_mvf = s->ref->tab_mvf;
1638 RefPicList *refPicList = s->ref->refPicList;
1639 HEVCFrame *ref0, *ref1;
1641 int tmpstride = MAX_PB_SIZE;
1643 uint8_t *dst0 = POS(0, x0, y0);
1644 uint8_t *dst1 = POS(1, x0, y0);
1645 uint8_t *dst2 = POS(2, x0, y0);
1646 int log2_min_cb_size = s->sps->log2_min_cb_size;
1647 int min_cb_width = s->sps->min_cb_width;
1648 int x_cb = x0 >> log2_min_cb_size;
1649 int y_cb = y0 >> log2_min_cb_size;
1653 int skip_flag = SAMPLE_CTB(s->skip_flag, x_cb, y_cb);
1656 lc->pu.merge_flag = ff_hevc_merge_flag_decode(s);
1658 if (skip_flag || lc->pu.merge_flag) {
1659 if (s->sh.max_num_merge_cand > 1)
1660 merge_idx = ff_hevc_merge_idx_decode(s);
1664 ff_hevc_luma_mv_merge_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1665 partIdx, merge_idx, ¤t_mv);
1667 enum InterPredIdc inter_pred_idc = PRED_L0;
1670 ff_hevc_set_neighbour_available(s, x0, y0, nPbW, nPbH);
1671 if (s->sh.slice_type == B_SLICE)
1672 inter_pred_idc = ff_hevc_inter_pred_idc_decode(s, nPbW, nPbH);
1674 if (inter_pred_idc != PRED_L1) {
1675 if (s->sh.nb_refs[L0]) {
1676 current_mv.ref_idx[0]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L0]);
1678 current_mv.pred_flag[0] = 1;
1679 hls_mvd_coding(s, x0, y0, 0);
1680 mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
1681 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1682 partIdx, merge_idx, ¤t_mv,
1684 current_mv.mv[0].x += lc->pu.mvd.x;
1685 current_mv.mv[0].y += lc->pu.mvd.y;
1688 if (inter_pred_idc != PRED_L0) {
1689 if (s->sh.nb_refs[L1]) {
1690 current_mv.ref_idx[1]= ff_hevc_ref_idx_lx_decode(s, s->sh.nb_refs[L1]);
1693 if (s->sh.mvd_l1_zero_flag == 1 && inter_pred_idc == PRED_BI) {
1694 AV_ZERO32(&lc->pu.mvd);
1696 hls_mvd_coding(s, x0, y0, 1);
1699 current_mv.pred_flag[1] = 1;
1700 mvp_flag = ff_hevc_mvp_lx_flag_decode(s);
1701 ff_hevc_luma_mv_mvp_mode(s, x0, y0, nPbW, nPbH, log2_cb_size,
1702 partIdx, merge_idx, ¤t_mv,
1704 current_mv.mv[1].x += lc->pu.mvd.x;
1705 current_mv.mv[1].y += lc->pu.mvd.y;
1709 x_pu = x0 >> s->sps->log2_min_pu_size;
1710 y_pu = y0 >> s->sps->log2_min_pu_size;
1712 for (j = 0; j < nPbH >> s->sps->log2_min_pu_size; j++)
1713 for (i = 0; i < nPbW >> s->sps->log2_min_pu_size; i++)
1714 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i] = current_mv;
1716 if (current_mv.pred_flag[0]) {
1717 ref0 = refPicList[0].ref[current_mv.ref_idx[0]];
1720 hevc_await_progress(s, ref0, ¤t_mv.mv[0], y0, nPbH);
1722 if (current_mv.pred_flag[1]) {
1723 ref1 = refPicList[1].ref[current_mv.ref_idx[1]];
1726 hevc_await_progress(s, ref1, ¤t_mv.mv[1], y0, nPbH);
1729 if (current_mv.pred_flag[0] && !current_mv.pred_flag[1]) {
1730 DECLARE_ALIGNED(16, int16_t, tmp[MAX_PB_SIZE * MAX_PB_SIZE]);
1731 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1733 luma_mc(s, tmp, tmpstride, ref0->frame,
1734 ¤t_mv.mv[0], x0, y0, nPbW, nPbH);
1736 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1737 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1738 s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1739 s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1740 s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1741 dst0, s->frame->linesize[0], tmp,
1742 tmpstride, nPbW, nPbH);
1744 s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1746 chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1747 ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1749 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1750 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1751 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1752 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1753 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1754 dst1, s->frame->linesize[1], tmp, tmpstride,
1755 nPbW / 2, nPbH / 2);
1756 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1757 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1758 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1759 dst2, s->frame->linesize[2], tmp2, tmpstride,
1760 nPbW / 2, nPbH / 2);
1762 s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1763 s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1765 } else if (!current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1766 DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1767 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1769 luma_mc(s, tmp, tmpstride, ref1->frame,
1770 ¤t_mv.mv[1], x0, y0, nPbW, nPbH);
1772 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1773 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1774 s->hevcdsp.weighted_pred(s->sh.luma_log2_weight_denom,
1775 s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1776 s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1777 dst0, s->frame->linesize[0], tmp, tmpstride,
1780 s->hevcdsp.put_unweighted_pred(dst0, s->frame->linesize[0], tmp, tmpstride, nPbW, nPbH);
1783 chroma_mc(s, tmp, tmp2, tmpstride, ref1->frame,
1784 ¤t_mv.mv[1], x0/2, y0/2, nPbW/2, nPbH/2);
