lavc: AV-prefix all codec flags
[libav.git] / libavcodec / h264.c
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 Libav.
6 *
7 * Libav 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 * Libav 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 Libav; 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 / MPEG4 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/timer.h"
34 #include "internal.h"
35 #include "cabac.h"
36 #include "cabac_functions.h"
37 #include "error_resilience.h"
38 #include "avcodec.h"
39 #include "h264.h"
40 #include "h264data.h"
41 #include "h264chroma.h"
42 #include "h264_mvpred.h"
43 #include "golomb.h"
44 #include "mathops.h"
45 #include "me_cmp.h"
46 #include "mpegutils.h"
47 #include "rectangle.h"
48 #include "svq3.h"
49 #include "thread.h"
50
51 #include <assert.h>
52
53 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
54
55 static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
56 int (*mv)[2][4][2],
57 int mb_x, int mb_y, int mb_intra, int mb_skipped)
58 {
59 H264Context *h = opaque;
60 H264SliceContext *sl = &h->slice_ctx[0];
61
62 sl->mb_x = mb_x;
63 sl->mb_y = mb_y;
64 sl->mb_xy = mb_x + mb_y * h->mb_stride;
65 memset(sl->non_zero_count_cache, 0, sizeof(sl->non_zero_count_cache));
66 assert(ref >= 0);
67 /* FIXME: It is possible albeit uncommon that slice references
68 * differ between slices. We take the easy approach and ignore
69 * it for now. If this turns out to have any relevance in
70 * practice then correct remapping should be added. */
71 if (ref >= sl->ref_count[0])
72 ref = 0;
73 fill_rectangle(&h->cur_pic.ref_index[0][4 * sl->mb_xy],
74 2, 2, 2, ref, 1);
75 fill_rectangle(&sl->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
76 fill_rectangle(sl->mv_cache[0][scan8[0]], 4, 4, 8,
77 pack16to32((*mv)[0][0][0], (*mv)[0][0][1]), 4);
78 assert(!FRAME_MBAFF(h));
79 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);
80 }
81
82 void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl,
83 int y, int height)
84 {
85 AVCodecContext *avctx = h->avctx;
86 const AVFrame *src = h->cur_pic.f;
87 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
88 int vshift = desc->log2_chroma_h;
89 const int field_pic = h->picture_structure != PICT_FRAME;
90 if (field_pic) {
91 height <<= 1;
92 y <<= 1;
93 }
94
95 height = FFMIN(height, avctx->height - y);
96
97 if (field_pic && h->first_field && !(avctx->slice_flags & SLICE_FLAG_ALLOW_FIELD))
98 return;
99
100 if (avctx->draw_horiz_band) {
101 int offset[AV_NUM_DATA_POINTERS];
102 int i;
103
104 offset[0] = y * src->linesize[0];
105 offset[1] =
106 offset[2] = (y >> vshift) * src->linesize[1];
107 for (i = 3; i < AV_NUM_DATA_POINTERS; i++)
108 offset[i] = 0;
109
110 emms_c();
111
112 avctx->draw_horiz_band(avctx, src, offset,
113 y, h->picture_structure, height);
114 }
115 }
116
117 /**
118 * Check if the top & left blocks are available if needed and
119 * change the dc mode so it only uses the available blocks.
120 */
121 int ff_h264_check_intra4x4_pred_mode(const H264Context *h, H264SliceContext *sl)
122 {
123 static const int8_t top[12] = {
124 -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
125 };
126 static const int8_t left[12] = {
127 0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
128 };
129 int i;
130
131 if (!(sl->top_samples_available & 0x8000)) {
132 for (i = 0; i < 4; i++) {
133 int status = top[sl->intra4x4_pred_mode_cache[scan8[0] + i]];
134 if (status < 0) {
135 av_log(h->avctx, AV_LOG_ERROR,
136 "top block unavailable for requested intra4x4 mode %d at %d %d\n",
137 status, sl->mb_x, sl->mb_y);
138 return AVERROR_INVALIDDATA;
139 } else if (status) {
140 sl->intra4x4_pred_mode_cache[scan8[0] + i] = status;
141 }
142 }
143 }
144
145 if ((sl->left_samples_available & 0x8888) != 0x8888) {
146 static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
147 for (i = 0; i < 4; i++)
148 if (!(sl->left_samples_available & mask[i])) {
149 int status = left[sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
150 if (status < 0) {
151 av_log(h->avctx, AV_LOG_ERROR,
152 "left block unavailable for requested intra4x4 mode %d at %d %d\n",
153 status, sl->mb_x, sl->mb_y);
154 return AVERROR_INVALIDDATA;
155 } else if (status) {
156 sl->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
157 }
158 }
159 }
160
161 return 0;
162 } // FIXME cleanup like ff_h264_check_intra_pred_mode
163
164 /**
165 * Check if the top & left blocks are available if needed and
166 * change the dc mode so it only uses the available blocks.
167 */
168 int ff_h264_check_intra_pred_mode(const H264Context *h, H264SliceContext *sl,
169 int mode, int is_chroma)
170 {
171 static const int8_t top[4] = { LEFT_DC_PRED8x8, 1, -1, -1 };
172 static const int8_t left[5] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
173
174 if (mode > 3U) {
175 av_log(h->avctx, AV_LOG_ERROR,
176 "out of range intra chroma pred mode at %d %d\n",
177 sl->mb_x, sl->mb_y);
178 return AVERROR_INVALIDDATA;
179 }
180
181 if (!(sl->top_samples_available & 0x8000)) {
182 mode = top[mode];
183 if (mode < 0) {
184 av_log(h->avctx, AV_LOG_ERROR,
185 "top block unavailable for requested intra mode at %d %d\n",
186 sl->mb_x, sl->mb_y);
187 return AVERROR_INVALIDDATA;
188 }
189 }
190
191 if ((sl->left_samples_available & 0x8080) != 0x8080) {
192 mode = left[mode];
193 if (is_chroma && (sl->left_samples_available & 0x8080)) {
194 // mad cow disease mode, aka MBAFF + constrained_intra_pred
195 mode = ALZHEIMER_DC_L0T_PRED8x8 +
196 (!(sl->left_samples_available & 0x8000)) +
197 2 * (mode == DC_128_PRED8x8);
198 }
199 if (mode < 0) {
200 av_log(h->avctx, AV_LOG_ERROR,
201 "left block unavailable for requested intra mode at %d %d\n",
202 sl->mb_x, sl->mb_y);
203 return AVERROR_INVALIDDATA;
204 }
205 }
206
207 return mode;
208 }
209
210 const uint8_t *ff_h264_decode_nal(H264Context *h, H264SliceContext *sl,
211 const uint8_t *src,
212 int *dst_length, int *consumed, int length)
213 {
214 int i, si, di;
215 uint8_t *dst;
