FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
h264.c
Go to the documentation of this file.
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 / MPEG4 part10 codec.
25  * @author Michael Niedermayer <michaelni@gmx.at>
26  */
27 
28 #define UNCHECKED_BITSTREAM_READER 1
29 
30 #include "libavutil/imgutils.h"
31 #include "libavutil/opt.h"
32 #include "internal.h"
33 #include "cabac.h"
34 #include "cabac_functions.h"
35 #include "dsputil.h"
36 #include "avcodec.h"
37 #include "mpegvideo.h"
38 #include "h264.h"
39 #include "h264data.h"
40 #include "h264_mvpred.h"
41 #include "golomb.h"
42 #include "mathops.h"
43 #include "rectangle.h"
44 #include "thread.h"
45 #include "vdpau_internal.h"
46 #include "libavutil/avassert.h"
47 
48 // #undef NDEBUG
49 #include <assert.h>
50 
51 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
52 
53 static const uint8_t rem6[QP_MAX_NUM + 1] = {
54  0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
55  3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
56  0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
57  3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
58  0, 1, 2, 3,
59 };
60 
61 static const uint8_t div6[QP_MAX_NUM + 1] = {
62  0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
63  3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
64  7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
65  10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
66  14,14,14,14,
67 };
68 
75 };
76 
78 {
79  H264Context *h = avctx->priv_data;
80  return h ? h->sps.num_reorder_frames : 0;
81 }
82 
83 /**
84  * Check if the top & left blocks are available if needed and
85  * change the dc mode so it only uses the available blocks.
86  */
88 {
89  MpegEncContext *const s = &h->s;
90  static const int8_t top[12] = {
91  -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
92  };
93  static const int8_t left[12] = {
94  0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
95  };
96  int i;
97 
98  if (!(h->top_samples_available & 0x8000)) {
99  for (i = 0; i < 4; i++) {
100  int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
101  if (status < 0) {
103  "top block unavailable for requested intra4x4 mode %d at %d %d\n",
104  status, s->mb_x, s->mb_y);
105  return -1;
106  } else if (status) {
108  }
109  }
110  }
111 
112  if ((h->left_samples_available & 0x8888) != 0x8888) {
113  static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
114  for (i = 0; i < 4; i++)
115  if (!(h->left_samples_available & mask[i])) {
116  int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
117  if (status < 0) {
119  "left block unavailable for requested intra4x4 mode %d at %d %d\n",
120  status, s->mb_x, s->mb_y);
121  return -1;
122  } else if (status) {
123  h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
124  }
125  }
126  }
127 
128  return 0;
129 } // FIXME cleanup like ff_h264_check_intra_pred_mode
130 
131 /**
132  * Check if the top & left blocks are available if needed and
133  * change the dc mode so it only uses the available blocks.
134  */
135 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
136 {
137  MpegEncContext *const s = &h->s;
138  static const int8_t top[7] = { LEFT_DC_PRED8x8, 1, -1, -1 };
139  static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
140 
141  if (mode > 6U) {
143  "out of range intra chroma pred mode at %d %d\n",
144  s->mb_x, s->mb_y);
145  return -1;
146  }
147 
148  if (!(h->top_samples_available & 0x8000)) {
149  mode = top[mode];
150  if (mode < 0) {
152  "top block unavailable for requested intra mode at %d %d\n",
153  s->mb_x, s->mb_y);
154  return -1;
155  }
156  }
157 
158  if ((h->left_samples_available & 0x8080) != 0x8080) {
159  mode = left[mode];
160  if (is_chroma && (h->left_samples_available & 0x8080)) {
161  // mad cow disease mode, aka MBAFF + constrained_intra_pred
162  mode = ALZHEIMER_DC_L0T_PRED8x8 +
163  (!(h->left_samples_available & 0x8000)) +
164  2 * (mode == DC_128_PRED8x8);
165  }
166  if (mode < 0) {
168  "left block unavailable for requested intra mode at %d %d\n",
169  s->mb_x, s->mb_y);
170  return -1;
171  }
172  }
173 
174  return mode;
175 }
176 
178  int *dst_length, int *consumed, int length)
179 {
180  int i, si, di;
181  uint8_t *dst;
182  int bufidx;
183 
184  // src[0]&0x80; // forbidden bit
185  h->nal_ref_idc = src[0] >> 5;
186  h->nal_unit_type = src[0] & 0x1F;
187 
188  src++;
189  length--;
190 
191 #define STARTCODE_TEST \
192  if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) { \
193  if (src[i + 2] != 3) { \
194  /* startcode, so we must be past the end */ \
195  length = i; \
196  } \
197  break; \
198  }
199 #if HAVE_FAST_UNALIGNED
200 #define FIND_FIRST_ZERO \
201  if (i > 0 && !src[i]) \
202  i--; \
203  while (src[i]) \
204  i++
205 #if HAVE_FAST_64BIT
206  for (i = 0; i + 1 < length; i += 9) {
207  if (!((~AV_RN64A(src + i) &
208  (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
209  0x8000800080008080ULL))
210  continue;
211  FIND_FIRST_ZERO;
213  i -= 7;
214  }
215 #else
216  for (i = 0; i + 1 < length; i += 5) {
217  if (!((~AV_RN32A(src + i) &
218  (AV_RN32A(src + i) - 0x01000101U)) &
219  0x80008080U))
220  continue;
221  FIND_FIRST_ZERO;
223  i -= 3;
224  }
225 #endif
226 #else
227  for (i = 0; i + 1 < length; i += 2) {
228  if (src[i])
229  continue;
230  if (i > 0 && src[i - 1] == 0)
231  i--;
233  }
234 #endif
235 
236  // use second escape buffer for inter data
237  bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
238 
239  si = h->rbsp_buffer_size[bufidx];
240  av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
241  dst = h->rbsp_buffer[bufidx];
242 
243  if (dst == NULL)
244  return NULL;
245 
246  if(i>=length-1){ //no escaped 0
247  *dst_length= length;
248  *consumed= length+1; //+1 for the header
249  if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
250  return src;
251  }else{
252  memcpy(dst, src, length);
253  return dst;
254  }
255  }
256 
257  memcpy(dst, src, i);
258  si = di = i;
259  while (si + 2 < length) {
260  // remove escapes (very rare 1:2^22)
261  if (src[si + 2] > 3) {
262  dst[di++] = src[si++];
263  dst[di++] = src[si++];
264  } else if (src[si] == 0 && src[si + 1] == 0) {
265  if (src[si + 2] == 3) { // escape
266  dst[di++] = 0;
267  dst[di++] = 0;
268  si += 3;
269  continue;
270  } else // next start code
271  goto nsc;
272  }
273 
274  dst[di++] = src[si++];
275  }
276  while (si < length)
277  dst[di++] = src[si++];
278 nsc:
279 
280  memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
281 
282  *dst_length = di;
283  *consumed = si + 1; // +1 for the header
284  /* FIXME store exact number of bits in the getbitcontext
285  * (it is needed for decoding) */
286  return dst;
287 }
288 
289 /**
290  * Identify the exact end of the bitstream
291  * @return the length of the trailing, or 0 if damaged
292  */
293 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
294 {
295  int v = *src;
296  int r;
297 
298  tprintf(h->s.avctx, "rbsp trailing %X\n", v);
299 
300  for (r = 1; r < 9; r++) {
301  if (v & 1)
302  return r;
303  v >>= 1;
304  }
305  return 0;
306 }
307 
308 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
309  int height, int y_offset, int list)
310 {
311  int raw_my = h->mv_cache[list][scan8[n]][1];
312  int filter_height_up = (raw_my & 3) ? 2 : 0;
313  int filter_height_down = (raw_my & 3) ? 3 : 0;
314  int full_my = (raw_my >> 2) + y_offset;
315  int top = full_my - filter_height_up;
316  int bottom = full_my + filter_height_down + height;
317 
318  return FFMAX(abs(top), bottom);
319 }
320 
321 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
322  int height, int y_offset, int list0,
323  int list1, int *nrefs)
324 {
325  MpegEncContext *const s = &h->s;
326  int my;
327 
328  y_offset += 16 * (s->mb_y >> MB_FIELD);
329 
330  if (list0) {
331  int ref_n = h->ref_cache[0][scan8[n]];
332  Picture *ref = &h->ref_list[0][ref_n];
333 
334  // Error resilience puts the current picture in the ref list.
335  // Don't try to wait on these as it will cause a deadlock.
336  // Fields can wait on each other, though.
337  if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
338  (ref->f.reference & 3) != s->picture_structure) {
339  my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
340  if (refs[0][ref_n] < 0)
341  nrefs[0] += 1;
342  refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
343  }
344  }
345 
346  if (list1) {
347  int ref_n = h->ref_cache[1][scan8[n]];
348  Picture *ref = &h->ref_list[1][ref_n];
349 
350  if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
351  (ref->f.reference & 3) != s->picture_structure) {
352  my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
353  if (refs[1][ref_n] < 0)
354  nrefs[1] += 1;
355  refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
356  }
357  }
358 }
359 
360 /**
361  * Wait until all reference frames are available for MC operations.
362  *
363  * @param h the H264 context
364  */
366 {
367  MpegEncContext *const s = &h->s;
368  const int mb_xy = h->mb_xy;
369  const int mb_type = s->current_picture.f.mb_type[mb_xy];
370  int refs[2][48];
371  int nrefs[2] = { 0 };
372  int ref, list;
373 
374  memset(refs, -1, sizeof(refs));
375 
376  if (IS_16X16(mb_type)) {
377  get_lowest_part_y(h, refs, 0, 16, 0,
378  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
379  } else if (IS_16X8(mb_type)) {
380  get_lowest_part_y(h, refs, 0, 8, 0,
381  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
382  get_lowest_part_y(h, refs, 8, 8, 8,
383  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
384  } else if (IS_8X16(mb_type)) {
385  get_lowest_part_y(h, refs, 0, 16, 0,
386  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
387  get_lowest_part_y(h, refs, 4, 16, 0,
388  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
389  } else {
390  int i;
391 
392  av_assert2(IS_8X8(mb_type));
393 
394  for (i = 0; i < 4; i++) {
395  const int sub_mb_type = h->sub_mb_type[i];
396  const int n = 4 * i;
397  int y_offset = (i & 2) << 2;
398 
399  if (IS_SUB_8X8(sub_mb_type)) {
400  get_lowest_part_y(h, refs, n, 8, y_offset,
401  IS_DIR(sub_mb_type, 0, 0),
402  IS_DIR(sub_mb_type, 0, 1),
403  nrefs);
404  } else if (IS_SUB_8X4(sub_mb_type)) {
405  get_lowest_part_y(h, refs, n, 4, y_offset,
406  IS_DIR(sub_mb_type, 0, 0),
407  IS_DIR(sub_mb_type, 0, 1),
408  nrefs);
409  get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
410  IS_DIR(sub_mb_type, 0, 0),
411  IS_DIR(sub_mb_type, 0, 1),
412  nrefs);
413  } else if (IS_SUB_4X8(sub_mb_type)) {
414  get_lowest_part_y(h, refs, n, 8, y_offset,
415  IS_DIR(sub_mb_type, 0, 0),
416  IS_DIR(sub_mb_type, 0, 1),
417  nrefs);
418  get_lowest_part_y(h, refs, n + 1, 8, y_offset,
419  IS_DIR(sub_mb_type, 0, 0),
420  IS_DIR(sub_mb_type, 0, 1),
421  nrefs);
422  } else {
423  int j;
424  av_assert2(IS_SUB_4X4(sub_mb_type));
425  for (j = 0; j < 4; j++) {
426  int sub_y_offset = y_offset + 2 * (j & 2);
427  get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
428  IS_DIR(sub_mb_type, 0, 0),
429  IS_DIR(sub_mb_type, 0, 1),
430  nrefs);
431  }
432  }
433  }
434  }
435 
436  for (list = h->list_count - 1; list >= 0; list--)
437  for (ref = 0; ref < 48 && nrefs[list]; ref++) {
438  int row = refs[list][ref];
439  if (row >= 0) {
440  Picture *ref_pic = &h->ref_list[list][ref];
441  int ref_field = ref_pic->f.reference - 1;
442  int ref_field_picture = ref_pic->field_picture;
443  int pic_height = 16 * s->mb_height >> ref_field_picture;
444 
445  row <<= MB_MBAFF;
446  nrefs[list]--;
447 
448  if (!FIELD_PICTURE && ref_field_picture) { // frame referencing two fields
449  ff_thread_await_progress(&ref_pic->f,
450  FFMIN((row >> 1) - !(row & 1),
451  pic_height - 1),
452  1);
453  ff_thread_await_progress(&ref_pic->f,
454  FFMIN((row >> 1), pic_height - 1),
455  0);
456  } else if (FIELD_PICTURE && !ref_field_picture) { // field referencing one field of a frame
457  ff_thread_await_progress(&ref_pic->f,
458  FFMIN(row * 2 + ref_field,
459  pic_height - 1),
460  0);
461  } else if (FIELD_PICTURE) {
462  ff_thread_await_progress(&ref_pic->f,
463  FFMIN(row, pic_height - 1),
464  ref_field);
465  } else {
466  ff_thread_await_progress(&ref_pic->f,
467  FFMIN(row, pic_height - 1),
468  0);
469  }
470  }
471  }
472 }
473 
475  int n, int square, int height,
476  int delta, int list,
477  uint8_t *dest_y, uint8_t *dest_cb,
478  uint8_t *dest_cr,
479  int src_x_offset, int src_y_offset,
480  qpel_mc_func *qpix_op,
481  h264_chroma_mc_func chroma_op,
482  int pixel_shift, int chroma_idc)
483 {
484  MpegEncContext *const s = &h->s;
485  const int mx = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
486  int my = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
487  const int luma_xy = (mx & 3) + ((my & 3) << 2);
488  int offset = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
489  uint8_t *src_y = pic->f.data[0] + offset;
490  uint8_t *src_cb, *src_cr;
491  int extra_width = h->emu_edge_width;
492  int extra_height = h->emu_edge_height;
493  int emu = 0;
494  const int full_mx = mx >> 2;
495  const int full_my = my >> 2;
496  const int pic_width = 16 * s->mb_width;
497  const int pic_height = 16 * s->mb_height >> MB_FIELD;
498  int ysh;
499 
500  if (mx & 7)
501  extra_width -= 3;
502  if (my & 7)
503  extra_height -= 3;
504 
505  if (full_mx < 0 - extra_width ||
506  full_my < 0 - extra_height ||
507  full_mx + 16 /*FIXME*/ > pic_width + extra_width ||
508  full_my + 16 /*FIXME*/ > pic_height + extra_height) {
510  src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
511  h->mb_linesize,
512  16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
513  full_my - 2, pic_width, pic_height);