1786 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1787 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1788 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1789 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1790 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1791 dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1792 s->hevcdsp.weighted_pred(s->sh.chroma_log2_weight_denom,
1793 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1794 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1795 dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1797 s->hevcdsp.put_unweighted_pred(dst1, s->frame->linesize[1], tmp, tmpstride, nPbW/2, nPbH/2);
1798 s->hevcdsp.put_unweighted_pred(dst2, s->frame->linesize[2], tmp2, tmpstride, nPbW/2, nPbH/2);
1800 } else if (current_mv.pred_flag[0] && current_mv.pred_flag[1]) {
1801 DECLARE_ALIGNED(16, int16_t, tmp [MAX_PB_SIZE * MAX_PB_SIZE]);
1802 DECLARE_ALIGNED(16, int16_t, tmp2[MAX_PB_SIZE * MAX_PB_SIZE]);
1803 DECLARE_ALIGNED(16, int16_t, tmp3[MAX_PB_SIZE * MAX_PB_SIZE]);
1804 DECLARE_ALIGNED(16, int16_t, tmp4[MAX_PB_SIZE * MAX_PB_SIZE]);
1806 luma_mc(s, tmp, tmpstride, ref0->frame,
1807 ¤t_mv.mv[0], x0, y0, nPbW, nPbH);
1808 luma_mc(s, tmp2, tmpstride, ref1->frame,
1809 ¤t_mv.mv[1], x0, y0, nPbW, nPbH);
1811 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1812 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1813 s->hevcdsp.weighted_pred_avg(s->sh.luma_log2_weight_denom,
1814 s->sh.luma_weight_l0[current_mv.ref_idx[0]],
1815 s->sh.luma_weight_l1[current_mv.ref_idx[1]],
1816 s->sh.luma_offset_l0[current_mv.ref_idx[0]],
1817 s->sh.luma_offset_l1[current_mv.ref_idx[1]],
1818 dst0, s->frame->linesize[0],
1819 tmp, tmp2, tmpstride, nPbW, nPbH);
1821 s->hevcdsp.put_weighted_pred_avg(dst0, s->frame->linesize[0],
1822 tmp, tmp2, tmpstride, nPbW, nPbH);
1825 chroma_mc(s, tmp, tmp2, tmpstride, ref0->frame,
1826 ¤t_mv.mv[0], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1827 chroma_mc(s, tmp3, tmp4, tmpstride, ref1->frame,
1828 ¤t_mv.mv[1], x0 / 2, y0 / 2, nPbW / 2, nPbH / 2);
1830 if ((s->sh.slice_type == P_SLICE && s->pps->weighted_pred_flag) ||
1831 (s->sh.slice_type == B_SLICE && s->pps->weighted_bipred_flag)) {
1832 s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1833 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][0],
1834 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][0],
1835 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][0],
1836 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][0],
1837 dst1, s->frame->linesize[1], tmp, tmp3,
1838 tmpstride, nPbW / 2, nPbH / 2);
1839 s->hevcdsp.weighted_pred_avg(s->sh.chroma_log2_weight_denom,
1840 s->sh.chroma_weight_l0[current_mv.ref_idx[0]][1],
1841 s->sh.chroma_weight_l1[current_mv.ref_idx[1]][1],
1842 s->sh.chroma_offset_l0[current_mv.ref_idx[0]][1],
1843 s->sh.chroma_offset_l1[current_mv.ref_idx[1]][1],
1844 dst2, s->frame->linesize[2], tmp2, tmp4,
1845 tmpstride, nPbW / 2, nPbH / 2);
1847 s->hevcdsp.put_weighted_pred_avg(dst1, s->frame->linesize[1], tmp, tmp3, tmpstride, nPbW/2, nPbH/2);
1848 s->hevcdsp.put_weighted_pred_avg(dst2, s->frame->linesize[2], tmp2, tmp4, tmpstride, nPbW/2, nPbH/2);
1856 static int luma_intra_pred_mode(HEVCContext *s, int x0, int y0, int pu_size,
1857 int prev_intra_luma_pred_flag)
1859 HEVCLocalContext *lc = &s->HEVClc;
1860 int x_pu = x0 >> s->sps->log2_min_pu_size;
1861 int y_pu = y0 >> s->sps->log2_min_pu_size;
1862 int min_pu_width = s->sps->min_pu_width;
1863 int size_in_pus = pu_size >> s->sps->log2_min_pu_size;
1864 int x0b = x0 & ((1 << s->sps->log2_ctb_size) - 1);
1865 int y0b = y0 & ((1 << s->sps->log2_ctb_size) - 1);
1867 int cand_up = (lc->ctb_up_flag || y0b) ?
1868 s->tab_ipm[(y_pu - 1) * min_pu_width + x_pu] : INTRA_DC;
1869 int cand_left = (lc->ctb_left_flag || x0b) ?
1870 s->tab_ipm[y_pu * min_pu_width + x_pu - 1] : INTRA_DC;
1872 int y_ctb = (y0 >> (s->sps->log2_ctb_size)) << (s->sps->log2_ctb_size);
1874 MvField *tab_mvf = s->ref->tab_mvf;
1875 int intra_pred_mode;
1879 // intra_pred_mode prediction does not cross vertical CTB boundaries
1880 if ((y0 - 1) < y_ctb)
1883 if (cand_left == cand_up) {
1884 if (cand_left < 2) {
1885 candidate[0] = INTRA_PLANAR;
1886 candidate[1] = INTRA_DC;
1887 candidate[2] = INTRA_ANGULAR_26;
1889 candidate[0] = cand_left;
1890 candidate[1] = 2 + ((cand_left - 2 - 1 + 32) & 31);
1891 candidate[2] = 2 + ((cand_left - 2 + 1) & 31);
1894 candidate[0] = cand_left;
1895 candidate[1] = cand_up;
1896 if (candidate[0] != INTRA_PLANAR && candidate[1] != INTRA_PLANAR) {
1897 candidate[2] = INTRA_PLANAR;
1898 } else if (candidate[0] != INTRA_DC && candidate[1] != INTRA_DC) {
1899 candidate[2] = INTRA_DC;
1901 candidate[2] = INTRA_ANGULAR_26;
1905 if (prev_intra_luma_pred_flag) {
1906 intra_pred_mode = candidate[lc->pu.mpm_idx];
1908 if (candidate[0] > candidate[1])
1909 FFSWAP(uint8_t, candidate[0], candidate[1]);
1910 if (candidate[0] > candidate[2])
1911 FFSWAP(uint8_t, candidate[0], candidate[2]);
1912 if (candidate[1] > candidate[2])
1913 FFSWAP(uint8_t, candidate[1], candidate[2]);
1915 intra_pred_mode = lc->pu.rem_intra_luma_pred_mode;
1916 for (i = 0; i < 3; i++)
1917 if (intra_pred_mode >= candidate[i])
1921 /* write the intra prediction units into the mv array */
1924 for (i = 0; i < size_in_pus; i++) {
