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 int x, y;
186
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;
190
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];
205
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)
210
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;
214
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;
223
224 h->mb2b_xy[mb_xy] = b_xy;
225 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
226 }
227
228 return 0;
229
230 fail:
231 return AVERROR(ENOMEM);
232 }
233
234 /**
235 * Init context
236 * Allocate buffers which are not shared amongst multiple threads.
237 */
ff_h264_slice_context_init(H264Context * h,H264SliceContext * sl)238 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
239 {
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;
245 int x, y, i;
246
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;
253
254 if (sl != h->slice_ctx) {
255 memset(er, 0, sizeof(*er));
256 } else
257 if (CONFIG_ERROR_RESILIENCE) {
258
259 /* init ER */
260 er->avctx = h->avctx;
261 er->decode_mb = h264_er_decode_mb;
262 er->opaque = h;
263 er->quarter_sample = 1;
264
265 er->mb_num = h->mb_num;
266 er->mb_width = h->mb_width;
267 er->mb_height = h->mb_height;
268 er->mb_stride = h->mb_stride;
269 er->b8_stride = h->mb_width * 2 + 1;
270
271 // error resilience code looks cleaner with this
272 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
273 (h->mb_num + 1) * sizeof(int), fail);
274
275 for (y = 0; y < h->mb_height; y++)
276 for (x = 0; x < h->mb_width; x++)
277 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
278
279 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
280 h->mb_stride + h->mb_width;
281
282 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
283 mb_array_size * sizeof(uint8_t), fail);
284
285 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
286 h->mb_height * h->mb_stride * (4*sizeof(int) + 1), fail);
287
288 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
289 yc_size * sizeof(int16_t), fail);
290 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
291 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
292 er->dc_val[2] = er->dc_val[1] + c_size;
293 for (i = 0; i < yc_size; i++)
294 sl->dc_val_base[i] = 1024;
295 }
296
297 return 0;
298
299 fail:
300 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
301 }
302
h264_init_context(AVCodecContext * avctx,H264Context * h)303 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
304 {
305 int i;
306
307 h->avctx = avctx;
308 h->cur_chroma_format_idc = -1;
309
310 h->width_from_caller = avctx->width;
311 h->height_from_caller = avctx->height;
312
313 h->workaround_bugs = avctx->workaround_bugs;
314 h->flags = avctx->flags;
315 h->poc.prev_poc_msb = 1 << 16;
316 h->recovery_frame = -1;
317 h->frame_recovered = 0;
318 h->poc.prev_frame_num = -1;
319 h->sei.frame_packing.arrangement_cancel_flag = -1;
320 h->sei.unregistered.x264_build = -1;
321
322 h->next_outputed_poc = INT_MIN;
323 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
324 h->last_pocs[i] = INT_MIN;
325
326 ff_h264_sei_uninit(&h->sei);
327
328 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
329 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
330 if (!h->slice_ctx) {
331 h->nb_slice_ctx = 0;
332 return AVERROR(ENOMEM);
333 }
334
335 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
336 h->DPB[i].f = av_frame_alloc();
337 if (!h->DPB[i].f)
338 return AVERROR(ENOMEM);
339 }
340
341 h->cur_pic.f = av_frame_alloc();
342 if (!h->cur_pic.f)
343 return AVERROR(ENOMEM);
344
345 h->last_pic_for_ec.f = av_frame_alloc();
346 if (!h->last_pic_for_ec.f)
347 return AVERROR(ENOMEM);
348
349 for (i = 0; i < h->nb_slice_ctx; i++)
350 h->slice_ctx[i].h264 = h;
351
352 return 0;
353 }
354
h264_decode_end(AVCodecContext * avctx)355 static av_cold int h264_decode_end(AVCodecContext *avctx)
356 {
357 H264Context *h = avctx->priv_data;