514  src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
515  emu = 1;
516  }
517 
518  qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
519  if (!square)
520  qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
521 
522  if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
523  return;
524 
525  if (chroma_idc == 3 /* yuv444 */) {
526  src_cb = pic->f.data[1] + offset;
527  if (emu) {
529  src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
530  h->mb_linesize,
531  16 + 5, 16 + 5 /*FIXME*/,
532  full_mx - 2, full_my - 2,
533  pic_width, pic_height);
534  src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
535  }
536  qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
537  if (!square)
538  qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
539 
540  src_cr = pic->f.data[2] + offset;
541  if (emu) {
543  src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
544  h->mb_linesize,
545  16 + 5, 16 + 5 /*FIXME*/,
546  full_mx - 2, full_my - 2,
547  pic_width, pic_height);
548  src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
549  }
550  qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
551  if (!square)
552  qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
553  return;
554  }
555 
556  ysh = 3 - (chroma_idc == 2 /* yuv422 */);
557  if (chroma_idc == 1 /* yuv420 */ && MB_FIELD) {
558  // chroma offset when predicting from a field of opposite parity
559  my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
560  emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
561  }
562 
563  src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
564  (my >> ysh) * h->mb_uvlinesize;
565  src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
566  (my >> ysh) * h->mb_uvlinesize;
567 
568  if (emu) {
570  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
571  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
572  src_cb = s->edge_emu_buffer;
573  }
574  chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
575  height >> (chroma_idc == 1 /* yuv420 */),
576  mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
577 
578  if (emu) {
580  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
581  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
582  src_cr = s->edge_emu_buffer;
583  }
584  chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
585  mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
586 }
587 
588 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
589  int height, int delta,
590  uint8_t *dest_y, uint8_t *dest_cb,
591  uint8_t *dest_cr,
592  int x_offset, int y_offset,
593  qpel_mc_func *qpix_put,
594  h264_chroma_mc_func chroma_put,
595  qpel_mc_func *qpix_avg,
596  h264_chroma_mc_func chroma_avg,
597  int list0, int list1,
598  int pixel_shift, int chroma_idc)
599 {
600  MpegEncContext *const s = &h->s;
601  qpel_mc_func *qpix_op = qpix_put;
602  h264_chroma_mc_func chroma_op = chroma_put;
603 
604  dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
605  if (chroma_idc == 3 /* yuv444 */) {
606  dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
607  dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
608  } else if (chroma_idc == 2 /* yuv422 */) {
609  dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
610  dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
611  } else { /* yuv420 */
612  dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
613  dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
614  }
615  x_offset += 8 * s->mb_x;
616  y_offset += 8 * (s->mb_y >> MB_FIELD);
617 
618  if (list0) {
619  Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
620  mc_dir_part(h, ref, n, square, height, delta, 0,
621  dest_y, dest_cb, dest_cr, x_offset, y_offset,
622  qpix_op, chroma_op, pixel_shift, chroma_idc);
623 
624  qpix_op = qpix_avg;
625  chroma_op = chroma_avg;
626  }
627 
628  if (list1) {
629  Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
630  mc_dir_part(h, ref, n, square, height, delta, 1,
631  dest_y, dest_cb, dest_cr, x_offset, y_offset,
632  qpix_op, chroma_op, pixel_shift, chroma_idc);
633  }
634 }
635 
637  int height, int delta,
638  uint8_t *dest_y, uint8_t *dest_cb,
639  uint8_t *dest_cr,
640  int x_offset, int y_offset,
641  qpel_mc_func *qpix_put,
642  h264_chroma_mc_func chroma_put,
643  h264_weight_func luma_weight_op,
644  h264_weight_func chroma_weight_op,
645  h264_biweight_func luma_weight_avg,
646  h264_biweight_func chroma_weight_avg,
647  int list0, int list1,
648  int pixel_shift, int chroma_idc)
649 {
650  MpegEncContext *const s = &h->s;
651  int chroma_height;
652 
653  dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
654  if (chroma_idc == 3 /* yuv444 */) {
655  chroma_height = height;
656  chroma_weight_avg = luma_weight_avg;
657  chroma_weight_op = luma_weight_op;
658  dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
659  dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
660  } else if (chroma_idc == 2 /* yuv422 */) {
661  chroma_height = height;
662  dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
663  dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
664  } else { /* yuv420 */
665  chroma_height = height >> 1;
666  dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
667  dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
668  }
669  x_offset += 8 * s->mb_x;
670  y_offset += 8 * (s->mb_y >> MB_FIELD);
671 
672  if (list0 && list1) {
673  /* don't optimize for luma-only case, since B-frames usually
674  * use implicit weights => chroma too. */
675  uint8_t *tmp_cb = h->bipred_scratchpad;
676  uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
677  uint8_t *tmp_y = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
678  int refn0 = h->ref_cache[0][scan8[n]];
679  int refn1 = h->ref_cache[1][scan8[n]];
680 
681  mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
682  dest_y, dest_cb, dest_cr,
683  x_offset, y_offset, qpix_put, chroma_put,
684  pixel_shift, chroma_idc);
685  mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
686  tmp_y, tmp_cb, tmp_cr,
687  x_offset, y_offset, qpix_put, chroma_put,
688  pixel_shift, chroma_idc);
689 
690  if (h->use_weight == 2) {
691  int weight0 = h->implicit_weight[refn0][refn1][s->mb_y & 1];
692  int weight1 = 64 - weight0;
693  luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
694  height, 5, weight0, weight1, 0);
695  chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
696  chroma_height, 5, weight0, weight1, 0);
697  chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
698  chroma_height, 5, weight0, weight1, 0);
699  } else {
700  luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
702  h->luma_weight[refn0][0][0],
703  h->luma_weight[refn1][1][0],
704  h->luma_weight[refn0][0][1] +
705  h->luma_weight[refn1][1][1]);
706  chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
708  h->chroma_weight[refn0][0][0][0],
709  h->chroma_weight[refn1][1][0][0],
710  h->chroma_weight[refn0][0][0][1] +
711  h->chroma_weight[refn1][1][0][1]);
712  chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
714  h->chroma_weight[refn0][0][1][0],
715  h->chroma_weight[refn1][1][1][0],
716  h->chroma_weight[refn0][0][1][1] +
717  h->chroma_weight[refn1][1][1][1]);
718  }
719  } else {
720  int list = list1 ? 1 : 0;
721  int refn = h->ref_cache[list][scan8[n]];
722  Picture *ref = &h->ref_list[list][refn];
723  mc_dir_part(h, ref, n, square, height, delta, list,
724  dest_y, dest_cb, dest_cr, x_offset, y_offset,
725  qpix_put, chroma_put, pixel_shift, chroma_idc);
726 
727  luma_weight_op(dest_y, h->mb_linesize, height,
729  h->luma_weight[refn][list][0],
730  h->luma_weight[refn][list][1]);
731  if (h->use_weight_chroma) {
732  chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
734  h->chroma_weight[refn][list][0][0],
735  h->chroma_weight[refn][list][0][1]);
736  chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
738  h->chroma_weight[refn][list][1][0],
739  h->chroma_weight[refn][list][1][1]);
740  }
741  }
742 }
743 
745  int pixel_shift, int chroma_idc)
746 {
747  /* fetch pixels for estimated mv 4 macroblocks ahead
748  * optimized for 64byte cache lines */
749  MpegEncContext *const s = &h->s;
750  const int refn = h->ref_cache[list][scan8[0]];
751  if (refn >= 0) {
752  const int mx = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * s->mb_x + 8;
753  const int my = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * s->mb_y;
754  uint8_t **src = h->ref_list[list][refn].f.data;
755  int off = (mx << pixel_shift) +
756  (my + (s->mb_x & 3) * 4) * h->mb_linesize +
757  (64 << pixel_shift);
758  s->vdsp.prefetch(src[0] + off, s->linesize, 4);
759  if (chroma_idc == 3 /* yuv444 */) {
760  s->vdsp.prefetch(src[1] + off, s->linesize, 4);
761  s->vdsp.prefetch(src[2] + off, s->linesize, 4);
762  } else {
763  off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
764  s->vdsp.prefetch(src[1] + off, src[2] - src[1], 2);
765  }
766  }
767 }
768 
769 static void free_tables(H264Context *h, int free_rbsp)
770 {
771  int i;
772  H264Context *hx;
773 
776  av_freep(&h->cbp_table);
777  av_freep(&h->mvd_table[0]);
778  av_freep(&h->mvd_table[1]);
779  av_freep(&h->direct_table);
782  h->slice_table = NULL;
783  av_freep(&h->list_counts);
784 
785  av_freep(&h->mb2b_xy);
786  av_freep(&h->mb2br_xy);
787 
788  for (i = 0; i < MAX_THREADS; i++) {
789  hx = h->thread_context[i];
790  if (!hx)
791  continue;
792  av_freep(&hx->top_borders[1]);
793  av_freep(&hx->top_borders[0]);
795  if (free_rbsp) {
796  av_freep(&hx->rbsp_buffer[1]);
797  av_freep(&hx->rbsp_buffer[0]);
798  hx->rbsp_buffer_size[0] = 0;
799  hx->rbsp_buffer_size[1] = 0;
800  }
801  if (i)
802  av_freep(&h->thread_context[i]);
803  }
804 }
805 
807 {
808  int i, j, q, x;
809  const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
810 
811  for (i = 0; i < 6; i++) {
812  h->dequant8_coeff[i] = h->dequant8_buffer[i];
813  for (j = 0; j < i; j++)
814  if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
815  64 * sizeof(uint8_t))) {
816  h->dequant8_coeff[i] = h->dequant8_buffer[j];
817  break;
818  }
819  if (j < i)
820  continue;
821 
822  for (q = 0; q < max_qp + 1; q++) {
823  int shift = div6[q];
824  int idx = rem6[q];
825  for (x = 0; x < 64; x++)
826  h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
827  ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
828  h->pps.scaling_matrix8[i][x]) << shift;
829  }
830  }
831 }
832 
834 {
835  int i, j, q, x;
836  const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
837  for (i = 0; i < 6; i++) {
838  h->dequant4_coeff[i] = h->dequant4_buffer[i];
839  for (j = 0; j < i; j++)
840  if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
841  16 * sizeof(uint8_t))) {
842  h->dequant4_coeff[i] = h->dequant4_buffer[j];
843  break;
844  }
845  if (j < i)
846  continue;
847 
848  for (q = 0; q < max_qp + 1; q++) {
849  int shift = div6[q] + 2;
850  int idx = rem6[q];
851  for (x = 0; x < 16; x++)
852  h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
853  ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
854  h->pps.scaling_matrix4[i][x]) << shift;
855  }
856  }
857 }
858 
860 {
861  int i, x;
863  if (h->pps.transform_8x8_mode)
865  if (h->sps.transform_bypass) {
866  for (i = 0; i < 6; i++)
867  for (x = 0; x < 16; x++)
868  h->dequant4_coeff[i][0][x] = 1 << 6;
870  for (i = 0; i < 6; i++)
871  for (x = 0; x < 64; x++)
872  h->dequant8_coeff[i][0][x] = 1 << 6;
873  }
874 }
875 
877 {
878  MpegEncContext *const s = &h->s;
879  const int big_mb_num = s->mb_stride * (s->mb_height + 1);
880  const int row_mb_num = 2*s->mb_stride*FFMAX(s->avctx->thread_count, 1);
881  int x, y;
882 
884  row_mb_num * 8 * sizeof(uint8_t), fail)
886  big_mb_num * 48 * sizeof(uint8_t), fail)
888  (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base), fail)
890  big_mb_num * sizeof(uint16_t), fail)
892  big_mb_num * sizeof(uint8_t), fail)
894  16 * row_mb_num * sizeof(uint8_t), fail);
896  16 * row_mb_num * sizeof(uint8_t), fail);
898  4 * big_mb_num * sizeof(uint8_t), fail);
900  big_mb_num * sizeof(uint8_t), fail)
901 
902  memset(h->slice_table_base, -1,
903  (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base));
904  h->slice_table = h->slice_table_base + s->mb_stride * 2 + 1;
905 
907  big_mb_num * sizeof(uint32_t), fail);
909  big_mb_num * sizeof(uint32_t), fail);
910  for (y = 0; y < s->mb_height; y++)
911  for (x = 0; x < s->mb_width; x++) {
912  const int mb_xy = x + y * s->mb_stride;
913  const int b_xy = 4 * x + 4 * y * h->b_stride;
914 
915  h->mb2b_xy[mb_xy] = b_xy;
916  h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * s->mb_stride)));
917  }
918 
919  if (!h->dequant4_coeff[0])
921 
922  return 0;
923 
924 fail:
925  free_tables(h, 1);
926  return -1;
927 }
928 
929 /**
930  * Mimic alloc_tables(), but for every context thread.
931  */
932 static void clone_tables(H264Context *dst, H264Context *src, int i)
933 {
934  MpegEncContext *const s = &src->s;
935  dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
936  dst->non_zero_count = src->non_zero_count;
937  dst->slice_table = src->slice_table;
938  dst->cbp_table = src->cbp_table;
939  dst->mb2b_xy = src->mb2b_xy;
940  dst->mb2br_xy = src->mb2br_xy;
942  dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
943  dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
944  dst->direct_table = src->direct_table;
945  dst->list_counts = src->list_counts;
946  dst->bipred_scratchpad = NULL;
947  ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
948  src->sps.chroma_format_idc);
949 }
950 
951 /**
952  * Init context
953  * Allocate buffers which are not shared amongst multiple threads.
