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 #include "libavutil/avassert.h"
29 #include "libavutil/display.h"
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/stereo3d.h"
33 #include "libavutil/video_enc_params.h"
34
35 #include "internal.h"
36 #include "bytestream.h"
37 #include "cabac.h"
38 #include "cabac_functions.h"
39 #include "error_resilience.h"
40 #include "avcodec.h"
41 #include "h264.h"
42 #include "h264dec.h"
43 #include "h2645_parse.h"
44 #include "h264data.h"
45 #include "h264chroma.h"
46 #include "h264_mvpred.h"
47 #include "h264_ps.h"
48 #include "golomb.h"
49 #include "hwconfig.h"
50 #include "mathops.h"
51 #include "me_cmp.h"
52 #include "mpegutils.h"
53 #include "profiles.h"
54 #include "rectangle.h"
55 #include "thread.h"
56
57 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
58
avpriv_h264_has_num_reorder_frames(AVCodecContext * avctx)59 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
60 {
61 H264Context *h = avctx->priv_data;
62 return h && h->ps.sps ? h->ps.sps->num_reorder_frames : 0;
63 }
64
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)65 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
66 int (*mv)[2][4][2],
67 int mb_x, int mb_y, int mb_intra, int mb_skipped)
68 {
69 H264Context *h = opaque;
70 H264SliceContext *sl = &h->slice_ctx[0];
71
72 sl->mb_x = mb_x;
73 sl->mb_y = mb_y;
74 sl->mb_xy = mb_x + mb_y * h->mb_stride;
75 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
76 av_assert1(ref >= 0);
77 /* FIXME: It is possible albeit uncommon that slice references
78 * differ between slices. We take the easy approach and ignore
79 * it for now. If this turns out to have any relevance in
80 * practice then correct remapping should be added. */
81 if (ref >= sl->ref_count[0])
82 ref = 0;
83 if (!sl->ref_list[0][ref].data[0]) {
84 av_log(h->avctx, AV_LOG_DEBUG, "Reference not available for error concealing\n");
85 ref = 0;
86 }
87 if ((sl->ref_list[0][ref].reference&3) != 3) {
88 av_log(h->avctx, AV_LOG_DEBUG, "Reference invalid\n");
89 return;
90 }
91 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
92 2, 2, 2, ref, 1);
93 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
94 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
95 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
96 sl->mb_mbaff =
97 sl->mb_field_decoding_flag = 0;
98 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
99 }
100
ff_h264_draw_horiz_band(const H264Context * h,H264SliceContext * sl,int y,int height)101 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
102 int y, int height)
103 {
104 AVCodecContext *avctx = h->avctx;
105 const AVFrame *src = h->cur_pic.f;
106 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
107 int vshift = desc->log2_chroma_h;
108 const int field_pic = h->picture_structure != PICT_FRAME;
109 if (field_pic) {
110 height <<= 1;
111 y <<= 1;
112 }
113
114 height = FFMIN(height, avctx->height - y);
115
116 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
117 return;
118
119 if (avctx->draw_horiz_band) {
120 int offset[AV_NUM_DATA_POINTERS];
121 int i;
122
123 offset[0] = y * src->linesize[0];
124 offset[1] =
125 offset[2] = (y >> vshift) * src->linesize[1];
126 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
127 offset[i] = 0;
128
129 emms_c();
130
131 avctx->draw_horiz_band(avctx, src, offset,
132 y, h->picture_structure, height);
133 }
134 }
135
ff_h264_free_tables(H264Context * h)136 void ff_h264_free_tables(H264Context *h)
137 {
138 int i;
139
140 av_freep(&h->intra4x4_pred_mode);
141 av_freep(&h->chroma_pred_mode_table);
142 av_freep(&h->cbp_table);
143 av_freep(&h->mvd_table[0]);
144 av_freep(&h->mvd_table[1]);
145 av_freep(&h->direct_table);
146 av_freep(&h->non_zero_count);
147 av_freep(&h->slice_table_base);
148 h->slice_table = NULL;
149 av_freep(&h->list_counts);
150
151 av_freep(&h->mb2b_xy);
152 av_freep(&h->mb2br_xy);
153
154 av_buffer_pool_uninit(&h->qscale_table_pool);
155 av_buffer_pool_uninit(&h->mb_type_pool);
156 av_buffer_pool_uninit(&h->motion_val_pool);
157 av_buffer_pool_uninit(&h->ref_index_pool);
158
159 for (i = 0; i < h->nb_slice_ctx; i++) {
