2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #define UNCHECKED_BITSTREAM_READER 1
30 #include "libavutil/avassert.h"
31 #include "libavutil/display.h"
32 #include "libavutil/imgutils.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/stereo3d.h"
35 #include "libavutil/video_enc_params.h"
38 #include "bytestream.h"
40 #include "cabac_functions.h"
41 #include "error_resilience.h"
45 #include "h2645_parse.h"
47 #include "h264chroma.h"
48 #include "h264_mvpred.h"
54 #include "mpegutils.h"
56 #include "rectangle.h"
59 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
61 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
63 H264Context *h = avctx->priv_data;
64 return h && h->ps.sps ? h->ps.sps->num_reorder_frames : 0;
67 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
69 int mb_x, int mb_y, int mb_intra, int mb_skipped)
71 H264Context *h = opaque;
72 H264SliceContext *sl = &h->slice_ctx[0];
76 sl->mb_xy = mb_x + mb_y * h->mb_stride;
77 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
79 /* FIXME: It is possible albeit uncommon that slice references
80 * differ between slices. We take the easy approach and ignore
81 * it for now. If this turns out to have any relevance in
82 * practice then correct remapping should be added. */
83 if (ref >= sl->ref_count[0])
85 if (!sl->ref_list[0][ref].data[0]) {
86 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
89 if ((sl->ref_list[0][ref].reference&3) != 3) {
90 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
93 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
95 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
96 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
97 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
99 sl->mb_field_decoding_flag = 0;
100 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
103 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
106 AVCodecContext *avctx = h->avctx;
107 const AVFrame *src = h->cur_pic.f;
108 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
109 int vshift = desc->log2_chroma_h;
110 const int field_pic = h->picture_structure != PICT_FRAME;
116 height = FFMIN(height, avctx->height - y);
118 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
121 if (avctx->draw_horiz_band) {
122 int offset[AV_NUM_DATA_POINTERS];
125 offset[0] = y * src->linesize[0];
127 offset[2] = (y >> vshift) * src->linesize[1];
128 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
133 avctx->draw_horiz_band(avctx, src, offset,
134 y, h->picture_structure, height);
138 void ff_h264_free_tables(H264Context *h)
142 av_freep(&h->intra4x4_pred_mode);
143 av_freep(&h->chroma_pred_mode_table);
144 av_freep(&h->cbp_table);
145 av_freep(&h->mvd_table[0]);
146 av_freep(&h->mvd_table[1]);
147 av_freep(&h->direct_table);
148 av_freep(&h->non_zero_count);
149 av_freep(&h->slice_table_base);
150 h->slice_table = NULL;
151 av_freep(&h->list_counts);
153 av_freep(&h->mb2b_xy);
154 av_freep(&h->mb2br_xy);
156 av_buffer_pool_uninit(&h->qscale_table_pool);
157 av_buffer_pool_uninit(&h->mb_type_pool);
158 av_buffer_pool_uninit(&h->motion_val_pool);
159 av_buffer_pool_uninit(&h->ref_index_pool);
161 for (i = 0; i < h->nb_slice_ctx; i++) {
162 H264SliceContext *sl = &h->slice_ctx[i];
164 av_freep(&sl->dc_val_base);
165 av_freep(&sl->er.mb_index2xy);
166 av_freep(&sl->er.error_status_table);
167 av_freep(&sl->er.er_temp_buffer);
169 av_freep(&sl->bipred_scratchpad);
170 av_freep(&sl->edge_emu_buffer);
171 av_freep(&sl->top_borders[0]);
172 av_freep(&sl->top_borders[1]);
174 sl->bipred_scratchpad_allocated = 0;
175 sl->edge_emu_buffer_allocated = 0;
176 sl->top_borders_allocated[0] = 0;
177 sl->top_borders_allocated[1] = 0;
181 int ff_h264_alloc_tables(H264Context *h)
183 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
184 const int row_mb_num = 2*h->mb_stride*FFMAX(h->nb_slice_ctx, 1);
187 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
188 row_mb_num, 8 * sizeof(uint8_t), fail)
189 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
191 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
192 big_mb_num * 48 * sizeof(uint8_t), fail)
193 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
194 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
195 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
196 big_mb_num * sizeof(uint16_t), fail)
197 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
198 big_mb_num * sizeof(uint8_t), fail)
199 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[0],
200 row_mb_num, 16 * sizeof(uint8_t), fail);
201 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[1],
202 row_mb_num, 16 * sizeof(uint8_t), fail);
203 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
204 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
206 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
207 4 * big_mb_num * sizeof(uint8_t), fail);
208 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
209 big_mb_num * sizeof(uint8_t), fail)
211 memset(h->slice_table_base, -1,
212 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
213 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
215 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
216 big_mb_num * sizeof(uint32_t), fail);
217 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
218 big_mb_num * sizeof(uint32_t), fail);
219 for (y = 0; y < h->mb_height; y++)
220 for (x = 0; x < h->mb_width; x++) {
221 const int mb_xy = x + y * h->mb_stride;
222 const int b_xy = 4 * x + 4 * y * h->b_stride;
224 h->mb2b_xy[mb_xy] = b_xy;
225 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
231 return AVERROR(ENOMEM);
236 * Allocate buffers which are not shared amongst multiple threads.