1925 memset(&s->tab_ipm[(y_pu + i) * min_pu_width + x_pu],
1926 intra_pred_mode, size_in_pus);
1928 for (j = 0; j < size_in_pus; j++) {
1929 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].is_intra = 1;
1930 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[0] = 0;
1931 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].pred_flag[1] = 0;
1932 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[0] = 0;
1933 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].ref_idx[1] = 0;
1934 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].x = 0;
1935 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[0].y = 0;
1936 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].x = 0;
1937 tab_mvf[(y_pu + j) * min_pu_width + x_pu + i].mv[1].y = 0;
1941 return intra_pred_mode;
1944 static av_always_inline void set_ct_depth(HEVCContext *s, int x0, int y0,
1945 int log2_cb_size, int ct_depth)
1947 int length = (1 << log2_cb_size) >> s->sps->log2_min_cb_size;
1948 int x_cb = x0 >> s->sps->log2_min_cb_size;
1949 int y_cb = y0 >> s->sps->log2_min_cb_size;
1952 for (y = 0; y < length; y++)
1953 memset(&s->tab_ct_depth[(y_cb + y) * s->sps->min_cb_width + x_cb],
1957 static void intra_prediction_unit(HEVCContext *s, int x0, int y0,
1960 HEVCLocalContext *lc = &s->HEVClc;
1961 static const uint8_t intra_chroma_table[4] = { 0, 26, 10, 1 };
1962 uint8_t prev_intra_luma_pred_flag[4];
1963 int split = lc->cu.part_mode == PART_NxN;
1964 int pb_size = (1 << log2_cb_size) >> split;
1965 int side = split + 1;
1969 for (i = 0; i < side; i++)
1970 for (j = 0; j < side; j++)
1971 prev_intra_luma_pred_flag[2 * i + j] = ff_hevc_prev_intra_luma_pred_flag_decode(s);
1973 for (i = 0; i < side; i++) {
1974 for (j = 0; j < side; j++) {
1975 if (prev_intra_luma_pred_flag[2 * i + j])
1976 lc->pu.mpm_idx = ff_hevc_mpm_idx_decode(s);
1978 lc->pu.rem_intra_luma_pred_mode = ff_hevc_rem_intra_luma_pred_mode_decode(s);
1980 lc->pu.intra_pred_mode[2 * i + j] =
1981 luma_intra_pred_mode(s, x0 + pb_size * j, y0 + pb_size * i, pb_size,
1982 prev_intra_luma_pred_flag[2 * i + j]);
1986 chroma_mode = ff_hevc_intra_chroma_pred_mode_decode(s);
1987 if (chroma_mode != 4) {
1988 if (lc->pu.intra_pred_mode[0] == intra_chroma_table[chroma_mode])
1989 lc->pu.intra_pred_mode_c = 34;
1991 lc->pu.intra_pred_mode_c = intra_chroma_table[chroma_mode];
1993 lc->pu.intra_pred_mode_c = lc->pu.intra_pred_mode[0];
1997 static void intra_prediction_unit_default_value(HEVCContext *s,
2001 HEVCLocalContext *lc = &s->HEVClc;
2002 int pb_size = 1 << log2_cb_size;
2003 int size_in_pus = pb_size >> s->sps->log2_min_pu_size;
2004 int min_pu_width = s->sps->min_pu_width;
2005 MvField *tab_mvf = s->ref->tab_mvf;
2006 int x_pu = x0 >> s->sps->log2_min_pu_size;
2007 int y_pu = y0 >> s->sps->log2_min_pu_size;
2010 if (size_in_pus == 0)
2012 for (j = 0; j < size_in_pus; j++) {
2013 memset(&s->tab_ipm[(y_pu + j) * min_pu_width + x_pu], INTRA_DC, size_in_pus);
2014 for (k = 0; k < size_in_pus; k++)
2015 tab_mvf[(y_pu + j) * min_pu_width + x_pu + k].is_intra = lc->cu.pred_mode == MODE_INTRA;
2019 static int hls_coding_unit(HEVCContext *s, int x0, int y0, int log2_cb_size)
2021 int cb_size = 1 << log2_cb_size;
2022 HEVCLocalContext *lc = &s->HEVClc;
2023 int log2_min_cb_size = s->sps->log2_min_cb_size;
2024 int length = cb_size >> log2_min_cb_size;
2025 int min_cb_width = s->sps->min_cb_width;
2026 int x_cb = x0 >> log2_min_cb_size;
2027 int y_cb = y0 >> log2_min_cb_size;
2032 lc->cu.pred_mode = MODE_INTRA;
2033 lc->cu.part_mode = PART_2Nx2N;
2034 lc->cu.intra_split_flag = 0;
2036 SAMPLE_CTB(s->skip_flag, x_cb, y_cb) = 0;
2037 for (x = 0; x < 4; x++)
2038 lc->pu.intra_pred_mode[x] = 1;
2039 if (s->pps->transquant_bypass_enable_flag) {
2040 lc->cu.cu_transquant_bypass_flag = ff_hevc_cu_transquant_bypass_flag_decode(s);
2041 if (lc->cu.cu_transquant_bypass_flag)
2042 set_deblocking_bypass(s, x0, y0, log2_cb_size);
2044 lc->cu.cu_transquant_bypass_flag = 0;
2046 if (s->sh.slice_type != I_SLICE) {
2047 uint8_t skip_flag = ff_hevc_skip_flag_decode(s, x0, y0, x_cb, y_cb);
2049 x = y_cb * min_cb_width + x_cb;
2050 for (y = 0; y < length; y++) {
2051 memset(&s->skip_flag[x], skip_flag, length);
2054 lc->cu.pred_mode = skip_flag ? MODE_SKIP : MODE_INTER;
2057 if (SAMPLE_CTB(s->skip_flag, x_cb, y_cb)) {
2058 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
2059 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
2061 if (!s->sh.disable_deblocking_filter_flag)
2062 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
2066 if (s->sh.slice_type != I_SLICE)
2067 lc->cu.pred_mode = ff_hevc_pred_mode_decode(s);
2068 if (lc->cu.pred_mode != MODE_INTRA ||
2069 log2_cb_size == s->sps->log2_min_cb_size) {
2070 lc->cu.part_mode = ff_hevc_part_mode_decode(s, log2_cb_size);
2071 lc->cu.intra_split_flag = lc->cu.part_mode == PART_NxN &&
2072 lc->cu.pred_mode == MODE_INTRA;
2075 if (lc->cu.pred_mode == MODE_INTRA) {
2076 if (lc->cu.part_mode == PART_2Nx2N && s->sps->pcm_enabled_flag &&
2077 log2_cb_size >= s->sps->pcm.log2_min_pcm_cb_size &&
2078 log2_cb_size <= s->sps->pcm.log2_max_pcm_cb_size) {
2079 pcm_flag = ff_hevc_pcm_flag_decode(s);
