1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 /**
23 * @file
24 * H.264 / AVC / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28 #define UNCHECKED_BITSTREAM_READER 1
29
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"
36
37 #include "internal.h"
38 #include "bytestream.h"
39 #include "cabac.h"
40 #include "cabac_functions.h"
41 #include "error_resilience.h"
42 #include "avcodec.h"
43 #include "h264.h"
44 #include "h264dec.h"
45 #include "h2645_parse.h"
46 #include "h264data.h"
47 #include "h264chroma.h"
48 #include "h264_mvpred.h"
49 #include "h264_ps.h"
50 #include "golomb.h"
51 #include "hwconfig.h"
52 #include "mathops.h"
53 #include "me_cmp.h"
54 #include "mpegutils.h"
55 #include "profiles.h"
56 #include "rectangle.h"
57 #include "thread.h"
58
59 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
60
avpriv_h264_has_num_reorder_frames(AVCodecContext * avctx)61 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
62 {
63 H264Context *h = avctx->priv_data;
64 return h && h->ps.sps ? h->ps.sps->num_reorder_frames : 0;
65 }
66
h264_er_decode_mb(void * opaque,int ref,int mv_dir,int mv_type,int (* mv)[2][4][2],int mb_x,int mb_y,int mb_intra,int mb_skipped)67 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
68 int (*mv)[2][4][2],
69 int mb_x, int mb_y, int mb_intra, int mb_skipped)
70 {
71 H264Context *h = opaque;
72 H264SliceContext *sl = &h->slice_ctx[0];
73
74 sl->mb_x = mb_x;
75 sl->mb_y = mb_y;
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));
78 av_assert1(ref >= 0);
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])
84 ref = 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");
87 ref = 0;
88 }
89 if ((sl->ref_list[0][ref].reference&3) != 3) {
90 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
91 return;
92 }
93 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
94 2, 2, 2, ref, 1);
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);
98 sl->mb_mbaff =
99 sl->mb_field_decoding_flag = 0;
100 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
101 }
102
ff_h264_draw_horiz_band(const H264Context * h,H264SliceContext * sl,int y,int height)103 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
104 int y, int height)
105 {
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;
111 if (field_pic) {
112 height <<= 1;
113 y <<= 1;
114 }
115
116 height = FFMIN(height, avctx->height - y);
117
118 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
119 return;
120
121 if (avctx->draw_horiz_band) {
122 int offset[AV_NUM_DATA_POINTERS];
123 int i;
124
125 offset[0] = y * src->linesize[0];
126 offset[1] =
127 offset[2] = (y >> vshift) * src->linesize[1];
128 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
129 offset[i] = 0;
130
131 emms_c();
132
133 avctx->draw_horiz_band(avctx, src, offset,
134 y, h->picture_structure, height);
135 }
136 }
137
ff_h264_free_tables(H264Context * h)138 void ff_h264_free_tables(H264Context *h)
139 {
140 int i;
141
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);
152
153 av_freep(&h->mb2b_xy);
154 av_freep(&h->mb2br_xy);
155
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);
160
161 for (i = 0; i < h->nb_slice_ctx; i++) {
162 H264SliceContext *sl = &h->slice_ctx[i];
163
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);
168
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]);
173
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;
178 }
179 }
180
ff_h264_alloc_tables(H264Context * h)181 int ff_h264_alloc_tables(H264Context *h)
182 {
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);
185 const int st_size = big_mb_num + h->mb_stride;
186 int x, y;
187
188 if (!FF_ALLOCZ_TYPED_ARRAY(h->intra4x4_pred_mode, row_mb_num * 8) ||
189 !FF_ALLOCZ_TYPED_ARRAY(h->non_zero_count, big_mb_num) ||
190 !FF_ALLOCZ_TYPED_ARRAY(h->slice_table_base, st_size) ||
191 !FF_ALLOCZ_TYPED_ARRAY(h->cbp_table, big_mb_num) ||
192 !FF_ALLOCZ_TYPED_ARRAY(h->chroma_pred_mode_table, big_mb_num) ||
193 !FF_ALLOCZ_TYPED_ARRAY(h->mvd_table[0], row_mb_num * 8) ||
194 !FF_ALLOCZ_TYPED_ARRAY(h->mvd_table[1], row_mb_num * 8) ||
195 !FF_ALLOCZ_TYPED_ARRAY(h->direct_table, big_mb_num * 4) ||
196 !FF_ALLOCZ_TYPED_ARRAY(h->list_counts, big_mb_num) ||
197 !FF_ALLOCZ_TYPED_ARRAY(h->mb2b_xy, big_mb_num) ||
198 !FF_ALLOCZ_TYPED_ARRAY(h->mb2br_xy, big_mb_num))
199 return AVERROR(ENOMEM);
200 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
201 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
202 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
203 memset(h->slice_table_base, -1,
204 st_size * sizeof(*h->slice_table_base));
205 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
206 for (y = 0; y < h->mb_height; y++)
207 for (x = 0; x < h->mb_width; x++) {
208 const int mb_xy = x + y * h->mb_stride;
209 const int b_xy = 4 * x + 4 * y * h->b_stride;
210
211 h->mb2b_xy[mb_xy] = b_xy;
212 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
213 }
214
215 return 0;
216 }
217
218 /**
219 * Init context
220 * Allocate buffers which are not shared amongst multiple threads.