954  */
955 static int context_init(H264Context *h)
956 {
958  h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
960  h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
961 
962  h->ref_cache[0][scan8[5] + 1] =
963  h->ref_cache[0][scan8[7] + 1] =
964  h->ref_cache[0][scan8[13] + 1] =
965  h->ref_cache[1][scan8[5] + 1] =
966  h->ref_cache[1][scan8[7] + 1] =
967  h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
968 
969  return 0;
970 
971 fail:
972  return -1; // free_tables will clean up for us
973 }
974 
975 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
976  int parse_extradata);
977 
979 {
980  MpegEncContext *const s = &h->s;
981 
982  s->width = s->avctx->width;
983  s->height = s->avctx->height;
984  s->codec_id = s->avctx->codec->id;
985 
986  s->avctx->bits_per_raw_sample = 8;
987  h->cur_chroma_format_idc = 1;
988 
993 
994  h->dequant_coeff_pps = -1;
995  s->unrestricted_mv = 1;
996 
997  s->dsp.dct_bits = 16;
998  /* needed so that IDCT permutation is known early */
999  ff_dsputil_init(&s->dsp, s->avctx);
1000  ff_videodsp_init(&s->vdsp, 8);
1001 
1002  memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1003  memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1004 }
1005 
1007 {
1008  AVCodecContext *avctx = h->s.avctx;
1009 
1010  if (!buf || size <= 0)
1011  return -1;
1012 
1013  if (buf[0] == 1) {
1014  int i, cnt, nalsize;
1015  const unsigned char *p = buf;
1016 
1017  h->is_avc = 1;
1018 
1019  if (size < 7) {
1020  av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1021  return -1;
1022  }
1023  /* sps and pps in the avcC always have length coded with 2 bytes,
1024  * so put a fake nal_length_size = 2 while parsing them */
1025  h->nal_length_size = 2;
1026  // Decode sps from avcC
1027  cnt = *(p + 5) & 0x1f; // Number of sps
1028  p += 6;
1029  for (i = 0; i < cnt; i++) {
1030  nalsize = AV_RB16(p) + 2;
1031  if(nalsize > size - (p-buf))
1032  return -1;
1033  if (decode_nal_units(h, p, nalsize, 1) < 0) {
1034  av_log(avctx, AV_LOG_ERROR,
1035  "Decoding sps %d from avcC failed\n", i);
1036  return -1;
1037  }
1038  p += nalsize;
1039  }
1040  // Decode pps from avcC
1041  cnt = *(p++); // Number of pps
1042  for (i = 0; i < cnt; i++) {
1043  nalsize = AV_RB16(p) + 2;
1044  if(nalsize > size - (p-buf))
1045  return -1;
1046  if (decode_nal_units(h, p, nalsize, 1) < 0) {
1047  av_log(avctx, AV_LOG_ERROR,
1048  "Decoding pps %d from avcC failed\n", i);
1049  return -1;
1050  }
1051  p += nalsize;
1052  }
1053  // Now store right nal length size, that will be used to parse all other nals
1054  h->nal_length_size = (buf[4] & 0x03) + 1;
1055  } else {
1056  h->is_avc = 0;
1057  if (decode_nal_units(h, buf, size, 1) < 0)
1058  return -1;
1059  }
1060  return size;
1061 }
1062 
1064 {
1065  H264Context *h = avctx->priv_data;
1066  MpegEncContext *const s = &h->s;
1067  int i;
1068 
1070 
1071  s->avctx = avctx;
1072  common_init(h);
1073 
1074  s->out_format = FMT_H264;
1075  s->workaround_bugs = avctx->workaround_bugs;
1076 
1077  /* set defaults */
1078  // s->decode_mb = ff_h263_decode_mb;
1079  s->quarter_sample = 1;
1080  if (!avctx->has_b_frames)
1081  s->low_delay = 1;
1082 
1084 
1086 
1087  h->pixel_shift = 0;
1088  h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1089 
1090  h->thread_context[0] = h;
1091  h->outputed_poc = h->next_outputed_poc = INT_MIN;
1092  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1093  h->last_pocs[i] = INT_MIN;
1094  h->prev_poc_msb = 1 << 16;
1095  h->prev_frame_num = -1;
1096  h->x264_build = -1;
1097  ff_h264_reset_sei(h);
1098  if (avctx->codec_id == AV_CODEC_ID_H264) {
1099  if (avctx->ticks_per_frame == 1) {
1100  if(s->avctx->time_base.den < INT_MAX/2) {
1101  s->avctx->time_base.den *= 2;
1102  } else
1103  s->avctx->time_base.num /= 2;
1104  }
1105  avctx->ticks_per_frame = 2;
1106  }
1107 
1108  if (avctx->extradata_size > 0 && avctx->extradata &&
1109  ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1111  return -1;
1112  }
1113 
1117  s->low_delay = 0;
1118  }
1119 
1121 
1122  return 0;
1123 }
1124 
1125 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1126 
1127 static void copy_picture_range(Picture **to, Picture **from, int count,
1128  MpegEncContext *new_base,
1129  MpegEncContext *old_base)
1130 {
1131  int i;
1132 
1133  for (i = 0; i < count; i++) {
1134  assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1135  IN_RANGE(from[i], old_base->picture,
1136  sizeof(Picture) * old_base->picture_count) ||
1137  !from[i]));
1138  to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1139  }
1140 }
1141 
1142 static void copy_parameter_set(void **to, void **from, int count, int size)
1143 {
1144  int i;
1145 
1146  for (i = 0; i < count; i++) {
1147  if (to[i] && !from[i])
1148  av_freep(&to[i]);
1149  else if (from[i] && !to[i])
1150  to[i] = av_malloc(size);
1151 
1152  if (from[i])
1153  memcpy(to[i], from[i], size);
1154  }
1155 }
1156 
1158 {
1159  H264Context *h = avctx->priv_data;
1160 
1161  if (!avctx->internal->is_copy)
1162  return 0;
1163  memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1164  memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1165 
1166  h->s.context_initialized = 0;
1167 
1168  return 0;
1169 }
1170 
1171 #define copy_fields(to, from, start_field, end_field) \
1172  memcpy(&to->start_field, &from->start_field, \
1173  (char *)&to->end_field - (char *)&to->start_field)
1174 
1175 static int h264_slice_header_init(H264Context *, int);
1176 
1178 
1180  const AVCodecContext *src)
1181 {
1182  H264Context *h = dst->priv_data, *h1 = src->priv_data;
1183  MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1184  int inited = s->context_initialized, err;
1185  int i;
1186 
1187  if (dst == src)
1188  return 0;
1189 
1190  if (inited &&
1191  (s->width != s1->width ||
1192  s->height != s1->height ||
1193  s->mb_width != s1->mb_width ||
1194  s->mb_height != s1->mb_height ||
1195  h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
1196  h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1197  h->sps.colorspace != h1->sps.colorspace)) {
1198 
1200 
1201  s->width = s1->width;
1202  s->height = s1->height;
1203  s->mb_height = s1->mb_height;
1204  h->b_stride = h1->b_stride;
1205  // SPS/PPS
1206  copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1207  MAX_SPS_COUNT, sizeof(SPS));
1208  h->sps = h1->sps;
1209  copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1210  MAX_PPS_COUNT, sizeof(PPS));
1211  h->pps = h1->pps;
1212 
1213  if ((err = h264_slice_header_init(h, 1)) < 0) {
1214  av_log(h->s.avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1215  return err;
1216  }
1217  h->context_reinitialized = 1;
1218 
1219  /* update linesize on resize for h264. The h264 decoder doesn't
1220  * necessarily call ff_MPV_frame_start in the new thread */
1221  s->linesize = s1->linesize;
1222  s->uvlinesize = s1->uvlinesize;
1223 
1224  /* copy block_offset since frame_start may not be called */
1225  memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1227  }
1228 
1229  err = ff_mpeg_update_thread_context(dst, src);
1230  if (err)
1231  return err;
1232 
1233  if (!inited) {
1234  for (i = 0; i < MAX_SPS_COUNT; i++)
1235  av_freep(h->sps_buffers + i);
1236 
1237  for (i = 0; i < MAX_PPS_COUNT; i++)
1238  av_freep(h->pps_buffers + i);
1239 
1240  // copy all fields after MpegEnc
1241  memcpy(&h->s + 1, &h1->s + 1,
1242  sizeof(H264Context) - sizeof(MpegEncContext));
1243  memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1244  memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1245 
1246  if (s1->context_initialized) {
1247  if (ff_h264_alloc_tables(h) < 0) {
1248  av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1249  return AVERROR(ENOMEM);
1250  }
1251  context_init(h);
1252 
1253  }
1254 
1255  for (i = 0; i < 2; i++) {
1256  h->rbsp_buffer[i] = NULL;
1257  h->rbsp_buffer_size[i] = 0;
1258  }
1259  h->bipred_scratchpad = NULL;
1260 
1261  h->thread_context[0] = h;
1262 
1263  s->dsp.clear_blocks(h->mb);
1264  s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1265  }
1266 
1267  /* frame_start may not be called for the next thread (if it's decoding
1268  * a bottom field) so this has to be allocated here */
1269  if (!h->bipred_scratchpad && s->linesize)
1270  h->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1271 
1272  // extradata/NAL handling
1273  h->is_avc = h1->is_avc;
1274 
1275  // SPS/PPS
1276  copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1277  MAX_SPS_COUNT, sizeof(SPS));
1278  h->sps = h1->sps;
1279  copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1280  MAX_PPS_COUNT, sizeof(PPS));
1281  h->pps = h1->pps;
1282 
1283  // Dequantization matrices
1284  // FIXME these are big - can they be only copied when PPS changes?
1285  copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1286 
1287  for (i = 0; i < 6; i++)
1288  h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1289  (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1290 
1291  for (i = 0; i < 6; i++)
1292  h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1293  (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1294 
1295  h->dequant_coeff_pps = h1->dequant_coeff_pps;
1296 
1297  // POC timing
1298  copy_fields(h, h1, poc_lsb, redundant_pic_count);
1299 
1300  // reference lists
1301  copy_fields(h, h1, ref_count, list_count);
1302  copy_fields(h, h1, ref_list, intra_gb);
1303  copy_fields(h, h1, short_ref, cabac_init_idc);
1304 
1305  copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1306  copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1307  copy_picture_range(h->delayed_pic, h1->delayed_pic,
1308  MAX_DELAYED_PIC_COUNT + 2, s, s1);
1309 
1310  h->last_slice_type = h1->last_slice_type;
1311  h->sync = h1->sync;
1312 
1313  if (!s->current_picture_ptr)
1314  return 0;
1315 
1316  if (!s->droppable) {
1318  h->prev_poc_msb = h->poc_msb;
1319  h->prev_poc_lsb = h->poc_lsb;
1320  }
1322  h->prev_frame_num = h->frame_num;
1324 
1325  return err;
1326 }
1327 
1329 {
1330  MpegEncContext *const s = &h->s;
1331  int i;
1332  const int pixel_shift = h->pixel_shift;
1333 
1334  if (ff_MPV_frame_start(s, s->avctx) < 0)
1335  return -1;
1336  ff_er_frame_start(s);
1337  /*
1338  * ff_MPV_frame_start uses pict_type to derive key_frame.
1339  * This is incorrect for H.264; IDR markings must be used.
1340  * Zero here; IDR markings per slice in frame or fields are ORed in later.
1341  * See decode_nal_units().
1342  */
1344  s->current_picture_ptr->sync = 0;
1346 
1347  assert(s->linesize && s->uvlinesize);
1348 
1349  for (i = 0; i < 16; i++) {
1350  h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1351  h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1352  }
1353  for (i = 0; i < 16; i++) {
1354  h->block_offset[16 + i] =
1355  h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1356  h->block_offset[48 + 16 + i] =
1357  h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1358  }
1359 
1360  /* can't be in alloc_tables because linesize isn't known there.
1361  * FIXME: redo bipred weight to not require extra buffer? */
1362  for (i = 0; i < s->slice_context_count; i++)
1363  if (h->thread_context[i] && !h->thread_context[i]->bipred_scratchpad)
1364  h->thread_context[i]->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1365 
1366  /* Some macroblocks can be accessed before they're available in case
1367  * of lost slices, MBAFF or threading. */
1368  memset(h->slice_table, -1,
1369  (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1370 
1371  // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1372  // s->current_picture.f.reference /* || h->contains_intra */ || 1;
1373 
1374  /* We mark the current picture as non-reference after allocating it, so
1375  * that if we break out due to an error it can be released automatically
1376  * in the next ff_MPV_frame_start().
1377  * SVQ3 as well as most other codecs have only last/next/current and thus
1378  * get released even with set reference, besides SVQ3 and others do not
1379  * mark frames as reference later "naturally". */
1380  if (s->codec_id != AV_CODEC_ID_SVQ3)
1382 
1384  s->current_picture_ptr->field_poc[1] = INT_MAX;
1385 
1386  h->next_output_pic = NULL;
1387 
1388  assert(s->current_picture_ptr->long_ref == 0);
1389 
1390  return 0;
1391 }
1392 
1393 /**
1394  * Run setup operations that must be run after slice header decoding.
1395  * This includes finding the next displayed frame.
1396  *
1397  * @param h h264 master context
1398  * @param setup_finished enough NALs have been read that we can call
1399  * ff_thread_finish_setup()
1400  */
1401 static void decode_postinit(H264Context *h, int setup_finished)
1402 {
1403  MpegEncContext *const s = &h->s;
1404  Picture *out = s->current_picture_ptr;
1405  Picture *cur = s->current_picture_ptr;
1406  int i, pics, out_of_order, out_idx;
1407 
1410 
1411  if (h->next_output_pic)
1412  return;
1413 
1414  if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1415  /* FIXME: if we have two PAFF fields in one packet, we can't start
1416  * the next thread here. If we have one field per packet, we can.
1417  * The check in decode_nal_units() is not good enough to find this
1418  * yet, so we assume the worst for now. */
1419  // if (setup_finished)
1420  // ff_thread_finish_setup(s->avctx);
1421  return;
1422  }
1423 
1424  cur->f.interlaced_frame = 0;
1425  cur->f.repeat_pict = 0;
1426 
1427  /* Signal interlacing information externally. */
1428  /* Prioritize picture timing SEI information over used
1429  * decoding process if it exists. */
1430 
1431  if (h->sps.pic_struct_present_flag) {
1432  switch (h->sei_pic_struct) {
1433  case SEI_PIC_STRUCT_FRAME:
1434  break;
1437  cur->f.interlaced_frame = 1;
1438  break;
1442  cur->f.interlaced_frame = 1;
1443  else
1444  // try to flag soft telecine progressive
1446  break;
1449  /* Signal the possibility of telecined film externally
1450  * (pic_struct 5,6). From these hints, let the applications
1451  * decide if they apply deinterlacing. */
1452  cur->f.repeat_pict = 1;
1453  break;
1455  // Force progressive here, doubling interlaced frame is a bad idea.
1456  cur->f.repeat_pict = 2;
1457  break;
1459  cur->f.repeat_pict = 4;
1460  break;
1461  }
1462 
1463  if ((h->sei_ct_type & 3) &&
1465  cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1466  } else {
1467  /* Derive interlacing flag from used decoding process. */
1469  }
1471 
1472  if (cur->field_poc[0] != cur->field_poc[1]) {
1473  /* Derive top_field_first from field pocs. */
1474  cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1475  } else {
1476  if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1477  /* Use picture timing SEI information. Even if it is a
1478  * information of a past frame, better than nothing. */
1481  cur->f.top_field_first = 1;
1482  else
1483  cur->f.top_field_first = 0;
1484  } else {
1485  /* Most likely progressive */
1486  cur->f.top_field_first = 0;
1487  }
1488  }
1489 
1490  cur->mmco_reset = h->mmco_reset;
1491  h->mmco_reset = 0;
1492  // FIXME do something with unavailable reference frames
1493 
1494  /* Sort B-frames into display order */
1495 
1499  s->low_delay = 0;
1500  }
1501 
1505  s->low_delay = 0;
1506  }
1507 
1508  for (i = 0; 1; i++) {
1509  if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1510  if(i)
1511  h->last_pocs[i-1] = cur->poc;
1512  break;
1513  } else if(i) {
1514  h->last_pocs[i-1]= h->last_pocs[i];
1515  }
1516  }
1517  out_of_order = MAX_DELAYED_PIC_COUNT - i;
1518  if( cur->f.pict_type == AV_PICTURE_TYPE_B
1520  out_of_order = FFMAX(out_of_order, 1);
1521  if (out_of_order == MAX_DELAYED_PIC_COUNT) {
1522  av_log(s->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
1523  for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
1524  h->last_pocs[i] = INT_MIN;
1525  h->last_pocs[0] = cur->poc;
1526  cur->mmco_reset = 1;
1527  } else if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1528  av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1529  s->avctx->has_b_frames = out_of_order;
1530  s->low_delay = 0;
1531  }
1532 
1533  pics = 0;
1534  while (h->delayed_pic[pics])
1535  pics++;
1536 
1538 
1539  h->delayed_pic[pics++] = cur;
1540  if (cur->f.reference == 0)
1541  cur->f.reference = DELAYED_PIC_REF;
1542 
1543  out = h->delayed_pic[0];
1544  out_idx = 0;
1545  for (i = 1; h->delayed_pic[i] &&
1546  !h->delayed_pic[i]->f.key_frame &&
1547  !h->delayed_pic[i]->mmco_reset;
1548  i++)
1549  if (h->delayed_pic[i]->poc < out->poc) {
1550  out = h->delayed_pic[i];
1551  out_idx = i;
1552  }
1553  if (s->avctx->has_b_frames == 0 &&
1554  (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1555  h->next_outputed_poc = INT_MIN;
1556  out_of_order = out->poc < h->next_outputed_poc;
1557 
1558  if (out_of_order || pics > s->avctx->has_b_frames) {
1559  out->f.reference &= ~DELAYED_PIC_REF;
1560  // for frame threading, the owner must be the second field's thread or
1561  // else the first thread can release the picture and reuse it unsafely
1562  out->owner2 = s;
1563  for (i = out_idx; h->delayed_pic[i]; i++)
1564  h->delayed_pic[i] = h->delayed_pic[i + 1];
1565  }
1566  if (!out_of_order && pics > s->avctx->has_b_frames) {
1567  h->next_output_pic = out;
1568  if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1569  h->next_outputed_poc = INT_MIN;
1570  } else
1571  h->next_outputed_poc = out->poc;
1572  } else {