238 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
240 ERContext *er = &sl->er;
241 int mb_array_size = h->mb_height * h->mb_stride;
242 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
243 int c_size = h->mb_stride * (h->mb_height + 1);
244 int yc_size = y_size + 2 * c_size;
247 sl->ref_cache[0][scan8[5] + 1] =
248 sl->ref_cache[0][scan8[7] + 1] =
249 sl->ref_cache[0][scan8[13] + 1] =
250 sl->ref_cache[1][scan8[5] + 1] =
251 sl->ref_cache[1][scan8[7] + 1] =
252 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
254 if (sl != h->slice_ctx) {
255 memset(er, 0, sizeof(*er));
256 } else if (CONFIG_ERROR_RESILIENCE) {
259 er->avctx = h->avctx;
260 er->decode_mb = h264_er_decode_mb;
262 er->quarter_sample = 1;
264 er->mb_num = h->mb_num;
265 er->mb_width = h->mb_width;
266 er->mb_height = h->mb_height;
267 er->mb_stride = h->mb_stride;
268 er->b8_stride = h->mb_width * 2 + 1;
270 // error resilience code looks cleaner with this
271 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
272 (h->mb_num + 1) * sizeof(int), fail);
274 for (y = 0; y < h->mb_height; y++)
275 for (x = 0; x < h->mb_width; x++)
276 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
278 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
279 h->mb_stride + h->mb_width;
281 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
282 mb_array_size * sizeof(uint8_t), fail);
284 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
285 h->mb_height * h->mb_stride * (4*sizeof(int) + 1), fail);
287 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
288 yc_size * sizeof(int16_t), fail);
289 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
290 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
291 er->dc_val[2] = er->dc_val[1] + c_size;
292 for (i = 0; i < yc_size; i++)
293 sl->dc_val_base[i] = 1024;
299 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
302 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
307 h->cur_chroma_format_idc = -1;
309 h->width_from_caller = avctx->width;
310 h->height_from_caller = avctx->height;
312 h->workaround_bugs = avctx->workaround_bugs;
313 h->flags = avctx->flags;
314 h->poc.prev_poc_msb = 1 << 16;
315 h->recovery_frame = -1;
316 h->frame_recovered = 0;
317 h->poc.prev_frame_num = -1;
318 h->sei.frame_packing.arrangement_cancel_flag = -1;
319 h->sei.unregistered.x264_build = -1;
321 h->next_outputed_poc = INT_MIN;
322 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
323 h->last_pocs[i] = INT_MIN;
325 ff_h264_sei_uninit(&h->sei);
327 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
328 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
331 return AVERROR(ENOMEM);
334 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
335 h->DPB[i].f = av_frame_alloc();
337 return AVERROR(ENOMEM);
340 h->cur_pic.f = av_frame_alloc();
342 return AVERROR(ENOMEM);
344 h->last_pic_for_ec.f = av_frame_alloc();
345 if (!h->last_pic_for_ec.f)
346 return AVERROR(ENOMEM);
348 for (i = 0; i < h->nb_slice_ctx; i++)
349 h->slice_ctx[i].h264 = h;
354 static av_cold int h264_decode_end(AVCodecContext *avctx)
356 H264Context *h = avctx->priv_data;
359 ff_h264_remove_all_refs(h);
360 ff_h264_free_tables(h);
362 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
363 ff_h264_unref_picture(h, &h->DPB[i]);
364 av_frame_free(&h->DPB[i].f);
366 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
368 h->cur_pic_ptr = NULL;
370 av_freep(&h->slice_ctx);
373 ff_h264_sei_uninit(&h->sei);
374 ff_h264_ps_uninit(&h->ps);
376 ff_h2645_packet_uninit(&h->pkt);