221 */
ff_h264_slice_context_init(H264Context * h,H264SliceContext * sl)222 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
223 {
224 ERContext *er = &sl->er;
225 int mb_array_size = h->mb_height * h->mb_stride;
226 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
227 int c_size = h->mb_stride * (h->mb_height + 1);
228 int yc_size = y_size + 2 * c_size;
229 int x, y, i;
230
231 sl->ref_cache[0][scan8[5] + 1] =
232 sl->ref_cache[0][scan8[7] + 1] =
233 sl->ref_cache[0][scan8[13] + 1] =
234 sl->ref_cache[1][scan8[5] + 1] =
235 sl->ref_cache[1][scan8[7] + 1] =
236 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
237
238 if (sl != h->slice_ctx) {
239 memset(er, 0, sizeof(*er));
240 } else if (CONFIG_ERROR_RESILIENCE) {
241 const int er_size = h->mb_height * h->mb_stride * (4*sizeof(int) + 1);
242
243 /* init ER */
244 er->avctx = h->avctx;
245 er->decode_mb = h264_er_decode_mb;
246 er->opaque = h;
247 er->quarter_sample = 1;
248
249 er->mb_num = h->mb_num;
250 er->mb_width = h->mb_width;
251 er->mb_height = h->mb_height;
252 er->mb_stride = h->mb_stride;
253 er->b8_stride = h->mb_width * 2 + 1;
254
255 // error resilience code looks cleaner with this
256 if (!FF_ALLOCZ_TYPED_ARRAY(er->mb_index2xy, h->mb_num + 1) ||
257 !FF_ALLOCZ_TYPED_ARRAY(er->error_status_table, mb_array_size) ||
258 !FF_ALLOCZ_TYPED_ARRAY(er->er_temp_buffer, er_size) ||
259 !FF_ALLOCZ_TYPED_ARRAY(sl->dc_val_base, yc_size))
260 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
261
262 for (y = 0; y < h->mb_height; y++)
263 for (x = 0; x < h->mb_width; x++)
264 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
265
266 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
267 h->mb_stride + h->mb_width;
268 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
269 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
270 er->dc_val[2] = er->dc_val[1] + c_size;
271 for (i = 0; i < yc_size; i++)
272 sl->dc_val_base[i] = 1024;
273 }
274
275 return 0;
276 }
277
h264_init_context(AVCodecContext * avctx,H264Context * h)278 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
279 {
280 int i;
281
282 h->avctx = avctx;
283 h->cur_chroma_format_idc = -1;
284
285 h->width_from_caller = avctx->width;
286 h->height_from_caller = avctx->height;
287
288 h->workaround_bugs = avctx->workaround_bugs;
289 h->flags = avctx->flags;
290 h->poc.prev_poc_msb = 1 << 16;
291 h->recovery_frame = -1;
292 h->frame_recovered = 0;
293 h->poc.prev_frame_num = -1;
294 h->sei.frame_packing.arrangement_cancel_flag = -1;
295 h->sei.unregistered.x264_build = -1;
296
297 h->next_outputed_poc = INT_MIN;
298 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
299 h->last_pocs[i] = INT_MIN;
300
301 ff_h264_sei_uninit(&h->sei);
302
303 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
304 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
305 if (!h->slice_ctx) {
306 h->nb_slice_ctx = 0;
307 return AVERROR(ENOMEM);
308 }
309
310 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