358 int i;
359
360 ff_h264_remove_all_refs(h);
361 ff_h264_free_tables(h);
362
363 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
364 ff_h264_unref_picture(h, &h->DPB[i]);
365 av_frame_free(&h->DPB[i].f);
366 }
367 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
368
369 h->cur_pic_ptr = NULL;
370
371 av_freep(&h->slice_ctx);
372 h->nb_slice_ctx = 0;
373
374 ff_h264_sei_uninit(&h->sei);
375 ff_h264_ps_uninit(&h->ps);
376
377 ff_h2645_packet_uninit(&h->pkt);
378
379 ff_h264_unref_picture(h, &h->cur_pic);
380 av_frame_free(&h->cur_pic.f);
381 ff_h264_unref_picture(h, &h->last_pic_for_ec);
382 av_frame_free(&h->last_pic_for_ec.f);
383
384 return 0;
385 }
386
387 static AVOnce h264_vlc_init = AV_ONCE_INIT;
388
h264_decode_init(AVCodecContext * avctx)389 static av_cold int h264_decode_init(AVCodecContext *avctx)
390 {
391 H264Context *h = avctx->priv_data;
392 int ret;
393
394 ret = h264_init_context(avctx, h);
395 if (ret < 0)
396 return ret;
397
398 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
399 if (ret != 0) {
400 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
401 return AVERROR_UNKNOWN;
402 }
403
404 if (avctx->ticks_per_frame == 1) {
405 if(h->avctx->time_base.den < INT_MAX/2) {
406 h->avctx->time_base.den *= 2;
407 } else
408 h->avctx->time_base.num /= 2;
409 }
410 avctx->ticks_per_frame = 2;
411
412 if (!avctx->internal->is_copy) {
413 if (avctx->extradata_size > 0 && avctx->extradata) {
414 ret = ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
415 &h->ps, &h->is_avc, &h->nal_length_size,
416 avctx->err_recognition, avctx);
417 if (ret < 0) {
418 int explode = avctx->err_recognition & AV_EF_EXPLODE;
419 av_log(avctx, explode ? AV_LOG_ERROR: AV_LOG_WARNING,
420 "Error decoding the extradata\n");
421 if (explode) {
422 return ret;
423 }
424 ret = 0;
425 }
426 }
427 }
428
429 if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
430 h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
431 h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
432 }
433
434 ff_h264_flush_change(h);
435
436 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
437 h->enable_er = 0;
438
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");
443 }
444
445 return 0;
446 }
447
448 /**
449 * instantaneous decoder refresh.
450 */
idr(H264Context * h)451 static void idr(H264Context *h)
452 {
453 int i;
454 ff_h264_remove_all_refs(h);
455 h->poc.prev_frame_num =
456 h->poc.prev_frame_num_offset = 0;
457 h->poc.prev_poc_msb = 1<<16;
458 h->poc.prev_poc_lsb = -1;
459 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
460 h->last_pocs[i] = INT_MIN;
461 }
462
463 /* forget old pics after a seek */
ff_h264_flush_change(H264Context * h)464 void ff_h264_flush_change(H264Context *h)
465 {
466 int i, j;
467
468 h->next_outputed_poc = INT_MIN;
469 h->prev_interlaced_frame = 1;
470 idr(h);
471
472 h->poc.prev_frame_num = -1;
473 if (h->cur_pic_ptr) {
474 h->cur_pic_ptr->reference = 0;
475 for (j=i=0; h->delayed_pic[i]; i++)
476 if (h->delayed_pic[i] != h->cur_pic_ptr)
477 h->delayed_pic[j++] = h->delayed_pic[i];
478 h->delayed_pic[j] = NULL;
479 }
480 ff_h264_unref_picture(h, &h->last_pic_for_ec);
481
482 h->first_field = 0;
483 h->recovery_frame = -1;
484 h->frame_recovered = 0;
485 h->current_slice = 0;
486 h->mmco_reset = 1;
487 }
488
h264_decode_flush(AVCodecContext * avctx)489 static void h264_decode_flush(AVCodecContext *avctx)
490 {
491 H264Context *h = avctx->priv_data;
492 int i;
493
494 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
495
496 ff_h264_flush_change(h);
497 ff_h264_sei_uninit(&h->sei);
498
499 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
500 ff_h264_unref_picture(h, &h->DPB[i]);
501 h->cur_pic_ptr = NULL;
502 ff_h264_unref_picture(h, &h->cur_pic);
503
504 h->mb_y = 0;
505
506 ff_h264_free_tables(h);
507 h->context_initialized = 0;
508 }
509
get_last_needed_nal(H264Context * h)510 static int get_last_needed_nal(H264Context *h)