378 ff_h264_unref_picture(h, &h->cur_pic);
379 av_frame_free(&h->cur_pic.f);
380 ff_h264_unref_picture(h, &h->last_pic_for_ec);
381 av_frame_free(&h->last_pic_for_ec.f);
386 static AVOnce h264_vlc_init = AV_ONCE_INIT;
388 static av_cold int h264_decode_init(AVCodecContext *avctx)
390 H264Context *h = avctx->priv_data;
393 ret = h264_init_context(avctx, h);
397 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
399 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
400 return AVERROR_UNKNOWN;
403 if (avctx->ticks_per_frame == 1) {
404 if(h->avctx->time_base.den < INT_MAX/2) {
405 h->avctx->time_base.den *= 2;
407 h->avctx->time_base.num /= 2;
409 avctx->ticks_per_frame = 2;
411 if (!avctx->internal->is_copy) {
412 if (avctx->extradata_size > 0 && avctx->extradata) {
413 ret = ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
414 &h->ps, &h->is_avc, &h->nal_length_size,
415 avctx->err_recognition, avctx);
417 int explode = avctx->err_recognition & AV_EF_EXPLODE;
418 av_log(avctx, explode ? AV_LOG_ERROR: AV_LOG_WARNING,
419 "Error decoding the extradata\n");
428 if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
429 h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
430 h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
433 ff_h264_flush_change(h);
435 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
438 if (h->enable_er && (avctx->active_thread_type & FF_THREAD_SLICE)) {
439 av_log(avctx, AV_LOG_WARNING,
440 "Error resilience with slice threads is enabled. It is unsafe and unsupported and may crash. "
441 "Use it at your own risk\n");
448 * instantaneous decoder refresh.
450 static void idr(H264Context *h)
453 ff_h264_remove_all_refs(h);
454 h->poc.prev_frame_num =
455 h->poc.prev_frame_num_offset = 0;
456 h->poc.prev_poc_msb = 1<<16;
457 h->poc.prev_poc_lsb = -1;
458 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
459 h->last_pocs[i] = INT_MIN;
462 /* forget old pics after a seek */
463 void ff_h264_flush_change(H264Context *h)
467 h->next_outputed_poc = INT_MIN;
468 h->prev_interlaced_frame = 1;
471 h->poc.prev_frame_num = -1;
472 if (h->cur_pic_ptr) {
473 h->cur_pic_ptr->reference = 0;
474 for (j=i=0; h->delayed_pic[i]; i++)
475 if (h->delayed_pic[i] != h->cur_pic_ptr)
476 h->delayed_pic[j++] = h->delayed_pic[i];
477 h->delayed_pic[j] = NULL;
479 ff_h264_unref_picture(h, &h->last_pic_for_ec);
482 h->recovery_frame = -1;
483 h->frame_recovered = 0;
484 h->current_slice = 0;
488 static void h264_decode_flush(AVCodecContext *avctx)
490 H264Context *h = avctx->priv_data;
493 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
495 ff_h264_flush_change(h);
496 ff_h264_sei_uninit(&h->sei);
498 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
499 ff_h264_unref_picture(h, &h->DPB[i]);
500 h->cur_pic_ptr = NULL;
501 ff_h264_unref_picture(h, &h->cur_pic);
505 ff_h264_free_tables(h);
506 h->context_initialized = 0;
509 static int get_last_needed_nal(H264Context *h)
515 for (i = 0; i < h->pkt.nb_nals; i++) {
516 H2645NAL *nal = &h->pkt.nals[i];
519 /* packets can sometimes contain multiple PPS/SPS,
520 * e.g. two PAFF field pictures in one packet, or a demuxer
521 * which splits NALs strangely if so, when frame threading we
522 * can't start the next thread until we've read all of them */
529 case H264_NAL_IDR_SLICE:
531 ret = init_get_bits8(&gb, nal->data + 1, nal->size - 1);
533 av_log(h->avctx, AV_LOG_ERROR, "Invalid zero-sized VCL NAL unit\n");
534 if (h->avctx->err_recognition & AV_EF_EXPLODE)
539 if (!get_ue_golomb_long(&gb) || // first_mb_in_slice