311 h->DPB[i].f = av_frame_alloc();
312 if (!h->DPB[i].f)
313 return AVERROR(ENOMEM);
314 }
315
316 h->cur_pic.f = av_frame_alloc();
317 if (!h->cur_pic.f)
318 return AVERROR(ENOMEM);
319
320 h->last_pic_for_ec.f = av_frame_alloc();
321 if (!h->last_pic_for_ec.f)
322 return AVERROR(ENOMEM);
323
324 for (i = 0; i < h->nb_slice_ctx; i++)
325 h->slice_ctx[i].h264 = h;
326
327 return 0;
328 }
329
h264_decode_end(AVCodecContext * avctx)330 static av_cold int h264_decode_end(AVCodecContext *avctx)
331 {
332 H264Context *h = avctx->priv_data;
333 int i;
334
335 ff_h264_remove_all_refs(h);
336 ff_h264_free_tables(h);
337
338 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
339 ff_h264_unref_picture(h, &h->DPB[i]);
340 av_frame_free(&h->DPB[i].f);
341 }
342 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
343
344 h->cur_pic_ptr = NULL;
345
346 av_freep(&h->slice_ctx);
347 h->nb_slice_ctx = 0;
348
349 ff_h264_sei_uninit(&h->sei);
350 ff_h264_ps_uninit(&h->ps);
351
352 ff_h2645_packet_uninit(&h->pkt);
353
354 ff_h264_unref_picture(h, &h->cur_pic);
355 av_frame_free(&h->cur_pic.f);
356 ff_h264_unref_picture(h, &h->last_pic_for_ec);
357 av_frame_free(&h->last_pic_for_ec.f);
358
359 return 0;
360 }
361
362 static AVOnce h264_vlc_init = AV_ONCE_INIT;
363
h264_decode_init(AVCodecContext * avctx)364 static av_cold int h264_decode_init(AVCodecContext *avctx)
365 {
366 H264Context *h = avctx->priv_data;
367 int ret;
368
369 ret = h264_init_context(avctx, h);
370 if (ret < 0)
371 return ret;
372
373 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
374 if (ret != 0) {
375 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
376 return AVERROR_UNKNOWN;
377 }
378
379 if (avctx->ticks_per_frame == 1) {
380 if(h->avctx->time_base.den < INT_MAX/2) {
381 h->avctx->time_base.den *= 2;
382 } else
383 h->avctx->time_base.num /= 2;
384 }
385 avctx->ticks_per_frame = 2;
386
387 if (!avctx->internal->is_copy) {
388 if (avctx->extradata_size > 0 && avctx->extradata) {
389 ret = ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
390 &h->ps, &h->is_avc, &h->nal_length_size,
391 avctx->err_recognition, avctx);
392 if (ret < 0) {
393 int explode = avctx->err_recognition & AV_EF_EXPLODE;
394 av_log(avctx, explode ? AV_LOG_ERROR: AV_LOG_WARNING,
395 "Error decoding the extradata\n");
396 if (explode) {
397 return ret;
398 }
399 ret = 0;
400 }
401 }
402 }
403
404 if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
405 h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
406 h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
407 }
408
409 ff_h264_flush_change(h);
410
411 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
412 h->enable_er = 0;
413
414 if (h->enable_er && (avctx->active_thread_type & FF_THREAD_SLICE)) {
415 av_log(avctx, AV_LOG_WARNING,
416 "Error resilience with slice threads is enabled. It is unsafe and unsupported and may crash. "
417 "Use it at your own risk\n");
418 }
419
420 return 0;
421 }
422
423 /**
424 * instantaneous decoder refresh.
