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/timer.h"
37 #include "bytestream.h"
39 #include "cabac_functions.h"
40 #include "error_resilience.h"
44 #include "h2645_parse.h"
46 #include "h264chroma.h"
47 #include "h264_mvpred.h"
52 #include "mpegutils.h"
54 #include "rectangle.h"
56 #include "vdpau_compat.h"
58 static int h264_decode_end(AVCodecContext *avctx);
60 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
62 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
64 H264Context *h = avctx->priv_data;
65 return h && h->ps.sps ? h->ps.sps->num_reorder_frames : 0;
68 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
70 int mb_x, int mb_y, int mb_intra, int mb_skipped)
72 H264Context *h = opaque;
73 H264SliceContext *sl = &h->slice_ctx[0];
77 sl->mb_xy = mb_x + mb_y * h->mb_stride;
78 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
80 /* FIXME: It is possible albeit uncommon that slice references
81 * differ between slices. We take the easy approach and ignore
82 * it for now. If this turns out to have any relevance in
83 * practice then correct remapping should be added. */
84 if (ref >= sl->ref_count[0])
86 if (!sl->ref_list[0][ref].data[0]) {
87 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
90 if ((sl->ref_list[0][ref].reference&3) != 3) {
91 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
94 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
96 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
97 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
98 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
100 sl->mb_field_decoding_flag = 0;
101 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
104 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
107 AVCodecContext *avctx = h->avctx;
108 const AVFrame *src = h->cur_pic.f;
109 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
110 int vshift = desc->log2_chroma_h;
111 const int field_pic = h->picture_structure != PICT_FRAME;
117 height = FFMIN(height, avctx->height - y);
119 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
122 if (avctx->draw_horiz_band) {
123 int offset[AV_NUM_DATA_POINTERS];
126 offset[0] = y * src->linesize[0];
128 offset[2] = (y >> vshift) * src->linesize[1];
129 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
134 avctx->draw_horiz_band(avctx, src, offset,
135 y, h->picture_structure, height);
139 void ff_h264_free_tables(H264Context *h)
143 av_freep(&h->intra4x4_pred_mode);
144 av_freep(&h->chroma_pred_mode_table);
145 av_freep(&h->cbp_table);
146 av_freep(&h->mvd_table[0]);
147 av_freep(&h->mvd_table[1]);
148 av_freep(&h->direct_table);
149 av_freep(&h->non_zero_count);
150 av_freep(&h->slice_table_base);
151 h->slice_table = NULL;
152 av_freep(&h->list_counts);
154 av_freep(&h->mb2b_xy);
155 av_freep(&h->mb2br_xy);
157 av_buffer_pool_uninit(&h->qscale_table_pool);
158 av_buffer_pool_uninit(&h->mb_type_pool);
159 av_buffer_pool_uninit(&h->motion_val_pool);
160 av_buffer_pool_uninit(&h->ref_index_pool);
162 for (i = 0; i < h->nb_slice_ctx; i++) {
163 H264SliceContext *sl = &h->slice_ctx[i];
165 av_freep(&sl->dc_val_base);
166 av_freep(&sl->er.mb_index2xy);
167 av_freep(&sl->er.error_status_table);
168 av_freep(&sl->er.er_temp_buffer);
170 av_freep(&sl->bipred_scratchpad);
171 av_freep(&sl->edge_emu_buffer);
172 av_freep(&sl->top_borders[0]);
173 av_freep(&sl->top_borders[1]);
175 sl->bipred_scratchpad_allocated = 0;
176 sl->edge_emu_buffer_allocated = 0;
177 sl->top_borders_allocated[0] = 0;
178 sl->top_borders_allocated[1] = 0;
182 int ff_h264_alloc_tables(H264Context *h)
184 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
185 const int row_mb_num = 2*h->mb_stride*FFMAX(h->nb_slice_ctx, 1);
188 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
189 row_mb_num, 8 * sizeof(uint8_t), fail)
190 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
192 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
193 big_mb_num * 48 * sizeof(uint8_t), fail)
194 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
195 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
196 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
197 big_mb_num * sizeof(uint16_t), fail)
198 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
199 big_mb_num * sizeof(uint8_t), fail)
200 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[0],
201 row_mb_num, 16 * sizeof(uint8_t), fail);
202 FF_ALLOCZ_ARRAY_OR_GOTO(h->avctx, h->mvd_table[1],
203 row_mb_num, 16 * sizeof(uint8_t), fail);
204 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
205 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
207 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
208 4 * big_mb_num * sizeof(uint8_t), fail);
209 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
210 big_mb_num * sizeof(uint8_t), fail)
212 memset(h->slice_table_base, -1,
213 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
214 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
216 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
217 big_mb_num * sizeof(uint32_t), fail);
218 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
219 big_mb_num * sizeof(uint32_t), fail);
220 for (y = 0; y < h->mb_height; y++)
221 for (x = 0; x < h->mb_width; x++) {
222 const int mb_xy = x + y * h->mb_stride;
223 const int b_xy = 4 * x + 4 * y * h->b_stride;
225 h->mb2b_xy[mb_xy] = b_xy;
226 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
232 ff_h264_free_tables(h);
233 return AVERROR(ENOMEM);
238 * Allocate buffers which are not shared amongst multiple threads.