2082 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
2083 ret = hls_pcm_sample(s, x0, y0, log2_cb_size);
2084 if (s->sps->pcm.loop_filter_disable_flag)
2085 set_deblocking_bypass(s, x0, y0, log2_cb_size);
2090 intra_prediction_unit(s, x0, y0, log2_cb_size);
2093 intra_prediction_unit_default_value(s, x0, y0, log2_cb_size);
2094 switch (lc->cu.part_mode) {
2096 hls_prediction_unit(s, x0, y0, cb_size, cb_size, log2_cb_size, 0);
2099 hls_prediction_unit(s, x0, y0, cb_size, cb_size / 2, log2_cb_size, 0);
2100 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size, cb_size / 2, log2_cb_size, 1);
2103 hls_prediction_unit(s, x0, y0, cb_size / 2, cb_size, log2_cb_size, 0);
2104 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size, log2_cb_size, 1);
2107 hls_prediction_unit(s, x0, y0, cb_size, cb_size / 4, log2_cb_size, 0);
2108 hls_prediction_unit(s, x0, y0 + cb_size / 4, cb_size, cb_size * 3 / 4, log2_cb_size, 1);
2111 hls_prediction_unit(s, x0, y0, cb_size, cb_size * 3 / 4, log2_cb_size, 0);
2112 hls_prediction_unit(s, x0, y0 + cb_size * 3 / 4, cb_size, cb_size / 4, log2_cb_size, 1);
2115 hls_prediction_unit(s, x0, y0, cb_size / 4, cb_size, log2_cb_size, 0);
2116 hls_prediction_unit(s, x0 + cb_size / 4, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 1);
2119 hls_prediction_unit(s, x0, y0, cb_size * 3 / 4, cb_size, log2_cb_size, 0);
2120 hls_prediction_unit(s, x0 + cb_size * 3 / 4, y0, cb_size / 4, cb_size, log2_cb_size, 1);
2123 hls_prediction_unit(s, x0, y0, cb_size / 2, cb_size / 2, log2_cb_size, 0);
2124 hls_prediction_unit(s, x0 + cb_size / 2, y0, cb_size / 2, cb_size / 2, log2_cb_size, 1);
2125 hls_prediction_unit(s, x0, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 2);
2126 hls_prediction_unit(s, x0 + cb_size / 2, y0 + cb_size / 2, cb_size / 2, cb_size / 2, log2_cb_size, 3);
2132 int rqt_root_cbf = 1;
2134 if (lc->cu.pred_mode != MODE_INTRA &&
2135 !(lc->cu.part_mode == PART_2Nx2N && lc->pu.merge_flag)) {
2136 rqt_root_cbf = ff_hevc_no_residual_syntax_flag_decode(s);
2139 lc->cu.max_trafo_depth = lc->cu.pred_mode == MODE_INTRA ?
2140 s->sps->max_transform_hierarchy_depth_intra + lc->cu.intra_split_flag :
2141 s->sps->max_transform_hierarchy_depth_inter;
2142 ret = hls_transform_tree(s, x0, y0, x0, y0, x0, y0,
2144 log2_cb_size, 0, 0, 0, 0);
2148 if (!s->sh.disable_deblocking_filter_flag)
2149 ff_hevc_deblocking_boundary_strengths(s, x0, y0, log2_cb_size);
2154 if (s->pps->cu_qp_delta_enabled_flag && lc->tu.is_cu_qp_delta_coded == 0)
2155 ff_hevc_set_qPy(s, x0, y0, x0, y0, log2_cb_size);
2157 x = y_cb * min_cb_width + x_cb;
2158 for (y = 0; y < length; y++) {
2159 memset(&s->qp_y_tab[x], lc->qp_y, length);
2163 set_ct_depth(s, x0, y0, log2_cb_size, lc->ct.depth);
2168 static int hls_coding_quadtree(HEVCContext *s, int x0, int y0,
2169 int log2_cb_size, int cb_depth)
2171 HEVCLocalContext *lc = &s->HEVClc;
2172 const int cb_size = 1 << log2_cb_size;
2175 lc->ct.depth = cb_depth;
2176 if (x0 + cb_size <= s->sps->width &&
2177 y0 + cb_size <= s->sps->height &&
2178 log2_cb_size > s->sps->log2_min_cb_size) {
2179 split_cu = ff_hevc_split_coding_unit_flag_decode(s, cb_depth, x0, y0);
2181 split_cu = (log2_cb_size > s->sps->log2_min_cb_size);
2183 if (s->pps->cu_qp_delta_enabled_flag &&
2184 log2_cb_size >= s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
2185 lc->tu.is_cu_qp_delta_coded = 0;
2186 lc->tu.cu_qp_delta = 0;
2190 const int cb_size_split = cb_size >> 1;
2191 const int x1 = x0 + cb_size_split;
2192 const int y1 = y0 + cb_size_split;
2197 #define SUBDIVIDE(x, y) \
2199 if (x < s->sps->width && y < s->sps->height) { \
2200 int ret = hls_coding_quadtree(s, x, y, log2_cb_size, cb_depth);\
2211 int ret = hls_coding_unit(s, x0, y0, log2_cb_size);
2219 static void hls_decode_neighbour(HEVCContext *s, int x_ctb, int y_ctb,
2222 HEVCLocalContext *lc = &s->HEVClc;
2223 int ctb_size = 1 << s->sps->log2_ctb_size;
2224 int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2225 int ctb_addr_in_slice = ctb_addr_rs - s->sh.slice_addr;
2227 s->tab_slice_address[ctb_addr_rs] = s->sh.slice_addr;
2229 if (s->pps->entropy_coding_sync_enabled_flag) {
2230 if (x_ctb == 0 && (y_ctb & (ctb_size - 1)) == 0)
2231 lc->first_qp_group = 1;
2232 lc->end_of_tiles_x = s->sps->width;
2233 } else if (s->pps->tiles_enabled_flag) {
2234 if (ctb_addr_ts && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[ctb_addr_ts - 1]) {
2235 int idxX = s->pps->col_idxX[x_ctb >> s->sps->log2_ctb_size];
2236 lc->start_of_tiles_x = x_ctb;
2237 lc->end_of_tiles_x = x_ctb + (s->pps->column_width[idxX] << s->sps->log2_ctb_size);
2238 lc->first_qp_group = 1;
2241 lc->end_of_tiles_x = s->sps->width;
2244 lc->end_of_tiles_y = FFMIN(y_ctb + ctb_size, s->sps->height);
2246 lc->boundary_flags = 0;
2247 if (s->pps->tiles_enabled_flag) {
2248 if (x_ctb > 0 && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - 1]])
2249 lc->boundary_flags |= BOUNDARY_LEFT_TILE;
2250 if (x_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - 1])
2251 lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
2252 if (y_ctb > 0 && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]])