425 */
idr(H264Context * h)426 static void idr(H264Context *h)
427 {
428 int i;
429 ff_h264_remove_all_refs(h);
430 h->poc.prev_frame_num =
431 h->poc.prev_frame_num_offset = 0;
432 h->poc.prev_poc_msb = 1<<16;
433 h->poc.prev_poc_lsb = -1;
434 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
435 h->last_pocs[i] = INT_MIN;
436 }
437
438 /* forget old pics after a seek */
ff_h264_flush_change(H264Context * h)439 void ff_h264_flush_change(H264Context *h)
440 {
441 int i, j;
442
443 h->next_outputed_poc = INT_MIN;
444 h->prev_interlaced_frame = 1;
445 idr(h);
446
447 h->poc.prev_frame_num = -1;
448 if (h->cur_pic_ptr) {
449 h->cur_pic_ptr->reference = 0;
450 for (j=i=0; h->delayed_pic[i]; i++)
451 if (h->delayed_pic[i] != h->cur_pic_ptr)
452 h->delayed_pic[j++] = h->delayed_pic[i];
453 h->delayed_pic[j] = NULL;
454 }
455 ff_h264_unref_picture(h, &h->last_pic_for_ec);
456
457 h->first_field = 0;
458 h->recovery_frame = -1;
459 h->frame_recovered = 0;
460 h->current_slice = 0;
461 h->mmco_reset = 1;
462 }
463
h264_decode_flush(AVCodecContext * avctx)464 static void h264_decode_flush(AVCodecContext *avctx)
465 {
466 H264Context *h = avctx->priv_data;
467 int i;
468
469 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
470
471 ff_h264_flush_change(h);
472 ff_h264_sei_uninit(&h->sei);
473
474 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
475 ff_h264_unref_picture(h, &h->DPB[i]);
476 h->cur_pic_ptr = NULL;
477 ff_h264_unref_picture(h, &h->cur_pic);
478
479 h->mb_y = 0;
480
481 ff_h264_free_tables(h);
482 h->context_initialized = 0;
483 }
484
get_last_needed_nal(H264Context * h)485 static int get_last_needed_nal(H264Context *h)
486 {
487 int nals_needed = 0;
488 int first_slice = 0;
489 int i, ret;
490
491 for (i = 0; i < h->pkt.nb_nals; i++) {
492 H2645NAL *nal = &h->pkt.nals[i];
493 GetBitContext gb;
494
495 /* packets can sometimes contain multiple PPS/SPS,
496 * e.g. two PAFF field pictures in one packet, or a demuxer
497 * which splits NALs strangely if so, when frame threading we
498 * can't start the next thread until we've read all of them */
499 switch (nal->type) {
500 case H264_NAL_SPS:
501 case H264_NAL_PPS:
502 nals_needed = i;
503 break;
504 case H264_NAL_DPA:
505 case H264_NAL_IDR_SLICE:
506 case H264_NAL_SLICE:
507 ret = init_get_bits8(&gb, nal->data + 1, nal->size - 1);
508 if (ret < 0) {
509 av_log(h->avctx, AV_LOG_ERROR, "Invalid zero-sized VCL NAL unit\n");
510 if (h->avctx->err_recognition & AV_EF_EXPLODE)
511 return ret;
512
513 break;
514 }
515 if (!get_ue_golomb_long(&gb) || // first_mb_in_slice
516 !first_slice ||
517 first_slice != nal->type)
518 nals_needed = i;
519 if (!first_slice)
520 first_slice = nal->type;
521 }
522 }
523
524 return nals_needed;
525 }
526
debug_green_metadata(const H264SEIGreenMetaData * gm,void * logctx)527 static void debug_green_metadata(const H264SEIGreenMetaData *gm, void *logctx)
528 {
529 av_log(logctx, AV_LOG_DEBUG, "Green Metadata Info SEI message\n");
530 av_log(logctx, AV_LOG_DEBUG, " green_metadata_type: %d\n", gm->green_metadata_type);
531
532 if (gm->green_metadata_type == 0) {
533 av_log(logctx, AV_LOG_DEBUG, " green_metadata_period_type: %d\n", gm->period_type);
534
535 if (gm->period_type == 2)
536 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_seconds: %d\n", gm->num_seconds);
537 else if (gm->period_type == 3)
538 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_pictures: %d\n", gm->num_pictures);
539
540 av_log(logctx, AV_LOG_DEBUG, " SEI GREEN Complexity Metrics: %f %f %f %f\n",
541 (float)gm->percent_non_zero_macroblocks/255,
542 (float)gm->percent_intra_coded_macroblocks/255,
543 (float)gm->percent_six_tap_filtering/255,
544 (float)gm->percent_alpha_point_deblocking_instance/255);
545
546 } else if (gm->green_metadata_type == 1) {
547 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_type: %d\n", gm->xsd_metric_type);
548
549 if (gm->xsd_metric_type == 0)
550 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_value: %f\n",
551 (float)gm->xsd_metric_value/100);
552 }
553 }
554
decode_nal_units(H264Context * h,const uint8_t * buf,int buf_size)555 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
556 {
557 AVCodecContext *const avctx = h->avctx;
558 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
559 int idr_cleared=0;
560 int i, ret = 0;
561
562 h->has_slice = 0;
563 h->nal_unit_type= 0;
564
565 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
566 h->current_slice = 0;
567 if (!h->first_field) {
568 h->cur_pic_ptr = NULL;
569 ff_h264_sei_uninit(&h->sei);
570 }
571 }
572
573 if (h->nal_length_size == 4) {
574 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
575 h->is_avc = 0;
576 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
577 h->is_avc = 1;