216
217 // src[0]&0x80; // forbidden bit
218 h->nal_ref_idc = src[0] >> 5;
219 h->nal_unit_type = src[0] & 0x1F;
220
221 src++;
222 length--;
223
224 #define STARTCODE_TEST \
225 if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
226 if (src[i + 2] != 3) { \
227 /* startcode, so we must be past the end */ \
228 length = i; \
229 } \
230 break; \
231 }
232
233 #if HAVE_FAST_UNALIGNED
234 #define FIND_FIRST_ZERO \
235 if (i > 0 && !src[i]) \
236 i--; \
237 while (src[i]) \
238 i++
239
240 #if HAVE_FAST_64BIT
241 for (i = 0; i + 1 < length; i += 9) {
242 if (!((~AV_RN64A(src + i) &
243 (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
244 0x8000800080008080ULL))
245 continue;
246 FIND_FIRST_ZERO;
247 STARTCODE_TEST;
248 i -= 7;
249 }
250 #else
251 for (i = 0; i + 1 < length; i += 5) {
252 if (!((~AV_RN32A(src + i) &
253 (AV_RN32A(src + i) - 0x01000101U)) &
254 0x80008080U))
255 continue;
256 FIND_FIRST_ZERO;
257 STARTCODE_TEST;
258 i -= 3;
259 }
260 #endif
261 #else
262 for (i = 0; i + 1 < length; i += 2) {
263 if (src[i])
264 continue;
265 if (i > 0 && src[i - 1] == 0)
266 i--;
267 STARTCODE_TEST;
268 }
269 #endif
270
271 if (i >= length - 1) { // no escaped 0
272 *dst_length = length;
273 *consumed = length + 1; // +1 for the header
274 return src;
275 }
276
277 av_fast_malloc(&sl->rbsp_buffer, &sl->rbsp_buffer_size,
278 length + FF_INPUT_BUFFER_PADDING_SIZE);
279 dst = sl->rbsp_buffer;
280
281 if (!dst)
282 return NULL;
283
284 memcpy(dst, src, i);
285 si = di = i;
286 while (si + 2 < length) {
287 // remove escapes (very rare 1:2^22)
288 if (src[si + 2] > 3) {
289 dst[di++] = src[si++];
290 dst[di++] = src[si++];
291 } else if (src[si] == 0 && src[si + 1] == 0) {
292 if (src[si + 2] == 3) { // escape
293 dst[di++] = 0;
294 dst[di++] = 0;
295 si += 3;
296 continue;
297 } else // next start code
298 goto nsc;
299 }
300
301 dst[di++] = src[si++];
302 }
303 while (si < length)
304 dst[di++] = src[si++];
305
306 nsc:
307 memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
308
309 *dst_length = di;
310 *consumed = si + 1; // +1 for the header
311 /* FIXME store exact number of bits in the getbitcontext
312 * (it is needed for decoding) */
313 return dst;
314 }
315
316 /**
317 * Identify the exact end of the bitstream
318 * @return the length of the trailing, or 0 if damaged
319 */
320 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
321 {
322 int v = *src;
323 int r;
324
325 ff_tlog(h->avctx, "rbsp trailing %X\n", v);
326
327 for (r = 1; r < 9; r++) {
328 if (v & 1)
329 return r;
330 v >>= 1;
331 }
332 return 0;
333 }
334
335 void ff_h264_free_tables(H264Context *h)
336 {
337 int i;
338
339 av_freep(&h->intra4x4_pred_mode);
340 av_freep(&h->chroma_pred_mode_table);
341 av_freep(&h->cbp_table);
342 av_freep(&h->mvd_table[0]);
343 av_freep(&h->mvd_table[1]);
344 av_freep(&h->direct_table);
345 av_freep(&h->non_zero_count);
346 av_freep(&h->slice_table_base);
347 h->slice_table = NULL;
348 av_freep(&h->list_counts);
349
350 av_freep(&h->mb2b_xy);
351 av_freep(&h->mb2br_xy);
352
353 av_buffer_pool_uninit(&h->qscale_table_pool);
354 av_buffer_pool_uninit(&h->mb_type_pool);
355 av_buffer_pool_uninit(&h->motion_val_pool);
356 av_buffer_pool_uninit(&h->ref_index_pool);
357
358 for (i = 0; i < h->nb_slice_ctx; i++) {
359 H264SliceContext *sl = &h->slice_ctx[i];
360
361 av_freep(&sl->dc_val_base);
362 av_freep(&sl->er.mb_index2xy);
363 av_freep(&sl->er.error_status_table);
364 av_freep(&sl->er.er_temp_buffer);
365
366 av_freep(&sl->bipred_scratchpad);
367 av_freep(&sl->edge_emu_buffer);
368 av_freep(&sl->top_borders[0]);
369 av_freep(&sl->top_borders[1]);
370
371 sl->bipred_scratchpad_allocated = 0;
372 sl->edge_emu_buffer_allocated = 0;
373 sl->top_borders_allocated[0] = 0;
374 sl->top_borders_allocated[1] = 0;
375 }
376 }
377
378 int ff_h264_alloc_tables(H264Context *h)
379 {
380 const int big_mb_num = h->mb_stride * (h->mb_height + 1);
381 const int row_mb_num = h->mb_stride * 2 * h->avctx->thread_count;
382 int x, y;
383
384 FF_ALLOCZ_OR_GOTO(h->avctx, h->intra4x4_pred_mode,
385 row_mb_num * 8 * sizeof(uint8_t), fail)
386 h->slice_ctx[0].intra4x4_pred_mode = h->intra4x4_pred_mode;
387
388 FF_ALLOCZ_OR_GOTO(h->avctx, h->non_zero_count,
389 big_mb_num * 48 * sizeof(uint8_t), fail)
390 FF_ALLOCZ_OR_GOTO(h->avctx, h->slice_table_base,
391 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base), fail)
392 FF_ALLOCZ_OR_GOTO(h->avctx, h->cbp_table,
393 big_mb_num * sizeof(uint16_t), fail)
394 FF_ALLOCZ_OR_GOTO(h->avctx, h->chroma_pred_mode_table,
395 big_mb_num * sizeof(uint8_t), fail)
396 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[0],
397 16 * row_mb_num * sizeof(uint8_t), fail);
398 FF_ALLOCZ_OR_GOTO(h->avctx, h->mvd_table[1],
399 16 * row_mb_num * sizeof(uint8_t), fail);
400 h->slice_ctx[0].mvd_table[0] = h->mvd_table[0];
401 h->slice_ctx[0].mvd_table[1] = h->mvd_table[1];
402
403 FF_ALLOCZ_OR_GOTO(h->avctx, h->direct_table,
404 4 * big_mb_num * sizeof(uint8_t), fail);
405 FF_ALLOCZ_OR_GOTO(h->avctx, h->list_counts,
406 big_mb_num * sizeof(uint8_t), fail)
407
408 memset(h->slice_table_base, -1,
409 (big_mb_num + h->mb_stride) * sizeof(*h->slice_table_base));
410 h->slice_table = h->slice_table_base + h->mb_stride * 2 + 1;
411
412 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2b_xy,
413 big_mb_num * sizeof(uint32_t), fail);
414 FF_ALLOCZ_OR_GOTO(h->avctx, h->mb2br_xy,
415 big_mb_num * sizeof(uint32_t), fail);
416 for (y = 0; y < h->mb_height; y++)
417 for (x = 0; x < h->mb_width; x++) {
418 const int mb_xy = x + y * h->mb_stride;
419 const int b_xy = 4 * x + 4 * y * h->b_stride;
420
421 h->mb2b_xy[mb_xy] = b_xy;
422 h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * h->mb_stride)));
423 }
424
425 if (!h->dequant4_coeff[0])
426 h264_init_dequant_tables(h);
427
428 return 0;
429
430 fail:
431 ff_h264_free_tables(h);
432 return AVERROR(ENOMEM);
433 }
434
435 /**
436 * Init context
437 * Allocate buffers which are not shared amongst multiple threads.