240 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
242 ERContext *er = &sl->er;
243 int mb_array_size = h->mb_height * h->mb_stride;
244 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
245 int c_size = h->mb_stride * (h->mb_height + 1);
246 int yc_size = y_size + 2 * c_size;
249 sl->ref_cache[0][scan8[5] + 1] =
250 sl->ref_cache[0][scan8[7] + 1] =
251 sl->ref_cache[0][scan8[13] + 1] =
252 sl->ref_cache[1][scan8[5] + 1] =
253 sl->ref_cache[1][scan8[7] + 1] =
254 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
256 if (sl != h->slice_ctx) {
257 memset(er, 0, sizeof(*er));
259 if (CONFIG_ERROR_RESILIENCE) {
262 er->avctx = h->avctx;
263 er->decode_mb = h264_er_decode_mb;
265 er->quarter_sample = 1;
267 er->mb_num = h->mb_num;
268 er->mb_width = h->mb_width;
269 er->mb_height = h->mb_height;
270 er->mb_stride = h->mb_stride;
271 er->b8_stride = h->mb_width * 2 + 1;
273 // error resilience code looks cleaner with this
274 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
275 (h->mb_num + 1) * sizeof(int), fail);
277 for (y = 0; y < h->mb_height; y++)
278 for (x = 0; x < h->mb_width; x++)
279 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
281 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
282 h->mb_stride + h->mb_width;
284 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
285 mb_array_size * sizeof(uint8_t), fail);
287 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
288 h->mb_height * h->mb_stride, fail);
290 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
291 yc_size * sizeof(int16_t), fail);
292 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
293 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
294 er->dc_val[2] = er->dc_val[1] + c_size;
295 for (i = 0; i < yc_size; i++)
296 sl->dc_val_base[i] = 1024;
302 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
305 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
310 h->backup_width = -1;
311 h->backup_height = -1;
312 h->backup_pix_fmt = AV_PIX_FMT_NONE;
313 h->cur_chroma_format_idc = -1;
315 h->picture_structure = PICT_FRAME;
316 h->workaround_bugs = avctx->workaround_bugs;
317 h->flags = avctx->flags;
318 h->poc.prev_poc_msb = 1 << 16;
319 h->recovery_frame = -1;
320 h->frame_recovered = 0;
321 h->poc.prev_frame_num = -1;
322 h->sei.frame_packing.frame_packing_arrangement_cancel_flag = -1;
323 h->sei.unregistered.x264_build = -1;
325 h->next_outputed_poc = INT_MIN;
326 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
327 h->last_pocs[i] = INT_MIN;
329 ff_h264_sei_uninit(&h->sei);
331 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
333 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
334 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
337 return AVERROR(ENOMEM);
340 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
341 h->DPB[i].f = av_frame_alloc();
343 return AVERROR(ENOMEM);
346 h->cur_pic.f = av_frame_alloc();
348 return AVERROR(ENOMEM);
350 h->last_pic_for_ec.f = av_frame_alloc();
351 if (!h->last_pic_for_ec.f)