578 }
579
580 ret = ff_h2645_packet_split(&h->pkt, buf, buf_size, avctx, h->is_avc, h->nal_length_size,
581 avctx->codec_id, 0, 0);
582 if (ret < 0) {
583 av_log(avctx, AV_LOG_ERROR,
584 "Error splitting the input into NAL units.\n");
585 return ret;
586 }
587
588 if (avctx->active_thread_type & FF_THREAD_FRAME)
589 nals_needed = get_last_needed_nal(h);
590 if (nals_needed < 0)
591 return nals_needed;
592
593 for (i = 0; i < h->pkt.nb_nals; i++) {
594 H2645NAL *nal = &h->pkt.nals[i];
595 int max_slice_ctx, err;
596
597 if (avctx->skip_frame >= AVDISCARD_NONREF &&
598 nal->ref_idc == 0 && nal->type != H264_NAL_SEI)
599 continue;
600
601 // FIXME these should stop being context-global variables
602 h->nal_ref_idc = nal->ref_idc;
603 h->nal_unit_type = nal->type;
604
605 err = 0;
606 switch (nal->type) {
607 case H264_NAL_IDR_SLICE:
608 if ((nal->data[1] & 0xFC) == 0x98) {
609 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
610 h->next_outputed_poc = INT_MIN;
611 ret = -1;
612 goto end;
613 }
614 if(!idr_cleared) {
615 idr(h); // FIXME ensure we don't lose some frames if there is reordering
616 }
617 idr_cleared = 1;
618 h->has_recovery_point = 1;
619 case H264_NAL_SLICE:
620 h->has_slice = 1;
621
622 if ((err = ff_h264_queue_decode_slice(h, nal))) {
623 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
624 sl->ref_count[0] = sl->ref_count[1] = 0;
625 break;
626 }
627
628 if (h->current_slice == 1) {
629 if (avctx->active_thread_type & FF_THREAD_FRAME &&
630 i >= nals_needed && !h->setup_finished && h->cur_pic_ptr) {
631 ff_thread_finish_setup(avctx);
632 h->setup_finished = 1;
633 }
634
635 if (h->avctx->hwaccel &&
636 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
637 goto end;
638 }
639
640 max_slice_ctx = avctx->hwaccel ? 1 : h->nb_slice_ctx;
641 if (h->nb_slice_ctx_queued == max_slice_ctx) {
642 if (h->avctx->hwaccel) {
643 ret = avctx->hwaccel->decode_slice(avctx, nal->raw_data, nal->raw_size);
644 h->nb_slice_ctx_queued = 0;
645 } else
646 ret = ff_h264_execute_decode_slices(h);
647 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
648 goto end;
649 }
650 break;
651 case H264_NAL_DPA:
652 case H264_NAL_DPB:
653 case H264_NAL_DPC:
654 avpriv_request_sample(avctx, "data partitioning");
655 break;
656 case H264_NAL_SEI:
657 ret = ff_h264_sei_decode(&h->sei, &nal->gb, &h->ps, avctx);
658 h->has_recovery_point = h->has_recovery_point || h->sei.recovery_point.recovery_frame_cnt != -1;
659 if (avctx->debug & FF_DEBUG_GREEN_MD)
660 debug_green_metadata(&h->sei.green_metadata, h->avctx);
661 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
662 goto end;
663 break;
664 case H264_NAL_SPS: {
665 GetBitContext tmp_gb = nal->gb;
666 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
667 ret = avctx->hwaccel->decode_params(avctx,
668 nal->type,
669 nal->raw_data,
670 nal->raw_size);
671 if (ret < 0)
672 goto end;
673 }
674 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
675 break;
676 av_log(h->avctx, AV_LOG_DEBUG,
677 "SPS decoding failure, trying again with the complete NAL\n");
678 init_get_bits8(&tmp_gb, nal->raw_data + 1, nal->raw_size - 1);
679 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
680 break;
681 ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps, 1);
682 break;
683 }
684 case H264_NAL_PPS:
685 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
686 ret = avctx->hwaccel->decode_params(avctx,
687 nal->type,
688 nal->raw_data,
689 nal->raw_size);
690 if (ret < 0)
691 goto end;
692 }
693 ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
694 nal->size_bits);
695 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
696 goto end;
697 break;
698 case H264_NAL_AUD:
699 case H264_NAL_END_SEQUENCE:
700 case H264_NAL_END_STREAM:
701 case H264_NAL_FILLER_DATA:
702 case H264_NAL_SPS_EXT:
703 case H264_NAL_AUXILIARY_SLICE:
704 break;
705 default:
706 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
707 nal->type, nal->size_bits);
708 }
709
710 if (err < 0) {
711 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
712 }
713 }
714
715 ret = ff_h264_execute_decode_slices(h);
716 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
717 goto end;
718
719 // set decode_error_flags to allow users to detect concealed decoding errors
720 if ((ret < 0 || h->slice_ctx->er.error_occurred) && h->cur_pic_ptr) {
721 h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
722 }
723
724 ret = 0;
725 end:
726
727 #if CONFIG_ERROR_RESILIENCE
728 /*
729 * FIXME: Error handling code does not seem to support interlaced
730 * when slices span multiple rows
731 * The ff_er_add_slice calls don't work right for bottom
732 * fields; they cause massive erroneous error concealing
733 * Error marking covers both fields (top and bottom).