160 H264SliceContext *sl = &h->slice_ctx[i];
161
162 av_freep(&sl->dc_val_base);
163 av_freep(&sl->er.mb_index2xy);
164 av_freep(&sl->er.error_status_table);
165 av_freep(&sl->er.er_temp_buffer);
166
167 av_freep(&sl->bipred_scratchpad);
168 av_freep(&sl->edge_emu_buffer);
169 av_freep(&sl->top_borders[0]);
170 av_freep(&sl->top_borders[1]);
171
172 sl->bipred_scratchpad_allocated = 0;
173 sl->edge_emu_buffer_allocated = 0;
174 sl->top_borders_allocated[0] = 0;
175 sl->top_borders_allocated[1] = 0;
176 }
177 }
178
ff_h264_alloc_tables(H264Context * h)179 int ff_h264_alloc_tables(H264Context *h)
180 {
181 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
182 const int row_mb_num = 2*h->mb_stride*FFMAX(h->nb_slice_ctx, 1);
183 const int st_size = big_mb_num + h->mb_stride;
184 int x, y;
185
186 if (!FF_ALLOCZ_TYPED_ARRAY(h->intra4x4_pred_mode, row_mb_num * 8) ||
187 !FF_ALLOCZ_TYPED_ARRAY(h->non_zero_count, big_mb_num) ||
188 !FF_ALLOCZ_TYPED_ARRAY(h->slice_table_base, st_size) ||
189 !FF_ALLOCZ_TYPED_ARRAY(h->cbp_table, big_mb_num) ||
190 !FF_ALLOCZ_TYPED_ARRAY(h->chroma_pred_mode_table, big_mb_num) ||
191 !FF_ALLOCZ_TYPED_ARRAY(h->mvd_table[0], row_mb_num * 8) ||
192 !FF_ALLOCZ_TYPED_ARRAY(h->mvd_table[1], row_mb_num * 8) ||
193 !FF_ALLOCZ_TYPED_ARRAY(h->direct_table, big_mb_num * 4) ||
194 !FF_ALLOCZ_TYPED_ARRAY(h->list_counts, big_mb_num) ||
195 !FF_ALLOCZ_TYPED_ARRAY(h->mb2b_xy, big_mb_num) ||
196 !FF_ALLOCZ_TYPED_ARRAY(h->mb2br_xy, big_mb_num))
197 return AVERROR(ENOMEM);
198 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
199 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
200 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
201 memset(h->slice_table_base, -1,
202 st_size * sizeof(*h->slice_table_base));
203 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
204 for (y = 0; y < h->mb_height; y++)
205 for (x = 0; x < h->mb_width; x++) {
206 const int mb_xy = x + y * h->mb_stride;
207 const int b_xy = 4 * x + 4 * y * h->b_stride;
208
209 h->mb2b_xy[mb_xy] = b_xy;
210 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
211 }
212
213 return 0;
214 }
215
216 /**
217 * Init context
218 * Allocate buffers which are not shared amongst multiple threads.
219 */
ff_h264_slice_context_init(H264Context * h,H264SliceContext * sl)220 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
221 {
222 ERContext *er = &sl->er;
223 int mb_array_size = h->mb_height * h->mb_stride;
224 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
225 int c_size = h->mb_stride * (h->mb_height + 1);
226 int yc_size = y_size + 2 * c_size;
227 int x, y, i;
228
229 sl->ref_cache[0][scan8[5] + 1] =
230 sl->ref_cache[0][scan8[7] + 1] =
231 sl->ref_cache[0][scan8[13] + 1] =
232 sl->ref_cache[1][scan8[5] + 1] =
233 sl->ref_cache[1][scan8[7] + 1] =
234 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
235
236 if (sl != h->slice_ctx) {
237 memset(er, 0, sizeof(*er));
238 } else if (CONFIG_ERROR_RESILIENCE) {
239 const int er_size = h->mb_height * h->mb_stride * (4*sizeof(int) + 1);
240
241 /* init ER */
242 er->avctx = h->avctx;
243 er->decode_mb = h264_er_decode_mb;
244 er->opaque = h;
245 er->quarter_sample = 1;
246
247 er->mb_num = h->mb_num;
248 er->mb_width = h->mb_width;
249 er->mb_height = h->mb_height;
250 er->mb_stride = h->mb_stride;
251 er->b8_stride = h->mb_width * 2 + 1;
252
253 // error resilience code looks cleaner with this
254 if (!FF_ALLOCZ_TYPED_ARRAY(er->mb_index2xy, h->mb_num + 1) ||
255 !FF_ALLOCZ_TYPED_ARRAY(er->error_status_table, mb_array_size) ||
256 !FF_ALLOCZ_TYPED_ARRAY(er->er_temp_buffer, er_size) ||
257 !FF_ALLOCZ_TYPED_ARRAY(sl->dc_val_base, yc_size))
258 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
259
260 for (y = 0; y < h->mb_height; y++)
261 for (x = 0; x < h->mb_width; x++)
262 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
263
264 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
265 h->mb_stride + h->mb_width;
266 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
267 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
268 er->dc_val[2] = er->dc_val[1] + c_size;
269 for (i = 0; i < yc_size; i++)
270 sl->dc_val_base[i] = 1024;