438 */
439 int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl)
440 {
441 ERContext *er = &sl->er;
442 int mb_array_size = h->mb_height * h->mb_stride;
443 int y_size = (2 * h->mb_width + 1) * (2 * h->mb_height + 1);
444 int c_size = h->mb_stride * (h->mb_height + 1);
445 int yc_size = y_size + 2 * c_size;
446 int x, y, i;
447
448 sl->ref_cache[0][scan8[5] + 1] =
449 sl->ref_cache[0][scan8[7] + 1] =
450 sl->ref_cache[0][scan8[13] + 1] =
451 sl->ref_cache[1][scan8[5] + 1] =
452 sl->ref_cache[1][scan8[7] + 1] =
453 sl->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
454
455 if (CONFIG_ERROR_RESILIENCE) {
456 /* init ER */
457 er->avctx = h->avctx;
458 er->decode_mb = h264_er_decode_mb;
459 er->opaque = h;
460 er->quarter_sample = 1;
461
462 er->mb_num = h->mb_num;
463 er->mb_width = h->mb_width;
464 er->mb_height = h->mb_height;
465 er->mb_stride = h->mb_stride;
466 er->b8_stride = h->mb_width * 2 + 1;
467
468 // error resilience code looks cleaner with this
469 FF_ALLOCZ_OR_GOTO(h->avctx, er->mb_index2xy,
470 (h->mb_num + 1) * sizeof(int), fail);
471
472 for (y = 0; y < h->mb_height; y++)
473 for (x = 0; x < h->mb_width; x++)
474 er->mb_index2xy[x + y * h->mb_width] = x + y * h->mb_stride;
475
476 er->mb_index2xy[h->mb_height * h->mb_width] = (h->mb_height - 1) *
477 h->mb_stride + h->mb_width;
478
479 FF_ALLOCZ_OR_GOTO(h->avctx, er->error_status_table,
480 mb_array_size * sizeof(uint8_t), fail);
481
482 FF_ALLOC_OR_GOTO(h->avctx, er->er_temp_buffer,
483 h->mb_height * h->mb_stride, fail);
484
485 FF_ALLOCZ_OR_GOTO(h->avctx, sl->dc_val_base,
486 yc_size * sizeof(int16_t), fail);
487 er->dc_val[0] = sl->dc_val_base + h->mb_width * 2 + 2;
488 er->dc_val[1] = sl->dc_val_base + y_size + h->mb_stride + 1;
489 er->dc_val[2] = er->dc_val[1] + c_size;
490 for (i = 0; i < yc_size; i++)
491 sl->dc_val_base[i] = 1024;
492 }
493
494 return 0;
495
496 fail:
497 return AVERROR(ENOMEM); // ff_h264_free_tables will clean up for us
498 }
499
500 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
501 int parse_extradata);
502
503 int ff_h264_decode_extradata(H264Context *h)
504 {
505 AVCodecContext *avctx = h->avctx;
506 int ret;
507
508 if (avctx->extradata[0] == 1) {
509 int i, cnt, nalsize;
510 unsigned char *p = avctx->extradata;
511
512 h->is_avc = 1;
513
514 if (avctx->extradata_size < 7) {
515 av_log(avctx, AV_LOG_ERROR,
516 "avcC %d too short\n", avctx->extradata_size);
517 return AVERROR_INVALIDDATA;
518 }
519 /* sps and pps in the avcC always have length coded with 2 bytes,
520 * so put a fake nal_length_size = 2 while parsing them */
521 h->nal_length_size = 2;
522 // Decode sps from avcC
523 cnt = *(p + 5) & 0x1f; // Number of sps
524 p += 6;
525 for (i = 0; i < cnt; i++) {
526 nalsize = AV_RB16(p) + 2;
527 if (p - avctx->extradata + nalsize > avctx->extradata_size)
528 return AVERROR_INVALIDDATA;
529 ret = decode_nal_units(h, p, nalsize, 1);
530 if (ret < 0) {
531 av_log(avctx, AV_LOG_ERROR,
532 "Decoding sps %d from avcC failed\n", i);
533 return ret;
534 }
535 p += nalsize;
536 }
537 // Decode pps from avcC
538 cnt = *(p++); // Number of pps
539 for (i = 0; i < cnt; i++) {
540 nalsize = AV_RB16(p) + 2;
541 if (p - avctx->extradata + nalsize > avctx->extradata_size)
542 return AVERROR_INVALIDDATA;
543 ret = decode_nal_units(h, p, nalsize, 1);
544 if (ret < 0) {
545 av_log(avctx, AV_LOG_ERROR,
546 "Decoding pps %d from avcC failed\n", i);
547 return ret;
548 }
549 p += nalsize;
550 }
551 // Store right nal length size that will be used to parse all other nals
552 h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
553 } else {
554 h->is_avc = 0;
555 ret = decode_nal_units(h, avctx->extradata, avctx->extradata_size, 1);
556 if (ret < 0)
557 return ret;
558 }
559 return 0;
560 }
561
562 static int h264_init_context(AVCodecContext *avctx, H264Context *h)
563 {
564 int i;
565
566 h->avctx = avctx;
567 h->dequant_coeff_pps = -1;
568
569 h->picture_structure = PICT_FRAME;
570 h->slice_context_count = 1;
571 h->workaround_bugs = avctx->workaround_bugs;
572 h->flags = avctx->flags;
573 h->prev_poc_msb = 1 << 16;
574 h->x264_build = -1;
575 h->recovery_frame = -1;
576 h->frame_recovered = 0;
577
578 h->next_outputed_poc = INT_MIN;
579 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
580 h->last_pocs[i] = INT_MIN;
581
582 ff_h264_reset_sei(h);
583
584 avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
585
586 h->nb_slice_ctx = (avctx->active_thread_type & FF_THREAD_SLICE) ? H264_MAX_THREADS : 1;
587 h->slice_ctx = av_mallocz_array(h->nb_slice_ctx, sizeof(*h->slice_ctx));
588 if (!h->slice_ctx) {
589 h->nb_slice_ctx = 0;
590 return AVERROR(ENOMEM);
591 }
592
593 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
594 h->DPB[i].f = av_frame_alloc();
595 if (!h->DPB[i].f)
596 return AVERROR(ENOMEM);
597 }
598
599 h->cur_pic.f = av_frame_alloc();
600 if (!h->cur_pic.f)
601 return AVERROR(ENOMEM);
602
603 for (i = 0; i < h->nb_slice_ctx; i++)
604 h->slice_ctx[i].h264 = h;
605
606 return 0;
607 }
608
609 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
610 {
611 H264Context *h = avctx->priv_data;
612 int ret;
613
614 ret = h264_init_context(avctx, h);
615 if (ret < 0)
616 return ret;
617
618 /* set defaults */
619 if (!avctx->has_b_frames)
620 h->low_delay = 1;
621
622 ff_h264_decode_init_vlc();
623
624 ff_init_cabac_states();
625
626 if (avctx->codec_id == AV_CODEC_ID_H264) {
627 if (avctx->ticks_per_frame == 1)
628 h->avctx->framerate.num *= 2;
629 avctx->ticks_per_frame = 2;
630 }
631
632 if (avctx->extradata_size > 0 && avctx->extradata) {
633 ret = ff_h264_decode_extradata(h);
634 if (ret < 0) {
635 ff_h264_free_context(h);
636 return ret;
637 }
638 }
639
640 if (h->sps.bitstream_restriction_flag &&
641 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
642 h->avctx->has_b_frames = h->sps.num_reorder_frames;
643 h->low_delay = 0;
644 }
645
646 avctx->internal->allocate_progress = 1;
647
648 if (h->enable_er) {
649 av_log(avctx, AV_LOG_WARNING,
650 "Error resilience is enabled. It is unsafe and unsupported and may crash. "
651 "Use it at your own risk\n");
652 }
653
654 return 0;
655 }
656
657 static int decode_init_thread_copy(AVCodecContext *avctx)
658 {
659 H264Context *h = avctx->priv_data;
660 int ret;
661
662 if (!avctx->internal->is_copy)
663 return 0;
664
665 memset(h, 0, sizeof(*h));
666
667 ret = h264_init_context(avctx, h);
668 if (ret < 0)
669 return ret;
670
671 h->context_initialized = 0;
672
673 return 0;
674 }
675
676 /**
677 * Run setup operations that must be run after slice header decoding.
678 * This includes finding the next displayed frame.
679 *
680 * @param h h264 master context
681 * @param setup_finished enough NALs have been read that we can call
682 * ff_thread_finish_setup()
683 */
684 static void decode_postinit(H264Context *h, int setup_finished)
685 {
686 H264Picture *out = h->cur_pic_ptr;
687 H264Picture *cur = h->cur_pic_ptr;
688 int i, pics, out_of_order, out_idx;
689 int invalid = 0, cnt = 0;
690
691 h->cur_pic_ptr->f->pict_type = h->pict_type;
692
693 if (h->next_output_pic)
694 return;
695
696 if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
697 /* FIXME: if we have two PAFF fields in one packet, we can't start
698 * the next thread here. If we have one field per packet, we can.