352 return AVERROR(ENOMEM);
354 for (i = 0; i < h->nb_slice_ctx; i++)
355 h->slice_ctx[i].h264 = h;
360 static av_cold int h264_decode_end(AVCodecContext *avctx)
362 H264Context *h = avctx->priv_data;
365 ff_h264_remove_all_refs(h);
366 ff_h264_free_tables(h);
368 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
369 ff_h264_unref_picture(h, &h->DPB[i]);
370 av_frame_free(&h->DPB[i].f);
372 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
374 h->cur_pic_ptr = NULL;
376 av_freep(&h->slice_ctx);
379 ff_h264_sei_uninit(&h->sei);
380 ff_h264_ps_uninit(&h->ps);
382 ff_h2645_packet_uninit(&h->pkt);
384 ff_h264_unref_picture(h, &h->cur_pic);
385 av_frame_free(&h->cur_pic.f);
386 ff_h264_unref_picture(h, &h->last_pic_for_ec);
387 av_frame_free(&h->last_pic_for_ec.f);
392 static AVOnce h264_vlc_init = AV_ONCE_INIT;
394 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
396 H264Context *h = avctx->priv_data;
399 ret = h264_init_context(avctx, h);
403 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
405 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
406 return AVERROR_UNKNOWN;
409 if (avctx->ticks_per_frame == 1) {
410 if(h->avctx->time_base.den < INT_MAX/2) {
411 h->avctx->time_base.den *= 2;
413 h->avctx->time_base.num /= 2;
415 avctx->ticks_per_frame = 2;
417 if (avctx->extradata_size > 0 && avctx->extradata) {
418 ret = ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
419 &h->ps, &h->is_avc, &h->nal_length_size,
420 avctx->err_recognition, avctx);
422 h264_decode_end(avctx);
427 if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
428 h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
429 h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
432 avctx->internal->allocate_progress = 1;
434 ff_h264_flush_change(h);
436 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
439 if (h->enable_er && (avctx->active_thread_type & FF_THREAD_SLICE)) {
440 av_log(avctx, AV_LOG_WARNING,
441 "Error resilience with slice threads is enabled. It is unsafe and unsupported and may crash. "
442 "Use it at your own risk\n");
449 static int decode_init_thread_copy(AVCodecContext *avctx)
451 H264Context *h = avctx->priv_data;
454 if (!avctx->internal->is_copy)
457 memset(h, 0, sizeof(*h));
459 ret = h264_init_context(avctx, h);
463 h->context_initialized = 0;
470 * Run setup operations that must be run after slice header decoding.
471 * This includes finding the next displayed frame.
473 * @param h h264 master context
474 * @param setup_finished enough NALs have been read that we can call
475 * ff_thread_finish_setup()
477 static void decode_postinit(H264Context *h, int setup_finished)
479 const SPS *sps = h->ps.sps;
480 H264Picture *out = h->cur_pic_ptr;
481 H264Picture *cur = h->cur_pic_ptr;
482 int i, pics, out_of_order, out_idx;
484 if (h->next_output_pic)
487 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
488 /* FIXME: if we have two PAFF fields in one packet, we can't start
489 * the next thread here. If we have one field per packet, we can.