2253 lc->boundary_flags |= BOUNDARY_UPPER_TILE;
2254 if (y_ctb > 0 && s->tab_slice_address[ctb_addr_rs] != s->tab_slice_address[ctb_addr_rs - s->sps->ctb_width])
2255 lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
2257 if (!ctb_addr_in_slice > 0)
2258 lc->boundary_flags |= BOUNDARY_LEFT_SLICE;
2259 if (ctb_addr_in_slice < s->sps->ctb_width)
2260 lc->boundary_flags |= BOUNDARY_UPPER_SLICE;
2263 lc->ctb_left_flag = ((x_ctb > 0) && (ctb_addr_in_slice > 0) && !(lc->boundary_flags & BOUNDARY_LEFT_TILE));
2264 lc->ctb_up_flag = ((y_ctb > 0) && (ctb_addr_in_slice >= s->sps->ctb_width) && !(lc->boundary_flags & BOUNDARY_UPPER_TILE));
2265 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]]));
2266 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]]));
2269 static int hls_slice_data(HEVCContext *s)
2271 int ctb_size = 1 << s->sps->log2_ctb_size;
2275 int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[s->sh.slice_ctb_addr_rs];
2278 while (more_data && ctb_addr_ts < s->sps->ctb_size) {
2279 int ctb_addr_rs = s->pps->ctb_addr_ts_to_rs[ctb_addr_ts];
2281 x_ctb = (ctb_addr_rs % ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2282 y_ctb = (ctb_addr_rs / ((s->sps->width + ctb_size - 1) >> s->sps->log2_ctb_size)) << s->sps->log2_ctb_size;
2283 hls_decode_neighbour(s, x_ctb, y_ctb, ctb_addr_ts);
2285 ff_hevc_cabac_init(s, ctb_addr_ts);
2287 hls_sao_param(s, x_ctb >> s->sps->log2_ctb_size, y_ctb >> s->sps->log2_ctb_size);
2289 s->deblock[ctb_addr_rs].beta_offset = s->sh.beta_offset;
2290 s->deblock[ctb_addr_rs].tc_offset = s->sh.tc_offset;
2291 s->filter_slice_edges[ctb_addr_rs] = s->sh.slice_loop_filter_across_slices_enabled_flag;
2293 ret = hls_coding_quadtree(s, x_ctb, y_ctb, s->sps->log2_ctb_size, 0);
2296 more_data = !ff_hevc_end_of_slice_flag_decode(s);
2299 ff_hevc_save_states(s, ctb_addr_ts);
2300 ff_hevc_hls_filters(s, x_ctb, y_ctb, ctb_size);
2303 if (x_ctb + ctb_size >= s->sps->width &&
2304 y_ctb + ctb_size >= s->sps->height)
2305 ff_hevc_hls_filter(s, x_ctb, y_ctb);
2311 * @return AVERROR_INVALIDDATA if the packet is not a valid NAL unit,
2312 * 0 if the unit should be skipped, 1 otherwise
2314 static int hls_nal_unit(HEVCContext *s)
2316 GetBitContext *gb = &s->HEVClc.gb;
2319 if (get_bits1(gb) != 0)
2320 return AVERROR_INVALIDDATA;
2322 s->nal_unit_type = get_bits(gb, 6);
2324 nuh_layer_id = get_bits(gb, 6);
2325 s->temporal_id = get_bits(gb, 3) - 1;
2326 if (s->temporal_id < 0)
2327 return AVERROR_INVALIDDATA;
2329 av_log(s->avctx, AV_LOG_DEBUG,
2330 "nal_unit_type: %d, nuh_layer_id: %dtemporal_id: %d\n",
2331 s->nal_unit_type, nuh_layer_id, s->temporal_id);
2333 return nuh_layer_id == 0;
2336 static void restore_tqb_pixels(HEVCContext *s)
2338 int min_pu_size = 1 << s->sps->log2_min_pu_size;
2341 for (c_idx = 0; c_idx < 3; c_idx++) {
2342 ptrdiff_t stride = s->frame->linesize[c_idx];
2343 int hshift = s->sps->hshift[c_idx];
2344 int vshift = s->sps->vshift[c_idx];
2345 for (y = 0; y < s->sps->min_pu_height; y++) {
2346 for (x = 0; x < s->sps->min_pu_width; x++) {
2347 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
2349 int len = min_pu_size >> hshift;
2350 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)];
2351 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)];
2352 for (n = 0; n < (min_pu_size >> vshift); n++) {
2353 memcpy(dst, src, len);
2363 static int set_side_data(HEVCContext *s)
2365 AVFrame *out = s->ref->frame;
2367 if (s->sei_frame_packing_present &&
2368 s->frame_packing_arrangement_type >= 3 &&
2369 s->frame_packing_arrangement_type <= 5 &&
2370 s->content_interpretation_type > 0 &&
2371 s->content_interpretation_type < 3) {
2372 AVStereo3D *stereo = av_stereo3d_create_side_data(out);
2374 return AVERROR(ENOMEM);
2376 switch (s->frame_packing_arrangement_type) {
2378 if (s->quincunx_subsampling)
2379 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
2381 stereo->type = AV_STEREO3D_SIDEBYSIDE;
2384 stereo->type = AV_STEREO3D_TOPBOTTOM;
2387 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
2391 if (s->content_interpretation_type == 2)
2392 stereo->flags = AV_STEREO3D_FLAG_INVERT;
2395 if (s->sei_display_orientation_present &&
2396 (s->sei_anticlockwise_rotation || s->sei_hflip || s->sei_vflip)) {
2397 double angle = s->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
2398 AVFrameSideData *rotation = av_frame_new_side_data(out,
2399 AV_FRAME_DATA_DISPLAYMATRIX,
2400 sizeof(int32_t) * 9);
2402 return AVERROR(ENOMEM);
2404 av_display_rotation_set((int32_t *)rotation->data, angle);
2405 av_display_matrix_flip((int32_t *)rotation->data,
2406 s->sei_hflip, s->sei_vflip);
2412 static int hevc_frame_start(HEVCContext *s)
2414 HEVCLocalContext *lc = &s->HEVClc;
2417 memset(s->horizontal_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2418 memset(s->vertical_bs, 0, 2 * s->bs_width * (s->bs_height + 1));
2419 memset(s->cbf_luma, 0, s->sps->min_tb_width * s->sps->min_tb_height);
2420 memset(s->is_pcm, 0, s->sps->min_pu_width * s->sps->min_pu_height);
2422 lc->start_of_tiles_x = 0;
2424 s->first_nal_type = s->nal_unit_type;
2426 if (s->pps->tiles_enabled_flag)