1573  av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1574  }
1575 
1576  if (h->next_output_pic && h->next_output_pic->sync) {
1577  h->sync |= 2;
1578  }
1579 
1580  if (setup_finished)
1582 }
1583 
1585  uint8_t *src_cb, uint8_t *src_cr,
1586  int linesize, int uvlinesize,
1587  int simple)
1588 {
1589  MpegEncContext *const s = &h->s;
1590  uint8_t *top_border;
1591  int top_idx = 1;
1592  const int pixel_shift = h->pixel_shift;
1593  int chroma444 = CHROMA444;
1594  int chroma422 = CHROMA422;
1595 
1596  src_y -= linesize;
1597  src_cb -= uvlinesize;
1598  src_cr -= uvlinesize;
1599 
1600  if (!simple && FRAME_MBAFF) {
1601  if (s->mb_y & 1) {
1602  if (!MB_MBAFF) {
1603  top_border = h->top_borders[0][s->mb_x];
1604  AV_COPY128(top_border, src_y + 15 * linesize);
1605  if (pixel_shift)
1606  AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1607  if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1608  if (chroma444) {
1609  if (pixel_shift) {
1610  AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1611  AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1612  AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1613  AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1614  } else {
1615  AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1616  AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1617  }
1618  } else if (chroma422) {
1619  if (pixel_shift) {
1620  AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1621  AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1622  } else {
1623  AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1624  AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1625  }
1626  } else {
1627  if (pixel_shift) {
1628  AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1629  AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1630  } else {
1631  AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1632  AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1633  }
1634  }
1635  }
1636  }
1637  } else if (MB_MBAFF) {
1638  top_idx = 0;
1639  } else
1640  return;
1641  }
1642 
1643  top_border = h->top_borders[top_idx][s->mb_x];
1644  /* There are two lines saved, the line above the top macroblock
1645  * of a pair, and the line above the bottom macroblock. */
1646  AV_COPY128(top_border, src_y + 16 * linesize);
1647  if (pixel_shift)
1648  AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1649 
1650  if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1651  if (chroma444) {
1652  if (pixel_shift) {
1653  AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1654  AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1655  AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1656  AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1657  } else {
1658  AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1659  AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1660  }
1661  } else if (chroma422) {
1662  if (pixel_shift) {
1663  AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1664  AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1665  } else {
1666  AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1667  AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1668  }
1669  } else {
1670  if (pixel_shift) {
1671  AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1672  AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1673  } else {
1674  AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1675  AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1676  }
1677  }
1678  }
1679 }
1680 
1682  uint8_t *src_cb, uint8_t *src_cr,
1683  int linesize, int uvlinesize,
1684  int xchg, int chroma444,
1685  int simple, int pixel_shift)
1686 {
1687  MpegEncContext *const s = &h->s;
1688  int deblock_topleft;
1689  int deblock_top;
1690  int top_idx = 1;
1691  uint8_t *top_border_m1;
1692  uint8_t *top_border;
1693 
1694  if (!simple && FRAME_MBAFF) {
1695  if (s->mb_y & 1) {
1696  if (!MB_MBAFF)
1697  return;
1698  } else {
1699  top_idx = MB_MBAFF ? 0 : 1;
1700  }
1701  }
1702 
1703  if (h->deblocking_filter == 2) {
1704  deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1705  deblock_top = h->top_type;
1706  } else {
1707  deblock_topleft = (s->mb_x > 0);
1708  deblock_top = (s->mb_y > !!MB_FIELD);
1709  }
1710 
1711  src_y -= linesize + 1 + pixel_shift;
1712  src_cb -= uvlinesize + 1 + pixel_shift;
1713  src_cr -= uvlinesize + 1 + pixel_shift;
1714 
1715  top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1716  top_border = h->top_borders[top_idx][s->mb_x];
1717 
1718 #define XCHG(a, b, xchg) \
1719  if (pixel_shift) { \
1720  if (xchg) { \
1721  AV_SWAP64(b + 0, a + 0); \
1722  AV_SWAP64(b + 8, a + 8); \
1723  } else { \
1724  AV_COPY128(b, a); \
1725  } \
1726  } else if (xchg) \
1727  AV_SWAP64(b, a); \
1728  else \
1729  AV_COPY64(b, a);
1730 
1731  if (deblock_top) {
1732  if (deblock_topleft) {
1733  XCHG(top_border_m1 + (8 << pixel_shift),
1734  src_y - (7 << pixel_shift), 1);
1735  }
1736  XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1737  XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1738  if (s->mb_x + 1 < s->mb_width) {
1739  XCHG(h->top_borders[top_idx][s->mb_x + 1],
1740  src_y + (17 << pixel_shift), 1);
1741  }
1742  }
1743  if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1744  if (chroma444) {
1745  if (deblock_topleft) {
1746  XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1747  XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1748  }
1749  XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1750  XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1751  XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1752  XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1753  if (s->mb_x + 1 < s->mb_width) {
1754  XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1755  XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1756  }
1757  } else {
1758  if (deblock_top) {
1759  if (deblock_topleft) {
1760  XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1761  XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1762  }
1763  XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1764  XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1765  }
1766  }
1767  }
1768 }
1769 
1770 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1771  int index)
1772 {
1773  if (high_bit_depth) {
1774  return AV_RN32A(((int32_t *)mb) + index);
1775  } else
1776  return AV_RN16A(mb + index);
1777 }
1778 
1779 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1780  int index, int value)
1781 {
1782  if (high_bit_depth) {
1783  AV_WN32A(((int32_t *)mb) + index, value);
1784  } else
1785  AV_WN16A(mb + index, value);
1786 }
1787 
1789  int mb_type, int is_h264,
1790  int simple,
1791  int transform_bypass,
1792  int pixel_shift,
1793  int *block_offset,
1794  int linesize,
1795  uint8_t *dest_y, int p)
1796 {
1797  MpegEncContext *const s = &h->s;
1798  void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1799  void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1800  int i;
1801  int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1802  block_offset += 16 * p;
1803  if (IS_INTRA4x4(mb_type)) {
1804  if (simple || !s->encoding) {
1805  if (IS_8x8DCT(mb_type)) {
1806  if (transform_bypass) {
1807  idct_dc_add =
1808  idct_add = s->dsp.add_pixels8;
1809  } else {
1810  idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1812  }
1813  for (i = 0; i < 16; i += 4) {
1814  uint8_t *const ptr = dest_y + block_offset[i];
1815  const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1816  if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1817  h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1818  } else {
1819  const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1820  h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1821  (h->topright_samples_available << i) & 0x4000, linesize);
1822  if (nnz) {
1823  if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1824  idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1825  else
1826  idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1827  }
1828  }
1829  }
1830  } else {
1831  if (transform_bypass) {
1832  idct_dc_add =
1833  idct_add = s->dsp.add_pixels4;
1834  } else {
1835  idct_dc_add = h->h264dsp.h264_idct_dc_add;
1837  }
1838  for (i = 0; i < 16; i++) {
1839  uint8_t *const ptr = dest_y + block_offset[i];
1840  const int dir = h->intra4x4_pred_mode_cache[scan8[i]];
1841 
1842  if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1843  h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1844  } else {
1845  uint8_t *topright;
1846  int nnz, tr;
1847  uint64_t tr_high;
1848  if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1849  const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1850  av_assert2(s->mb_y || linesize <= block_offset[i]);
1851  if (!topright_avail) {
1852  if (pixel_shift) {
1853  tr_high = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1854  topright = (uint8_t *)&tr_high;
1855  } else {
1856  tr = ptr[3 - linesize] * 0x01010101u;
1857  topright = (uint8_t *)&tr;
1858  }
1859  } else
1860  topright = ptr + (4 << pixel_shift) - linesize;
1861  } else
1862  topright = NULL;
1863 
1864  h->hpc.pred4x4[dir](ptr, topright, linesize);
1865  nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1866  if (nnz) {
1867  if (is_h264) {
1868  if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1869  idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1870  else
1871  idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1872  } else if (CONFIG_SVQ3_DECODER)
1873  ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1874  }
1875  }
1876  }
1877  }
1878  }
1879  } else {
1880  h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1881  if (is_h264) {
1883  if (!transform_bypass)
1884  h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1885  h->mb_luma_dc[p],
1886  h->dequant4_coeff[p][qscale][0]);
1887  else {
1888  static const uint8_t dc_mapping[16] = {
1889  0 * 16, 1 * 16, 4 * 16, 5 * 16,
1890  2 * 16, 3 * 16, 6 * 16, 7 * 16,
1891  8 * 16, 9 * 16, 12 * 16, 13 * 16,
1892  10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1893  for (i = 0; i < 16; i++)
1894  dctcoef_set(h->mb + (p * 256 << pixel_shift),
1895  pixel_shift, dc_mapping[i],
1896  dctcoef_get(h->mb_luma_dc[p],
1897  pixel_shift, i));
1898  }
1899  }
1900  } else if (CONFIG_SVQ3_DECODER)
1901  ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1902  h->mb_luma_dc[p], qscale);
1903  }
1904 }
1905 
1907  int is_h264, int simple,
1908  int transform_bypass,
1909  int pixel_shift,
1910  int *block_offset,
1911  int linesize,
1912  uint8_t *dest_y, int p)
1913 {
1914  MpegEncContext *const s = &h->s;
1915  void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1916  int i;
1917  block_offset += 16 * p;
1918  if (!IS_INTRA4x4(mb_type)) {
1919  if (is_h264) {
1920  if (IS_INTRA16x16(mb_type)) {
1921  if (transform_bypass) {
1922  if (h->sps.profile_idc == 244 &&
1925  h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1926  h->mb + (p * 256 << pixel_shift),
1927  linesize);
1928  } else {
1929  for (i = 0; i < 16; i++)
1930  if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1931  dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1932  s->dsp.add_pixels4(dest_y + block_offset[i],
1933  h->mb + (i * 16 + p * 256 << pixel_shift),
1934  linesize);
1935  }
1936  } else {
1937  h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1938  h->mb + (p * 256 << pixel_shift),
1939  linesize,
1940  h->non_zero_count_cache + p * 5 * 8);
1941  }
1942  } else if (h->cbp & 15) {
1943  if (transform_bypass) {
1944  const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1945  idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1946  : s->dsp.add_pixels4;
1947  for (i = 0; i < 16; i += di)
1948  if (h->non_zero_count_cache[scan8[i + p * 16]])
1949  idct_add(dest_y + block_offset[i],
1950  h->mb + (i * 16 + p * 256 << pixel_shift),
1951  linesize);
1952  } else {
1953  if (IS_8x8DCT(mb_type))
1954  h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1955  h->mb + (p * 256 << pixel_shift),
1956  linesize,
1957  h->non_zero_count_cache + p * 5 * 8);
1958  else
1959  h->h264dsp.h264_idct_add16(dest_y, block_offset,
1960  h->mb + (p * 256 << pixel_shift),
1961  linesize,
1962  h->non_zero_count_cache + p * 5 * 8);
1963  }
1964  }
1965  } else if (CONFIG_SVQ3_DECODER) {
1966  for (i = 0; i < 16; i++)
1967  if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1968  // FIXME benchmark weird rule, & below
1969  uint8_t *const ptr = dest_y + block_offset[i];
1970  ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1971  s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1972  }
1973  }
1974  }
1975 }
1976 
1977 #define BITS 8
1978 #define SIMPLE 1
1979 #include "h264_mb_template.c"
1980 
1981 #undef BITS
1982 #define BITS 16
1983 #include "h264_mb_template.c"
1984 
1985 #undef SIMPLE
1986 #define SIMPLE 0
1987 #include "h264_mb_template.c"
1988 
1990 {
1991  MpegEncContext *const s = &h->s;
1992  const int mb_xy = h->mb_xy;
1993  const int mb_type = s->current_picture.f.mb_type[mb_xy];
1994  int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1995 
1996  if (CHROMA444) {
1997  if (is_complex || h->pixel_shift)
1998  hl_decode_mb_444_complex(h);
1999  else
2000  hl_decode_mb_444_simple_8(h);
2001  } else if (is_complex) {
2002  hl_decode_mb_complex(h);
2003  } else if (h->pixel_shift) {
2004  hl_decode_mb_simple_16(h);
2005  } else
2006  hl_decode_mb_simple_8(h);
2007 }
2008 
2010 {
2011  MpegEncContext *const s = &h->s;
2012  int list, i;
2013  int luma_def, chroma_def;
2014 
2015  h->use_weight = 0;
2016  h->use_weight_chroma = 0;
2018  if (h->sps.chroma_format_idc)
2020  luma_def = 1 << h->luma_log2_weight_denom;
2021  chroma_def = 1 << h->chroma_log2_weight_denom;
2022 
2023  for (list = 0; list < 2; list++) {
2024  h->luma_weight_flag[list] = 0;
2025  h->chroma_weight_flag[list] = 0;
2026  for (i = 0; i < h->ref_count[list]; i++) {
2027  int luma_weight_flag, chroma_weight_flag;
2028 
2029  luma_weight_flag = get_bits1(&s->gb);
2030  if (luma_weight_flag) {
2031  h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
2032  h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
2033  if (h->luma_weight[i][list][0] != luma_def ||
2034  h->luma_weight[i][list][1] != 0) {
2035  h->use_weight = 1;
2036  h->luma_weight_flag[list] = 1;
2037  }
2038  } else {
2039  h->luma_weight[i][list][0] = luma_def;
2040  h->luma_weight[i][list][1] = 0;
2041  }
2042 
2043  if (h->sps.chroma_format_idc) {
2044  chroma_weight_flag = get_bits1(&s->gb);
2045  if (chroma_weight_flag) {
2046  int j;
2047  for (j = 0; j < 2; j++) {
2048  h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
2049  h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
2050  if (h->chroma_weight[i][list][j][0] != chroma_def ||
2051  h->chroma_weight[i][list][j][1] != 0) {
2052  h->use_weight_chroma = 1;
2053  h->chroma_weight_flag[list] = 1;
2054  }
2055  }
2056  } else {
2057  int j;
2058  for (j = 0; j < 2; j++) {
2059  h->chroma_weight[i][list][j][0] = chroma_def;
2060  h->chroma_weight[i][list][j][1] = 0;
2061  }
2062  }
2063  }
2064  }
2066  break;
2067  }
2068  h->use_weight = h->use_weight || h->use_weight_chroma;
2069  return 0;
2070 }
2071 
2072 /**
2073  * Initialize implicit_weight table.
2074  * @param field 0/1 initialize the weight for interlaced MBAFF
2075  * -1 initializes the rest
2076  */
2077 static void implicit_weight_table(H264Context *h, int field)
2078 {
2079  MpegEncContext *const s = &h->s;
2080  int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2081 
2082  for (i = 0; i < 2; i++) {
2083  h->luma_weight_flag[i] = 0;
2084  h->chroma_weight_flag[i] = 0;
2085  }
2086 
2087  if (field < 0) {
2088  if (s->picture_structure == PICT_FRAME) {
2089  cur_poc = s->current_picture_ptr->poc;
2090  } else {
2091  cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2092  }
2093  if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2094  h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2095  h->use_weight = 0;
2096  h->use_weight_chroma = 0;
2097  return;
2098  }
2099  ref_start = 0;
2100  ref_count0 = h->ref_count[0];
2101  ref_count1 = h->ref_count[1];
2102  } else {
2103  cur_poc = s->current_picture_ptr->field_poc[field];
2104  ref_start = 16;
2105  ref_count0 = 16 + 2 * h->ref_count[0];
2106  ref_count1 = 16 + 2 * h->ref_count[1];
2107  }
2108 
2109  h->use_weight = 2;
2110  h->use_weight_chroma = 2;
2111  h->luma_log2_weight_denom = 5;
2112  h->chroma_log2_weight_denom = 5;
2113 
2114  for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2115  int poc0 = h->ref_list[0][ref0].poc;
2116  for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2117  int w = 32;
2118  if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2119  int poc1 = h->ref_list[1][ref1].poc;
2120  int td = av_clip(poc1 - poc0, -128, 127);
2121  if (td) {
2122  int tb = av_clip(cur_poc - poc0, -128, 127);
2123  int tx = (16384 + (FFABS(td) >> 1)) / td;
2124  int dist_scale_factor = (tb * tx + 32) >> 8;
2125  if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2126  w = 64 - dist_scale_factor;
2127  }
2128  }
2129  if (field < 0) {
2130  h->implicit_weight[ref0][ref1][0] =
2131  h->implicit_weight[ref0][ref1][1] = w;
2132  } else {
2133  h->implicit_weight[ref0][ref1][field] = w;
2134  }
2135  }
2136  }
2137 }
2138 
2139 /**
2140  * instantaneous decoder refresh.