734 * This causes a mismatched s->error_count
735 * and a bad error table. Further, the error count goes to
736 * INT_MAX when called for bottom field, because mb_y is
737 * past end by one (callers fault) and resync_mb_y != 0
738 * causes problems for the first MB line, too.
739 */
740 if (!FIELD_PICTURE(h) && h->current_slice && h->enable_er) {
741
742 H264SliceContext *sl = h->slice_ctx;
743 int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
744
745 ff_h264_set_erpic(&sl->er.cur_pic, h->cur_pic_ptr);
746
747 if (use_last_pic) {
748 ff_h264_set_erpic(&sl->er.last_pic, &h->last_pic_for_ec);
749 sl->ref_list[0][0].parent = &h->last_pic_for_ec;
750 memcpy(sl->ref_list[0][0].data, h->last_pic_for_ec.f->data, sizeof(sl->ref_list[0][0].data));
751 memcpy(sl->ref_list[0][0].linesize, h->last_pic_for_ec.f->linesize, sizeof(sl->ref_list[0][0].linesize));
752 sl->ref_list[0][0].reference = h->last_pic_for_ec.reference;
753 } else if (sl->ref_count[0]) {
754 ff_h264_set_erpic(&sl->er.last_pic, sl->ref_list[0][0].parent);
755 } else
756 ff_h264_set_erpic(&sl->er.last_pic, NULL);
757
758 if (sl->ref_count[1])
759 ff_h264_set_erpic(&sl->er.next_pic, sl->ref_list[1][0].parent);
760
761 sl->er.ref_count = sl->ref_count[0];
762
763 ff_er_frame_end(&sl->er);
764 if (use_last_pic)
765 memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
766 }
767 #endif /* CONFIG_ERROR_RESILIENCE */
768 /* clean up */
769 if (h->cur_pic_ptr && !h->droppable && h->has_slice) {
770 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
771 h->picture_structure == PICT_BOTTOM_FIELD);
772 }
773
774 return (ret < 0) ? ret : buf_size;
775 }
776
777 /**
778 * Return the number of bytes consumed for building the current frame.
779 */
get_consumed_bytes(int pos,int buf_size)780 static int get_consumed_bytes(int pos, int buf_size)
781 {
782 if (pos == 0)
783 pos = 1; // avoid infinite loops (I doubt that is needed but...)