490 * The check in decode_nal_units() is not good enough to find this
491 * yet, so we assume the worst for now. */
492 // if (setup_finished)
493 // ff_thread_finish_setup(h->avctx);
494 if (cur->field_poc[0] == INT_MAX && cur->field_poc[1] == INT_MAX)
496 if (h->avctx->hwaccel || h->missing_fields <=1)
500 cur->mmco_reset = h->mmco_reset;
503 // FIXME do something with unavailable reference frames
505 /* Sort B-frames into display order */
506 if (sps->bitstream_restriction_flag ||
507 h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
508 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
511 for (i = 0; 1; i++) {
512 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
514 h->last_pocs[i-1] = cur->poc;
517 h->last_pocs[i-1]= h->last_pocs[i];
520 out_of_order = MAX_DELAYED_PIC_COUNT - i;
521 if( cur->f->pict_type == AV_PICTURE_TYPE_B
522 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
523 out_of_order = FFMAX(out_of_order, 1);
524 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
525 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
526 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
527 h->last_pocs[i] = INT_MIN;
528 h->last_pocs[0] = cur->poc;
530 } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
531 av_log(h->avctx, AV_LOG_INFO, "Increasing reorder buffer to %d\n", out_of_order);
532 h->avctx->has_b_frames = out_of_order;
536 while (h->delayed_pic[pics])
539 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
541 h->delayed_pic[pics++] = cur;
542 if (cur->reference == 0)
543 cur->reference = DELAYED_PIC_REF;
545 out = h->delayed_pic[0];
547 for (i = 1; h->delayed_pic[i] &&
548 !h->delayed_pic[i]->f->key_frame &&
549 !h->delayed_pic[i]->mmco_reset;
551 if (h->delayed_pic[i]->poc < out->poc) {
552 out = h->delayed_pic[i];
555 if (h->avctx->has_b_frames == 0 &&
556 (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
557 h->next_outputed_poc = INT_MIN;
558 out_of_order = out->poc < h->next_outputed_poc;
560 if (out_of_order || pics > h->avctx->has_b_frames) {
561 out->reference &= ~DELAYED_PIC_REF;
562 for (i = out_idx; h->delayed_pic[i]; i++)
563 h->delayed_pic[i] = h->delayed_pic[i + 1];
565 if (!out_of_order && pics > h->avctx->has_b_frames) {
566 h->next_output_pic = out;
567 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
568 h->next_outputed_poc = INT_MIN;
570 h->next_outputed_poc = out->poc;
572 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
575 if (h->next_output_pic) {
576 if (h->next_output_pic->recovered) {
577 // We have reached an recovery point and all frames after it in
578 // display order are "recovered".
579 h->frame_recovered |= FRAME_RECOVERED_SEI;
581 h->next_output_pic->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
584 if (setup_finished && !h->avctx->hwaccel) {
585 ff_thread_finish_setup(h->avctx);
587 if (h->avctx->active_thread_type & FF_THREAD_FRAME)
588 h->setup_finished = 1;
593 * instantaneous decoder refresh.
595 static void idr(H264Context *h)
598 ff_h264_remove_all_refs(h);
599 h->poc.prev_frame_num =
600 h->poc.prev_frame_num_offset = 0;
601 h->poc.prev_poc_msb = 1<<16;
602 h->poc.prev_poc_lsb = 0;
603 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
604 h->last_pocs[i] = INT_MIN;
607 /* forget old pics after a seek */
608 void ff_h264_flush_change(H264Context *h)
612 h->next_outputed_poc = INT_MIN;
613 h->prev_interlaced_frame = 1;
616 h->poc.prev_frame_num = -1;
617 if (h->cur_pic_ptr) {
618 h->cur_pic_ptr->reference = 0;
619 for (j=i=0; h->delayed_pic[i]; i++)
620 if (h->delayed_pic[i] != h->cur_pic_ptr)
621 h->delayed_pic[j++] = h->delayed_pic[i];
622 h->delayed_pic[j] = NULL;
624 ff_h264_unref_picture(h, &h->last_pic_for_ec);
627 ff_h264_sei_uninit(&h->sei);
628 h->recovery_frame = -1;
629 h->frame_recovered = 0;
630 h->current_slice = 0;
634 /* forget old pics after a seek */
635 static void flush_dpb(AVCodecContext *avctx)
637 H264Context *h = avctx->priv_data;
640 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
642 ff_h264_flush_change(h);
644 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
645 ff_h264_unref_picture(h, &h->DPB[i]);
646 h->cur_pic_ptr = NULL;