541 first_slice != nal->type)
544 first_slice = nal->type;
551 static void debug_green_metadata(const H264SEIGreenMetaData *gm, void *logctx)
553 av_log(logctx, AV_LOG_DEBUG, "Green Metadata Info SEI message\n");
554 av_log(logctx, AV_LOG_DEBUG, " green_metadata_type: %d\n", gm->green_metadata_type);
556 if (gm->green_metadata_type == 0) {
557 av_log(logctx, AV_LOG_DEBUG, " green_metadata_period_type: %d\n", gm->period_type);
559 if (gm->period_type == 2)
560 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_seconds: %d\n", gm->num_seconds);
561 else if (gm->period_type == 3)
562 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_pictures: %d\n", gm->num_pictures);
564 av_log(logctx, AV_LOG_DEBUG, " SEI GREEN Complexity Metrics: %f %f %f %f\n",
565 (float)gm->percent_non_zero_macroblocks/255,
566 (float)gm->percent_intra_coded_macroblocks/255,
567 (float)gm->percent_six_tap_filtering/255,
568 (float)gm->percent_alpha_point_deblocking_instance/255);
570 } else if (gm->green_metadata_type == 1) {
571 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_type: %d\n", gm->xsd_metric_type);
573 if (gm->xsd_metric_type == 0)
574 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_value: %f\n",
575 (float)gm->xsd_metric_value/100);
579 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
581 AVCodecContext *const avctx = h->avctx;
582 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
589 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
590 h->current_slice = 0;
591 if (!h->first_field) {
592 h->cur_pic_ptr = NULL;
593 ff_h264_sei_uninit(&h->sei);
597 if (h->nal_length_size == 4) {
598 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
600 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
604 ret = ff_h2645_packet_split(&h->pkt, buf, buf_size, avctx, h->is_avc, h->nal_length_size,
605 avctx->codec_id, 0, 0);
607 av_log(avctx, AV_LOG_ERROR,
608 "Error splitting the input into NAL units.\n");
612 if (avctx->active_thread_type & FF_THREAD_FRAME)
613 nals_needed = get_last_needed_nal(h);
617 for (i = 0; i < h->pkt.nb_nals; i++) {
618 H2645NAL *nal = &h->pkt.nals[i];
619 int max_slice_ctx, err;
621 if (avctx->skip_frame >= AVDISCARD_NONREF &&
622 nal->ref_idc == 0 && nal->type != H264_NAL_SEI)
625 // FIXME these should stop being context-global variables
626 h->nal_ref_idc = nal->ref_idc;
627 h->nal_unit_type = nal->type;
631 case H264_NAL_IDR_SLICE:
632 if ((nal->data[1] & 0xFC) == 0x98) {
633 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
634 h->next_outputed_poc = INT_MIN;
639 idr(h); // FIXME ensure we don't lose some frames if there is reordering
642 h->has_recovery_point = 1;
646 if ((err = ff_h264_queue_decode_slice(h, nal))) {
647 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
648 sl->ref_count[0] = sl->ref_count[1] = 0;
652 if (h->current_slice == 1) {
653 if (avctx->active_thread_type & FF_THREAD_FRAME &&
654 i >= nals_needed && !h->setup_finished && h->cur_pic_ptr) {
655 ff_thread_finish_setup(avctx);
656 h->setup_finished = 1;
659 if (h->avctx->hwaccel &&
660 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
664 max_slice_ctx = avctx->hwaccel ? 1 : h->nb_slice_ctx;
665 if (h->nb_slice_ctx_queued == max_slice_ctx) {
666 if (h->avctx->hwaccel) {
667 ret = avctx->hwaccel->decode_slice(avctx, nal->raw_data, nal->raw_size);
668 h->nb_slice_ctx_queued = 0;
670 ret = ff_h264_execute_decode_slices(h);
671 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
678 avpriv_request_sample(avctx, "data partitioning");