2427 lc->end_of_tiles_x = s->pps->column_width[0] << s->sps->log2_ctb_size;
2429 ret = ff_hevc_set_new_ref(s, s->sps->sao_enabled ? &s->sao_frame : &s->frame,
2434 ret = ff_hevc_frame_rps(s);
2436 av_log(s->avctx, AV_LOG_ERROR, "Error constructing the frame RPS.\n");
2440 s->ref->frame->key_frame = IS_IRAP(s);
2442 ret = set_side_data(s);
2446 av_frame_unref(s->output_frame);
2447 ret = ff_hevc_output_frame(s, s->output_frame, 0);
2451 ff_thread_finish_setup(s->avctx);
2457 ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2462 static int decode_nal_unit(HEVCContext *s, const uint8_t *nal, int length)
2464 HEVCLocalContext *lc = &s->HEVClc;
2465 GetBitContext *gb = &lc->gb;
2466 int ctb_addr_ts, ret;
2468 ret = init_get_bits8(gb, nal, length);
2472 ret = hls_nal_unit(s);
2474 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit %d, skipping.\n",
2480 switch (s->nal_unit_type) {
2482 ret = ff_hevc_decode_nal_vps(s);
2487 ret = ff_hevc_decode_nal_sps(s);
2492 ret = ff_hevc_decode_nal_pps(s);
2496 case NAL_SEI_PREFIX:
2497 case NAL_SEI_SUFFIX:
2498 ret = ff_hevc_decode_nal_sei(s);
2509 case NAL_BLA_W_RADL:
2511 case NAL_IDR_W_RADL:
2518 ret = hls_slice_header(s);
2522 if (s->max_ra == INT_MAX) {
2523 if (s->nal_unit_type == NAL_CRA_NUT || IS_BLA(s)) {
2527 s->max_ra = INT_MIN;
2531 if ((s->nal_unit_type == NAL_RASL_R || s->nal_unit_type == NAL_RASL_N) &&
2532 s->poc <= s->max_ra) {
2536 if (s->nal_unit_type == NAL_RASL_R && s->poc > s->max_ra)
2537 s->max_ra = INT_MIN;
2540 if (s->sh.first_slice_in_pic_flag) {
2541 ret = hevc_frame_start(s);
2544 } else if (!s->ref) {
2545 av_log(s->avctx, AV_LOG_ERROR, "First slice in a frame missing.\n");
2549 if (s->nal_unit_type != s->first_nal_type) {
2550 av_log(s->avctx, AV_LOG_ERROR,
2551 "Non-matching NAL types of the VCL NALUs: %d %d\n",
2552 s->first_nal_type, s->nal_unit_type);
2553 return AVERROR_INVALIDDATA;
2556 if (!s->sh.dependent_slice_segment_flag &&
2557 s->sh.slice_type != I_SLICE) {
2558 ret = ff_hevc_slice_rpl(s);
2560 av_log(s->avctx, AV_LOG_WARNING,
2561 "Error constructing the reference lists for the current slice.\n");
2566 ctb_addr_ts = hls_slice_data(s);
2567 if (ctb_addr_ts >= (s->sps->ctb_width * s->sps->ctb_height)) {
2569 if ((s->pps->transquant_bypass_enable_flag ||
2570 (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) &&
2571 s->sps->sao_enabled)
2572 restore_tqb_pixels(s);
2575 if (ctb_addr_ts < 0) {
2582 s->seq_decode = (s->seq_decode + 1) & 0xff;
2583 s->max_ra = INT_MAX;
2589 av_log(s->avctx, AV_LOG_INFO,
2590 "Skipping NAL unit %d\n", s->nal_unit_type);
2595 if (s->avctx->err_recognition & AV_EF_EXPLODE)
2600 /* FIXME: This is adapted from ff_h264_decode_nal, avoiding duplication
2601 * between these functions would be nice. */
2602 static int extract_rbsp(const uint8_t *src, int length,
2608 #define STARTCODE_TEST \
2609 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
2610 if (src[i + 2] != 3) { \
2611 /* startcode, so we must be past the end */ \
2616 #if HAVE_FAST_UNALIGNED
2617 #define FIND_FIRST_ZERO \
2618 if (i > 0 && !src[i]) \
2623 for (i = 0; i + 1 < length; i += 9) {
2624 if (!((~AV_RN64A(src + i) &
2625 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
2626 0x8000800080008080ULL))
2633 for (i = 0; i + 1 < length; i += 5) {
2634 if (!((~AV_RN32A(src + i) &
2635 (AV_RN32A(src + i) - 0x01000101U)) &
2642 #endif /* HAVE_FAST_64BIT */
2644 for (i = 0; i + 1 < length; i += 2) {
2647 if (i > 0 && src[i - 1] == 0)
2651 #endif /* HAVE_FAST_UNALIGNED */
2653 if (i >= length - 1) { // no escaped 0
2659 av_fast_malloc(&nal->rbsp_buffer, &nal->rbsp_buffer_size,
2660 length + FF_INPUT_BUFFER_PADDING_SIZE);
2661 if (!nal->rbsp_buffer)
2662 return AVERROR(ENOMEM);
2664 dst = nal->rbsp_buffer;
2666 memcpy(dst, src, i);
2668 while (si + 2 < length) {
2669 // remove escapes (very rare 1:2^22)
2670 if (src[si + 2] > 3) {
2671 dst[di++] = src[si++];
2672 dst[di++] = src[si++];
2673 } else if (src[si] == 0 && src[si + 1] == 0) {
2674 if (src[si + 2] == 3) { // escape
2680 } else // next start code
2684 dst[di++] = src[si++];
2687 dst[di++] = src[si++];
2690 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2697 static int decode_nal_units(HEVCContext *s, const uint8_t *buf, int length)
2699 int i, consumed, ret = 0;
2704 /* split the input packet into NAL units, so we know the upper bound on the
2705 * number of slices in the frame */
2707 while (length >= 4) {
2709 int extract_length = 0;
2713 for (i = 0; i < s->nal_length_size; i++)
2714 extract_length = (extract_length << 8) | buf[i];
2715 buf += s->nal_length_size;
2716 length -= s->nal_length_size;
2718 if (extract_length > length) {
2719 av_log(s->avctx, AV_LOG_ERROR, "Invalid NAL unit size.\n");
2720 ret = AVERROR_INVALIDDATA;
2729 if (buf[0] != 0 || buf[1] != 0 || buf[2] != 1) {
2730 ret = AVERROR_INVALIDDATA;
2736 extract_length = length;
2739 if (s->nals_allocated < s->nb_nals + 1) {
2740 int new_size = s->nals_allocated + 1;
2741 HEVCNAL *tmp = av_realloc_array(s->nals, new_size, sizeof(*tmp));
2743 ret = AVERROR(ENOMEM);
2747 memset(s->nals + s->nals_allocated, 0,
2748 (new_size - s->nals_allocated) * sizeof(*tmp));