647 ff_h264_unref_picture(h, &h->cur_pic);
651 ff_h264_free_tables(h);
652 h->context_initialized = 0;
656 static const uint8_t start_code[] = { 0x00, 0x00, 0x01 };
659 static int get_last_needed_nal(H264Context *h)
666 for (i = 0; i < h->pkt.nb_nals; i++) {
667 H2645NAL *nal = &h->pkt.nals[i];
670 /* packets can sometimes contain multiple PPS/SPS,
671 * e.g. two PAFF field pictures in one packet, or a demuxer
672 * which splits NALs strangely if so, when frame threading we
673 * can't start the next thread until we've read all of them */
680 case H264_NAL_IDR_SLICE:
682 ret = init_get_bits8(&gb, nal->data + 1, (nal->size - 1));
685 if (!get_ue_golomb_long(&gb) || // first_mb_in_slice
687 first_slice != nal->type)
690 first_slice = nal->type;
697 static void debug_green_metadata(const H264SEIGreenMetaData *gm, void *logctx)
699 av_log(logctx, AV_LOG_DEBUG, "Green Metadata Info SEI message\n");
700 av_log(logctx, AV_LOG_DEBUG, " green_metadata_type: %d\n", gm->green_metadata_type);
702 if (gm->green_metadata_type == 0) {
703 av_log(logctx, AV_LOG_DEBUG, " green_metadata_period_type: %d\n", gm->period_type);
705 if (gm->period_type == 2)
706 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_seconds: %d\n", gm->num_seconds);
707 else if (gm->period_type == 3)
708 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_pictures: %d\n", gm->num_pictures);
710 av_log(logctx, AV_LOG_DEBUG, " SEI GREEN Complexity Metrics: %f %f %f %f\n",
711 (float)gm->percent_non_zero_macroblocks/255,
712 (float)gm->percent_intra_coded_macroblocks/255,
713 (float)gm->percent_six_tap_filtering/255,
714 (float)gm->percent_alpha_point_deblocking_instance/255);
716 } else if (gm->green_metadata_type == 1) {
717 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_type: %d\n", gm->xsd_metric_type);
719 if (gm->xsd_metric_type == 0)
720 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_value: %f\n",
721 (float)gm->xsd_metric_value/100);
725 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
727 AVCodecContext *const avctx = h->avctx;
728 unsigned context_count = 0;
729 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
735 h->max_contexts = h->nb_slice_ctx;
736 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
737 h->current_slice = 0;
739 h->cur_pic_ptr = NULL;
740 ff_h264_sei_uninit(&h->sei);
743 if (h->nal_length_size == 4) {
744 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
746 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
750 ret = ff_h2645_packet_split(&h->pkt, buf, buf_size, avctx, h->is_avc,
751 h->nal_length_size, avctx->codec_id);
753 av_log(avctx, AV_LOG_ERROR,
754 "Error splitting the input into NAL units.\n");
758 if (avctx->active_thread_type & FF_THREAD_FRAME)
759 nals_needed = get_last_needed_nal(h);
763 for (i = 0; i < h->pkt.nb_nals; i++) {
764 H2645NAL *nal = &h->pkt.nals[i];
765 H264SliceContext *sl = &h->slice_ctx[context_count];
768 if (avctx->skip_frame >= AVDISCARD_NONREF &&
769 nal->ref_idc == 0 && nal->type != H264_NAL_SEI)
773 // FIXME these should stop being context-global variables
774 h->nal_ref_idc = nal->ref_idc;
775 h->nal_unit_type = nal->type;
779 case H264_NAL_IDR_SLICE:
780 if ((nal->data[1] & 0xFC) == 0x98) {
781 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
782 h->next_outputed_poc = INT_MIN;
787 if (h->current_slice && (avctx->active_thread_type & FF_THREAD_SLICE)) {
788 av_log(h, AV_LOG_ERROR, "invalid mixed IDR / non IDR frames cannot be decoded in slice multithreading mode\n");
789 ret = AVERROR_INVALIDDATA;
792 idr(h); // FIXME ensure we don't lose some frames if there is reordering
795 h->has_recovery_point = 1;
799 if ((err = ff_h264_decode_slice_header(h, sl, nal)))
802 if (sl->redundant_pic_count > 0)
805 if (h->current_slice == 1) {
806 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS))
807 decode_postinit(h, i >= nals_needed);
809 if (h->avctx->hwaccel &&
810 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
813 if (CONFIG_H264_VDPAU_DECODER &&
814 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
815 ff_vdpau_h264_picture_start(h);
819 if (avctx->hwaccel) {
820 ret = avctx->hwaccel->decode_slice(avctx,
826 } else if (CONFIG_H264_VDPAU_DECODER &&