271 }
272
273 return 0;
274 }
275
h264_init_context(AVCodecContext * avctx,H264Context * h)276 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
277 {
278 int i;
279
280 h->avctx = avctx;
281 h->cur_chroma_format_idc = -1;
282
283 h->width_from_caller = avctx->width;
284 h->height_from_caller = avctx->height;
285
286 h->workaround_bugs = avctx->workaround_bugs;
287 h->flags = avctx->flags;
288 h->poc.prev_poc_msb = 1 << 16;
289 h->recovery_frame = -1;
290 h->frame_recovered = 0;
291 h->poc.prev_frame_num = -1;
292 h->sei.frame_packing.arrangement_cancel_flag = -1;
293 h->sei.unregistered.x264_build = -1;
294
295 h->next_outputed_poc = INT_MIN;
296 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
297 h->last_pocs[i] = INT_MIN;
298
299 ff_h264_sei_uninit(&h->sei);
300
301 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? avctx->thread_count : 1;
302 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
303 if (!h->slice_ctx) {
304 h->nb_slice_ctx = 0;
305 return AVERROR(ENOMEM);
306 }
307
308 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
309 h->DPB[i].f = av_frame_alloc();
310 if (!h->DPB[i].f)
311 return AVERROR(ENOMEM);
312 }
313
314 h->cur_pic.f = av_frame_alloc();
315 if (!h->cur_pic.f)
316 return AVERROR(ENOMEM);
317
318 h->last_pic_for_ec.f = av_frame_alloc();
319 if (!h->last_pic_for_ec.f)
320 return AVERROR(ENOMEM);
321
322 for (i = 0; i < h->nb_slice_ctx; i++)
323 h->slice_ctx[i].h264 = h;
324
325 return 0;
326 }
327
h264_decode_end(AVCodecContext * avctx)328 static av_cold int h264_decode_end(AVCodecContext *avctx)
329 {
330 H264Context *h = avctx->priv_data;
331 int i;
332
333 ff_h264_remove_all_refs(h);
334 ff_h264_free_tables(h);
335
336 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
337 ff_h264_unref_picture(h, &h->DPB[i]);
338 av_frame_free(&h->DPB[i].f);
339 }
340 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
341
342 h->cur_pic_ptr = NULL;
343
344 av_freep(&h->slice_ctx);
345 h->nb_slice_ctx = 0;
346
347 ff_h264_sei_uninit(&h->sei);
348 ff_h264_ps_uninit(&h->ps);
349
350 ff_h2645_packet_uninit(&h->pkt);
351
352 ff_h264_unref_picture(h, &h->cur_pic);
353 av_frame_free(&h->cur_pic.f);
354 ff_h264_unref_picture(h, &h->last_pic_for_ec);
355 av_frame_free(&h->last_pic_for_ec.f);
356
357 return 0;
358 }
359
360 static AVOnce h264_vlc_init = AV_ONCE_INIT;
361
h264_decode_init(AVCodecContext * avctx)362 static av_cold int h264_decode_init(AVCodecContext *avctx)
363 {
364 H264Context *h = avctx->priv_data;
365 int ret;
366
367 ret = h264_init_context(avctx, h);
368 if (ret < 0)
369 return ret;
370
371 ret = ff_thread_once(&h264_vlc_init, ff_h264_decode_init_vlc);
372 if (ret != 0) {
373 av_log(avctx, AV_LOG_ERROR, "pthread_once has failed.");
374 return AVERROR_UNKNOWN;
375 }
376
377 if (avctx->ticks_per_frame == 1) {
378 if(h->avctx->time_base.den < INT_MAX/2) {
379 h->avctx->time_base.den *= 2;
380 } else
381 h->avctx->time_base.num /= 2;
382 }
383 avctx->ticks_per_frame = 2;
384
385 if (!avctx->internal->is_copy) {
386 if (avctx->extradata_size > 0 && avctx->extradata) {
387 ret = ff_h264_decode_extradata(avctx->extradata, avctx->extradata_size,
388 &h->ps, &h->is_avc, &h->nal_length_size,
389 avctx->err_recognition, avctx);
390 if (ret < 0) {
391 int explode = avctx->err_recognition & AV_EF_EXPLODE;
392 av_log(avctx, explode ? AV_LOG_ERROR: AV_LOG_WARNING,
393 "Error decoding the extradata\n");
394 if (explode) {
395 return ret;
396 }
397 ret = 0;
398 }
399 }
400 }
401
402 if (h->ps.sps && h->ps.sps->bitstream_restriction_flag &&
403 h->avctx->has_b_frames < h->ps.sps->num_reorder_frames) {
404 h->avctx->has_b_frames = h->ps.sps->num_reorder_frames;
405 }
406
407 ff_h264_flush_change(h);
408
409 if (h->enable_er < 0 && (avctx->active_thread_type & FF_THREAD_SLICE))
410 h->enable_er = 0;
411
412 if (h->enable_er && (avctx->active_thread_type & FF_THREAD_SLICE)) {
413 av_log(avctx, AV_LOG_WARNING,
414 "Error resilience with slice threads is enabled. It is unsafe and unsupported and may crash. "
415 "Use it at your own risk\n");
416 }
417
418 return 0;
419 }
420
421 /**
422 * instantaneous decoder refresh.