784 if (pos + 10 > buf_size)
785 pos = buf_size; // oops ;)
786
787 return pos;
788 }
789
h264_export_enc_params(AVFrame * f,H264Picture * p)790 static int h264_export_enc_params(AVFrame *f, H264Picture *p)
791 {
792 AVVideoEncParams *par;
793 unsigned int nb_mb = p->mb_height * p->mb_width;
794 unsigned int x, y;
795
796 par = av_video_enc_params_create_side_data(f, AV_VIDEO_ENC_PARAMS_H264, nb_mb);
797 if (!par)
798 return AVERROR(ENOMEM);
799
800 par->qp = p->pps->init_qp;
801
802 par->delta_qp[1][0] = p->pps->chroma_qp_index_offset[0];
803 par->delta_qp[1][1] = p->pps->chroma_qp_index_offset[0];
804 par->delta_qp[2][0] = p->pps->chroma_qp_index_offset[1];
805 par->delta_qp[2][1] = p->pps->chroma_qp_index_offset[1];
806
807 for (y = 0; y < p->mb_height; y++)
808 for (x = 0; x < p->mb_width; x++) {
809 const unsigned int block_idx = y * p->mb_width + x;
810 const unsigned int mb_xy = y * p->mb_stride + x;
811 AVVideoBlockParams *b = av_video_enc_params_block(par, block_idx);
812
813 b->src_x = x * 16;
814 b->src_y = y * 16;
815 b->w = 16;
816 b->h = 16;
817
818 b->delta_qp = p->qscale_table[mb_xy] - par->qp;
819 }
820
821 return 0;
822 }
823
output_frame(H264Context * h,AVFrame * dst,H264Picture * srcp)824 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
825 {
826 AVFrame *src = srcp->f;
827 int ret;
828
829 ret = av_frame_ref(dst, src);
830 if (ret < 0)
831 return ret;
832
833 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.frame_packing), 0);
834
835 if (srcp->sei_recovery_frame_cnt == 0)
836 dst->key_frame = 1;
837
838 if (h->avctx->export_side_data & AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS) {
839 ret = h264_export_enc_params(dst, srcp);
840 if (ret < 0)
841 goto fail;
842 }
843
844 return 0;
845 fail:
846 av_frame_unref(dst);
847 return ret;
848 }
849
is_avcc_extradata(const uint8_t * buf,int buf_size)850 static int is_avcc_extradata(const uint8_t *buf, int buf_size)
851 {
852 int cnt= buf[5]&0x1f;
853 const uint8_t *p= buf+6;
854 if (!cnt)
855 return 0;
856 while(cnt--){
857 int nalsize= AV_RB16(p) + 2;
858 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 7)
859 return 0;
860 p += nalsize;
861 }
862 cnt = *(p++);
863 if(!cnt)
864 return 0;
865 while(cnt--){
866 int nalsize= AV_RB16(p) + 2;
867 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 8)
868 return 0;
869 p += nalsize;
870 }
871 return 1;
872 }
873
finalize_frame(H264Context * h,AVFrame * dst,H264Picture * out,int * got_frame)874 static int finalize_frame(H264Context *h, AVFrame *dst, H264Picture *out, int *got_frame)
875 {
876 int ret;
877
878 if (((h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
879 (h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
880 out->recovered)) {
881
882 if (!h->avctx->hwaccel &&
883 (out->field_poc[0] == INT_MAX ||
884 out->field_poc[1] == INT_MAX)
885 ) {
886 int p;
887 AVFrame *f = out->f;
888 int field = out->field_poc[0] == INT_MAX;
889 uint8_t *dst_data[4];
890 int linesizes[4];
891 const uint8_t *src_data[4];
892
893 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
894
895 for (p = 0; p<4; p++) {
896 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
897 src_data[p] = f->data[p] + field *f->linesize[p];
898 linesizes[p] = 2*f->linesize[p];
899 }
900
901 av_image_copy(dst_data, linesizes, src_data, linesizes,
902 f->format, f->width, f->height>>1);
903 }
904
905 ret = output_frame(h, dst, out);
906 if (ret < 0)
907 return ret;
908
909 *got_frame = 1;
910
911 if (CONFIG_MPEGVIDEO) {
912 ff_print_debug_info2(h->avctx, dst, NULL,
913 out->mb_type,
914 out->qscale_table,
915 out->motion_val,
916 NULL,
917 out->mb_width, out->mb_height, out->mb_stride, 1);
918 }
919 }
920
921 return 0;
922 }
923
send_next_delayed_frame(H264Context * h,AVFrame * dst_frame,int * got_frame,int buf_index)924 static int send_next_delayed_frame(H264Context *h, AVFrame *dst_frame,
925 int *got_frame, int buf_index)
926 {
927 int ret, i, out_idx;
928 H264Picture *out = h->delayed_pic[0];
929
930 h->cur_pic_ptr = NULL;
931 h->first_field = 0;
932
933 out_idx = 0;
934 for (i = 1;
935 h->delayed_pic[i] &&
936 !h->delayed_pic[i]->f->key_frame &&
937 !h->delayed_pic[i]->mmco_reset;
938 i++)
939 if (h->delayed_pic[i]->poc < out->poc) {
940 out = h->delayed_pic[i];
941 out_idx = i;
942 }
943
944 for (i = out_idx; h->delayed_pic[i]; i++)
945 h->delayed_pic[i] = h->delayed_pic[i + 1];
946
947 if (out) {
948 out->reference &= ~DELAYED_PIC_REF;
949 ret = finalize_frame(h, dst_frame, out, got_frame);