827 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU) {
828 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f->data[0],
831 ff_vdpau_add_data_chunk(h->cur_pic_ptr->f->data[0],
841 avpriv_request_sample(avctx, "data partitioning");
844 ret = ff_h264_sei_decode(&h->sei, &nal->gb, &h->ps, avctx);
845 h->has_recovery_point = h->has_recovery_point || h->sei.recovery_point.recovery_frame_cnt != -1;
846 if (avctx->debug & FF_DEBUG_GREEN_MD)
847 debug_green_metadata(&h->sei.green_metadata, h->avctx);
849 FF_DISABLE_DEPRECATION_WARNINGS
850 h->avctx->dtg_active_format = h->sei.afd.active_format_description;
851 FF_ENABLE_DEPRECATION_WARNINGS
852 #endif /* FF_API_AFD */
853 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
857 GetBitContext tmp_gb = nal->gb;
858 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
860 av_log(h->avctx, AV_LOG_DEBUG,
861 "SPS decoding failure, trying again with the complete NAL\n");
862 init_get_bits8(&tmp_gb, nal->raw_data + 1, nal->raw_size - 1);
863 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
865 ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps, 1);
869 ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
871 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
875 case H264_NAL_END_SEQUENCE:
876 case H264_NAL_END_STREAM:
877 case H264_NAL_FILLER_DATA:
878 case H264_NAL_SPS_EXT:
879 case H264_NAL_AUXILIARY_SLICE:
882 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
883 nal->type, nal->size_bits);
886 if (context_count == h->max_contexts) {
887 ret = ff_h264_execute_decode_slices(h, context_count);
888 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
893 if (err < 0 || err == SLICE_SKIPED) {
895 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
896 sl->ref_count[0] = sl->ref_count[1] = sl->list_count = 0;
897 } else if (err == SLICE_SINGLETHREAD) {
898 if (context_count > 0) {
899 ret = ff_h264_execute_decode_slices(h, context_count);
900 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
904 /* Slice could not be decoded in parallel mode, restart. */
905 sl = &h->slice_ctx[0];
910 ret = ff_h264_execute_decode_slices(h, context_count);
911 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
918 #if CONFIG_ERROR_RESILIENCE
920 * FIXME: Error handling code does not seem to support interlaced
921 * when slices span multiple rows
922 * The ff_er_add_slice calls don't work right for bottom
923 * fields; they cause massive erroneous error concealing
924 * Error marking covers both fields (top and bottom).
925 * This causes a mismatched s->error_count
926 * and a bad error table. Further, the error count goes to
927 * INT_MAX when called for bottom field, because mb_y is
928 * past end by one (callers fault) and resync_mb_y != 0
929 * causes problems for the first MB line, too.
931 if (!FIELD_PICTURE(h) && h->current_slice &&
932 h->ps.sps == (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data &&
935 H264SliceContext *sl = h->slice_ctx;
936 int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
938 ff_h264_set_erpic(&sl->er.cur_pic, h->cur_pic_ptr);
941 ff_h264_set_erpic(&sl->er.last_pic, &h->last_pic_for_ec);
942 sl->ref_list[0][0].parent = &h->last_pic_for_ec;
943 memcpy(sl->ref_list[0][0].data, h->last_pic_for_ec.f->data, sizeof(sl->ref_list[0][0].data));
944 memcpy(sl->ref_list[0][0].linesize, h->last_pic_for_ec.f->linesize, sizeof(sl->ref_list[0][0].linesize));
945 sl->ref_list[0][0].reference = h->last_pic_for_ec.reference;
946 } else if (sl->ref_count[0]) {
947 ff_h264_set_erpic(&sl->er.last_pic, sl->ref_list[0][0].parent);
949 ff_h264_set_erpic(&sl->er.last_pic, NULL);
951 if (sl->ref_count[1])
952 ff_h264_set_erpic(&sl->er.next_pic, sl->ref_list[1][0].parent);
954 sl->er.ref_count = sl->ref_count[0];
956 ff_er_frame_end(&sl->er);
958 memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
960 #endif /* CONFIG_ERROR_RESILIENCE */
962 if (h->cur_pic_ptr && !h->droppable) {
963 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
964 h->picture_structure == PICT_BOTTOM_FIELD);
967 return (ret < 0) ? ret : buf_size;
971 * Return the number of bytes consumed for building the current frame.