423 */
idr(H264Context * h)424 static void idr(H264Context *h)
425 {
426 int i;
427 ff_h264_remove_all_refs(h);
428 h->poc.prev_frame_num =
429 h->poc.prev_frame_num_offset = 0;
430 h->poc.prev_poc_msb = 1<<16;
431 h->poc.prev_poc_lsb = -1;
432 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
433 h->last_pocs[i] = INT_MIN;
434 }
435
436 /* forget old pics after a seek */
ff_h264_flush_change(H264Context * h)437 void ff_h264_flush_change(H264Context *h)
438 {
439 int i, j;
440
441 h->next_outputed_poc = INT_MIN;
442 h->prev_interlaced_frame = 1;
443 idr(h);
444
445 h->poc.prev_frame_num = -1;
446 if (h->cur_pic_ptr) {
447 h->cur_pic_ptr->reference = 0;
448 for (j=i=0; h->delayed_pic[i]; i++)
449 if (h->delayed_pic[i] != h->cur_pic_ptr)
450 h->delayed_pic[j++] = h->delayed_pic[i];
451 h->delayed_pic[j] = NULL;
452 }
453 ff_h264_unref_picture(h, &h->last_pic_for_ec);
454
455 h->first_field = 0;
456 h->recovery_frame = -1;
457 h->frame_recovered = 0;
458 h->current_slice = 0;
459 h->mmco_reset = 1;
460 }
461
h264_decode_flush(AVCodecContext * avctx)462 static void h264_decode_flush(AVCodecContext *avctx)
463 {
464 H264Context *h = avctx->priv_data;
465 int i;
466
467 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
468
469 ff_h264_flush_change(h);
470 ff_h264_sei_uninit(&h->sei);
471
472 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
473 ff_h264_unref_picture(h, &h->DPB[i]);
474 h->cur_pic_ptr = NULL;
475 ff_h264_unref_picture(h, &h->cur_pic);
476
477 h->mb_y = 0;
478
479 ff_h264_free_tables(h);
480 h->context_initialized = 0;
481 }
482
get_last_needed_nal(H264Context * h)483 static int get_last_needed_nal(H264Context *h)
484 {
485 int nals_needed = 0;
486 int first_slice = 0;
487 int i, ret;
488
489 for (i = 0; i < h->pkt.nb_nals; i++) {
490 H2645NAL *nal = &h->pkt.nals[i];
491 GetBitContext gb;
492
493 /* packets can sometimes contain multiple PPS/SPS,
494 * e.g. two PAFF field pictures in one packet, or a demuxer
495 * which splits NALs strangely if so, when frame threading we
496 * can't start the next thread until we've read all of them */
497 switch (nal->type) {
498 case H264_NAL_SPS:
499 case H264_NAL_PPS:
500 nals_needed = i;
501 break;
502 case H264_NAL_DPA:
503 case H264_NAL_IDR_SLICE:
504 case H264_NAL_SLICE:
505 ret = init_get_bits8(&gb, nal->data + 1, nal->size - 1);
506 if (ret < 0) {
507 av_log(h->avctx, AV_LOG_ERROR, "Invalid zero-sized VCL NAL unit\n");
508 if (h->avctx->err_recognition & AV_EF_EXPLODE)
509 return ret;
510
511 break;
512 }
513 if (!get_ue_golomb_long(&gb) || // first_mb_in_slice
514 !first_slice ||
515 first_slice != nal->type)
516 nals_needed = i;
517 if (!first_slice)
518 first_slice = nal->type;
519 }
520 }
521
522 return nals_needed;
523 }
524
debug_green_metadata(const H264SEIGreenMetaData * gm,void * logctx)525 static void debug_green_metadata(const H264SEIGreenMetaData *gm, void *logctx)
526 {
527 av_log(logctx, AV_LOG_DEBUG, "Green Metadata Info SEI message\n");
528 av_log(logctx, AV_LOG_DEBUG, " green_metadata_type: %d\n", gm->green_metadata_type);
529
530 if (gm->green_metadata_type == 0) {
531 av_log(logctx, AV_LOG_DEBUG, " green_metadata_period_type: %d\n", gm->period_type);
532
533 if (gm->period_type == 2)
534 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_seconds: %d\n", gm->num_seconds);
535 else if (gm->period_type == 3)
536 av_log(logctx, AV_LOG_DEBUG, " green_metadata_num_pictures: %d\n", gm->num_pictures);
537
538 av_log(logctx, AV_LOG_DEBUG, " SEI GREEN Complexity Metrics: %f %f %f %f\n",
539 (float)gm->percent_non_zero_macroblocks/255,
540 (float)gm->percent_intra_coded_macroblocks/255,
541 (float)gm->percent_six_tap_filtering/255,
542 (float)gm->percent_alpha_point_deblocking_instance/255);
543
544 } else if (gm->green_metadata_type == 1) {
545 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_type: %d\n", gm->xsd_metric_type);
546
547 if (gm->xsd_metric_type == 0)
548 av_log(logctx, AV_LOG_DEBUG, " xsd_metric_value: %f\n",
549 (float)gm->xsd_metric_value/100);
550 }
551 }
552
decode_nal_units(H264Context * h,const uint8_t * buf,int buf_size)553 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
554 {
555 AVCodecContext *const avctx = h->avctx;
556 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
557 int idr_cleared=0;
558 int i, ret = 0;
559
560 h->has_slice = 0;
561 h->nal_unit_type= 0;
562
563 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
564 h->current_slice = 0;
565 if (!h->first_field) {
566 h->cur_pic_ptr = NULL;
567 ff_h264_sei_uninit(&h->sei);
568 }
569 }
570
571 if (h->nal_length_size == 4) {
572 if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
573 h->is_avc = 0;
574 }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
575 h->is_avc = 1;
576 }
577