2749 s->nals_allocated = new_size;
2751 nal = &s->nals[s->nb_nals++];
2753 consumed = extract_rbsp(buf, extract_length, nal);
2759 ret = init_get_bits8(&s->HEVClc.gb, nal->data, nal->size);
2764 if (s->nal_unit_type == NAL_EOB_NUT ||
2765 s->nal_unit_type == NAL_EOS_NUT)
2772 /* parse the NAL units */
2773 for (i = 0; i < s->nb_nals; i++) {
2774 int ret = decode_nal_unit(s, s->nals[i].data, s->nals[i].size);
2776 av_log(s->avctx, AV_LOG_WARNING,
2777 "Error parsing NAL unit #%d.\n", i);
2784 ff_thread_report_progress(&s->ref->tf, INT_MAX, 0);
2789 static void print_md5(void *log_ctx, int level, uint8_t md5[16])
2792 for (i = 0; i < 16; i++)
2793 av_log(log_ctx, level, "%02"PRIx8, md5[i]);
2796 static int verify_md5(HEVCContext *s, AVFrame *frame)
2798 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
2803 return AVERROR(EINVAL);
2805 pixel_shift = desc->comp[0].depth_minus1 > 7;
2807 av_log(s->avctx, AV_LOG_DEBUG, "Verifying checksum for frame with POC %d: ",
2810 /* the checksums are LE, so we have to byteswap for >8bpp formats
2813 if (pixel_shift && !s->checksum_buf) {
2814 av_fast_malloc(&s->checksum_buf, &s->checksum_buf_size,
2815 FFMAX3(frame->linesize[0], frame->linesize[1],
2816 frame->linesize[2]));
2817 if (!s->checksum_buf)
2818 return AVERROR(ENOMEM);
2822 for (i = 0; frame->data[i]; i++) {
2823 int width = s->avctx->coded_width;
2824 int height = s->avctx->coded_height;
2825 int w = (i == 1 || i == 2) ? (width >> desc->log2_chroma_w) : width;
2826 int h = (i == 1 || i == 2) ? (height >> desc->log2_chroma_h) : height;
2829 av_md5_init(s->md5_ctx);
2830 for (j = 0; j < h; j++) {
2831 const uint8_t *src = frame->data[i] + j * frame->linesize[i];
2834 s->bdsp.bswap16_buf((uint16_t *) s->checksum_buf,
2835 (const uint16_t *) src, w);
2836 src = s->checksum_buf;
2839 av_md5_update(s->md5_ctx, src, w << pixel_shift);
2841 av_md5_final(s->md5_ctx, md5);
2843 if (!memcmp(md5, s->md5[i], 16)) {
2844 av_log (s->avctx, AV_LOG_DEBUG, "plane %d - correct ", i);
2845 print_md5(s->avctx, AV_LOG_DEBUG, md5);
2846 av_log (s->avctx, AV_LOG_DEBUG, "; ");
2848 av_log (s->avctx, AV_LOG_ERROR, "mismatching checksum of plane %d - ", i);
2849 print_md5(s->avctx, AV_LOG_ERROR, md5);
2850 av_log (s->avctx, AV_LOG_ERROR, " != ");
2851 print_md5(s->avctx, AV_LOG_ERROR, s->md5[i]);
2852 av_log (s->avctx, AV_LOG_ERROR, "\n");
2853 return AVERROR_INVALIDDATA;
2857 av_log(s->avctx, AV_LOG_DEBUG, "\n");
2862 static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
2866 HEVCContext *s = avctx->priv_data;
2869 ret = ff_hevc_output_frame(s, data, 1);
2878 ret = decode_nal_units(s, avpkt->data, avpkt->size);
2882 /* verify the SEI checksum */
2883 if (avctx->err_recognition & AV_EF_CRCCHECK && s->is_decoded &&
2885 ret = verify_md5(s, s->ref->frame);
2886 if (ret < 0 && avctx->err_recognition & AV_EF_EXPLODE) {
2887 ff_hevc_unref_frame(s, s->ref, ~0);
2893 if (s->is_decoded) {
2894 av_log(avctx, AV_LOG_DEBUG, "Decoded frame with POC %d.\n", s->poc);
2898 if (s->output_frame->buf[0]) {
2899 av_frame_move_ref(data, s->output_frame);
2906 static int hevc_ref_frame(HEVCContext *s, HEVCFrame *dst, HEVCFrame *src)
2908 int ret = ff_thread_ref_frame(&dst->tf, &src->tf);
2912 dst->tab_mvf_buf = av_buffer_ref(src->tab_mvf_buf);
2913 if (!dst->tab_mvf_buf)
2915 dst->tab_mvf = src->tab_mvf;
2917 dst->rpl_tab_buf = av_buffer_ref(src->rpl_tab_buf);
2918 if (!dst->rpl_tab_buf)
2920 dst->rpl_tab = src->rpl_tab;
2922 dst->rpl_buf = av_buffer_ref(src->rpl_buf);
2926 dst->poc = src->poc;
2927 dst->ctb_count = src->ctb_count;
2928 dst->window = src->window;
2929 dst->flags = src->flags;
2930 dst->sequence = src->sequence;
2934 ff_hevc_unref_frame(s, dst, ~0);
2935 return AVERROR(ENOMEM);
2938 static av_cold int hevc_decode_free(AVCodecContext *avctx)
2940 HEVCContext *s = avctx->priv_data;
2945 av_freep(&s->md5_ctx);
2947 av_frame_free(&s->tmp_frame);
2948 av_frame_free(&s->output_frame);
2950 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2951 ff_hevc_unref_frame(s, &s->DPB[i], ~0);
2952 av_frame_free(&s->DPB[i].frame);
2955 for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++)
2956 av_buffer_unref(&s->vps_list[i]);
2957 for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++)
2958 av_buffer_unref(&s->sps_list[i]);
2959 for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++)
2960 av_buffer_unref(&s->pps_list[i]);
2962 for (i = 0; i < s->nals_allocated; i++)
2963 av_freep(&s->nals[i].rbsp_buffer);
2965 s->nals_allocated = 0;
2970 static av_cold int hevc_init_context(AVCodecContext *avctx)
2972 HEVCContext *s = avctx->priv_data;
2977 s->tmp_frame = av_frame_alloc();
2981 s->output_frame = av_frame_alloc();
2982 if (!s->output_frame)
2985 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
2986 s->DPB[i].frame = av_frame_alloc();
2987 if (!s->DPB[i].frame)
2989 s->DPB[i].tf.f = s->DPB[i].frame;
2992 s->max_ra = INT_MAX;
2994 s->md5_ctx = av_md5_alloc();
2998 ff_bswapdsp_init(&s->bdsp);
3000 s->context_initialized = 1;
3005 hevc_decode_free(avctx);
3006 return AVERROR(ENOMEM);
3009 static int hevc_update_thread_context(AVCodecContext *dst,