973 static int get_consumed_bytes(int pos, int buf_size)
976 pos = 1; // avoid infinite loops (I doubt that is needed but...)
977 if (pos + 10 > buf_size)
978 pos = buf_size; // oops ;)
983 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
985 AVFrame *src = srcp->f;
986 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(src->format);
988 int ret = av_frame_ref(dst, src);
992 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.frame_packing), 0);
994 h->backup_width = h->avctx->width;
995 h->backup_height = h->avctx->height;
996 h->backup_pix_fmt = h->avctx->pix_fmt;
998 h->avctx->width = dst->width;
999 h->avctx->height = dst->height;
1000 h->avctx->pix_fmt = dst->format;
1002 if (srcp->sei_recovery_frame_cnt == 0)
1007 for (i = 0; i < desc->nb_components; i++) {
1008 int hshift = (i > 0) ? desc->log2_chroma_w : 0;
1009 int vshift = (i > 0) ? desc->log2_chroma_h : 0;
1010 int off = ((srcp->crop_left >> hshift) << h->pixel_shift) +
1011 (srcp->crop_top >> vshift) * dst->linesize[i];
1012 dst->data[i] += off;
1017 static int is_extra(const uint8_t *buf, int buf_size)
1019 int cnt= buf[5]&0x1f;
1020 const uint8_t *p= buf+6;
1022 int nalsize= AV_RB16(p) + 2;
1023 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 7)
1031 int nalsize= AV_RB16(p) + 2;
1032 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 8)
1039 static int h264_decode_frame(AVCodecContext *avctx, void *data,
1040 int *got_frame, AVPacket *avpkt)
1042 const uint8_t *buf = avpkt->data;
1043 int buf_size = avpkt->size;
1044 H264Context *h = avctx->priv_data;
1045 AVFrame *pict = data;
1051 h->flags = avctx->flags;
1052 h->setup_finished = 0;
1054 if (h->backup_width != -1) {
1055 avctx->width = h->backup_width;
1056 h->backup_width = -1;
1058 if (h->backup_height != -1) {
1059 avctx->height = h->backup_height;
1060 h->backup_height = -1;
1062 if (h->backup_pix_fmt != AV_PIX_FMT_NONE) {
1063 avctx->pix_fmt = h->backup_pix_fmt;
1064 h->backup_pix_fmt = AV_PIX_FMT_NONE;
1067 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1069 /* end of stream, output what is still in the buffers */
1070 if (buf_size == 0) {
1073 h->cur_pic_ptr = NULL;
1076 // FIXME factorize this with the output code below
1077 out = h->delayed_pic[0];
1080 h->delayed_pic[i] &&
1081 !h->delayed_pic[i]->f->key_frame &&
1082 !h->delayed_pic[i]->mmco_reset;
1084 if (h->delayed_pic[i]->poc < out->poc) {
1085 out = h->delayed_pic[i];
1089 for (i = out_idx; h->delayed_pic[i]; i++)
1090 h->delayed_pic[i] = h->delayed_pic[i + 1];
1093 out->reference &= ~DELAYED_PIC_REF;
1094 ret = output_frame(h, pict, out);
1102 if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
1104 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
1105 if (is_extra(side, side_size))
1106 ff_h264_decode_extradata(side, side_size,
1107 &h->ps, &h->is_avc, &h->nal_length_size,
1108 avctx->err_recognition, avctx);
1110 if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
1111 if (is_extra(buf, buf_size))
1112 return ff_h264_decode_extradata(buf, buf_size,
1113 &h->ps, &h->is_avc, &h->nal_length_size,
1114 avctx->err_recognition, avctx);
1117 buf_index = decode_nal_units(h, buf, buf_size);
1119 return AVERROR_INVALIDDATA;
1121 if (!h->cur_pic_ptr && h->nal_unit_type == H264_NAL_END_SEQUENCE) {
1122 av_assert0(buf_index <= buf_size);
1126 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
1127 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1128 buf_size >= 4 && !memcmp("Q264", buf, 4))
1130 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1131 return AVERROR_INVALIDDATA;
1134 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1135 (h->mb_y >= h->mb_height && h->mb_height)) {