2141  */
2142 static void idr(H264Context *h)
2143 {
2144  int i;
2146  h->prev_frame_num = 0;
2147  h->prev_frame_num_offset = 0;
2148  h->prev_poc_msb = 1<<16;
2149  h->prev_poc_lsb = 0;
2150  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2151  h->last_pocs[i] = INT_MIN;
2152 }
2153 
2154 /* forget old pics after a seek */
2155 static void flush_change(H264Context *h)
2156 {
2157  h->outputed_poc = h->next_outputed_poc = INT_MIN;
2158  h->prev_interlaced_frame = 1;
2159  idr(h);
2160  h->prev_frame_num = -1;
2161  if (h->s.current_picture_ptr)
2162  h->s.current_picture_ptr->f.reference = 0;
2163  h->s.first_field = 0;
2164  memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2165  memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2166  memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2167  memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2168  ff_h264_reset_sei(h);
2169  h->recovery_frame= -1;
2170  h->sync= 0;
2171  h->list_count = 0;
2172  h->current_slice = 0;
2173 }
2174 
2175 /* forget old pics after a seek */
2176 static void flush_dpb(AVCodecContext *avctx)
2177 {
2178  H264Context *h = avctx->priv_data;
2179  int i;
2180 
2181  for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2182  if (h->delayed_pic[i])
2183  h->delayed_pic[i]->f.reference = 0;
2184  h->delayed_pic[i] = NULL;
2185  }
2186 
2187  flush_change(h);
2188  ff_mpeg_flush(avctx);
2189 }
2190 
2191 static int init_poc(H264Context *h)
2192 {
2193  MpegEncContext *const s = &h->s;
2194  const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2195  int field_poc[2];
2196  Picture *cur = s->current_picture_ptr;
2197 
2199  if (h->frame_num < h->prev_frame_num)
2200  h->frame_num_offset += max_frame_num;
2201 
2202  if (h->sps.poc_type == 0) {
2203  const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2204 
2205  if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2206  h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2207  else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2208  h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2209  else
2210  h->poc_msb = h->prev_poc_msb;
2211  field_poc[0] =
2212  field_poc[1] = h->poc_msb + h->poc_lsb;
2213  if (s->picture_structure == PICT_FRAME)
2214  field_poc[1] += h->delta_poc_bottom;
2215  } else if (h->sps.poc_type == 1) {
2216  int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2217  int i;
2218 
2219  if (h->sps.poc_cycle_length != 0)
2220  abs_frame_num = h->frame_num_offset + h->frame_num;
2221  else
2222  abs_frame_num = 0;
2223 
2224  if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2225  abs_frame_num--;
2226 
2227  expected_delta_per_poc_cycle = 0;
2228  for (i = 0; i < h->sps.poc_cycle_length; i++)
2229  // FIXME integrate during sps parse
2230  expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2231 
2232  if (abs_frame_num > 0) {
2233  int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2234  int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2235 
2236  expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2237  for (i = 0; i <= frame_num_in_poc_cycle; i++)
2238  expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2239  } else
2240  expectedpoc = 0;
2241 
2242  if (h->nal_ref_idc == 0)
2243  expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2244 
2245  field_poc[0] = expectedpoc + h->delta_poc[0];
2246  field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2247 
2248  if (s->picture_structure == PICT_FRAME)
2249  field_poc[1] += h->delta_poc[1];
2250  } else {
2251  int poc = 2 * (h->frame_num_offset + h->frame_num);
2252 
2253  if (!h->nal_ref_idc)
2254  poc--;
2255 
2256  field_poc[0] = poc;
2257  field_poc[1] = poc;
2258  }
2259 
2261  s->current_picture_ptr->field_poc[0] = field_poc[0];
2263  s->current_picture_ptr->field_poc[1] = field_poc[1];
2264  cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2265 
2266  return 0;
2267 }
2268 
2269 /**
2270  * initialize scan tables
2271  */
2273 {
2274  int i;
2275  for (i = 0; i < 16; i++) {
2276 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2277  h->zigzag_scan[i] = T(zigzag_scan[i]);
2278  h->field_scan[i] = T(field_scan[i]);
2279 #undef T
2280  }
2281  for (i = 0; i < 64; i++) {
2282 #define T(x) (x >> 3) | ((x & 7) << 3)
2283  h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2285  h->field_scan8x8[i] = T(field_scan8x8[i]);
2287 #undef T
2288  }
2289  if (h->sps.transform_bypass) { // FIXME same ugly
2290  memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2291  memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
2293  memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
2294  memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2296  } else {
2297  memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
2298  memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
2300  memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
2301  memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
2303  }
2304 }
2305 
2306 static int field_end(H264Context *h, int in_setup)
2307 {
2308  MpegEncContext *const s = &h->s;
2309  AVCodecContext *const avctx = s->avctx;
2310  int err = 0;
2311  s->mb_y = 0;
2312 
2313  if (!in_setup && !s->droppable)
2316 
2317  if (CONFIG_H264_VDPAU_DECODER &&
2320 
2321  if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2322  if (!s->droppable) {
2324  h->prev_poc_msb = h->poc_msb;
2325  h->prev_poc_lsb = h->poc_lsb;
2326  }
2328  h->prev_frame_num = h->frame_num;
2330  }
2331 
2332  if (avctx->hwaccel) {
2333  if (avctx->hwaccel->end_frame(avctx) < 0)
2334  av_log(avctx, AV_LOG_ERROR,
2335  "hardware accelerator failed to decode picture\n");
2336  }
2337 
2338  if (CONFIG_H264_VDPAU_DECODER &&
2341 
2342  /*
2343  * FIXME: Error handling code does not seem to support interlaced
2344  * when slices span multiple rows
2345  * The ff_er_add_slice calls don't work right for bottom
2346  * fields; they cause massive erroneous error concealing
2347  * Error marking covers both fields (top and bottom).
2348  * This causes a mismatched s->error_count
2349  * and a bad error table. Further, the error count goes to
2350  * INT_MAX when called for bottom field, because mb_y is
2351  * past end by one (callers fault) and resync_mb_y != 0
2352  * causes problems for the first MB line, too.
2353  */
2354  if (!FIELD_PICTURE && h->current_slice && !h->sps.new)
2355  ff_er_frame_end(s);
2356 
2357  ff_MPV_frame_end(s);
2358 
2359  h->current_slice = 0;
2360 
2361  return err;
2362 }
2363 
2364 /**
2365  * Replicate H264 "master" context to thread contexts.
2366  */
2368 {
2369  int ret;
2370 
2371  memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2373  dst->s.current_picture = src->s.current_picture;
2374  dst->s.linesize = src->s.linesize;
2375  dst->s.uvlinesize = src->s.uvlinesize;
2376  dst->s.first_field = src->s.first_field;
2377 
2378  if (!dst->s.edge_emu_buffer &&
2379  (ret = ff_mpv_frame_size_alloc(&dst->s, dst->s.linesize))) {
2380  av_log(dst->s.avctx, AV_LOG_ERROR,
2381  "Failed to allocate scratch buffers\n");
2382  return ret;
2383  }
2384 
2385  dst->prev_poc_msb = src->prev_poc_msb;
2386  dst->prev_poc_lsb = src->prev_poc_lsb;
2388  dst->prev_frame_num = src->prev_frame_num;
2389  dst->short_ref_count = src->short_ref_count;
2390 
2391  memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2392  memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2393  memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2394  memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2395 
2396  memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2397  memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2398 
2399  return 0;
2400 }
2401 
2402 /**
2403  * Compute profile from profile_idc and constraint_set?_flags.
2404  *
2405  * @param sps SPS
2406  *
2407  * @return profile as defined by FF_PROFILE_H264_*
2408  */
2410 {
2411  int profile = sps->profile_idc;
2412 
2413  switch (sps->profile_idc) {
2415  // constraint_set1_flag set to 1
2416  profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2417  break;
2421  // constraint_set3_flag set to 1
2422  profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2423  break;
2424  }
2425 
2426  return profile;
2427 }
2428 
2430 {
2431  MpegEncContext *s = &h->s;
2432 
2433  if (s->flags & CODEC_FLAG_LOW_DELAY ||
2435  !h->sps.num_reorder_frames)) {
2436  if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
2437  av_log(h->s.avctx, AV_LOG_WARNING, "Delayed frames seen. "
2438  "Reenabling low delay requires a codec flush.\n");
2439  else
2440  s->low_delay = 1;
2441  }
2442 
2443  if (s->avctx->has_b_frames < 2)
2444  s->avctx->has_b_frames = !s->low_delay;
2445 
2446  if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
2448  "Different bit depth between chroma and luma", 1);
2449  return AVERROR_PATCHWELCOME;
2450  }
2451 
2452  if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2454  if (s->avctx->codec &&
2456  (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2458  "VDPAU decoding does not support video colorspace.\n");
2459  return AVERROR_INVALIDDATA;
2460  }
2461  if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2462  h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2463  (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2466  h->pixel_shift = h->sps.bit_depth_luma > 8;
2467 
2469  h->sps.chroma_format_idc);
2471  h->sps.chroma_format_idc);
2472  s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2473  ff_dsputil_init(&s->dsp, s->avctx);
2475  } else {
2476  av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2477  h->sps.bit_depth_luma);
2478  return AVERROR_INVALIDDATA;
2479  }
2480  }
2481  return 0;
2482 }
2483 
2484 static enum PixelFormat get_pixel_format(H264Context *h, int force_callback)
2485 {
2486  MpegEncContext *const s = &h->s;
2487  switch (h->sps.bit_depth_luma) {
2488  case 9:
2489  if (CHROMA444) {
2490  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2491  return AV_PIX_FMT_GBRP9;
2492  } else
2493  return AV_PIX_FMT_YUV444P9;
2494  } else if (CHROMA422)
2495  return AV_PIX_FMT_YUV422P9;
2496  else
2497  return AV_PIX_FMT_YUV420P9;
2498  break;
2499  case 10:
2500  if (CHROMA444) {
2501  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2502  return AV_PIX_FMT_GBRP10;
2503  } else
2504  return AV_PIX_FMT_YUV444P10;
2505  } else if (CHROMA422)
2506  return AV_PIX_FMT_YUV422P10;
2507  else
2508  return AV_PIX_FMT_YUV420P10;
2509  break;
2510  case 12:
2511  if (CHROMA444) {
2512  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2513  return AV_PIX_FMT_GBRP12;
2514  } else
2515  return AV_PIX_FMT_YUV444P12;
2516  } else if (CHROMA422)
2517  return AV_PIX_FMT_YUV422P12;
2518  else
2519  return AV_PIX_FMT_YUV420P12;
2520  break;
2521  case 14:
2522  if (CHROMA444) {
2523  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2524  return AV_PIX_FMT_GBRP14;
2525  } else
2526  return AV_PIX_FMT_YUV444P14;
2527  } else if (CHROMA422)
2528  return AV_PIX_FMT_YUV422P14;
2529  else
2530  return AV_PIX_FMT_YUV420P14;
2531  break;
2532  case 8:
2533  if (CHROMA444) {
2534  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2535  av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2536  return AV_PIX_FMT_GBR24P;
2537  } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2538  av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2539  }
2542  } else if (CHROMA422) {
2545  } else {
2546  int i;
2547  const enum AVPixelFormat * fmt = s->avctx->codec->pix_fmts ?
2548  s->avctx->codec->pix_fmts :
2552 
2553  for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
2554  if (fmt[i] == s->avctx->pix_fmt && !force_callback)
2555  return fmt[i];
2556  return s->avctx->get_format(s->avctx, fmt);
2557  }
2558  break;
2559  default:
2561  "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2562  return AVERROR_INVALIDDATA;
2563  }
2564 }
2565 
2567 {
2568  MpegEncContext *const s = &h->s;
2569  int i, ret;
2570 
2571  if( FFALIGN(s->avctx->width , 16 ) == s->width
2572  && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2573  && !h->sps.crop_right && !h->sps.crop_bottom
2574  && (s->avctx->width != s->width || s->avctx->height && s->height)
2575  ) {
2576  av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2577  s->avctx->coded_width = s->width;
2578  s->avctx->coded_height = s->height;
2579  } else{
2581  s->avctx->width -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2582  s->avctx->height -= (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1) * (2 - h->sps.frame_mbs_only_flag);
2583  }
2584 
2585  s->avctx->sample_aspect_ratio = h->sps.sar;
2587 
2588  if (h->sps.timing_info_present_flag) {
2589  int64_t den = h->sps.time_scale;
2590  if (h->x264_build < 44U)
2591  den *= 2;
2593  h->sps.num_units_in_tick, den, 1 << 30);
2594  }
2595 
2597 
2598  if (reinit) {
2599  free_tables(h, 0);
2600  if ((ret = ff_MPV_common_frame_size_change(s)) < 0) {
2601  av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_frame_size_change() failed.\n");
2602  return ret;
2603  }
2604  } else {
2605  if ((ret = ff_MPV_common_init(s)) < 0) {
2606  av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2607  return ret;
2608  }
2609  }
2610  s->first_field = 0;
2611  h->prev_interlaced_frame = 1;
2612 
2613  init_scan_tables(h);
2614  if (ff_h264_alloc_tables(h) < 0) {
2615  av_log(h->s.avctx, AV_LOG_ERROR,
2616  "Could not allocate memory for h264\n");
2617  return AVERROR(ENOMEM);
2618  }
2619 
2620  if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2621  if (context_init(h) < 0) {
2622  av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2623  return -1;
2624  }
2625  } else {
2626  for (i = 1; i < s->slice_context_count; i++) {
2627  H264Context *c;
2628  c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2629  memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2630  memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2631  c->h264dsp = h->h264dsp;
2632  c->sps = h->sps;
2633  c->pps = h->pps;
2634  c->pixel_shift = h->pixel_shift;
2636  init_scan_tables(c);
2637  clone_tables(c, h, i);
2638  }
2639 
2640  for (i = 0; i < s->slice_context_count; i++)
2641  if (context_init(h->thread_context[i]) < 0) {
2642  av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2643  return -1;
2644  }
2645  }
2646 
2647  return 0;
2648 }
2649 
2650 /**
2651  * Decode a slice header.
2652  * This will also call ff_MPV_common_init() and frame_start() as needed.