950 if (ret < 0)
951 return ret;
952 }
953
954 return buf_index;
955 }
956
h264_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)957 static int h264_decode_frame(AVCodecContext *avctx, void *data,
958 int *got_frame, AVPacket *avpkt)
959 {
960 const uint8_t *buf = avpkt->data;
961 int buf_size = avpkt->size;
962 H264Context *h = avctx->priv_data;
963 AVFrame *pict = data;
964 int buf_index;
965 int ret;
966
967 h->flags = avctx->flags;
968 h->setup_finished = 0;
969 h->nb_slice_ctx_queued = 0;
970
971 ff_h264_unref_picture(h, &h->last_pic_for_ec);
972
973 /* end of stream, output what is still in the buffers */
974 if (buf_size == 0)
975 return send_next_delayed_frame(h, pict, got_frame, 0);
976
977 if (av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
978 buffer_size_t side_size;
979 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
980 ff_h264_decode_extradata(side, side_size,
981 &h->ps, &h->is_avc, &h->nal_length_size,
982 avctx->err_recognition, avctx);
983 }
984 if (h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC) {
985 if (is_avcc_extradata(buf, buf_size))
986 return ff_h264_decode_extradata(buf, buf_size,
987 &h->ps, &h->is_avc, &h->nal_length_size,
988 avctx->err_recognition, avctx);
989 }
990
991 buf_index = decode_nal_units(h, buf, buf_size);
992 if (buf_index < 0)
993 return AVERROR_INVALIDDATA;
994
995 if (!h->cur_pic_ptr && h->nal_unit_type == H264_NAL_END_SEQUENCE) {
996 av_assert0(buf_index <= buf_size);
997 return send_next_delayed_frame(h, pict, got_frame, buf_index);
998 }
999
1000 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && (!h->cur_pic_ptr || !h->has_slice)) {
1001 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1002 buf_size >= 4 && !memcmp("Q264", buf, 4))
1003 return buf_size;
1004 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1005 return AVERROR_INVALIDDATA;
1006 }
1007
1008 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1009 (h->mb_y >= h->mb_height && h->mb_height)) {
1010 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1011 return ret;
1012
1013 /* Wait for second field. */
1014 if (h->next_output_pic) {
1015 ret = finalize_frame(h, pict, h->next_output_pic, got_frame);
1016 if (ret < 0)
1017 return ret;
1018 }
1019 }
1020
1021 av_assert0(pict->buf[0] || !*got_frame);
1022
1023 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1024
1025 return get_consumed_bytes(buf_index, buf_size);
1026 }
1027
1028 #define OFFSET(x) offsetof(H264Context, x)
1029 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1030 static const AVOption h264_options[] = {
1031 { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0 },
1032 { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0 },
1033 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1034 { "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 },
1035 { NULL },
1036 };
1037
1038 static const AVClass h264_class = {
1039 .class_name = "H264 Decoder",
1040 .item_name = av_default_item_name,
1041 .option = h264_options,
1042 .version = LIBAVUTIL_VERSION_INT,
1043 };
1044
1045 AVCodec ff_h264_decoder = {
1046 .name = "h264",
1047 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1048 .type = AVMEDIA_TYPE_VIDEO,
1049 .id = AV_CODEC_ID_H264,
1050 .priv_data_size = sizeof(H264Context),
1051 .init = h264_decode_init,
1052 .close = h264_decode_end,
1053 .decode = h264_decode_frame,
1054 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1055 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1056 AV_CODEC_CAP_FRAME_THREADS,
1057 .hw_configs = (const AVCodecHWConfigInternal *const []) {
1058 #if CONFIG_H264_DXVA2_HWACCEL
1059 HWACCEL_DXVA2(h264),
1060 #endif
1061 #if CONFIG_H264_D3D11VA_HWACCEL
1062 HWACCEL_D3D11VA(h264),
1063 #endif
1064 #if CONFIG_H264_D3D11VA2_HWACCEL
1065 HWACCEL_D3D11VA2(h264),
1066 #endif
1067 #if CONFIG_H264_NVDEC_HWACCEL
1068 HWACCEL_NVDEC(h264),
1069 #endif
1070 #if CONFIG_H264_VAAPI_HWACCEL
1071 HWACCEL_VAAPI(h264),
1072 #endif
1073 #if CONFIG_H264_VDPAU_HWACCEL
1074 HWACCEL_VDPAU(h264),
1075 #endif
1076 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
1077 HWACCEL_VIDEOTOOLBOX(h264),
1078 #endif
1079 NULL
1080 },
1081 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_EXPORTS_CROPPING |
1082 FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
1083 .flush = h264_decode_flush,
1084 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1085 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1086 .priv_class = &h264_class,
1087 };
1088