578 ret = ff_h2645_packet_split(&h->pkt, buf, buf_size, avctx, h->is_avc, h->nal_length_size,
579 avctx->codec_id, 0, 0);
580 if (ret < 0) {
581 av_log(avctx, AV_LOG_ERROR,
582 "Error splitting the input into NAL units.\n");
583 return ret;
584 }
585
586 if (avctx->active_thread_type & FF_THREAD_FRAME)
587 nals_needed = get_last_needed_nal(h);
588 if (nals_needed < 0)
589 return nals_needed;
590
591 for (i = 0; i < h->pkt.nb_nals; i++) {
592 H2645NAL *nal = &h->pkt.nals[i];
593 int max_slice_ctx, err;
594
595 if (avctx->skip_frame >= AVDISCARD_NONREF &&
596 nal->ref_idc == 0 && nal->type != H264_NAL_SEI)
597 continue;
598
599 // FIXME these should stop being context-global variables
600 h->nal_ref_idc = nal->ref_idc;
601 h->nal_unit_type = nal->type;
602
603 err = 0;
604 switch (nal->type) {
605 case H264_NAL_IDR_SLICE:
606 if ((nal->data[1] & 0xFC) == 0x98) {
607 av_log(h->avctx, AV_LOG_ERROR, "Invalid inter IDR frame\n");
608 h->next_outputed_poc = INT_MIN;
609 ret = -1;
610 goto end;
611 }
612 if(!idr_cleared) {
613 idr(h); // FIXME ensure we don't lose some frames if there is reordering
614 }
615 idr_cleared = 1;
616 h->has_recovery_point = 1;
617 case H264_NAL_SLICE:
618 h->has_slice = 1;
619
620 if ((err = ff_h264_queue_decode_slice(h, nal))) {
621 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
622 sl->ref_count[0] = sl->ref_count[1] = 0;
623 break;
624 }
625
626 if (h->current_slice == 1) {
627 if (avctx->active_thread_type & FF_THREAD_FRAME &&
628 i >= nals_needed && !h->setup_finished && h->cur_pic_ptr) {
629 ff_thread_finish_setup(avctx);
630 h->setup_finished = 1;
631 }
632
633 if (h->avctx->hwaccel &&
634 (ret = h->avctx->hwaccel->start_frame(h->avctx, buf, buf_size)) < 0)
635 goto end;
636 }
637
638 max_slice_ctx = avctx->hwaccel ? 1 : h->nb_slice_ctx;
639 if (h->nb_slice_ctx_queued == max_slice_ctx) {
640 if (h->avctx->hwaccel) {
641 ret = avctx->hwaccel->decode_slice(avctx, nal->raw_data, nal->raw_size);
642 h->nb_slice_ctx_queued = 0;
643 } else
644 ret = ff_h264_execute_decode_slices(h);
645 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
646 goto end;
647 }
648 break;
649 case H264_NAL_DPA:
650 case H264_NAL_DPB:
651 case H264_NAL_DPC:
652 avpriv_request_sample(avctx, "data partitioning");
653 break;
654 case H264_NAL_SEI:
655 ret = ff_h264_sei_decode(&h->sei, &nal->gb, &h->ps, avctx);
656 h->has_recovery_point = h->has_recovery_point || h->sei.recovery_point.recovery_frame_cnt != -1;
657 if (avctx->debug & FF_DEBUG_GREEN_MD)
658 debug_green_metadata(&h->sei.green_metadata, h->avctx);
659 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
660 goto end;
661 break;
662 case H264_NAL_SPS: {
663 GetBitContext tmp_gb = nal->gb;
664 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
665 ret = avctx->hwaccel->decode_params(avctx,
666 nal->type,
667 nal->raw_data,
668 nal->raw_size);
669 if (ret < 0)
670 goto end;
671 }
672 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
673 break;
674 av_log(h->avctx, AV_LOG_DEBUG,
675 "SPS decoding failure, trying again with the complete NAL\n");
676 init_get_bits8(&tmp_gb, nal->raw_data + 1, nal->raw_size - 1);
677 if (ff_h264_decode_seq_parameter_set(&tmp_gb, avctx, &h->ps, 0) >= 0)
678 break;
679 ff_h264_decode_seq_parameter_set(&nal->gb, avctx, &h->ps, 1);
680 break;
681 }
682 case H264_NAL_PPS:
683 if (avctx->hwaccel && avctx->hwaccel->decode_params) {
684 ret = avctx->hwaccel->decode_params(avctx,
685 nal->type,
686 nal->raw_data,
687 nal->raw_size);
688 if (ret < 0)
689 goto end;
690 }
691 ret = ff_h264_decode_picture_parameter_set(&nal->gb, avctx, &h->ps,
692 nal->size_bits);
693 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
694 goto end;
695 break;
696 case H264_NAL_AUD:
697 case H264_NAL_END_SEQUENCE:
698 case H264_NAL_END_STREAM:
699 case H264_NAL_FILLER_DATA:
700 case H264_NAL_SPS_EXT:
701 case H264_NAL_AUXILIARY_SLICE:
702 break;
703 default:
704 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
705 nal->type, nal->size_bits);
706 }
707
708 if (err < 0) {
709 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
710 }
711 }
712
713 ret = ff_h264_execute_decode_slices(h);
714 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
715 goto end;
716
717 // set decode_error_flags to allow users to detect concealed decoding errors
718 if ((ret < 0 || h->slice_ctx->er.error_occurred) && h->cur_pic_ptr) {
719 h->cur_pic_ptr->f->decode_error_flags |= FF_DECODE_ERROR_DECODE_SLICES;
720 }
721
722 ret = 0;
723 end:
724
725 #if CONFIG_ERROR_RESILIENCE
726 /*
727 * FIXME: Error handling code does not seem to support interlaced
728 * when slices span multiple rows
729 * The ff_er_add_slice calls don't work right for bottom
730 * fields; they cause massive erroneous error concealing
731 * Error marking covers both fields (top and bottom).