699 * The check in decode_nal_units() is not good enough to find this
700 * yet, so we assume the worst for now. */
701 // if (setup_finished)
702 // ff_thread_finish_setup(h->avctx);
703 return;
704 }
705
706 cur->f->interlaced_frame = 0;
707 cur->f->repeat_pict = 0;
708
709 /* Signal interlacing information externally. */
710 /* Prioritize picture timing SEI information over used
711 * decoding process if it exists. */
712
713 if (h->sps.pic_struct_present_flag) {
714 switch (h->sei_pic_struct) {
715 case SEI_PIC_STRUCT_FRAME:
716 break;
717 case SEI_PIC_STRUCT_TOP_FIELD:
718 case SEI_PIC_STRUCT_BOTTOM_FIELD:
719 cur->f->interlaced_frame = 1;
720 break;
721 case SEI_PIC_STRUCT_TOP_BOTTOM:
722 case SEI_PIC_STRUCT_BOTTOM_TOP:
723 if (FIELD_OR_MBAFF_PICTURE(h))
724 cur->f->interlaced_frame = 1;
725 else
726 // try to flag soft telecine progressive
727 cur->f->interlaced_frame = h->prev_interlaced_frame;
728 break;
729 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
730 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
731 /* Signal the possibility of telecined film externally
732 * (pic_struct 5,6). From these hints, let the applications
733 * decide if they apply deinterlacing. */
734 cur->f->repeat_pict = 1;
735 break;
736 case SEI_PIC_STRUCT_FRAME_DOUBLING:
737 cur->f->repeat_pict = 2;
738 break;
739 case SEI_PIC_STRUCT_FRAME_TRIPLING:
740 cur->f->repeat_pict = 4;
741 break;
742 }
743
744 if ((h->sei_ct_type & 3) &&
745 h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
746 cur->f->interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
747 } else {
748 /* Derive interlacing flag from used decoding process. */
749 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
750 }
751 h->prev_interlaced_frame = cur->f->interlaced_frame;
752
753 if (cur->field_poc[0] != cur->field_poc[1]) {
754 /* Derive top_field_first from field pocs. */
755 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
756 } else {
757 if (cur->f->interlaced_frame || h->sps.pic_struct_present_flag) {
758 /* Use picture timing SEI information. Even if it is a
759 * information of a past frame, better than nothing. */
760 if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
761 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
762 cur->f->top_field_first = 1;
763 else
764 cur->f->top_field_first = 0;
765 } else {
766 /* Most likely progressive */
767 cur->f->top_field_first = 0;
768 }
769 }
770
771 if (h->sei_frame_packing_present &&
772 h->frame_packing_arrangement_type >= 0 &&
773 h->frame_packing_arrangement_type <= 6 &&
774 h->content_interpretation_type > 0 &&
775 h->content_interpretation_type < 3) {
776 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
777 if (!stereo)
778 return;
779
780 switch (h->frame_packing_arrangement_type) {
781 case 0:
782 stereo->type = AV_STEREO3D_CHECKERBOARD;
783 break;
784 case 1:
785 stereo->type = AV_STEREO3D_COLUMNS;
786 break;
787 case 2:
788 stereo->type = AV_STEREO3D_LINES;
789 break;
790 case 3:
791 if (h->quincunx_subsampling)
792 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
793 else
794 stereo->type = AV_STEREO3D_SIDEBYSIDE;
795 break;
796 case 4:
797 stereo->type = AV_STEREO3D_TOPBOTTOM;
798 break;
799 case 5:
800 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
801 break;
802 case 6:
803 stereo->type = AV_STEREO3D_2D;
804 break;
805 }
806
807 if (h->content_interpretation_type == 2)
808 stereo->flags = AV_STEREO3D_FLAG_INVERT;
809 }
810
811 if (h->sei_display_orientation_present &&
812 (h->sei_anticlockwise_rotation || h->sei_hflip || h->sei_vflip)) {
813 double angle = h->sei_anticlockwise_rotation * 360 / (double) (1 << 16);
814 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
815 AV_FRAME_DATA_DISPLAYMATRIX,
816 sizeof(int32_t) * 9);
817 if (!rotation)
818 return;
819
820 av_display_rotation_set((int32_t *)rotation->data, angle);
821 av_display_matrix_flip((int32_t *)rotation->data,
822 h->sei_hflip, h->sei_vflip);
823 }
824
825 if (h->sei_reguserdata_afd_present) {
826 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
827 sizeof(uint8_t));
828 if (!sd)
829 return;
830
831 *sd->data = h->active_format_description;
832 h->sei_reguserdata_afd_present = 0;
833 }
834
835 if (h->a53_caption) {
836 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
837 AV_FRAME_DATA_A53_CC,
838 h->a53_caption_size);
839 if (!sd)
840 return;
841
842 memcpy(sd->data, h->a53_caption, h->a53_caption_size);
843 av_freep(&h->a53_caption);
844 h->a53_caption_size = 0;
845 }
846
847 // FIXME do something with unavailable reference frames
848
849 /* Sort B-frames into display order */
850
851 if (h->sps.bitstream_restriction_flag &&
852 h->avctx->has_b_frames < h->sps.num_reorder_frames) {
853 h->avctx->has_b_frames = h->sps.num_reorder_frames;
854 h->low_delay = 0;
855 }
856
857 if (h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
858 !h->sps.bitstream_restriction_flag) {
859 h->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
860 h->low_delay = 0;
861 }
862
863 pics = 0;
864 while (h->delayed_pic[pics])
865 pics++;
866
867 assert(pics <= MAX_DELAYED_PIC_COUNT);
868
869 h->delayed_pic[pics++] = cur;
870 if (cur->reference == 0)
871 cur->reference = DELAYED_PIC_REF;
872
873 /* Frame reordering. This code takes pictures from coding order and sorts
874 * them by their incremental POC value into display order. It supports POC
875 * gaps, MMCO reset codes and random resets.
876 * A "display group" can start either with a IDR frame (f.key_frame = 1),
877 * and/or can be closed down with a MMCO reset code. In sequences where
878 * there is no delay, we can't detect that (since the frame was already
879 * output to the user), so we also set h->mmco_reset to detect the MMCO
880 * reset code.
881 * FIXME: if we detect insufficient delays (as per h->avctx->has_b_frames),
882 * we increase the delay between input and output. All frames affected by
883 * the lag (e.g. those that should have been output before another frame
884 * that we already returned to the user) will be dropped. This is a bug
885 * that we will fix later. */
886 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
887 cnt += out->poc < h->last_pocs[i];
888 invalid += out->poc == INT_MIN;
889 }
890 if (!h->mmco_reset && !cur->f->key_frame &&
891 cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
892 h->mmco_reset = 2;
893 if (pics > 1)
894 h->delayed_pic[pics - 2]->mmco_reset = 2;
895 }
896 if (h->mmco_reset || cur->f->key_frame) {
897 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
898 h->last_pocs[i] = INT_MIN;
899 cnt = 0;
900 invalid = MAX_DELAYED_PIC_COUNT;
901 }
902 out = h->delayed_pic[0];
903 out_idx = 0;
904 for (i = 1; i < MAX_DELAYED_PIC_COUNT &&
905 h->delayed_pic[i] &&
906 !h->delayed_pic[i - 1]->mmco_reset &&
907 !h->delayed_pic[i]->f->key_frame;
908 i++)
909 if (h->delayed_pic[i]->poc < out->poc) {
910 out = h->delayed_pic[i];
911 out_idx = i;
912 }
913 if (h->avctx->has_b_frames == 0 &&
914 (h->delayed_pic[0]->f->key_frame || h->mmco_reset))
915 h->next_outputed_poc = INT_MIN;
916 out_of_order = !out->f->key_frame && !h->mmco_reset &&
917 (out->poc < h->next_outputed_poc);
918
919 if (h->sps.bitstream_restriction_flag &&
920 h->avctx->has_b_frames >= h->sps.num_reorder_frames) {
921 } else if (out_of_order && pics - 1 == h->avctx->has_b_frames &&
922 h->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
923 if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
924 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, cnt);
925 }
926 h->low_delay = 0;
927 } else if (h->low_delay &&
928 ((h->next_outputed_poc != INT_MIN &&
929 out->poc > h->next_outputed_poc + 2) ||
930 cur->f->pict_type == AV_PICTURE_TYPE_B)) {
931 h->low_delay = 0;
932 h->avctx->has_b_frames++;
933 }
934
935 if (pics > h->avctx->has_b_frames) {
936 out->reference &= ~DELAYED_PIC_REF;
937 // for frame threading, the owner must be the second field's thread or
938 // else the first thread can release the picture and reuse it unsafely
939 for (i = out_idx; h->delayed_pic[i]; i++)
940 h->delayed_pic[i] = h->delayed_pic[i + 1];
941 }
942 memmove(h->last_pocs, &h->last_pocs[1],
943 sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
944 h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
945 if (!out_of_order && pics > h->avctx->has_b_frames) {
946 h->next_output_pic = out;
947 if (out->mmco_reset) {
948 if (out_idx > 0) {
949 h->next_outputed_poc = out->poc;
950 h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
951 } else {
952 h->next_outputed_poc = INT_MIN;
953 }
954 } else {
955 if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f->key_frame) {
956 h->next_outputed_poc = INT_MIN;
957 } else {
958 h->next_outputed_poc = out->poc;
959 }
960 }
961 h->mmco_reset = 0;
962 } else {
963 av_log(h->avctx, AV_LOG_DEBUG, "no picture\n");
964 }
965
966 if (h->next_output_pic) {
967 if (h->next_output_pic->recovered) {
968 // We have reached an recovery point and all frames after it in
969 // display order are "recovered".