1136 if (avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)
1137 decode_postinit(h, 1);
1139 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1142 /* Wait for second field. */
1144 if (h->next_output_pic && ((avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
1145 (avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
1146 h->next_output_pic->recovered)) {
1147 if (!h->next_output_pic->recovered)
1148 h->next_output_pic->f->flags |= AV_FRAME_FLAG_CORRUPT;
1150 if (!h->avctx->hwaccel &&
1151 (h->next_output_pic->field_poc[0] == INT_MAX ||
1152 h->next_output_pic->field_poc[1] == INT_MAX)
1155 AVFrame *f = h->next_output_pic->f;
1156 int field = h->next_output_pic->field_poc[0] == INT_MAX;
1157 uint8_t *dst_data[4];
1159 const uint8_t *src_data[4];
1161 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
1163 for (p = 0; p<4; p++) {
1164 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
1165 src_data[p] = f->data[p] + field *f->linesize[p];
1166 linesizes[p] = 2*f->linesize[p];
1169 av_image_copy(dst_data, linesizes, src_data, linesizes,
1170 f->format, f->width, f->height>>1);
1173 ret = output_frame(h, pict, h->next_output_pic);
1177 if (CONFIG_MPEGVIDEO) {
1178 ff_print_debug_info2(h->avctx, pict, NULL,
1179 h->next_output_pic->mb_type,
1180 h->next_output_pic->qscale_table,
1181 h->next_output_pic->motion_val,
1183 h->mb_width, h->mb_height, h->mb_stride, 1);
1188 av_assert0(pict->buf[0] || !*got_frame);
1190 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1192 return get_consumed_bytes(buf_index, buf_size);
1195 #define OFFSET(x) offsetof(H264Context, x)
1196 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1197 static const AVOption h264_options[] = {
1198 {"is_avc", "is avc", offsetof(H264Context, is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0},
1199 {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
1200 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1204 static const AVClass h264_class = {
1205 .class_name = "H264 Decoder",
1206 .item_name = av_default_item_name,
1207 .option = h264_options,
1208 .version = LIBAVUTIL_VERSION_INT,
1211 AVCodec ff_h264_decoder = {
1213 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1214 .type = AVMEDIA_TYPE_VIDEO,
1215 .id = AV_CODEC_ID_H264,
1216 .priv_data_size = sizeof(H264Context),
1217 .init = ff_h264_decode_init,
1218 .close = h264_decode_end,
1219 .decode = h264_decode_frame,
1220 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1221 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1222 AV_CODEC_CAP_FRAME_THREADS,
1223 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
1225 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
1226 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1227 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1228 .priv_class = &h264_class,
1231 #if CONFIG_H264_VDPAU_DECODER && FF_API_VDPAU
1232 static const AVClass h264_vdpau_class = {
1233 .class_name = "H264 VDPAU Decoder",
1234 .item_name = av_default_item_name,
1235 .option = h264_options,
1236 .version = LIBAVUTIL_VERSION_INT,
1239 AVCodec ff_h264_vdpau_decoder = {
1240 .name = "h264_vdpau",
1241 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
1242 .type = AVMEDIA_TYPE_VIDEO,
1243 .id = AV_CODEC_ID_H264,
1244 .priv_data_size = sizeof(H264Context),
1245 .init = ff_h264_decode_init,
1246 .close = h264_decode_end,
1247 .decode = h264_decode_frame,
1248 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HWACCEL_VDPAU,
1250 .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
1252 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1253 .priv_class = &h264_vdpau_class,