732 * This causes a mismatched s->error_count
733 * and a bad error table. Further, the error count goes to
734 * INT_MAX when called for bottom field, because mb_y is
735 * past end by one (callers fault) and resync_mb_y != 0
736 * causes problems for the first MB line, too.
737 */
738 if (!FIELD_PICTURE(h) && h->current_slice && h->enable_er) {
739
740 H264SliceContext *sl = h->slice_ctx;
741 int use_last_pic = h->last_pic_for_ec.f->buf[0] && !sl->ref_count[0];
742
743 ff_h264_set_erpic(&sl->er.cur_pic, h->cur_pic_ptr);
744
745 if (use_last_pic) {
746 ff_h264_set_erpic(&sl->er.last_pic, &h->last_pic_for_ec);
747 sl->ref_list[0][0].parent = &h->last_pic_for_ec;
748 memcpy(sl->ref_list[0][0].data, h->last_pic_for_ec.f->data, sizeof(sl->ref_list[0][0].data));
749 memcpy(sl->ref_list[0][0].linesize, h->last_pic_for_ec.f->linesize, sizeof(sl->ref_list[0][0].linesize));
750 sl->ref_list[0][0].reference = h->last_pic_for_ec.reference;
751 } else if (sl->ref_count[0]) {
752 ff_h264_set_erpic(&sl->er.last_pic, sl->ref_list[0][0].parent);
753 } else
754 ff_h264_set_erpic(&sl->er.last_pic, NULL);
755
756 if (sl->ref_count[1])
757 ff_h264_set_erpic(&sl->er.next_pic, sl->ref_list[1][0].parent);
758
759 sl->er.ref_count = sl->ref_count[0];
760
761 ff_er_frame_end(&sl->er);
762 if (use_last_pic)
763 memset(&sl->ref_list[0][0], 0, sizeof(sl->ref_list[0][0]));
764 }
765 #endif /* CONFIG_ERROR_RESILIENCE */
766 /* clean up */
767 if (h->cur_pic_ptr && !h->droppable && h->has_slice) {
768 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
769 h->picture_structure == PICT_BOTTOM_FIELD);
770 }
771
772 return (ret < 0) ? ret : buf_size;
773 }
774
775 /**
776 * Return the number of bytes consumed for building the current frame.
777 */
get_consumed_bytes(int pos,int buf_size)778 static int get_consumed_bytes(int pos, int buf_size)
779 {
780 if (pos == 0)
781 pos = 1; // avoid infinite loops (I doubt that is needed but...)