970 h->frame_recovered |= FRAME_RECOVERED_SEI;
971 }
972 h->next_output_pic->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
973 }
974
975 if (setup_finished && !h->avctx->hwaccel) {
976 ff_thread_finish_setup(h->avctx);
977
978 if (h->avctx->active_thread_type & FF_THREAD_FRAME)
979 h->setup_finished = 1;
980 }
981 }
982
983 int ff_pred_weight_table(H264Context *h, H264SliceContext *sl)
984 {
985 int list, i;
986 int luma_def, chroma_def;
987
988 sl->use_weight = 0;
989 sl->use_weight_chroma = 0;
990 sl->luma_log2_weight_denom = get_ue_golomb(&sl->gb);
991 if (h->sps.chroma_format_idc)
992 sl->chroma_log2_weight_denom = get_ue_golomb(&sl->gb);
993 luma_def = 1 << sl->luma_log2_weight_denom;
994 chroma_def = 1 << sl->chroma_log2_weight_denom;
995
996 for (list = 0; list < 2; list++) {
997 sl->luma_weight_flag[list] = 0;
998 sl->chroma_weight_flag[list] = 0;
999 for (i = 0; i < sl->ref_count[list]; i++) {
1000 int luma_weight_flag, chroma_weight_flag;
1001
1002 luma_weight_flag = get_bits1(&sl->gb);
1003 if (luma_weight_flag) {
1004 sl->luma_weight[i][list][0] = get_se_golomb(&sl->gb);
1005 sl->luma_weight[i][list][1] = get_se_golomb(&sl->gb);
1006 if (sl->luma_weight[i][list][0] != luma_def ||
1007 sl->luma_weight[i][list][1] != 0) {
1008 sl->use_weight = 1;
1009 sl->luma_weight_flag[list] = 1;
1010 }
1011 } else {
1012 sl->luma_weight[i][list][0] = luma_def;
1013 sl->luma_weight[i][list][1] = 0;
1014 }
1015
1016 if (h->sps.chroma_format_idc) {
1017 chroma_weight_flag = get_bits1(&sl->gb);
1018 if (chroma_weight_flag) {
1019 int j;
1020 for (j = 0; j < 2; j++) {
1021 sl->chroma_weight[i][list][j][0] = get_se_golomb(&sl->gb);
1022 sl->chroma_weight[i][list][j][1] = get_se_golomb(&sl->gb);
1023 if (sl->chroma_weight[i][list][j][0] != chroma_def ||
1024 sl->chroma_weight[i][list][j][1] != 0) {
1025 sl->use_weight_chroma = 1;
1026 sl->chroma_weight_flag[list] = 1;
1027 }
1028 }
1029 } else {
1030 int j;
1031 for (j = 0; j < 2; j++) {
1032 sl->chroma_weight[i][list][j][0] = chroma_def;
1033 sl->chroma_weight[i][list][j][1] = 0;
1034 }
1035 }
1036 }
1037 }
1038 if (sl->slice_type_nos != AV_PICTURE_TYPE_B)
1039 break;
1040 }
1041 sl->use_weight = sl->use_weight || sl->use_weight_chroma;
1042 return 0;
1043 }
1044
1045 /**
1046 * instantaneous decoder refresh.
1047 */
1048 static void idr(H264Context *h)
1049 {
1050 ff_h264_remove_all_refs(h);
1051 h->prev_frame_num =
1052 h->prev_frame_num_offset =
1053 h->prev_poc_msb =
1054 h->prev_poc_lsb = 0;
1055 }
1056
1057 /* forget old pics after a seek */
1058 void ff_h264_flush_change(H264Context *h)
1059 {
1060 int i;
1061 for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1062 h->last_pocs[i] = INT_MIN;
1063 h->next_outputed_poc = INT_MIN;
1064 h->prev_interlaced_frame = 1;
1065 idr(h);
1066 if (h->cur_pic_ptr)
1067 h->cur_pic_ptr->reference = 0;
1068 h->first_field = 0;
1069 ff_h264_reset_sei(h);
1070 h->recovery_frame = -1;
1071 h->frame_recovered = 0;
1072 }
1073
1074 /* forget old pics after a seek */
1075 static void flush_dpb(AVCodecContext *avctx)
1076 {
1077 H264Context *h = avctx->priv_data;
1078 int i;
1079
1080 memset(h->delayed_pic, 0, sizeof(h->delayed_pic));
1081
1082 ff_h264_flush_change(h);
1083
1084 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++)
1085 ff_h264_unref_picture(h, &h->DPB[i]);
1086 h->cur_pic_ptr = NULL;
1087 ff_h264_unref_picture(h, &h->cur_pic);
1088
1089 h->mb_y = 0;
1090
1091 ff_h264_free_tables(h);
1092 h->context_initialized = 0;
1093 }
1094
1095 int ff_init_poc(H264Context *h, int pic_field_poc[2], int *pic_poc)
1096 {
1097 const int max_frame_num = 1 << h->sps.log2_max_frame_num;
1098 int field_poc[2];
1099
1100 h->frame_num_offset = h->prev_frame_num_offset;
1101 if (h->frame_num < h->prev_frame_num)
1102 h->frame_num_offset += max_frame_num;
1103
1104 if (h->sps.poc_type == 0) {
1105 const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
1106
1107 if (h->poc_lsb < h->prev_poc_lsb &&
1108 h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
1109 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
1110 else if (h->poc_lsb > h->prev_poc_lsb &&
1111 h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
1112 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
1113 else
1114 h->poc_msb = h->prev_poc_msb;
1115 field_poc[0] =
1116 field_poc[1] = h->poc_msb + h->poc_lsb;
1117 if (h->picture_structure == PICT_FRAME)
1118 field_poc[1] += h->delta_poc_bottom;
1119 } else if (h->sps.poc_type == 1) {
1120 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
1121 int i;
1122
1123 if (h->sps.poc_cycle_length != 0)
1124 abs_frame_num = h->frame_num_offset + h->frame_num;
1125 else
1126 abs_frame_num = 0;
1127
1128 if (h->nal_ref_idc == 0 && abs_frame_num > 0)
1129 abs_frame_num--;
1130
1131 expected_delta_per_poc_cycle = 0;
1132 for (i = 0; i < h->sps.poc_cycle_length; i++)
1133 // FIXME integrate during sps parse
1134 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
1135
1136 if (abs_frame_num > 0) {
1137 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
1138 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
1139
1140 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
1141 for (i = 0; i <= frame_num_in_poc_cycle; i++)
1142 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
1143 } else
1144 expectedpoc = 0;
1145
1146 if (h->nal_ref_idc == 0)
1147 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
1148
1149 field_poc[0] = expectedpoc + h->delta_poc[0];
1150 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
1151
1152 if (h->picture_structure == PICT_FRAME)
1153 field_poc[1] += h->delta_poc[1];
1154 } else {
1155 int poc = 2 * (h->frame_num_offset + h->frame_num);
1156
1157 if (!h->nal_ref_idc)
1158 poc--;
1159
1160 field_poc[0] = poc;
1161 field_poc[1] = poc;
1162 }
1163
1164 if (h->picture_structure != PICT_BOTTOM_FIELD)
1165 pic_field_poc[0] = field_poc[0];
1166 if (h->picture_structure != PICT_TOP_FIELD)
1167 pic_field_poc[1] = field_poc[1];
1168 *pic_poc = FFMIN(pic_field_poc[0], pic_field_poc[1]);
1169
1170 return 0;
1171 }
1172
1173 /**
1174 * Compute profile from profile_idc and constraint_set?_flags.