511 {
512 int nals_needed = 0;
513 int first_slice = 0;
514 int i, ret;
515
516 for (i = 0; i < h->pkt.nb_nals; i++) {
517 H2645NAL *nal = &h->pkt.nals[i];
518 GetBitContext gb;
519
520 /* packets can sometimes contain multiple PPS/SPS,
521 * e.g. two PAFF field pictures in one packet, or a demuxer
522 * which splits NALs strangely if so, when frame threading we
523 * can't start the next thread until we've read all of them */
524 switch (nal->type) {
525 case H264_NAL_SPS:
526 case H264_NAL_PPS:
527 nals_needed = i;
528 break;
529 case H264_NAL_DPA:
530 case H264_NAL_IDR_SLICE:
531 case H264_NAL_SLICE:
532 ret = init_get_bits8(&gb, nal->data + 1, nal->size - 1);
533 if (ret < 0) {
534 av_log(h->avctx, AV_LOG_ERROR, "Invalid zero-sized VCL NAL unit\n");
535 if (h->avctx->err_recognition & AV_EF_EXPLODE)
536 return ret;
537
538 break;
539 }
540 if (!get_ue_golomb_long(&gb) || // first_mb_in_slice
541 !first_slice ||
542 first_slice != nal->type)
543 nals_needed = i;
544 if (!first_slice)
545 first_slice = nal->type;
546 }
547 }
548
549 return nals_needed;
550 }
551
debug_green_metadata(const H264SEIGreenMetaData * gm,void * logctx)552 static void debug_green_metadata(const H264SEIGreenMetaData *gm, void *logctx)
553 {
554 av_log(logctx, AV_LOG_DEBUG, "Green Metadata Info SEI message\n");
555 av_log(logctx, AV_LOG_DEBUG, " green_metadata_type: %d\n", gm->green_metadata_type);
556
557 if (gm->green_metadata_type == 0) {
558 av_log(logctx, AV_LOG_DEBUG, " green_metadata_period_type: %d\n", gm->period_type);
559
560 if (gm->period_type == 2)
561 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_seconds: %d\n", gm->num_seconds);
562 else if (gm->period_type == 3)
563 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_pictures: %d\n", gm->num_pictures);
564
565 av_log(logctx, AV_LOG_DEBUG, " SEI GREEN Complexity Metrics: %f %f %f %f\n",
566 (float)gm->percent_non_zero_macroblocks/255,
567 (float)gm->percent_intra_coded_macroblocks/255,
568 (float)gm->percent_six_tap_filtering/255,
569 (float)gm->percent_alpha_point_deblocking_instance/255);
570
571 } else if (gm->green_metadata_type == 1) {
572 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_type: %d\n", gm->xsd_metric_type);
573
574 if (gm->xsd_metric_type == 0)
575 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_value: %f\n",
576 (float)gm->xsd_metric_value/100);
577 }
578 }
579
decode_nal_units(H264Context * h,const uint8_t * buf,int buf_size)580 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
581 {
582 AVCodecContext *const avctx = h->avctx;
583 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
584 int idr_cleared=0;
585 int i, ret = 0;
586
587 h->has_slice = 0;
588 h->nal_unit_type= 0;
589
590 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
591 h->current_slice = 0;
592 if (!h->first_field) {
593 h->cur_pic_ptr = NULL;
594 ff_h264_sei_uninit(&h->sei);
595 }
596 }
597
598 if (h->nal_length_size == 4) {
599 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
600 h->is_avc = 0;
601 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
602 h->is_avc = 1;
603 }
604
605 ret = ff_h2645_packet_split(&h->pkt, buf, buf_size, avctx, h->is_avc, h->nal_length_size,
606 avctx->codec_id, 0, 0);
607 if (ret < 0) {
608 av_log(avctx, AV_LOG_ERROR,
609 "Error splitting the input into NAL units.\n");
610 return ret;
611 }
612
613 if (avctx->active_thread_type & FF_THREAD_FRAME)
614 nals_needed = get_last_needed_nal(h);
615 if (nals_needed < 0)
616 return nals_needed;
617
618 for (i = 0; i < h->pkt.nb_nals; i++) {
619 H2645NAL *nal = &h->pkt.nals[i];
620 int max_slice_ctx, err;
621
622 if (avctx->skip_frame >= AVDISCARD_NONREF &&
623 nal->ref_idc == 0 && nal->type != H264_NAL_SEI)
624 continue;
625
626 // FIXME these should stop being context-global variables
627 h->nal_ref_idc = nal->ref_idc;
628 h->nal_unit_type = nal->type;
629
630 err = 0;
631 switch (nal->type) {