2653  *
2654  * @param h h264context
2655  * @param h0 h264 master context (differs from 'h' when doing sliced based
2656  * parallel decoding)
2657  *
2658  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2659  */
2661 {
2662  MpegEncContext *const s = &h->s;
2663  MpegEncContext *const s0 = &h0->s;
2664  unsigned int first_mb_in_slice;
2665  unsigned int pps_id;
2666  int num_ref_idx_active_override_flag, ret;
2667  unsigned int slice_type, tmp, i, j;
2668  int default_ref_list_done = 0;
2669  int last_pic_structure, last_pic_droppable;
2670  int must_reinit;
2671  int needs_reinit = 0;
2672 
2673  /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2674  if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2675  !h->nal_ref_idc && !h->pixel_shift) {
2678  } else {
2681  }
2682 
2683  first_mb_in_slice = get_ue_golomb_long(&s->gb);
2684 
2685  if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2686  if (h0->current_slice && FIELD_PICTURE) {
2687  field_end(h, 1);
2688  }
2689 
2690  h0->current_slice = 0;
2691  if (!s0->first_field) {
2692  if (s->current_picture_ptr && !s->droppable &&
2693  s->current_picture_ptr->owner2 == s) {
2696  }
2698  }
2699  }
2700 
2701  slice_type = get_ue_golomb_31(&s->gb);
2702  if (slice_type > 9) {
2703  av_log(h->s.avctx, AV_LOG_ERROR,
2704  "slice type too large (%d) at %d %d\n",
2705  slice_type, s->mb_x, s->mb_y);
2706  return -1;
2707  }
2708  if (slice_type > 4) {
2709  slice_type -= 5;
2710  h->slice_type_fixed = 1;
2711  } else
2712  h->slice_type_fixed = 0;
2713 
2714  slice_type = golomb_to_pict_type[slice_type];
2715  if (slice_type == AV_PICTURE_TYPE_I ||
2716  (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2717  default_ref_list_done = 1;
2718  }
2719  h->slice_type = slice_type;
2720  h->slice_type_nos = slice_type & 3;
2721 
2722  // to make a few old functions happy, it's wrong though
2723  s->pict_type = h->slice_type;
2724 
2725  pps_id = get_ue_golomb(&s->gb);
2726  if (pps_id >= MAX_PPS_COUNT) {
2727  av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2728  return -1;
2729  }
2730  if (!h0->pps_buffers[pps_id]) {
2731  av_log(h->s.avctx, AV_LOG_ERROR,
2732  "non-existing PPS %u referenced\n",
2733  pps_id);
2734  return -1;
2735  }
2736  h->pps = *h0->pps_buffers[pps_id];
2737 
2738  if (!h0->sps_buffers[h->pps.sps_id]) {
2739  av_log(h->s.avctx, AV_LOG_ERROR,
2740  "non-existing SPS %u referenced\n",
2741  h->pps.sps_id);
2742  return -1;
2743  }
2744 
2745  if (h->pps.sps_id != h->current_sps_id ||
2746  h->context_reinitialized ||
2747  h0->sps_buffers[h->pps.sps_id]->new) {
2748  SPS *new_sps = h0->sps_buffers[h->pps.sps_id];
2749 
2750  h0->sps_buffers[h->pps.sps_id]->new = 0;
2751 
2752  if (h->sps.chroma_format_idc != new_sps->chroma_format_idc ||
2753  h->sps.bit_depth_luma != new_sps->bit_depth_luma)
2754  needs_reinit = 1;
2755 
2756  h->current_sps_id = h->pps.sps_id;
2757  h->sps = *h0->sps_buffers[h->pps.sps_id];
2758 
2759  if (s->mb_width != h->sps.mb_width ||
2760  s->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
2763  )
2764  needs_reinit = 1;
2765 
2766  if ((ret = h264_set_parameter_from_sps(h)) < 0)
2767  return ret;
2768  }
2769 
2770  s->avctx->profile = ff_h264_get_profile(&h->sps);
2771  s->avctx->level = h->sps.level_idc;
2772  s->avctx->refs = h->sps.ref_frame_count;
2773 
2774  must_reinit = (s->context_initialized &&
2775  ( 16*h->sps.mb_width != s->avctx->coded_width
2776  || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2779  || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2780  if (h0->s.avctx->pix_fmt != get_pixel_format(h0, 0))
2781  must_reinit = 1;
2782 
2783  s->mb_width = h->sps.mb_width;
2784  s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2785 
2786  h->b_stride = s->mb_width * 4;
2787 
2788  s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2789 
2790  s->width = 16 * s->mb_width;
2791  s->height = 16 * s->mb_height;
2792 
2795  : AVCOL_RANGE_MPEG;
2797  if (s->avctx->colorspace != h->sps.colorspace)
2798  needs_reinit = 1;
2800  s->avctx->color_trc = h->sps.color_trc;
2801  s->avctx->colorspace = h->sps.colorspace;
2802  }
2803  }
2804 
2805  if (s->context_initialized &&
2806  (
2807  needs_reinit ||
2808  must_reinit)) {
2809 
2810  if (h != h0) {
2811  av_log(s->avctx, AV_LOG_ERROR, "changing width/height on "
2812  "slice %d\n", h0->current_slice + 1);
2813  return AVERROR_INVALIDDATA;
2814  }
2815 
2816  flush_change(h);
2817 
2818  if ((ret = get_pixel_format(h, 1)) < 0)
2819  return ret;
2820  s->avctx->pix_fmt = ret;
2821 
2822  av_log(h->s.avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
2823  "pix_fmt: %d\n", s->width, s->height, s->avctx->pix_fmt);
2824 
2825  if ((ret = h264_slice_header_init(h, 1)) < 0) {
2826  av_log(h->s.avctx, AV_LOG_ERROR,
2827  "h264_slice_header_init() failed\n");
2828  return ret;
2829  }
2830  h->context_reinitialized = 1;
2831  }
2832  if (!s->context_initialized) {
2833  if (h != h0) {
2834  av_log(h->s.avctx, AV_LOG_ERROR,
2835  "Cannot (re-)initialize context during parallel decoding.\n");
2836  return -1;
2837  }
2838 
2839  if ((ret = get_pixel_format(h, 1)) < 0)
2840  return ret;
2841  s->avctx->pix_fmt = ret;
2842 
2843  if ((ret = h264_slice_header_init(h, 0)) < 0) {
2844  av_log(h->s.avctx, AV_LOG_ERROR,
2845  "h264_slice_header_init() failed\n");
2846  return ret;
2847  }
2848  }
2849 
2850  if (h == h0 && h->dequant_coeff_pps != pps_id) {
2851  h->dequant_coeff_pps = pps_id;
2853  }
2854 
2855  h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2856 
2857  h->mb_mbaff = 0;
2858  h->mb_aff_frame = 0;
2859  last_pic_structure = s0->picture_structure;
2860  last_pic_droppable = s0->droppable;
2861  s->droppable = h->nal_ref_idc == 0;
2862  if (h->sps.frame_mbs_only_flag) {
2864  } else {
2865  if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2866  av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2867  return -1;
2868  }
2869  if (get_bits1(&s->gb)) { // field_pic_flag
2870  s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2871  } else {
2873  h->mb_aff_frame = h->sps.mb_aff;
2874  }
2875  }
2877 
2878  if (h0->current_slice != 0) {
2879  if (last_pic_structure != s->picture_structure ||
2880  last_pic_droppable != s->droppable) {
2881  av_log(h->s.avctx, AV_LOG_ERROR,
2882  "Changing field mode (%d -> %d) between slices is not allowed\n",
2883  last_pic_structure, s->picture_structure);
2884  s->picture_structure = last_pic_structure;
2885  s->droppable = last_pic_droppable;
2886  return AVERROR_INVALIDDATA;
2887  } else if (!s0->current_picture_ptr) {
2889  "unset current_picture_ptr on %d. slice\n",
2890  h0->current_slice + 1);
2891  return AVERROR_INVALIDDATA;
2892  }
2893  } else {
2894  /* Shorten frame num gaps so we don't have to allocate reference
2895  * frames just to throw them away */
2896  if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2897  int unwrap_prev_frame_num = h->prev_frame_num;
2898  int max_frame_num = 1 << h->sps.log2_max_frame_num;
2899 
2900  if (unwrap_prev_frame_num > h->frame_num)
2901  unwrap_prev_frame_num -= max_frame_num;
2902 
2903  if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2904  unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2905  if (unwrap_prev_frame_num < 0)
2906  unwrap_prev_frame_num += max_frame_num;
2907 
2908  h->prev_frame_num = unwrap_prev_frame_num;
2909  }
2910  }
2911 
2912  /* See if we have a decoded first field looking for a pair...
2913  * Here, we're using that to see if we should mark previously
2914  * decode frames as "finished".
2915  * We have to do that before the "dummy" in-between frame allocation,
2916  * since that can modify s->current_picture_ptr. */
2917  if (s0->first_field) {
2918  assert(s0->current_picture_ptr);
2919  assert(s0->current_picture_ptr->f.data[0]);
2921 
2922  /* Mark old field/frame as completed */
2923  if (!last_pic_droppable && s0->current_picture_ptr->owner2 == s0) {
2925  last_pic_structure == PICT_BOTTOM_FIELD);
2926  }
2927 
2928  /* figure out if we have a complementary field pair */
2929  if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2930  /* Previous field is unmatched. Don't display it, but let it
2931  * remain for reference if marked as such. */
2932  if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2934  last_pic_structure == PICT_TOP_FIELD);
2935  }
2936  } else {
2937  if (s0->current_picture_ptr->frame_num != h->frame_num) {
2938  /* This and previous field were reference, but had
2939  * different frame_nums. Consider this field first in
2940  * pair. Throw away previous field except for reference
2941  * purposes. */
2942  if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2944  last_pic_structure == PICT_TOP_FIELD);
2945  }
2946  } else {
2947  /* Second field in complementary pair */
2948  if (!((last_pic_structure == PICT_TOP_FIELD &&
2950  (last_pic_structure == PICT_BOTTOM_FIELD &&
2953  "Invalid field mode combination %d/%d\n",
2954  last_pic_structure, s->picture_structure);
2955  s->picture_structure = last_pic_structure;
2956  s->droppable = last_pic_droppable;
2957  return AVERROR_INVALIDDATA;
2958  } else if (last_pic_droppable != s->droppable) {
2960  "Cannot combine reference and non-reference fields in the same frame\n");
2962  s->picture_structure = last_pic_structure;
2963  s->droppable = last_pic_droppable;
2964  return AVERROR_PATCHWELCOME;
2965  }
2966 
2967  /* Take ownership of this buffer. Note that if another thread owned
2968  * the first field of this buffer, we're not operating on that pointer,
2969  * so the original thread is still responsible for reporting progress
2970  * on that first field (or if that was us, we just did that above).
2971  * By taking ownership, we assign responsibility to ourselves to
2972  * report progress on the second field. */
2973  s0->current_picture_ptr->owner2 = s0;
2974  }
2975  }
2976  }
2977 
2978  while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !s0->first_field &&
2979  h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2980  Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2981  av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2982  h->frame_num, h->prev_frame_num);
2984  for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
2985  h->last_pocs[i] = INT_MIN;
2986  if (ff_h264_frame_start(h) < 0)
2987  return -1;
2988  h->prev_frame_num++;
2989  h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2993  if ((ret = ff_generate_sliding_window_mmcos(h, 1)) < 0 &&
2995  return ret;
2996  if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2998  return AVERROR_INVALIDDATA;
2999  /* Error concealment: if a ref is missing, copy the previous ref in its place.
3000  * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
3001  * about there being no actual duplicates.
3002  * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
3003  * concealing a lost frame, this probably isn't noticeable by comparison, but it should
3004  * be fixed. */
3005  if (h->short_ref_count) {
3006  if (prev) {
3007  av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
3008  (const uint8_t **)prev->f.data, prev->f.linesize,
3009  s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
3010  h->short_ref[0]->poc = prev->poc + 2;
3011  }
3012  h->short_ref[0]->frame_num = h->prev_frame_num;
3013  }
3014  }
3015 
3016  /* See if we have a decoded first field looking for a pair...
3017  * We're using that to see whether to continue decoding in that
3018  * frame, or to allocate a new one. */
3019  if (s0->first_field) {
3020  assert(s0->current_picture_ptr);
3021  assert(s0->current_picture_ptr->f.data[0]);
3023 
3024  /* figure out if we have a complementary field pair */
3025  if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3026  /* Previous field is unmatched. Don't display it, but let it
3027  * remain for reference if marked as such. */
3028  s0->current_picture_ptr = NULL;
3029  s0->first_field = FIELD_PICTURE;
3030  } else {
3031  if (s0->current_picture_ptr->frame_num != h->frame_num) {
3034  /* This and the previous field had different frame_nums.
3035  * Consider this field first in pair. Throw away previous
3036  * one except for reference purposes. */
3037  s0->first_field = 1;
3038  s0->current_picture_ptr = NULL;
3039  } else {
3040  /* Second field in complementary pair */
3041  s0->first_field = 0;
3042  }
3043  }
3044  } else {
3045  /* Frame or first field in a potentially complementary pair */
3046  s0->first_field = FIELD_PICTURE;
3047  }
3048 
3049  if (!FIELD_PICTURE || s0->first_field) {
3050  if (ff_h264_frame_start(h) < 0) {
3051  s0->first_field = 0;
3052  return -1;
3053  }
3054  } else {
3056  }
3057  }
3058  if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3059  return ret;
3060 
3061  s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3062 
3063  av_assert1(s->mb_num == s->mb_width * s->mb_height);
3064  if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3065  first_mb_in_slice >= s->mb_num) {
3066  av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3067  return -1;
3068  }
3069  s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3070  s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3072  s->resync_mb_y = s->mb_y = s->mb_y + 1;
3073  av_assert1(s->mb_y < s->mb_height);
3074 
3075  if (s->picture_structure == PICT_FRAME) {
3076  h->curr_pic_num = h->frame_num;
3077  h->max_pic_num = 1 << h->sps.log2_max_frame_num;
3078  } else {
3079  h->curr_pic_num = 2 * h->frame_num + 1;
3080  h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
3081  }
3082 
3083  if (h->nal_unit_type == NAL_IDR_SLICE)
3084  get_ue_golomb(&s->gb); /* idr_pic_id */
3085 
3086  if (h->sps.poc_type == 0) {
3087  h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3088 
3089  if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3090  h->delta_poc_bottom = get_se_golomb(&s->gb);
3091  }
3092 
3093  if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3094  h->delta_poc[0] = get_se_golomb(&s->gb);
3095 
3096  if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3097  h->delta_poc[1] = get_se_golomb(&s->gb);
3098  }
3099 
3100  init_poc(h);
3101 
3104 
3105  // set defaults, might be overridden a few lines later
3106  h->ref_count[0] = h->pps.ref_count[0];
3107  h->ref_count[1] = h->pps.ref_count[1];
3108 
3109  if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3110  unsigned max[2];
3111  max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
3112 
3115  num_ref_idx_active_override_flag = get_bits1(&s->gb);
3116 
3117  if (num_ref_idx_active_override_flag) {
3118  h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
3119  if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3120  h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
3121  } else
3122  // full range is spec-ok in this case, even for frames
3123  h->ref_count[1] = 1;
3124  }
3125 
3126  if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3127  av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]);
3128  h->ref_count[0] = h->ref_count[1] = 0;
3129  return AVERROR_INVALIDDATA;
3130  }
3131 
3133  h->list_count = 2;
3134  else
3135  h->list_count = 1;
3136  } else {
3137  h->list_count = 0;
3138  h->ref_count[0] = h->ref_count[1] = 0;
3139  }
3140 
3141  if (!default_ref_list_done)
3143 
3144  if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
3146  h->ref_count[1] = h->ref_count[0] = 0;
3147  return -1;
3148  }
3149 
3150  if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3151  s->last_picture_ptr = &h->ref_list[0][0];
3152  s->last_picture_ptr->owner2 = s;
3154  }
3155  if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3156  s->next_picture_ptr = &h->ref_list[1][0];
3157  s->next_picture_ptr->owner2 = s;
3159  }
3160 
3161  if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3162  (h->pps.weighted_bipred_idc == 1 &&
3164  pred_weight_table(h);
3165  else if (h->pps.weighted_bipred_idc == 2 &&
3167  implicit_weight_table(h, -1);
3168  } else {
3169  h->use_weight = 0;
3170  for (i = 0; i < 2; i++) {
3171  h->luma_weight_flag[i] = 0;
3172  h->chroma_weight_flag[i] = 0;
3173  }
3174  }
3175 
3176  // If frame-mt is enabled, only update mmco tables for the first slice
3177  // in a field. Subsequent slices can temporarily clobber h->mmco_index
3178  // or h->mmco, which will cause ref list mix-ups and decoding errors
3179  // further down the line. This may break decoding if the first slice is
3180  // corrupt, thus we only do this if frame-mt is enabled.