782 if (pos + 10 > buf_size)
783 pos = buf_size; // oops ;)
784
785 return pos;
786 }
787
h264_export_enc_params(AVFrame * f,H264Picture * p)788 static int h264_export_enc_params(AVFrame *f, H264Picture *p)
789 {
790 AVVideoEncParams *par;
791 unsigned int nb_mb = p->mb_height * p->mb_width;
792 unsigned int x, y;
793
794 par = av_video_enc_params_create_side_data(f, AV_VIDEO_ENC_PARAMS_H264, nb_mb);
795 if (!par)
796 return AVERROR(ENOMEM);
797
798 par->qp = p->pps->init_qp;
799
800 par->delta_qp[1][0] = p->pps->chroma_qp_index_offset[0];
801 par->delta_qp[1][1] = p->pps->chroma_qp_index_offset[0];
802 par->delta_qp[2][0] = p->pps->chroma_qp_index_offset[1];
803 par->delta_qp[2][1] = p->pps->chroma_qp_index_offset[1];
804
805 for (y = 0; y < p->mb_height; y++)
806 for (x = 0; x < p->mb_width; x++) {
807 const unsigned int block_idx = y * p->mb_width + x;
808 const unsigned int mb_xy = y * p->mb_stride + x;
809 AVVideoBlockParams *b = av_video_enc_params_block(par, block_idx);
810
811 b->src_x = x * 16;
812 b->src_y = y * 16;
813 b->w = 16;
814 b->h = 16;
815
816 b->delta_qp = p->qscale_table[mb_xy] - par->qp;
817 }
818
819 return 0;
820 }
821
output_frame(H264Context * h,AVFrame * dst,H264Picture * srcp)822 static int output_frame(H264Context *h, AVFrame *dst, H264Picture *srcp)
823 {
824 AVFrame *src = srcp->f;
825 int ret;
826
827 ret = av_frame_ref(dst, src);
828 if (ret < 0)
829 return ret;
830
831 av_dict_set(&dst->metadata, "stereo_mode", ff_h264_sei_stereo_mode(&h->sei.frame_packing), 0);
832
833 if (srcp->sei_recovery_frame_cnt == 0)
834 dst->key_frame = 1;
835
836 if (h->avctx->export_side_data & AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS) {
837 ret = h264_export_enc_params(dst, srcp);
838 if (ret < 0)
839 goto fail;
840 }
841
842 return 0;
843 fail:
844 av_frame_unref(dst);
845 return ret;
846 }
847
is_avcc_extradata(const uint8_t * buf,int buf_size)848 static int is_avcc_extradata(const uint8_t *buf, int buf_size)
849 {
850 int cnt= buf[5]&0x1f;
851 const uint8_t *p= buf+6;
852 if (!cnt)
853 return 0;
854 while(cnt--){
855 int nalsize= AV_RB16(p) + 2;
856 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 7)
857 return 0;
858 p += nalsize;
859 }
860 cnt = *(p++);
861 if(!cnt)
862 return 0;
863 while(cnt--){
864 int nalsize= AV_RB16(p) + 2;
865 if(nalsize > buf_size - (p-buf) || (p[2] & 0x9F) != 8)
866 return 0;
867 p += nalsize;
868 }
869 return 1;
870 }
871
finalize_frame(H264Context * h,AVFrame * dst,H264Picture * out,int * got_frame)872 static int finalize_frame(H264Context *h, AVFrame *dst, H264Picture *out, int *got_frame)
873 {
874 int ret;
875
876 if (((h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
877 (h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL) ||
878 out->recovered)) {
879
880 if (!h->avctx->hwaccel &&
881 (out->field_poc[0] == INT_MAX ||
882 out->field_poc[1] == INT_MAX)
883 ) {
884 int p;
885 AVFrame *f = out->f;
886 int field = out->field_poc[0] == INT_MAX;
887 uint8_t *dst_data[4];
888 int linesizes[4];
889 const uint8_t *src_data[4];
890
891 av_log(h->avctx, AV_LOG_DEBUG, "Duplicating field %d to fill missing\n", field);
892 for (p = 0; p<4; p++) {
893 if(f->data[p] == NULL) {
894 dst_data[p] = NULL;
895 src_data[p] = NULL;
896 } else {
897 dst_data[p] = f->data[p] + (field^1)*f->linesize[p];
898 src_data[p] = f->data[p] + field *f->linesize[p];
899 }
900 linesizes[p] = 2*f->linesize[p];
901 }
902
903 av_image_copy(dst_data, linesizes, src_data, linesizes,
904 f->format, f->width, f->height>>1);
905 }
906
907 ret = output_frame(h, dst, out);
908 if (ret < 0)
909 return ret;
910
911 *got_frame = 1;
912
913 if (CONFIG_MPEGVIDEO) {
914 ff_print_debug_info2(h->avctx, dst, NULL,
915 out->mb_type,
916 out->qscale_table,
917 out->motion_val,
918 NULL,
919 h->mb_width, h->mb_height, h->mb_stride, 1);
920 }
921 }
922
923 return 0;
924 }
925
send_next_delayed_frame(H264Context * h,AVFrame * dst_frame,int * got_frame,int buf_index)926 static int send_next_delayed_frame(H264Context *h, AVFrame *dst_frame,
927 int *got_frame, int buf_index)
928 {
929 int ret, i, out_idx;
930 H264Picture *out = h->delayed_pic[0];
931
932 h->cur_pic_ptr = NULL;
933 h->first_field = 0;
934
935 out_idx = 0;
936 for (i = 1;
937 h->delayed_pic[i] &&
938 !h->delayed_pic[i]->f->key_frame &&
939 !h->delayed_pic[i]->mmco_reset;
940 i++)
941 if (h->delayed_pic[i]->poc < out->poc) {
942 out = h->delayed_pic[i];
943 out_idx = i;
944 }
945
946 for (i = out_idx; h->delayed_pic[i]; i++)
947 h->delayed_pic[i] = h->delayed_pic[i + 1];
948
949 if (out) {
950 out->reference &= ~DELAYED_PIC_REF;
951 ret = finalize_frame(h, dst_frame, out, got_frame);