681 ret = ff_h264_sei_decode(&h->sei, &nal->gb, &h->ps, avctx);
682 h->has_recovery_point = h->has_recovery_point || h->sei.recovery_point.recovery_frame_cnt != -1;
683 if (avctx->debug & FF_DEBUG_GREEN_MD)
684 debug_green_metadata(&h->sei.green_metadata, h->avctx);
685 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
689 GetBitContext tmp_gb = nal->gb;
690 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
691 ret = avctx->hwaccel->decode_params(avctx,
698 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
700 av_log(h->avctx, AV_LOG_DEBUG,
701 "SPS decoding failure, trying again with the complete NAL\n");
702 init_get_bits8(&tmp_gb, nal->raw_data + 1, nal->raw_size - 1);
703 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
705 ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps, 1);
709 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
710 ret = avctx->hwaccel->decode_params(avctx,
717 ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
719 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
723 case H264_NAL_END_SEQUENCE:
724 case H264_NAL_END_STREAM:
725 case H264_NAL_FILLER_DATA:
726 case H264_NAL_SPS_EXT:
727 case H264_NAL_AUXILIARY_SLICE:
730 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
731 nal->type, nal->size_bits);
735 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
739 ret = ff_h264_execute_decode_slices(h);
740 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
743 // set decode_error_flags to allow users to detect concealed decoding errors
744 if ((ret < 0 || h->slice_ctx->er.error_occurred) && h->cur_pic_ptr) {
745 h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
751 #if CONFIG_ERROR_RESILIENCE
753 * FIXME: Error handling code does not seem to support interlaced
754 * when slices span multiple rows
755 * The ff_er_add_slice calls don't work right for bottom
756 * fields; they cause massive erroneous error concealing
757 * Error marking covers both fields (top and bottom).
758 * This causes a mismatched s->error_count
759 * and a bad error table. Further, the error count goes to
760 * INT_MAX when called for bottom field, because mb_y is
761 * past end by one (callers fault) and resync_mb_y != 0
762 * causes problems for the first MB line, too.
764 if (!FIELD_PICTURE(h) && h->current_slice && h->enable_er) {
766 H264SliceContext *sl = h->slice_ctx;
767 int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
769 ff_h264_set_erpic(&sl->er.cur_pic, h->cur_pic_ptr);
772 ff_h264_set_erpic(&sl->er.last_pic, &h->last_pic_for_ec);
773 sl->ref_list[0][0].parent = &h->last_pic_for_ec;
774 memcpy(sl->ref_list[0][0].data, h->last_pic_for_ec.f->data, sizeof(sl->ref_list[0][0].data));
775 memcpy(sl->ref_list[0][0].linesize, h->last_pic_for_ec.f->linesize, sizeof(sl->ref_list[0][0].linesize));
776 sl->ref_list[0][0].reference = h->last_pic_for_ec.reference;
777 } else if (sl->ref_count[0]) {
778 ff_h264_set_erpic(&sl->er.last_pic, sl->ref_list[0][0].parent);
780 ff_h264_set_erpic(&sl->er.last_pic, NULL);
782 if (sl->ref_count[1])
783 ff_h264_set_erpic(&sl->er.next_pic, sl->ref_list[1][0].parent);
785 sl->er.ref_count = sl->ref_count[0];
787 ff_er_frame_end(&sl->er);
789 memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
791 #endif /* CONFIG_ERROR_RESILIENCE */
793 if (h->cur_pic_ptr && !h->droppable && h->has_slice) {
794 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
795 h->picture_structure == PICT_BOTTOM_FIELD);
798 return (ret < 0) ? ret : buf_size;
802 * Return the number of bytes consumed for building the current frame.