1175 *
1176 * @param sps SPS
1177 *
1178 * @return profile as defined by FF_PROFILE_H264_*
1179 */
1180 int ff_h264_get_profile(SPS *sps)
1181 {
1182 int profile = sps->profile_idc;
1183
1184 switch (sps->profile_idc) {
1185 case FF_PROFILE_H264_BASELINE:
1186 // constraint_set1_flag set to 1
1187 profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
1188 break;
1189 case FF_PROFILE_H264_HIGH_10:
1190 case FF_PROFILE_H264_HIGH_422:
1191 case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
1192 // constraint_set3_flag set to 1
1193 profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
1194 break;
1195 }
1196
1197 return profile;
1198 }
1199
1200 int ff_set_ref_count(H264Context *h, H264SliceContext *sl)
1201 {
1202 int ref_count[2], list_count;
1203 int num_ref_idx_active_override_flag, max_refs;
1204
1205 // set defaults, might be overridden a few lines later
1206 ref_count[0] = h->pps.ref_count[0];
1207 ref_count[1] = h->pps.ref_count[1];
1208
1209 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1210 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1211 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1212 num_ref_idx_active_override_flag = get_bits1(&sl->gb);
1213
1214 if (num_ref_idx_active_override_flag) {
1215 ref_count[0] = get_ue_golomb(&sl->gb) + 1;
1216 if (ref_count[0] < 1)
1217 return AVERROR_INVALIDDATA;
1218 if (sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1219 ref_count[1] = get_ue_golomb(&sl->gb) + 1;
1220 if (ref_count[1] < 1)
1221 return AVERROR_INVALIDDATA;
1222 }
1223 }
1224
1225 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1226 list_count = 2;
1227 else
1228 list_count = 1;
1229 } else {
1230 list_count = 0;
1231 ref_count[0] = ref_count[1] = 0;
1232 }
1233
1234 max_refs = h->picture_structure == PICT_FRAME ? 16 : 32;
1235
1236 if (ref_count[0] > max_refs || ref_count[1] > max_refs) {
1237 av_log(h->avctx, AV_LOG_ERROR, "reference overflow\n");
1238 sl->ref_count[0] = sl->ref_count[1] = 0;
1239 return AVERROR_INVALIDDATA;
1240 }
1241
1242 if (list_count != sl->list_count ||
1243 ref_count[0] != sl->ref_count[0] ||
1244 ref_count[1] != sl->ref_count[1]) {
1245 sl->ref_count[0] = ref_count[0];
1246 sl->ref_count[1] = ref_count[1];
1247 sl->list_count = list_count;
1248 return 1;
1249 }
1250
1251 return 0;
1252 }
1253
1254 static int find_start_code(const uint8_t *buf, int buf_size,
1255 int buf_index, int next_avc)
1256 {
1257 // start code prefix search
1258 for (; buf_index + 3 < next_avc; buf_index++)
1259 // This should always succeed in the first iteration.
1260 if (buf[buf_index] == 0 &&
1261 buf[buf_index + 1] == 0 &&
1262 buf[buf_index + 2] == 1)
1263 break;
1264
1265 if (buf_index + 3 >= buf_size)
1266 return buf_size;
1267
1268 return buf_index + 3;
1269 }
1270
1271 static int get_avc_nalsize(H264Context *h, const uint8_t *buf,
1272 int buf_size, int *buf_index)
1273 {
1274 int i, nalsize = 0;
1275
1276 if (*buf_index >= buf_size - h->nal_length_size)
1277 return -1;
1278
1279 for (i = 0; i < h->nal_length_size; i++)
1280 nalsize = (nalsize << 8) | buf[(*buf_index)++];
1281 if (nalsize <= 0 || nalsize > buf_size - *buf_index) {
1282 av_log(h->avctx, AV_LOG_ERROR,
1283 "AVC: nal size %d\n", nalsize);
1284 return -1;
1285 }
1286 return nalsize;
1287 }
1288
1289 static int get_bit_length(H264Context *h, const uint8_t *buf,
1290 const uint8_t *ptr, int dst_length,
1291 int i, int next_avc)
1292 {
1293 if ((h->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
1294 buf[i] == 0x00 && buf[i + 1] == 0x00 &&
1295 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
1296 h->workaround_bugs |= FF_BUG_TRUNCATED;
1297
1298 if (!(h->workaround_bugs & FF_BUG_TRUNCATED))
1299 while (dst_length > 0 && ptr[dst_length - 1] == 0)
1300 dst_length--;
1301
1302 if (!dst_length)
1303 return 0;
1304
1305 return 8 * dst_length - decode_rbsp_trailing(h, ptr + dst_length - 1);
1306 }
1307
1308 static int get_last_needed_nal(H264Context *h, const uint8_t *buf, int buf_size)
1309 {
1310 int next_avc = h->is_avc ? 0 : buf_size;
1311 int nal_index = 0;
1312 int buf_index = 0;
1313 int nals_needed = 0;
1314
1315 while(1) {
1316 GetBitContext gb;
1317 int nalsize = 0;
1318 int dst_length, bit_length, consumed;
1319 const uint8_t *ptr;
1320
1321 if (buf_index >= next_avc) {
1322 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1323 if (nalsize < 0)
1324 break;
1325 next_avc = buf_index + nalsize;
1326 } else {
1327 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1328 if (buf_index >= buf_size)
1329 break;
1330 }
1331
1332 ptr = ff_h264_decode_nal(h, &h->slice_ctx[0], buf + buf_index, &dst_length, &consumed,
1333 next_avc - buf_index);
1334
1335 if (!ptr || dst_length < 0)
1336 return AVERROR_INVALIDDATA;
1337
1338 buf_index += consumed;
1339
1340 bit_length = get_bit_length(h, buf, ptr, dst_length,
1341 buf_index, next_avc);
1342 nal_index++;
1343
1344 /* packets can sometimes contain multiple PPS/SPS,
1345 * e.g. two PAFF field pictures in one packet, or a demuxer
1346 * which splits NALs strangely if so, when frame threading we
1347 * can't start the next thread until we've read all of them */
1348 switch (h->nal_unit_type) {
1349 case NAL_SPS:
1350 case NAL_PPS:
1351 nals_needed = nal_index;
1352 break;
1353 case NAL_DPA:
1354 case NAL_IDR_SLICE:
1355 case NAL_SLICE:
1356 init_get_bits(&gb, ptr, bit_length);
1357 if (!get_ue_golomb(&gb))
1358 nals_needed = nal_index;
1359 }
1360 }
1361
1362 return nals_needed;
1363 }
1364
1365 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
1366 int parse_extradata)
1367 {
1368 AVCodecContext *const avctx = h->avctx;
1369 H264SliceContext *sl;
1370 int buf_index;
1371 unsigned context_count;
1372 int next_avc;
1373 int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
1374 int nal_index;
1375 int ret = 0;
1376
1377 h->max_contexts = h->slice_context_count;
1378 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)) {
1379 h->current_slice = 0;
1380 if (!h->first_field)
1381 h->cur_pic_ptr = NULL;
1382 ff_h264_reset_sei(h);
1383 }
1384
1385 if (avctx->active_thread_type & FF_THREAD_FRAME)
1386 nals_needed = get_last_needed_nal(h, buf, buf_size);
1387
1388 {
1389 buf_index = 0;
1390 context_count = 0;
1391 next_avc = h->is_avc ? 0 : buf_size;
1392 nal_index = 0;
1393 for (;;) {
1394 int consumed;
1395 int dst_length;
1396 int bit_length;