632 case H264_NAL_IDR_SLICE:
633 if ((nal->data[1] & 0xFC) == 0x98) {
634 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
635 h->next_outputed_poc = INT_MIN;
636 ret = -1;
637 goto end;
638 }
639 if(!idr_cleared) {
640 idr(h); // FIXME ensure we don't lose some frames if there is reordering
641 }
642 idr_cleared = 1;
643 h->has_recovery_point = 1;
644 case H264_NAL_SLICE:
645 h->has_slice = 1;
646
647 if ((err = ff_h264_queue_decode_slice(h, nal))) {
648 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
649 sl->ref_count[0] = sl->ref_count[1] = 0;
650 break;
651 }
652
653 if (h->current_slice == 1) {
654 if (avctx->active_thread_type & FF_THREAD_FRAME &&
655 i >= nals_needed && !h->setup_finished && h->cur_pic_ptr) {
656 ff_thread_finish_setup(avctx);
657 h->setup_finished = 1;
658 }
659
660 if (h->avctx->hwaccel &&
661 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
662 goto end;
663 }
664
665 max_slice_ctx = avctx->hwaccel ? 1 : h->nb_slice_ctx;
666 if (h->nb_slice_ctx_queued == max_slice_ctx) {
667 if (h->avctx->hwaccel) {
668 ret = avctx->hwaccel->decode_slice(avctx, nal->raw_data, nal->raw_size);
669 h->nb_slice_ctx_queued = 0;
670 } else
671 ret = ff_h264_execute_decode_slices(h);
672 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
673 goto end;
674 }
675 break;
676 case H264_NAL_DPA:
677 case H264_NAL_DPB:
678 case H264_NAL_DPC:
679 avpriv_request_sample(avctx, "data partitioning");
680 break;
681 case H264_NAL_SEI:
682 ret = ff_h264_sei_decode(&h->sei, &nal->gb, &h->ps, avctx);
683 h->has_recovery_point = h->has_recovery_point || h->sei.recovery_point.recovery_frame_cnt != -1;
684 if (avctx->debug & FF_DEBUG_GREEN_MD)
685 debug_green_metadata(&h->sei.green_metadata, h->avctx);
686 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
687 goto end;
688 break;
689 case H264_NAL_SPS: {
690 GetBitContext tmp_gb = nal->gb;
691 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
692 ret = avctx->hwaccel->decode_params(avctx,
693 nal->type,
694 nal->raw_data,
695 nal->raw_size);
696 if (ret < 0)
697 goto end;
698 }
699 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
700 break;
701 av_log(h->avctx, AV_LOG_DEBUG,
702 "SPS decoding failure, trying again with the complete NAL\n");
703 init_get_bits8(&tmp_gb, nal->raw_data + 1, nal->raw_size - 1);
704 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
705 break;
706 ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps, 1);
707 break;
708 }
709 case H264_NAL_PPS:
710 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
711 ret = avctx->hwaccel->decode_params(avctx,
712 nal->type,
713 nal->raw_data,
714 nal->raw_size);
715 if (ret < 0)
716 goto end;
717 }
718 ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
719 nal->size_bits);
720 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
721 goto end;
722 break;
723 case H264_NAL_AUD:
724 case H264_NAL_END_SEQUENCE:
725 case H264_NAL_END_STREAM:
726 case H264_NAL_FILLER_DATA:
727 case H264_NAL_SPS_EXT:
728 case H264_NAL_AUXILIARY_SLICE:
729 break;
730 default:
731 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
732 nal->type, nal->size_bits);
733 }
734
735 if (err < 0) {
736 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
737 }
738 }
739
740 ret = ff_h264_execute_decode_slices(h);
741 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
742 goto end;
743
744 // set decode_error_flags to allow users to detect concealed decoding errors
745 if ((ret < 0 || h->slice_ctx->er.error_occurred) && h->cur_pic_ptr) {
746 h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
747 }
748
749 ret = 0;
750 end:
751
752 #if CONFIG_ERROR_RESILIENCE
753 /*
754 * FIXME: Error handling code does not seem to support interlaced
755 * when slices span multiple rows
756 * The ff_er_add_slice calls don't work right for bottom
757 * fields; they cause massive erroneous error concealing
758 * Error marking covers both fields (top and bottom).