804 static int get_consumed_bytes(int pos, int buf_size)
807 pos = 1; // avoid infinite loops (I doubt that is needed but...)
808 if (pos + 10 > buf_size)
809 pos = buf_size; // oops ;)
814 static int h264_export_enc_params(AVFrame *f, H264Picture *p)
816 AVVideoEncParams *par;
817 unsigned int nb_mb = p->mb_height * p->mb_width;
820 par = av_video_enc_params_create_side_data(f, AV_VIDEO_ENC_PARAMS_H264, nb_mb);
822 return AVERROR(ENOMEM);
824 par->qp = p->pps->init_qp;
826 par->delta_qp[1][0] = p->pps->chroma_qp_index_offset[0];
827 par->delta_qp[1][1] = p->pps->chroma_qp_index_offset[0];
828 par->delta_qp[2][0] = p->pps->chroma_qp_index_offset[1];
829 par->delta_qp[2][1] = p->pps->chroma_qp_index_offset[1];
831 for (y = 0; y < p->mb_height; y++)
832 for (x = 0; x < p->mb_width; x++) {
833 const unsigned int block_idx = y * p->mb_width + x;
834 const unsigned int mb_xy = y * p->mb_stride + x;
835 AVVideoBlockParams *b = av_video_enc_params_block(par, block_idx);
842 b->delta_qp = p->qscale_table[mb_xy] - par->qp;
848 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
850 AVFrame *src = srcp->f;
853 ret = av_frame_ref(dst, src);
857 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.frame_packing), 0);
859 if (srcp->sei_recovery_frame_cnt == 0)
862 if (h->avctx->export_side_data & AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS) {
863 ret = h264_export_enc_params(dst, srcp);
874 static int is_extra(const uint8_t *buf, int buf_size)
876 int cnt= buf[5]&0x1f;
877 const uint8_t *p= buf+6;
881 int nalsize= AV_RB16(p) + 2;
882 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 7)
890 int nalsize= AV_RB16(p) + 2;
891 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 8)
898 static int finalize_frame(H264Context *h, AVFrame *dst, H264Picture *out, int *got_frame)
902 if (((h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
903 (h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
906 if (!h->avctx->hwaccel &&
907 (out->field_poc[0] == INT_MAX ||
908 out->field_poc[1] == INT_MAX)
912 int field = out->field_poc[0] == INT_MAX;
913 uint8_t *dst_data[4];
915 const uint8_t *src_data[4];
917 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
919 for (p = 0; p<4; p++) {
920 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
921 src_data[p] = f->data[p] + field *f->linesize[p];
922 linesizes[p] = 2*f->linesize[p];
925 av_image_copy(dst_data, linesizes, src_data, linesizes,
926 f->format, f->width, f->height>>1);
929 ret = output_frame(h, dst, out);
935 if (CONFIG_MPEGVIDEO) {
936 ff_print_debug_info2(h->avctx, dst, NULL,
941 h->mb_width, h->mb_height, h->mb_stride, 1);
948 static int send_next_delayed_frame(H264Context *h, AVFrame *dst_frame,
949 int *got_frame, int buf_index)
952 H264Picture *out = h->delayed_pic[0];
954 h->cur_pic_ptr = NULL;
960 !h->delayed_pic[i]->f->key_frame &&
961 !h->delayed_pic[i]->mmco_reset;
963 if (h->delayed_pic[i]->poc < out->poc) {
964 out = h->delayed_pic[i];
968 for (i = out_idx; h->delayed_pic[i]; i++)
969 h->delayed_pic[i] = h->delayed_pic[i + 1];
972 out->reference &= ~DELAYED_PIC_REF;
973 ret = finalize_frame(h, dst_frame, out, got_frame);
981 static int h264_decode_frame(AVCodecContext *avctx, void *data,
982 int *got_frame, AVPacket *avpkt)
984 const uint8_t *buf = avpkt->data;
985 int buf_size = avpkt->size;
986 H264Context *h = avctx->priv_data;
987 AVFrame *pict = data;
991 h->flags = avctx->flags;
992 h->setup_finished = 0;
993 h->nb_slice_ctx_queued = 0;
995 ff_h264_unref_picture(h, &h->last_pic_for_ec);
997 /* end of stream, output what is still in the buffers */