3181  if (h->nal_ref_idc &&
3184  h0->current_slice == 0) < 0 &&
3186  return AVERROR_INVALIDDATA;
3187 
3188  if (FRAME_MBAFF) {
3190 
3192  implicit_weight_table(h, 0);
3193  implicit_weight_table(h, 1);
3194  }
3195  }
3196 
3200 
3201  if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3202  tmp = get_ue_golomb_31(&s->gb);
3203  if (tmp > 2) {
3204  av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3205  return -1;
3206  }
3207  h->cabac_init_idc = tmp;
3208  }
3209 
3210  h->last_qscale_diff = 0;
3211  tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3212  if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3213  av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3214  return -1;
3215  }
3216  s->qscale = tmp;
3217  h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3218  h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3219  // FIXME qscale / qp ... stuff
3220  if (h->slice_type == AV_PICTURE_TYPE_SP)
3221  get_bits1(&s->gb); /* sp_for_switch_flag */
3222  if (h->slice_type == AV_PICTURE_TYPE_SP ||
3224  get_se_golomb(&s->gb); /* slice_qs_delta */
3225 
3226  h->deblocking_filter = 1;
3227  h->slice_alpha_c0_offset = 52;
3228  h->slice_beta_offset = 52;
3230  tmp = get_ue_golomb_31(&s->gb);
3231  if (tmp > 2) {
3233  "deblocking_filter_idc %u out of range\n", tmp);
3234  return -1;
3235  }
3236  h->deblocking_filter = tmp;
3237  if (h->deblocking_filter < 2)
3238  h->deblocking_filter ^= 1; // 1<->0
3239 
3240  if (h->deblocking_filter) {
3241  h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3242  h->slice_beta_offset += get_se_golomb(&s->gb) << 1;
3243  if (h->slice_alpha_c0_offset > 104U ||
3244  h->slice_beta_offset > 104U) {
3246  "deblocking filter parameters %d %d out of range\n",
3248  return -1;
3249  }
3250  }
3251  }
3252 
3253  if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3259  h->nal_ref_idc == 0))
3260  h->deblocking_filter = 0;
3261 
3262  if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3263  if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3264  /* Cheat slightly for speed:
3265  * Do not bother to deblock across slices. */
3266  h->deblocking_filter = 2;
3267  } else {
3268  h0->max_contexts = 1;
3269  if (!h0->single_decode_warning) {
3270  av_log(s->avctx, AV_LOG_INFO,
3271  "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3272  h0->single_decode_warning = 1;
3273  }
3274  if (h != h0) {
3275  av_log(h->s.avctx, AV_LOG_ERROR,
3276  "Deblocking switched inside frame.\n");
3277  return 1;
3278  }
3279  }
3280  }
3281  h->qp_thresh = 15 + 52 -
3283  FFMAX3(0,
3285  h->pps.chroma_qp_index_offset[1]) +
3286  6 * (h->sps.bit_depth_luma - 8);
3287 
3288  h0->last_slice_type = slice_type;
3289  h->slice_num = ++h0->current_slice;
3290 
3291  if (h->slice_num)
3292  h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3293  if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3294  && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3295  && h->slice_num >= MAX_SLICES) {
3296  //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3297  av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3298  }
3299 
3300  for (j = 0; j < 2; j++) {
3301  int id_list[16];
3302  int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3303  for (i = 0; i < 16; i++) {
3304  id_list[i] = 60;
3305  if (h->ref_list[j][i].f.data[0]) {
3306  int k;
3307  uint8_t *base = h->ref_list[j][i].f.base[0];
3308  for (k = 0; k < h->short_ref_count; k++)
3309  if (h->short_ref[k]->f.base[0] == base) {
3310  id_list[i] = k;
3311  break;
3312  }
3313  for (k = 0; k < h->long_ref_count; k++)
3314  if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3315  id_list[i] = h->short_ref_count + k;
3316  break;
3317  }
3318  }
3319  }
3320 
3321  ref2frm[0] =
3322  ref2frm[1] = -1;
3323  for (i = 0; i < 16; i++)
3324  ref2frm[i + 2] = 4 * id_list[i] +
3325  (h->ref_list[j][i].f.reference & 3);
3326  ref2frm[18 + 0] =
3327  ref2frm[18 + 1] = -1;
3328  for (i = 16; i < 48; i++)
3329  ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3330  (h->ref_list[j][i].f.reference & 3);
3331  }
3332 
3333  // FIXME: fix draw_edges + PAFF + frame threads
3335  (!h->sps.frame_mbs_only_flag &&
3337  ? 0 : 16;
3339 
3340  if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3341  av_log(h->s.avctx, AV_LOG_DEBUG,
3342  "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3343  h->slice_num,
3344  (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3345  first_mb_in_slice,
3347  h->slice_type_fixed ? " fix" : "",
3348  h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3349  pps_id, h->frame_num,
3352  h->ref_count[0], h->ref_count[1],
3353  s->qscale,
3354  h->deblocking_filter,
3355  h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3356  h->use_weight,
3357  h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3358  h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3359  }
3360 
3361  return 0;
3362 }
3363 
3365 {
3366  switch (h->slice_type) {
3367  case AV_PICTURE_TYPE_P:
3368  return 0;
3369  case AV_PICTURE_TYPE_B:
3370  return 1;
3371  case AV_PICTURE_TYPE_I:
3372  return 2;
3373  case AV_PICTURE_TYPE_SP:
3374  return 3;
3375  case AV_PICTURE_TYPE_SI:
3376  return 4;
3377  default:
3378  return -1;
3379  }
3380 }
3381 
3383  MpegEncContext *const s,
3384  int mb_type, int top_xy,
3385  int left_xy[LEFT_MBS],
3386  int top_type,
3387  int left_type[LEFT_MBS],
3388  int mb_xy, int list)
3389 {
3390  int b_stride = h->b_stride;
3391  int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3392  int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3393  if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3394  if (USES_LIST(top_type, list)) {
3395  const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
3396  const int b8_xy = 4 * top_xy + 2;
3397  int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3398  AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3399  ref_cache[0 - 1 * 8] =
3400  ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3401  ref_cache[2 - 1 * 8] =
3402  ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3403  } else {
3404  AV_ZERO128(mv_dst - 1 * 8);
3405  AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3406  }
3407 
3408  if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3409  if (USES_LIST(left_type[LTOP], list)) {
3410  const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
3411  const int b8_xy = 4 * left_xy[LTOP] + 1;
3412  int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3413  AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3414  AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3415  AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3416  AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3417  ref_cache[-1 + 0] =
3418  ref_cache[-1 + 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3419  ref_cache[-1 + 16] =
3420  ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3421  } else {
3422  AV_ZERO32(mv_dst - 1 + 0);
3423  AV_ZERO32(mv_dst - 1 + 8);
3424  AV_ZERO32(mv_dst - 1 + 16);
3425  AV_ZERO32(mv_dst - 1 + 24);
3426  ref_cache[-1 + 0] =
3427  ref_cache[-1 + 8] =
3428  ref_cache[-1 + 16] =
3429  ref_cache[-1 + 24] = LIST_NOT_USED;
3430  }
3431  }
3432  }
3433 
3434  if (!USES_LIST(mb_type, list)) {
3435  fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3436  AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3437  AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3438  AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3439  AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3440  return;
3441  }
3442 
3443  {
3444  int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3445  int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3446  uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3447  uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3448  AV_WN32A(&ref_cache[0 * 8], ref01);
3449  AV_WN32A(&ref_cache[1 * 8], ref01);
3450  AV_WN32A(&ref_cache[2 * 8], ref23);
3451  AV_WN32A(&ref_cache[3 * 8], ref23);
3452  }
3453 
3454  {
3455  int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3456  AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3457  AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3458  AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3459  AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3460  }
3461 }
3462 
3463 /**
3464  *
3465  * @return non zero if the loop filter can be skipped
3466  */
3467 static int fill_filter_caches(H264Context *h, int mb_type)
3468 {
3469  MpegEncContext *const s = &h->s;
3470  const int mb_xy = h->mb_xy;
3471  int top_xy, left_xy[LEFT_MBS];
3472  int top_type, left_type[LEFT_MBS];
3473  uint8_t *nnz;
3474  uint8_t *nnz_cache;
3475 
3476  top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3477 
3478  /* Wow, what a mess, why didn't they simplify the interlacing & intra
3479  * stuff, I can't imagine that these complex rules are worth it. */
3480 
3481  left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3482  if (FRAME_MBAFF) {
3483  const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3484  const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3485  if (s->mb_y & 1) {
3486  if (left_mb_field_flag != curr_mb_field_flag)
3487  left_xy[LTOP] -= s->mb_stride;
3488  } else {
3489  if (curr_mb_field_flag)
3490  top_xy += s->mb_stride &
3491  (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3492  if (left_mb_field_flag != curr_mb_field_flag)
3493  left_xy[LBOT] += s->mb_stride;
3494  }
3495  }
3496 
3497  h->top_mb_xy = top_xy;
3498  h->left_mb_xy[LTOP] = left_xy[LTOP];
3499  h->left_mb_xy[LBOT] = left_xy[LBOT];
3500  {
3501  /* For sufficiently low qp, filtering wouldn't do anything.
3502  * This is a conservative estimate: could also check beta_offset
3503  * and more accurate chroma_qp. */
3504  int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3505  int qp = s->current_picture.f.qscale_table[mb_xy];
3506  if (qp <= qp_thresh &&
3507  (left_xy[LTOP] < 0 ||
3508  ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3509  (top_xy < 0 ||
3510  ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3511  if (!FRAME_MBAFF)
3512  return 1;
3513  if ((left_xy[LTOP] < 0 ||
3514  ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3515  (top_xy < s->mb_stride ||
3516  ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3517  return 1;
3518  }
3519  }
3520 
3521  top_type = s->current_picture.f.mb_type[top_xy];
3522  left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3523  left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3524  if (h->deblocking_filter == 2) {
3525  if (h->slice_table[top_xy] != h->slice_num)
3526  top_type = 0;
3527  if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3528  left_type[LTOP] = left_type[LBOT] = 0;
3529  } else {
3530  if (h->slice_table[top_xy] == 0xFFFF)
3531  top_type = 0;
3532  if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3533  left_type[LTOP] = left_type[LBOT] = 0;
3534  }
3535  h->top_type = top_type;
3536  h->left_type[LTOP] = left_type[LTOP];
3537  h->left_type[LBOT] = left_type[LBOT];
3538 
3539  if (IS_INTRA(mb_type))
3540  return 0;
3541 
3542  fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3543  top_type, left_type, mb_xy, 0);
3544  if (h->list_count == 2)
3545  fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3546  top_type, left_type, mb_xy, 1);
3547 
3548  nnz = h->non_zero_count[mb_xy];
3549  nnz_cache = h->non_zero_count_cache;
3550  AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3551  AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3552  AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3553  AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3554  h->cbp = h->cbp_table[mb_xy];
3555 
3556  if (top_type) {
3557  nnz = h->non_zero_count[top_xy];
3558  AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3559  }
3560 
3561  if (left_type[LTOP]) {
3562  nnz = h->non_zero_count[left_xy[LTOP]];
3563  nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3564  nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3565  nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3566  nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3567  }
3568 
3569  /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3570  * from what the loop filter needs */
3571  if (!CABAC && h->pps.transform_8x8_mode) {
3572  if (IS_8x8DCT(top_type)) {
3573  nnz_cache[4 + 8 * 0] =
3574  nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3575  nnz_cache[6 + 8 * 0] =
3576  nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3577  }
3578  if (IS_8x8DCT(left_type[LTOP])) {
3579  nnz_cache[3 + 8 * 1] =
3580  nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3581  }
3582  if (IS_8x8DCT(left_type[LBOT])) {
3583  nnz_cache[3 + 8 * 3] =
3584  nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3585  }
3586 
3587  if (IS_8x8DCT(mb_type)) {
3588  nnz_cache[scan8[0]] =
3589  nnz_cache[scan8[1]] =
3590  nnz_cache[scan8[2]] =
3591  nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3592 
3593  nnz_cache[scan8[0 + 4]] =
3594  nnz_cache[scan8[1 + 4]] =
3595  nnz_cache[scan8[2 + 4]] =
3596  nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3597 
3598  nnz_cache[scan8[0 + 8]] =
3599  nnz_cache[scan8[1 + 8]] =
3600  nnz_cache[scan8[2 + 8]] =
3601  nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3602 
3603  nnz_cache[scan8[0 + 12]] =
3604  nnz_cache[scan8[1 + 12]] =
3605  nnz_cache[scan8[2 + 12]] =
3606  nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3607  }
3608  }
3609 
3610  return 0;
3611 }
3612 
3613 static void loop_filter(H264Context *h, int start_x, int end_x)
3614 {
3615  MpegEncContext *const s = &h->s;
3616  uint8_t *dest_y, *dest_cb, *dest_cr;
3617  int linesize, uvlinesize, mb_x, mb_y;
3618  const int end_mb_y = s->mb_y + FRAME_MBAFF;
3619  const int old_slice_type = h->slice_type;
3620  const int pixel_shift = h->pixel_shift;
3621  const int block_h = 16 >> s->chroma_y_shift;
3622 
3623  if (h->deblocking_filter) {
3624  for (mb_x = start_x; mb_x < end_x; mb_x++)
3625  for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3626  int mb_xy, mb_type;
3627  mb_xy = h->mb_xy = mb_x + mb_y * s->mb_stride;
3628  h->slice_num = h->slice_table[mb_xy];
3629  mb_type = s->current_picture.f.mb_type[mb_xy];
3630  h->list_count = h->list_counts[mb_xy];
3631 
3632  if (FRAME_MBAFF)
3633  h->mb_mbaff =
3634  h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3635 
3636  s->mb_x = mb_x;
3637  s->mb_y = mb_y;
3638  dest_y = s->current_picture.f.data[0] +
3639  ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3640  dest_cb = s->current_picture.f.data[1] +
3641  (mb_x << pixel_shift) * (8 << CHROMA444) +
3642  mb_y * s->uvlinesize * block_h;
3643  dest_cr = s->current_picture.f.data[2] +
3644  (mb_x << pixel_shift) * (8 << CHROMA444) +
3645  mb_y * s->uvlinesize * block_h;
3646  // FIXME simplify above
3647 
3648  if (MB_FIELD) {
3649  linesize = h->mb_linesize = s->linesize * 2;
3650  uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3651  if (mb_y & 1) { // FIXME move out of this function?
3652  dest_y -= s->linesize * 15;
3653  dest_cb -= s->uvlinesize * (block_h - 1);
3654  dest_cr -= s->uvlinesize * (block_h - 1);
3655  }
3656  } else {
3657  linesize = h->mb_linesize = s->linesize;
3658  uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3659  }
3660  backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3661  uvlinesize, 0);
3662  if (fill_filter_caches(h, mb_type))
3663  continue;
3664  h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3665  h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3666 
3667  if (FRAME_MBAFF) {
3668  ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3669  linesize, uvlinesize);
3670  } else {
3671  ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3672  dest_cr, linesize, uvlinesize);
3673  }
3674  }
3675  }
3676  h->slice_type = old_slice_type;
3677  s->mb_x = end_x;
3678  s->mb_y = end_mb_y - FRAME_MBAFF;
3679  h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3680  h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3681 }
3682 
3684 {
3685  MpegEncContext *const s = &h->s;
3686  const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3687  int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3688  s->current_picture.f.mb_type[mb_xy - 1] :
3689  (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3690  s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3691  h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3692 }
3693 
3694 /**
3695  * Draw edges and report progress for the last MB row.
3696  */
3698 {
3699  MpegEncContext *const s = &h->s;
3700  int top = 16 * (s->mb_y >> FIELD_PICTURE);
3701  int pic_height = 16 * s->mb_height >> FIELD_PICTURE;
3702  int height = 16 << FRAME_MBAFF;
3703  int deblock_border = (16 + 4) << FRAME_MBAFF;
3704 
3705  if (h->deblocking_filter) {
3706  if ((top + height) >= pic_height)
3707  height += deblock_border;
3708  top -= deblock_border;
3709  }
3710 
3711  if (top >= pic_height || (top + height) < h->emu_edge_height)
3712  return;
3713 
3714  height = FFMIN(height, pic_height - top);
3715  if (top < h->emu_edge_height) {
3716  height = top + height;
3717  top = 0;
3718  }
3719 
3720  ff_draw_horiz_band(s, top, height);
3721 
3722  if (s->droppable)
3723  return;
3724 
3725  ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3727 }
3728 
3729 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3730 {
3731  H264Context *h = *(void **)arg;
3732  MpegEncContext *const s = &h->s;
3733  const int part_mask = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3734  : 0x7F;
3735  int lf_x_start = s->mb_x;
3736 
3737  s->mb_skip_run = -1;
3738 
3740  s->codec_id != AV_CODEC_ID_H264 ||
3741  (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3742 
3743  if (h->pps.cabac) {
3744  /* realign */
3745  align_get_bits(&s->gb);
3746 
3747  /* init cabac */
3749  s->gb.buffer + get_bits_count(&s->gb) / 8,
3750  (get_bits_left(&s->gb) + 7) / 8);
3751 
3753 
3754  for (;;) {
3755  // START_TIMER
3756  int ret = ff_h264_decode_mb_cabac(h);
3757  int eos;
3758  // STOP_TIMER("decode_mb_cabac")
3759 
3760  if (ret >= 0)
3762 
3763  // FIXME optimal? or let mb_decode decode 16x32 ?