759 * This causes a mismatched s->error_count
760 * and a bad error table. Further, the error count goes to
761 * INT_MAX when called for bottom field, because mb_y is
762 * past end by one (callers fault) and resync_mb_y != 0
763 * causes problems for the first MB line, too.
764 */
765 if (!FIELD_PICTURE(h) && h->current_slice && h->enable_er) {
766
767 H264SliceContext *sl = h->slice_ctx;
768 int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
769
770 ff_h264_set_erpic(&sl->er.cur_pic, h->cur_pic_ptr);
771
772 if (use_last_pic) {
773 ff_h264_set_erpic(&sl->er.last_pic, &h->last_pic_for_ec);
774 sl->ref_list[0][0].parent = &h->last_pic_for_ec;
775 memcpy(sl->ref_list[0][0].data, h->last_pic_for_ec.f->data, sizeof(sl->ref_list[0][0].data));
776 memcpy(sl->ref_list[0][0].linesize, h->last_pic_for_ec.f->linesize, sizeof(sl->ref_list[0][0].linesize));
777 sl->ref_list[0][0].reference = h->last_pic_for_ec.reference;
778 } else if (sl->ref_count[0]) {
779 ff_h264_set_erpic(&sl->er.last_pic, sl->ref_list[0][0].parent);
780 } else
781 ff_h264_set_erpic(&sl->er.last_pic, NULL);
782
783 if (sl->ref_count[1])
784 ff_h264_set_erpic(&sl->er.next_pic, sl->ref_list[1][0].parent);
785
786 sl->er.ref_count = sl->ref_count[0];
787
788 ff_er_frame_end(&sl->er);
789 if (use_last_pic)
790 memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
791 }
792 #endif /* CONFIG_ERROR_RESILIENCE */
793 /* clean up */
794 if (h->cur_pic_ptr && !h->droppable && h->has_slice) {
795 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
796 h->picture_structure == PICT_BOTTOM_FIELD);
797 }
798
799 return (ret < 0) ? ret : buf_size;
800 }
801
802 /**
803 * Return the number of bytes consumed for building the current frame.
804 */
get_consumed_bytes(int pos,int buf_size)805 static int get_consumed_bytes(int pos, int buf_size)
806 {
807 if (pos == 0)
808 pos = 1; // avoid infinite loops (I doubt that is needed but...)
809 if (pos + 10 > buf_size)
810 pos = buf_size; // oops ;)
811
812 return pos;
813 }
814
h264_export_enc_params(AVFrame * f,H264Picture * p)815 static int h264_export_enc_params(AVFrame *f, H264Picture *p)
816 {
817 AVVideoEncParams *par;
818 unsigned int nb_mb = p->mb_height * p->mb_width;
819 unsigned int x, y;
820
821 par = av_video_enc_params_create_side_data(f, AV_VIDEO_ENC_PARAMS_H264, nb_mb);
822 if (!par)
823 return AVERROR(ENOMEM);
824
825 par->qp = p->pps->init_qp;
826
827 par->delta_qp[1][0] = p->pps->chroma_qp_index_offset[0];
828 par->delta_qp[1][1] = p->pps->chroma_qp_index_offset[0];
829 par->delta_qp[2][0] = p->pps->chroma_qp_index_offset[1];
830 par->delta_qp[2][1] = p->pps->chroma_qp_index_offset[1];
831
832 for (y = 0; y < p->mb_height; y++)
833 for (x = 0; x < p->mb_width; x++) {
834 const unsigned int block_idx = y * p->mb_width + x;
835 const unsigned int mb_xy = y * p->mb_stride + x;
836 AVVideoBlockParams *b = av_video_enc_params_block(par, block_idx);
837
838 b->src_x = x * 16;
839 b->src_y = y * 16;
840 b->w = 16;
841 b->h = 16;
842
843 b->delta_qp = p->qscale_table[mb_xy] - par->qp;
844 }
845
846 return 0;
847 }
848
output_frame(H264Context * h,AVFrame * dst,H264Picture * srcp)849 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
850 {
851 AVFrame *src = srcp->f;
852 int ret;
853
854 ret = av_frame_ref(dst, src);
855 if (ret < 0)
856 return ret;
857
858 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.frame_packing), 0);
859
860 if (srcp->sei_recovery_frame_cnt == 0)
861 dst->key_frame = 1;
862
863 if (h->avctx->export_side_data & AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS) {
864 ret = h264_export_enc_params(dst, srcp);
865 if (ret < 0)