1397 const uint8_t *ptr;
1398 int nalsize = 0;
1399 int err;
1400
1401 if (buf_index >= next_avc) {
1402 nalsize = get_avc_nalsize(h, buf, buf_size, &buf_index);
1403 if (nalsize < 0)
1404 break;
1405 next_avc = buf_index + nalsize;
1406 } else {
1407 buf_index = find_start_code(buf, buf_size, buf_index, next_avc);
1408 if (buf_index >= buf_size)
1409 break;
1410 if (buf_index >= next_avc)
1411 continue;
1412 }
1413
1414 sl = &h->slice_ctx[context_count];
1415
1416 ptr = ff_h264_decode_nal(h, sl, buf + buf_index, &dst_length,
1417 &consumed, next_avc - buf_index);
1418 if (!ptr || dst_length < 0) {
1419 ret = -1;
1420 goto end;
1421 }
1422
1423 bit_length = get_bit_length(h, buf, ptr, dst_length,
1424 buf_index + consumed, next_avc);
1425
1426 if (h->avctx->debug & FF_DEBUG_STARTCODE)
1427 av_log(h->avctx, AV_LOG_DEBUG,
1428 "NAL %d at %d/%d length %d\n",
1429 h->nal_unit_type, buf_index, buf_size, dst_length);
1430
1431 if (h->is_avc && (nalsize != consumed) && nalsize)
1432 av_log(h->avctx, AV_LOG_DEBUG,
1433 "AVC: Consumed only %d bytes instead of %d\n",
1434 consumed, nalsize);
1435
1436 buf_index += consumed;
1437 nal_index++;
1438
1439 if (avctx->skip_frame >= AVDISCARD_NONREF &&
1440 h->nal_ref_idc == 0 &&
1441 h->nal_unit_type != NAL_SEI)
1442 continue;
1443
1444 again:
1445 /* Ignore every NAL unit type except PPS and SPS during extradata
1446 * parsing. Decoding slices is not possible in codec init
1447 * with frame-mt */
1448 if (parse_extradata && HAVE_THREADS &&
1449 (h->avctx->active_thread_type & FF_THREAD_FRAME) &&
1450 (h->nal_unit_type != NAL_PPS &&
1451 h->nal_unit_type != NAL_SPS)) {
1452 if (h->nal_unit_type < NAL_AUD ||
1453 h->nal_unit_type > NAL_AUXILIARY_SLICE)
1454 av_log(avctx, AV_LOG_INFO,
1455 "Ignoring NAL unit %d during extradata parsing\n",
1456 h->nal_unit_type);
1457 h->nal_unit_type = NAL_FF_IGNORE;
1458 }
1459 err = 0;
1460 switch (h->nal_unit_type) {
1461 case NAL_IDR_SLICE:
1462 if (h->nal_unit_type != NAL_IDR_SLICE) {
1463 av_log(h->avctx, AV_LOG_ERROR,
1464 "Invalid mix of idr and non-idr slices\n");
1465 ret = -1;
1466 goto end;
1467 }
1468 idr(h); // FIXME ensure we don't lose some frames if there is reordering
1469 case NAL_SLICE:
1470 init_get_bits(&sl->gb, ptr, bit_length);
1471
1472 if ((err = ff_h264_decode_slice_header(h, sl)))
1473 break;
1474
1475 if (h->sei_recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
1476 h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
1477 ((1 << h->sps.log2_max_frame_num) - 1);
1478 }
1479
1480 h->cur_pic_ptr->f->key_frame |=
1481 (h->nal_unit_type == NAL_IDR_SLICE) ||
1482 (h->sei_recovery_frame_cnt >= 0);
1483
1484 if (h->nal_unit_type == NAL_IDR_SLICE ||
1485 h->recovery_frame == h->frame_num) {
1486 h->recovery_frame = -1;
1487 h->cur_pic_ptr->recovered = 1;
1488 }
1489 // If we have an IDR, all frames after it in decoded order are
1490 // "recovered".
1491 if (h->nal_unit_type == NAL_IDR_SLICE)
1492 h->frame_recovered |= FRAME_RECOVERED_IDR;
1493 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1494
1495 if (h->current_slice == 1) {
1496 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS))
1497 decode_postinit(h, nal_index >= nals_needed);
1498
1499 if (h->avctx->hwaccel &&
1500 (ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0)) < 0)
1501 return ret;
1502 }
1503
1504 if (sl->redundant_pic_count == 0 &&
1505 (avctx->skip_frame < AVDISCARD_NONREF ||
1506 h->nal_ref_idc) &&
1507 (avctx->skip_frame < AVDISCARD_BIDIR ||
1508 sl->slice_type_nos != AV_PICTURE_TYPE_B) &&
1509 (avctx->skip_frame < AVDISCARD_NONKEY ||
1510 sl->slice_type_nos == AV_PICTURE_TYPE_I) &&
1511 avctx->skip_frame < AVDISCARD_ALL) {
1512 if (avctx->hwaccel) {
1513 ret = avctx->hwaccel->decode_slice(avctx,
1514 &buf[buf_index - consumed],
1515 consumed);
1516 if (ret < 0)
1517 return ret;
1518 } else
1519 context_count++;
1520 }
1521 break;
1522 case NAL_DPA:
1523 case NAL_DPB:
1524 case NAL_DPC:
1525 avpriv_request_sample(avctx, "data partitioning");
1526 ret = AVERROR(ENOSYS);
1527 goto end;
1528 break;
1529 case NAL_SEI:
1530 init_get_bits(&h->gb, ptr, bit_length);
1531 ret = ff_h264_decode_sei(h);
1532 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1533 goto end;
1534 break;
1535 case NAL_SPS:
1536 init_get_bits(&h->gb, ptr, bit_length);
1537 ret = ff_h264_decode_seq_parameter_set(h);
1538 if (ret < 0 && h->is_avc && (nalsize != consumed) && nalsize) {
1539 av_log(h->avctx, AV_LOG_DEBUG,
1540 "SPS decoding failure, trying again with the complete NAL\n");
1541 init_get_bits(&h->gb, buf + buf_index + 1 - consumed,
1542 8 * (nalsize - 1));
1543 ff_h264_decode_seq_parameter_set(h);
1544 }
1545
1546 break;
1547 case NAL_PPS:
1548 init_get_bits(&h->gb, ptr, bit_length);
1549 ret = ff_h264_decode_picture_parameter_set(h, bit_length);
1550 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1551 goto end;
1552 break;
1553 case NAL_AUD:
1554 case NAL_END_SEQUENCE:
1555 case NAL_END_STREAM:
1556 case NAL_FILLER_DATA:
1557 case NAL_SPS_EXT:
1558 case NAL_AUXILIARY_SLICE:
1559 break;
1560 case NAL_FF_IGNORE:
1561 break;
1562 default:
1563 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
1564 h->nal_unit_type, bit_length);
1565 }
1566
1567 if (context_count == h->max_contexts) {
1568 ret = ff_h264_execute_decode_slices(h, context_count);
1569 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1570 goto end;
1571 context_count = 0;
1572 }
1573
1574 if (err < 0) {
1575 av_log(h->avctx, AV_LOG_ERROR, "decode_slice_header error\n");
1576 sl->ref_count[0] = sl->ref_count[1] = sl->list_count = 0;
1577 } else if (err == 1) {
1578 /* Slice could not be decoded in parallel mode, restart. Note
1579 * that rbsp_buffer is not transferred, but since we no longer
1580 * run in parallel mode this should not be an issue. */
1581 sl = &h->slice_ctx[0];
1582 goto again;
1583 }
1584 }
1585 }
1586 if (context_count) {
1587 ret = ff_h264_execute_decode_slices(h, context_count);
1588 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1589 goto end;
1590 }
1591
1592 ret = 0;
1593 end:
1594 /* clean up */
1595 if (h->cur_pic_ptr && !h->droppable) {
1596 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1597 h->picture_structure == PICT_BOTTOM_FIELD);
1598 }
1599
1600 return (ret < 0) ? ret : buf_index;
1601 }
1602
1603 /**
1604 * Return the number of bytes consumed for building the current frame.