999 return send_next_delayed_frame(h, pict, got_frame, 0);
1001 if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
1003 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
1004 if (is_extra(side, side_size))
1005 ff_h264_decode_extradata(side, side_size,
1006 &h->ps, &h->is_avc, &h->nal_length_size,
1007 avctx->err_recognition, avctx);
1009 if (h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC) {
1010 if (is_extra(buf, buf_size))
1011 return ff_h264_decode_extradata(buf, buf_size,
1012 &h->ps, &h->is_avc, &h->nal_length_size,
1013 avctx->err_recognition, avctx);
1016 buf_index = decode_nal_units(h, buf, buf_size);
1018 return AVERROR_INVALIDDATA;
1020 if (!h->cur_pic_ptr && h->nal_unit_type == H264_NAL_END_SEQUENCE) {
1021 av_assert0(buf_index <= buf_size);
1022 return send_next_delayed_frame(h, pict, got_frame, buf_index);
1025 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && (!h->cur_pic_ptr || !h->has_slice)) {
1026 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1027 buf_size >= 4 && !memcmp("Q264", buf, 4))
1029 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1030 return AVERROR_INVALIDDATA;
1033 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1034 (h->mb_y >= h->mb_height && h->mb_height)) {
1035 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1038 /* Wait for second field. */
1039 if (h->next_output_pic) {
1040 ret = finalize_frame(h, pict, h->next_output_pic, got_frame);
1046 av_assert0(pict->buf[0] || !*got_frame);
1048 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1050 return get_consumed_bytes(buf_index, buf_size);
1053 #define OFFSET(x) offsetof(H264Context, x)
1054 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1055 static const AVOption h264_options[] = {
1056 { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0 },
1057 { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0 },
1058 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1059 { "x264_build", "Assume this x264 version if no x264 version found in any SEI", OFFSET(x264_build), AV_OPT_TYPE_INT, {.i64 = -1}, -1, INT_MAX, VD },
1063 static const AVClass h264_class = {
1064 .class_name = "H264 Decoder",
1065 .item_name = av_default_item_name,
1066 .option = h264_options,
1067 .version = LIBAVUTIL_VERSION_INT,
1070 AVCodec ff_h264_decoder = {
1072 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1073 .type = AVMEDIA_TYPE_VIDEO,
1074 .id = AV_CODEC_ID_H264,
1075 .priv_data_size = sizeof(H264Context),
1076 .init = h264_decode_init,
1077 .close = h264_decode_end,
1078 .decode = h264_decode_frame,
1079 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1080 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1081 AV_CODEC_CAP_FRAME_THREADS,
1082 .hw_configs = (const AVCodecHWConfigInternal*[]) {
1083 #if CONFIG_H264_DXVA2_HWACCEL
1084 HWACCEL_DXVA2(h264),
1086 #if CONFIG_H264_D3D11VA_HWACCEL
1087 HWACCEL_D3D11VA(h264),
1089 #if CONFIG_H264_D3D11VA2_HWACCEL
1090 HWACCEL_D3D11VA2(h264),
1092 #if CONFIG_H264_NVDEC_HWACCEL
1093 HWACCEL_NVDEC(h264),
1095 #if CONFIG_H264_VAAPI_HWACCEL
1096 HWACCEL_VAAPI(h264),
1098 #if CONFIG_H264_VDPAU_HWACCEL
1099 HWACCEL_VDPAU(h264),
1101 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
1102 HWACCEL_VIDEOTOOLBOX(h264),
1106 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_EXPORTS_CROPPING |
1107 FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
1108 .flush = h264_decode_flush,
1109 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1110 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1111 .priv_class = &h264_class,