866 goto fail;
867 }
868
869 return 0;
870 fail:
871 av_frame_unref(dst);
872 return ret;
873 }
874
is_extra(const uint8_t * buf,int buf_size)875 static int is_extra(const uint8_t *buf, int buf_size)
876 {
877 int cnt= buf[5]&0x1f;
878 const uint8_t *p= buf+6;
879 if (!cnt)
880 return 0;
881 while(cnt--){
882 int nalsize= AV_RB16(p) + 2;
883 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 7)
884 return 0;
885 p += nalsize;
886 }
887 cnt = *(p++);
888 if(!cnt)
889 return 0;
890 while(cnt--){
891 int nalsize= AV_RB16(p) + 2;
892 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 8)
893 return 0;
894 p += nalsize;
895 }
896 return 1;
897 }
898
finalize_frame(H264Context * h,AVFrame * dst,H264Picture * out,int * got_frame)899 static int finalize_frame(H264Context *h, AVFrame *dst, H264Picture *out, int *got_frame)
900 {
901 int ret;
902
903 if (((h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
904 (h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
905 out->recovered)) {
906
907 if (!h->avctx->hwaccel &&
908 (out->field_poc[0] == INT_MAX ||
909 out->field_poc[1] == INT_MAX)
910 ) {
911 int p;
912 AVFrame *f = out->f;
913 int field = out->field_poc[0] == INT_MAX;
914 uint8_t *dst_data[4];
915 int linesizes[4];
916 const uint8_t *src_data[4];
917
918 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
919
920 for (p = 0; p<4; p++) {
921 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
922 src_data[p] = f->data[p] + field *f->linesize[p];
923 linesizes[p] = 2*f->linesize[p];
924 }
925
926 av_image_copy(dst_data, linesizes, src_data, linesizes,
927 f->format, f->width, f->height>>1);
928 }
929
930 ret = output_frame(h, dst, out);
931 if (ret < 0)
932 return ret;
933
934 *got_frame = 1;
935
936 if (CONFIG_MPEGVIDEO) {
937 ff_print_debug_info2(h->avctx, dst, NULL,
938 out->mb_type,
939 out->qscale_table,
940 out->motion_val,
941 NULL,
942 h->mb_width, h->mb_height, h->mb_stride, 1);
943 }
944 }
945
946 return 0;
947 }
948
send_next_delayed_frame(H264Context * h,AVFrame * dst_frame,int * got_frame,int buf_index)949 static int send_next_delayed_frame(H264Context *h, AVFrame *dst_frame,
950 int *got_frame, int buf_index)
951 {
952 int ret, i, out_idx;
953 H264Picture *out = h->delayed_pic[0];
954
955 h->cur_pic_ptr = NULL;
956 h->first_field = 0;
957
958 out_idx = 0;
959 for (i = 1;
960 h->delayed_pic[i] &&
961 !h->delayed_pic[i]->f->key_frame &&
962 !h->delayed_pic[i]->mmco_reset;
963 i++)
964 if (h->delayed_pic[i]->poc < out->poc) {
965 out = h->delayed_pic[i];
966 out_idx = i;
967 }
968
969 for (i = out_idx; h->delayed_pic[i]; i++)
970 h->delayed_pic[i] = h->delayed_pic[i + 1];
971
972 if (out) {
973 out->reference &= ~DELAYED_PIC_REF;
974 ret = finalize_frame(h, dst_frame, out, got_frame);
975 if (ret < 0)
976 return ret;
977 }
978
979 return buf_index;
980 }
981
h264_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)982 static int h264_decode_frame(AVCodecContext *avctx, void *data,
983 int *got_frame, AVPacket *avpkt)
984 {
985 const uint8_t *buf = avpkt->data;
986 int buf_size = avpkt->size;
987 H264Context *h = avctx->priv_data;
988 AVFrame *pict = data;
989 int buf_index;
990 int ret;
991
992 h->flags = avctx->flags;
993 h->setup_finished = 0;
994 h->nb_slice_ctx_queued = 0;
995
996 ff_h264_unref_picture(h, &h->last_pic_for_ec);
997
998 /* end of stream, output what is still in the buffers */
999 if (buf_size == 0)
1000 return send_next_delayed_frame(h, pict, got_frame, 0);
1001
1002 if (h->is_avc && av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
1003 int side_size;
1004 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
1005 if (is_extra(side, side_size))