1605 */
1606 static int get_consumed_bytes(int pos, int buf_size)
1607 {
1608 if (pos == 0)
1609 pos = 1; // avoid infinite loops (I doubt that is needed but...)
1610 if (pos + 10 > buf_size)
1611 pos = buf_size; // oops ;)
1612
1613 return pos;
1614 }
1615
1616 static int output_frame(H264Context *h, AVFrame *dst, AVFrame *src)
1617 {
1618 int i;
1619 int ret = av_frame_ref(dst, src);
1620 if (ret < 0)
1621 return ret;
1622
1623 if (!h->sps.crop)
1624 return 0;
1625
1626 for (i = 0; i < 3; i++) {
1627 int hshift = (i > 0) ? h->chroma_x_shift : 0;
1628 int vshift = (i > 0) ? h->chroma_y_shift : 0;
1629 int off = ((h->sps.crop_left >> hshift) << h->pixel_shift) +
1630 (h->sps.crop_top >> vshift) * dst->linesize[i];
1631 dst->data[i] += off;
1632 }
1633 return 0;
1634 }
1635
1636 static int h264_decode_frame(AVCodecContext *avctx, void *data,
1637 int *got_frame, AVPacket *avpkt)
1638 {
1639 const uint8_t *buf = avpkt->data;
1640 int buf_size = avpkt->size;
1641 H264Context *h = avctx->priv_data;
1642 AVFrame *pict = data;
1643 int buf_index = 0;
1644 int ret;
1645
1646 h->flags = avctx->flags;
1647 h->setup_finished = 0;
1648
1649 /* end of stream, output what is still in the buffers */
1650 out:
1651 if (buf_size == 0) {
1652 H264Picture *out;
1653 int i, out_idx;
1654
1655 h->cur_pic_ptr = NULL;
1656
1657 // FIXME factorize this with the output code below
1658 out = h->delayed_pic[0];
1659 out_idx = 0;
1660 for (i = 1;
1661 h->delayed_pic[i] &&
1662 !h->delayed_pic[i]->f->key_frame &&
1663 !h->delayed_pic[i]->mmco_reset;
1664 i++)
1665 if (h->delayed_pic[i]->poc < out->poc) {
1666 out = h->delayed_pic[i];
1667 out_idx = i;
1668 }
1669
1670 for (i = out_idx; h->delayed_pic[i]; i++)
1671 h->delayed_pic[i] = h->delayed_pic[i + 1];
1672
1673 if (out) {
1674 ret = output_frame(h, pict, out->f);
1675 if (ret < 0)
1676 return ret;
1677 *got_frame = 1;
1678 }
1679
1680 return buf_index;
1681 }
1682
1683 buf_index = decode_nal_units(h, buf, buf_size, 0);
1684 if (buf_index < 0)
1685 return AVERROR_INVALIDDATA;
1686
1687 if (!h->cur_pic_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
1688 buf_size = 0;
1689 goto out;
1690 }
1691
1692 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) && !h->cur_pic_ptr) {
1693 if (avctx->skip_frame >= AVDISCARD_NONREF)
1694 return 0;
1695 av_log(avctx, AV_LOG_ERROR, "no frame!\n");
1696 return AVERROR_INVALIDDATA;
1697 }
1698
1699 if (!(avctx->flags2 & AV_CODEC_FLAG2_CHUNKS) ||
1700 (h->mb_y >= h->mb_height && h->mb_height)) {
1701 if (avctx->flags2 & AV_CODEC_FLAG2_CHUNKS)
1702 decode_postinit(h, 1);
1703
1704 ff_h264_field_end(h, &h->slice_ctx[0], 0);
1705
1706 *got_frame = 0;
1707 if (h->next_output_pic && ((avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) ||
1708 h->next_output_pic->recovered)) {
1709 if (!h->next_output_pic->recovered)
1710 h->next_output_pic->f->flags |= AV_FRAME_FLAG_CORRUPT;
1711
1712 ret = output_frame(h, pict, h->next_output_pic->f);
1713 if (ret < 0)
1714 return ret;
1715 *got_frame = 1;
1716 }
1717 }
1718
1719 assert(pict->buf[0] || !*got_frame);
1720
1721 return get_consumed_bytes(buf_index, buf_size);
1722 }
1723
1724 av_cold void ff_h264_free_context(H264Context *h)
1725 {
1726 int i;
1727
1728 ff_h264_free_tables(h);
1729
1730 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
1731 ff_h264_unref_picture(h, &h->DPB[i]);
1732 av_frame_free(&h->DPB[i].f);
1733 }
1734
1735 h->cur_pic_ptr = NULL;
1736
1737 for (i = 0; i < h->nb_slice_ctx; i++)
1738 av_freep(&h->slice_ctx[i].rbsp_buffer);
1739 av_freep(&h->slice_ctx);
1740 h->nb_slice_ctx = 0;
1741
1742 for (i = 0; i < MAX_SPS_COUNT; i++)
1743 av_freep(h->sps_buffers + i);
1744
1745 for (i = 0; i < MAX_PPS_COUNT; i++)
1746 av_freep(h->pps_buffers + i);
1747 }
1748
1749 static av_cold int h264_decode_end(AVCodecContext *avctx)
1750 {
1751 H264Context *h = avctx->priv_data;
1752
1753 ff_h264_free_context(h);
1754
1755 ff_h264_unref_picture(h, &h->cur_pic);
1756 av_frame_free(&h->cur_pic.f);
1757
1758 return 0;
1759 }
1760
1761 #define OFFSET(x) offsetof(H264Context, x)
1762 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
1763 static const AVOption h264_options[] = {
1764 { "enable_er", "Enable error resilience on damaged frames (unsafe)", OFFSET(enable_er), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VD },
1765 { NULL },
1766 };
1767
1768 static const AVClass h264_class = {
1769 .class_name = "h264",
1770 .item_name = av_default_item_name,
1771 .option = h264_options,
1772 .version = LIBAVUTIL_VERSION_INT,
1773 };
1774
1775 static const AVProfile profiles[] = {
1776 { FF_PROFILE_H264_BASELINE, "Baseline" },
1777 { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
1778 { FF_PROFILE_H264_MAIN, "Main" },
1779 { FF_PROFILE_H264_EXTENDED, "Extended" },
1780 { FF_PROFILE_H264_HIGH, "High" },
1781 { FF_PROFILE_H264_HIGH_10, "High 10" },
1782 { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
1783 { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
1784 { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
1785 { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
1786 { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
1787 { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
1788 { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
1789 { FF_PROFILE_UNKNOWN },
1790 };
1791
1792 AVCodec ff_h264_decoder = {
1793 .name = "h264",
1794 .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
1795 .type = AVMEDIA_TYPE_VIDEO,
1796 .id = AV_CODEC_ID_H264,
1797 .priv_data_size = sizeof(H264Context),
1798 .init = ff_h264_decode_init,
1799 .close = h264_decode_end,
1800 .decode = h264_decode_frame,
1801 .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
1802 CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
1803 CODEC_CAP_FRAME_THREADS,
1804 .flush = flush_dpb,
1805 .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
1806 .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
1807 .profiles = NULL_IF_CONFIG_SMALL(profiles),
1808 .priv_class = &h264_class,
1809 };