952 if (ret < 0)
953 return ret;
954 }
955
956 return buf_index;
957 }
958
h264_decode_frame(AVCodecContext * avctx,void * data,int * got_frame,AVPacket * avpkt)959 static int h264_decode_frame(AVCodecContext *avctx, void *data,
960 int *got_frame, AVPacket *avpkt)
961 {
962 const uint8_t *buf = avpkt->data;
963 int buf_size = avpkt->size;
964 H264Context *h = avctx->priv_data;
965 AVFrame *pict = data;
966 int buf_index;
967 int ret;
968
969 h->flags = avctx->flags;
970 h->setup_finished = 0;
971 h->nb_slice_ctx_queued = 0;
972
973 ff_h264_unref_picture(h, &h->last_pic_for_ec);
974
975 /* end of stream, output what is still in the buffers */
976 if (buf_size == 0)
977 return send_next_delayed_frame(h, pict, got_frame, 0);
978
979 if (av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, NULL)) {
980 int side_size;
981 uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, &side_size);
982 ff_h264_decode_extradata(side, side_size,
983 &h->ps, &h->is_avc, &h->nal_length_size,
984 avctx->err_recognition, avctx);
985 }
986 if (h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC) {
987 if (is_avcc_extradata(buf, buf_size))
988 return ff_h264_decode_extradata(buf, buf_size,
989 &h->ps, &h->is_avc, &h->nal_length_size,
990 avctx->err_recognition, avctx);
991 }
992
993 buf_index = decode_nal_units(h, buf, buf_size);
994 if (buf_index < 0)
995 return AVERROR_INVALIDDATA;
996
997 if (!h->cur_pic_ptr && h->nal_unit_type == H264_NAL_END_SEQUENCE) {
998 av_assert0(buf_index <= buf_size);
999 return send_next_delayed_frame(h, pict, got_frame, buf_index);
1000 }
1001
1002 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && (!h->cur_pic_ptr || !h->has_slice)) {
1003 if (avctx->skip_frame >= AVDISCARD_NONREF ||
1004 buf_size >= 4 && !memcmp("Q264", buf, 4))
1005 return buf_size;
1006 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1007 return AVERROR_INVALIDDATA;
1008 }
1009
1010 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1011 (h->mb_y >= h->mb_height && h->mb_height)) {
1012 if ((ret = ff_h264_field_end(h, &h->slice_ctx[0], 0)) < 0)
1013 return ret;
1014
1015 /* Wait for second field. */
1016 if (h->next_output_pic) {
1017 ret = finalize_frame(h, pict, h->next_output_pic, got_frame);
1018 if (ret < 0)
1019 return ret;
1020 }
1021 }
1022
1023 av_assert0(pict->buf[0] || !*got_frame);
1024
1025 ff_h264_unref_picture(h, &h->last_pic_for_ec);
1026
1027 return get_consumed_bytes(buf_index, buf_size);
1028 }
1029
1030 #define OFFSET(x) offsetof(H264Context, x)
1031 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1032 static const AVOption h264_options[] = {
1033 { "is_avc", "is avc", OFFSET(is_avc), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, 0 },
1034 { "nal_length_size", "nal_length_size", OFFSET(nal_length_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0 },
1035 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD },
1036 { "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 },
1037 { NULL },
1038 };
1039
1040 static const AVClass h264_class = {
1041 .class_name = "H264 Decoder",
1042 .item_name = av_default_item_name,
1043 .option = h264_options,
1044 .version = LIBAVUTIL_VERSION_INT,
1045 };
1046
1047 AVCodec ff_h264_decoder = {
1048 .name = "h264",
1049 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1050 .type = AVMEDIA_TYPE_VIDEO,
1051 .id = AV_CODEC_ID_H264,
1052 .priv_data_size = sizeof(H264Context),
1053 .init = h264_decode_init,
1054 .close = h264_decode_end,
1055 .decode = h264_decode_frame,
1056 .capabilities = /*AV_CODEC_CAP_DRAW_HORIZ_BAND |*/ AV_CODEC_CAP_DR1 |
1057 AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SLICE_THREADS |
1058 AV_CODEC_CAP_FRAME_THREADS,
1059 .hw_configs = (const AVCodecHWConfigInternal*[]) {
1060 #if CONFIG_H264_DXVA2_HWACCEL
1061 HWACCEL_DXVA2(h264),
1062 #endif
1063 #if CONFIG_H264_D3D11VA_HWACCEL
1064 HWACCEL_D3D11VA(h264),
1065 #endif
1066 #if CONFIG_H264_D3D11VA2_HWACCEL
1067 HWACCEL_D3D11VA2(h264),
1068 #endif
1069 #if CONFIG_H264_NVDEC_HWACCEL
1070 HWACCEL_NVDEC(h264),
1071 #endif
1072 #if CONFIG_H264_VAAPI_HWACCEL
1073 HWACCEL_VAAPI(h264),
1074 #endif
1075 #if CONFIG_H264_VDPAU_HWACCEL
1076 HWACCEL_VDPAU(h264),
1077 #endif
1078 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
1079 HWACCEL_VIDEOTOOLBOX(h264),
1080 #endif
1081 NULL
1082 },
1083 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_EXPORTS_CROPPING |
1084 FF_CODEC_CAP_ALLOCATE_PROGRESS | FF_CODEC_CAP_INIT_CLEANUP,
1085 .flush = h264_decode_flush,
1086 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1087 .profiles = NULL_IF_CONFIG_SMALL(ff_h264_profiles),
1088 .priv_class = &h264_class,
1089 };
1090