1006 ff_h264_decode_extradata(side, side_size,
1007 &h->ps, &h->is_avc, &h->nal_length_size,
1008 avctx->err_recognition, avctx);
1009 }
1010 if (h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC) {
1011 if (is_extra(buf, buf_size))
1012 return ff_h264_decode_extradata(buf, buf_size,
1013 &h->ps, &h->is_avc, &h->nal_length_size,
1014 avctx->err_recognition, avctx);
1015 }
1016
1017 buf_index = decode_nal_units(h, buf, buf_size);
1018 if (buf_index < 0)
1019 return AVERROR_INVALIDDATA;
1020
1021 if (!h->cur_pic_ptr && h->nal_unit_type == H264_NAL_END_SEQUENCE) {
1022 av_assert0(buf_index <= buf_size);
1023 return send_next_delayed_frame(h, pict, got_frame, buf_index);
1024 }
1025
1026 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && (!h->cur_pic_ptr || !h->has_slice)) {
1027 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1028 buf_size >= 4 && !memcmp("Q264", buf, 4))
1029 return buf_size;
1030 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1031 return AVERROR_INVALIDDATA;
1032 }
1033
1034 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1035 (h->mb_y >= h->mb_height && h->mb_height)) {
1036 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1037 return ret;
1038
1039 /* Wait for second field. */
1040 if (h->next_output_pic) {
1041 ret = finalize_frame(h, pict, h->next_output_pic, got_frame);
1042 if (ret < 0)
1043 return ret;
1044 }
1045 }
1046
1047 av_assert0(pict->buf[0] || !*got_frame);
1048
1049 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1050
1051 return get_consumed_bytes(buf_index, buf_size);
1052 }
1053
1054 #define OFFSET(x) offsetof(H264Context, x)
1055 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1056 static const AVOption h264_options[] = {
1057 { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0 },
1058 { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0 },
1059 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1060 { "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 },
1061 { NULL },
1062 };
1063
1064 static const AVClass h264_class = {
1065 .class_name = "H264 Decoder",
1066 .item_name = av_default_item_name,
1067 .option = h264_options,
1068 .version = LIBAVUTIL_VERSION_INT,
1069 };
1070
1071 AVCodec ff_h264_decoder = {
1072 .name = "h264",
1073 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1074 .type = AVMEDIA_TYPE_VIDEO,
1075 .id = AV_CODEC_ID_H264,
1076 .priv_data_size = sizeof(H264Context),
1077 .init = h264_decode_init,
1078 .close = h264_decode_end,
1079 .decode = h264_decode_frame,
1080 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1081 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1082 AV_CODEC_CAP_FRAME_THREADS,
1083 .hw_configs = (const AVCodecHWConfigInternal*[]) {
1084 #if CONFIG_H264_DXVA2_HWACCEL
1085 HWACCEL_DXVA2(h264),
1086 #endif
1087 #if CONFIG_H264_D3D11VA_HWACCEL
1088 HWACCEL_D3D11VA(h264),
1089 #endif
1090 #if CONFIG_H264_D3D11VA2_HWACCEL
1091 HWACCEL_D3D11VA2(h264),
1092 #endif
1093 #if CONFIG_H264_NVDEC_HWACCEL
1094 HWACCEL_NVDEC(h264),
1095 #endif
1096 #if CONFIG_H264_VAAPI_HWACCEL
1097 HWACCEL_VAAPI(h264),
1098 #endif
1099 #if CONFIG_H264_VDPAU_HWACCEL
1100 HWACCEL_VDPAU(h264),
1101 #endif
1102 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
1103 HWACCEL_VIDEOTOOLBOX(h264),
1104 #endif
1105 NULL
1106 },
1107 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_EXPORTS_CROPPING |
1108 FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
1109 .flush = h264_decode_flush,
1110 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1111 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1112 .priv_class = &h264_class,
1113 };
1114