3010 const AVCodecContext *src)
3012 HEVCContext *s = dst->priv_data;
3013 HEVCContext *s0 = src->priv_data;
3016 if (!s->context_initialized) {
3017 ret = hevc_init_context(dst);
3022 for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
3023 ff_hevc_unref_frame(s, &s->DPB[i], ~0);
3024 if (s0->DPB[i].frame->buf[0]) {
3025 ret = hevc_ref_frame(s, &s->DPB[i], &s0->DPB[i]);
3031 for (i = 0; i < FF_ARRAY_ELEMS(s->vps_list); i++) {
3032 av_buffer_unref(&s->vps_list[i]);
3033 if (s0->vps_list[i]) {
3034 s->vps_list[i] = av_buffer_ref(s0->vps_list[i]);
3035 if (!s->vps_list[i])
3036 return AVERROR(ENOMEM);
3040 for (i = 0; i < FF_ARRAY_ELEMS(s->sps_list); i++) {
3041 av_buffer_unref(&s->sps_list[i]);
3042 if (s0->sps_list[i]) {
3043 s->sps_list[i] = av_buffer_ref(s0->sps_list[i]);
3044 if (!s->sps_list[i])
3045 return AVERROR(ENOMEM);
3049 for (i = 0; i < FF_ARRAY_ELEMS(s->pps_list); i++) {
3050 av_buffer_unref(&s->pps_list[i]);
3051 if (s0->pps_list[i]) {
3052 s->pps_list[i] = av_buffer_ref(s0->pps_list[i]);
3053 if (!s->pps_list[i])
3054 return AVERROR(ENOMEM);
3058 if (s->sps != s0->sps)
3059 ret = set_sps(s, s0->sps);
3061 s->seq_decode = s0->seq_decode;
3062 s->seq_output = s0->seq_output;
3063 s->pocTid0 = s0->pocTid0;
3064 s->max_ra = s0->max_ra;
3066 s->is_nalff = s0->is_nalff;
3067 s->nal_length_size = s0->nal_length_size;
3070 s->seq_decode = (s->seq_decode + 1) & 0xff;
3071 s->max_ra = INT_MAX;
3077 static int hevc_decode_extradata(HEVCContext *s)
3079 AVCodecContext *avctx = s->avctx;
3083 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
3085 if (avctx->extradata_size > 3 &&
3086 (avctx->extradata[0] || avctx->extradata[1] ||
3087 avctx->extradata[2] > 1)) {
3088 /* It seems the extradata is encoded as hvcC format.
3089 * Temporarily, we support configurationVersion==0 until 14496-15 3rd
3090 * is finalized. When finalized, configurationVersion will be 1 and we
3091 * can recognize hvcC by checking if avctx->extradata[0]==1 or not. */
3092 int i, j, num_arrays, nal_len_size;
3096 bytestream2_skip(&gb, 21);
3097 nal_len_size = (bytestream2_get_byte(&gb) & 3) + 1;
3098 num_arrays = bytestream2_get_byte(&gb);
3100 /* nal units in the hvcC always have length coded with 2 bytes,
3101 * so put a fake nal_length_size = 2 while parsing them */
3102 s->nal_length_size = 2;
3104 /* Decode nal units from hvcC. */
3105 for (i = 0; i < num_arrays; i++) {
3106 int type = bytestream2_get_byte(&gb) & 0x3f;
3107 int cnt = bytestream2_get_be16(&gb);
3109 for (j = 0; j < cnt; j++) {
3110 // +2 for the nal size field
3111 int nalsize = bytestream2_peek_be16(&gb) + 2;
3112 if (bytestream2_get_bytes_left(&gb) < nalsize) {
3113 av_log(s->avctx, AV_LOG_ERROR,
3114 "Invalid NAL unit size in extradata.\n");
3115 return AVERROR_INVALIDDATA;
3118 ret = decode_nal_units(s, gb.buffer, nalsize);
3120 av_log(avctx, AV_LOG_ERROR,
3121 "Decoding nal unit %d %d from hvcC failed\n",
3125 bytestream2_skip(&gb, nalsize);
3129 /* Now store right nal length size, that will be used to parse
3131 s->nal_length_size = nal_len_size;
3134 ret = decode_nal_units(s, avctx->extradata, avctx->extradata_size);
3141 static av_cold int hevc_decode_init(AVCodecContext *avctx)
3143 HEVCContext *s = avctx->priv_data;
3146 ff_init_cabac_states();
3148 avctx->internal->allocate_progress = 1;
3150 ret = hevc_init_context(avctx);
3154 if (avctx->extradata_size > 0 && avctx->extradata) {
3155 ret = hevc_decode_extradata(s);
3157 hevc_decode_free(avctx);
3165 static av_cold int hevc_init_thread_copy(AVCodecContext *avctx)
3167 HEVCContext *s = avctx->priv_data;
3170 memset(s, 0, sizeof(*s));
3172 ret = hevc_init_context(avctx);
3179 static void hevc_decode_flush(AVCodecContext *avctx)
3181 HEVCContext *s = avctx->priv_data;
3182 ff_hevc_flush_dpb(s);
3183 s->max_ra = INT_MAX;
3186 #define OFFSET(x) offsetof(HEVCContext, x)
3187 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
3189 static const AVProfile profiles[] = {
3190 { FF_PROFILE_HEVC_MAIN, "Main" },
3191 { FF_PROFILE_HEVC_MAIN_10, "Main 10" },
3192 { FF_PROFILE_HEVC_MAIN_STILL_PICTURE, "Main Still Picture" },
3193 { FF_PROFILE_UNKNOWN },
3196 static const AVOption options[] = {
3197 { "apply_defdispwin", "Apply default display window from VUI", OFFSET(apply_defdispwin),
3198 AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, PAR },
3202 static const AVClass hevc_decoder_class = {
3203 .class_name = "HEVC decoder",
3204 .item_name = av_default_item_name,
3206 .version = LIBAVUTIL_VERSION_INT,
3209 AVCodec ff_hevc_decoder = {
3211 .long_name = NULL_IF_CONFIG_SMALL("HEVC (High Efficiency Video Coding)"),
3212 .type = AVMEDIA_TYPE_VIDEO,
3213 .id = AV_CODEC_ID_HEVC,
3214 .priv_data_size = sizeof(HEVCContext),
3215 .priv_class = &hevc_decoder_class,
3216 .init = hevc_decode_init,
3217 .close = hevc_decode_free,
3218 .decode = hevc_decode_frame,
3219 .flush = hevc_decode_flush,
3220 .update_thread_context = hevc_update_thread_context,
3221 .init_thread_copy = hevc_init_thread_copy,
3222 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY |
3223 CODEC_CAP_FRAME_THREADS,
3224 .profiles = NULL_IF_CONFIG_SMALL(profiles),