3764  if (ret >= 0 && FRAME_MBAFF) {
3765  s->mb_y++;
3766 
3767  ret = ff_h264_decode_mb_cabac(h);
3768 
3769  if (ret >= 0)
3771  s->mb_y--;
3772  }
3773  eos = get_cabac_terminate(&h->cabac);
3774 
3775  if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3776  h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3777  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3778  s->mb_y, ER_MB_END & part_mask);
3779  if (s->mb_x >= lf_x_start)
3780  loop_filter(h, lf_x_start, s->mb_x + 1);
3781  return 0;
3782  }
3783  if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3784  av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3785  if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3786  av_log(h->s.avctx, AV_LOG_ERROR,
3787  "error while decoding MB %d %d, bytestream (%td)\n",
3788  s->mb_x, s->mb_y,
3791  s->mb_y, ER_MB_ERROR & part_mask);
3792  return -1;
3793  }
3794 
3795  if (++s->mb_x >= s->mb_width) {
3796  loop_filter(h, lf_x_start, s->mb_x);
3797  s->mb_x = lf_x_start = 0;
3798  decode_finish_row(h);
3799  ++s->mb_y;
3800  if (FIELD_OR_MBAFF_PICTURE) {
3801  ++s->mb_y;
3802  if (FRAME_MBAFF && s->mb_y < s->mb_height)
3804  }
3805  }
3806 
3807  if (eos || s->mb_y >= s->mb_height) {
3808  tprintf(s->avctx, "slice end %d %d\n",
3809  get_bits_count(&s->gb), s->gb.size_in_bits);
3810  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3811  s->mb_y, ER_MB_END & part_mask);
3812  if (s->mb_x > lf_x_start)
3813  loop_filter(h, lf_x_start, s->mb_x);
3814  return 0;
3815  }
3816  }
3817  } else {
3818  for (;;) {
3819  int ret = ff_h264_decode_mb_cavlc(h);
3820 
3821  if (ret >= 0)
3823 
3824  // FIXME optimal? or let mb_decode decode 16x32 ?
3825  if (ret >= 0 && FRAME_MBAFF) {
3826  s->mb_y++;
3827  ret = ff_h264_decode_mb_cavlc(h);
3828 
3829  if (ret >= 0)
3831  s->mb_y--;
3832  }
3833 
3834  if (ret < 0) {
3835  av_log(h->s.avctx, AV_LOG_ERROR,
3836  "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3838  s->mb_y, ER_MB_ERROR & part_mask);
3839  return -1;
3840  }
3841 
3842  if (++s->mb_x >= s->mb_width) {
3843  loop_filter(h, lf_x_start, s->mb_x);
3844  s->mb_x = lf_x_start = 0;
3845  decode_finish_row(h);
3846  ++s->mb_y;
3847  if (FIELD_OR_MBAFF_PICTURE) {
3848  ++s->mb_y;
3849  if (FRAME_MBAFF && s->mb_y < s->mb_height)
3851  }
3852  if (s->mb_y >= s->mb_height) {
3853  tprintf(s->avctx, "slice end %d %d\n",
3854  get_bits_count(&s->gb), s->gb.size_in_bits);
3855 
3856  if ( get_bits_left(&s->gb) == 0
3857  || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3859  s->mb_x - 1, s->mb_y,
3860  ER_MB_END & part_mask);
3861 
3862  return 0;
3863  } else {
3865  s->mb_x, s->mb_y,
3866  ER_MB_END & part_mask);
3867 
3868  return -1;
3869  }
3870  }
3871  }
3872 
3873  if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3874  tprintf(s->avctx, "slice end %d %d\n",
3875  get_bits_count(&s->gb), s->gb.size_in_bits);
3876  if (get_bits_left(&s->gb) == 0) {
3878  s->mb_x - 1, s->mb_y,
3879  ER_MB_END & part_mask);
3880  if (s->mb_x > lf_x_start)
3881  loop_filter(h, lf_x_start, s->mb_x);
3882 
3883  return 0;
3884  } else {
3886  s->mb_y, ER_MB_ERROR & part_mask);
3887 
3888  return -1;
3889  }
3890  }
3891  }
3892  }
3893 }
3894 
3895 /**
3896  * Call decode_slice() for each context.
3897  *
3898  * @param h h264 master context
3899  * @param context_count number of contexts to execute
3900  */
3901 static int execute_decode_slices(H264Context *h, int context_count)
3902 {
3903  MpegEncContext *const s = &h->s;
3904  AVCodecContext *const avctx = s->avctx;
3905  H264Context *hx;
3906  int i;
3907 
3908  if (s->avctx->hwaccel ||
3910  return 0;
3911  if (context_count == 1) {
3912  return decode_slice(avctx, &h);
3913  } else {
3914  for (i = 1; i < context_count; i++) {
3915  hx = h->thread_context[i];
3916  hx->s.err_recognition = avctx->err_recognition;
3917  hx->s.error_count = 0;
3918  hx->x264_build = h->x264_build;
3919  }
3920 
3921  avctx->execute(avctx, decode_slice, h->thread_context,
3922  NULL, context_count, sizeof(void *));
3923 
3924  /* pull back stuff from slices to master context */
3925  hx = h->thread_context[context_count - 1];
3926  s->mb_x = hx->s.mb_x;
3927  s->mb_y = hx->s.mb_y;
3928  s->droppable = hx->s.droppable;
3930  for (i = 1; i < context_count; i++)
3931  h->s.error_count += h->thread_context[i]->s.error_count;
3932  }
3933 
3934  return 0;
3935 }
3936 
3937 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
3938  int parse_extradata)
3939 {
3940  MpegEncContext *const s = &h->s;
3941  AVCodecContext *const avctx = s->avctx;
3942  H264Context *hx; ///< thread context
3943  int buf_index;
3944  int context_count;
3945  int next_avc;
3946  int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3947  int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3948  int nal_index;
3949  int idr_cleared=0;
3950  int first_slice = 0;
3951 
3952  h->nal_unit_type= 0;
3953 
3954  if(!s->slice_context_count)
3955  s->slice_context_count= 1;
3957  if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3958  h->current_slice = 0;
3959  if (!s->first_field)
3961  ff_h264_reset_sei(h);
3962  }
3963 
3964  if (h->nal_length_size == 4) {
3965  if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
3966  h->is_avc = 0;
3967  }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
3968  h->is_avc = 1;
3969  }
3970 
3971  for (; pass <= 1; pass++) {
3972  buf_index = 0;
3973  context_count = 0;
3974  next_avc = h->is_avc ? 0 : buf_size;
3975  nal_index = 0;
3976  for (;;) {
3977  int consumed;
3978  int dst_length;
3979  int bit_length;
3980  const uint8_t *ptr;
3981  int i, nalsize = 0;
3982  int err;
3983 
3984  if (buf_index >= next_avc) {
3985  if (buf_index >= buf_size - h->nal_length_size)
3986  break;
3987  nalsize = 0;
3988  for (i = 0; i < h->nal_length_size; i++)
3989  nalsize = (nalsize << 8) | buf[buf_index++];
3990  if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3991  av_log(h->s.avctx, AV_LOG_ERROR,
3992  "AVC: nal size %d\n", nalsize);
3993  break;
3994  }
3995  next_avc = buf_index + nalsize;
3996  } else {
3997  // start code prefix search
3998  for (; buf_index + 3 < next_avc; buf_index++)
3999  // This should always succeed in the first iteration.
4000  if (buf[buf_index] == 0 &&
4001  buf[buf_index + 1] == 0 &&
4002  buf[buf_index + 2] == 1)
4003  break;
4004 
4005  if (buf_index + 3 >= buf_size) {
4006  buf_index = buf_size;
4007  break;
4008  }
4009 
4010  buf_index += 3;
4011  if (buf_index >= next_avc)
4012  continue;
4013  }
4014 
4015  hx = h->thread_context[context_count];
4016 
4017  ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
4018  &consumed, next_avc - buf_index);
4019  if (ptr == NULL || dst_length < 0) {
4020  buf_index = -1;
4021  goto end;
4022  }
4023  i = buf_index + consumed;
4024  if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
4025  buf[i] == 0x00 && buf[i + 1] == 0x00 &&
4026  buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
4028 
4029  if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
4030  while(dst_length > 0 && ptr[dst_length - 1] == 0)
4031  dst_length--;
4032  bit_length = !dst_length ? 0
4033  : (8 * dst_length -
4034  decode_rbsp_trailing(h, ptr + dst_length - 1));
4035 
4036  if (s->avctx->debug & FF_DEBUG_STARTCODE)
4037  av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
4038 
4039  if (h->is_avc && (nalsize != consumed) && nalsize)
4040  av_log(h->s.avctx, AV_LOG_DEBUG,
4041  "AVC: Consumed only %d bytes instead of %d\n",
4042  consumed, nalsize);
4043 
4044  buf_index += consumed;
4045  nal_index++;
4046 
4047  if (pass == 0) {
4048  /* packets can sometimes contain multiple PPS/SPS,
4049  * e.g. two PAFF field pictures in one packet, or a demuxer
4050  * which splits NALs strangely if so, when frame threading we
4051  * can't start the next thread until we've read all of them */
4052  switch (hx->nal_unit_type) {
4053  case NAL_SPS:
4054  case NAL_PPS:
4055  nals_needed = nal_index;
4056  break;
4057  case NAL_DPA:
4058  case NAL_IDR_SLICE:
4059  case NAL_SLICE:
4060  init_get_bits(&hx->s.gb, ptr, bit_length);
4061  if (!get_ue_golomb(&hx->s.gb) || !first_slice)
4062  nals_needed = nal_index;
4063  if (!first_slice)
4064  first_slice = hx->nal_unit_type;
4065  }
4066  continue;
4067  }
4068 
4069  if (!first_slice)
4070  switch (hx->nal_unit_type) {
4071  case NAL_DPA:
4072  case NAL_IDR_SLICE:
4073  case NAL_SLICE:
4074  first_slice = hx->nal_unit_type;
4075  }
4076 
4077  // FIXME do not discard SEI id
4078  if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4079  continue;
4080 
4081 again:
4082  /* Ignore per frame NAL unit type during extradata
4083  * parsing. Decoding slices is not possible in codec init
4084  * with frame-mt */
4085  if (parse_extradata) {
4086  switch (hx->nal_unit_type) {
4087  case NAL_IDR_SLICE:
4088  case NAL_SLICE:
4089  case NAL_DPA:
4090  case NAL_DPB:
4091  case NAL_DPC:
4092  case NAL_AUXILIARY_SLICE:
4093  av_log(h->s.avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4095  }
4096  }
4097 
4098  err = 0;
4099 
4100  switch (hx->nal_unit_type) {
4101  case NAL_IDR_SLICE:
4102  if (first_slice != NAL_IDR_SLICE) {
4103  av_log(h->s.avctx, AV_LOG_ERROR,
4104  "Invalid mix of idr and non-idr slices\n");
4105  buf_index = -1;
4106  goto end;
4107  }
4108  if(!idr_cleared)
4109  idr(h); // FIXME ensure we don't lose some frames if there is reordering
4110  idr_cleared = 1;
4111  case NAL_SLICE:
4112  init_get_bits(&hx->s.gb, ptr, bit_length);
4113  hx->intra_gb_ptr =
4114  hx->inter_gb_ptr = &hx->s.gb;
4115  hx->s.data_partitioning = 0;
4116 
4117  if ((err = decode_slice_header(hx, h)))
4118  break;
4119 
4121  h->valid_recovery_point = 1;
4122 
4123  if ( h->sei_recovery_frame_cnt >= 0
4124  && ( h->recovery_frame<0
4125  || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4127  (1 << h->sps.log2_max_frame_num);
4128 
4129  if (!h->valid_recovery_point)
4130  h->recovery_frame = h->frame_num;
4131  }
4132 
4134  (hx->nal_unit_type == NAL_IDR_SLICE);
4135 
4136  if (h->recovery_frame == h->frame_num) {
4137  s->current_picture_ptr->sync |= 1;
4138  h->recovery_frame = -1;
4139  }
4140 
4141  h->sync |= !!s->current_picture_ptr->f.key_frame;
4142  h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
4143  s->current_picture_ptr->sync |= h->sync;
4144 
4145  if (h->current_slice == 1) {
4146  if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
4147  decode_postinit(h, nal_index >= nals_needed);
4148 
4149  if (s->avctx->hwaccel &&
4150  s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
4151  return -1;
4152  if (CONFIG_H264_VDPAU_DECODER &&
4155  }
4156 
4157  if (hx->redundant_pic_count == 0 &&
4158  (avctx->skip_frame < AVDISCARD_NONREF ||
4159  hx->nal_ref_idc) &&
4160  (avctx->skip_frame < AVDISCARD_BIDIR ||
4162  (avctx->skip_frame < AVDISCARD_NONKEY ||
4164  avctx->skip_frame < AVDISCARD_ALL) {
4165  if (avctx->hwaccel) {
4166  if (avctx->hwaccel->decode_slice(avctx,
4167  &buf[buf_index - consumed],
4168  consumed) < 0)
4169  return -1;
4170  } else if (CONFIG_H264_VDPAU_DECODER &&
4172  static const uint8_t start_code[] = {
4173  0x00, 0x00, 0x01 };
4174  ff_vdpau_add_data_chunk(s, start_code,
4175  sizeof(start_code));
4176  ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
4177  consumed);
4178  } else
4179  context_count++;
4180  }
4181  break;
4182  case NAL_DPA:
4183  init_get_bits(&hx->s.gb, ptr, bit_length);
4184  hx->intra_gb_ptr =
4185  hx->inter_gb_ptr = NULL;
4186 
4187  if ((err = decode_slice_header(hx, h)) < 0)
4188  break;
4189 
4190  hx->s.data_partitioning = 1;
4191  break;
4192  case NAL_DPB:
4193  init_get_bits(&hx->intra_gb, ptr, bit_length);
4194  hx->intra_gb_ptr = &hx->intra_gb;
4195  break;
4196  case NAL_DPC:
4197  init_get_bits(&hx->inter_gb, ptr, bit_length);
4198  hx->inter_gb_ptr = &hx->inter_gb;
4199 
4200  av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4201  break;
4202 
4203  if (hx->redundant_pic_count == 0 &&
4204  hx->intra_gb_ptr &&
4205  hx->s.data_partitioning &&
4206  s->current_picture_ptr &&
4207  s->context_initialized &&
4208  (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4209  (avctx->skip_frame < AVDISCARD_BIDIR ||
4211  (avctx->skip_frame < AVDISCARD_NONKEY ||
4213  avctx->skip_frame < AVDISCARD_ALL)
4214  context_count++;
4215  break;
4216  case NAL_SEI:
4217  init_get_bits(&s->gb, ptr, bit_length);
4218  ff_h264_decode_sei(h);
4219  break;
4220  case NAL_SPS:
4221  init_get_bits(&s->gb, ptr, bit_length);
4222  if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4223  av_log(h->s.avctx, AV_LOG_DEBUG,
4224  "SPS decoding failure, trying again with the complete NAL\n");
4225  if (h->is_avc)
4226  av_assert0(next_avc - buf_index + consumed == nalsize);
4227  if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4228  break;
4229  init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
4230  8*(next_avc - buf_index + consumed - 1));
4232  }
4233 
4234  break;
4235  case NAL_PPS:
4236  init_get_bits(&s->gb, ptr, bit_length);
4237  ff_h264_decode_picture_parameter_set(h, bit_length);
4238  break;
4239  case NAL_AUD:
4240  case NAL_END_SEQUENCE:
4241  case NAL_END_STREAM:
4242  case NAL_FILLER_DATA:
4243  case NAL_SPS_EXT:
4244  case NAL_AUXILIARY_SLICE:
4245  break;
4246  case NAL_FF_IGNORE:
4247  break;
4248  default:
4249  av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4250  hx->nal_unit_type, bit_length);
4251  }
4252 
4253  if (context_count == h->max_contexts) {
4254  execute_decode_slices(h, context_count);
4255  context_count = 0;
4256  }
4257 
4258  if (err < 0)
4259  av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4260  else if (err == 1) {
4261  /* Slice could not be decoded in parallel mode, copy down
4262  * NAL unit stuff to context 0 and restart. Note that
4263  * rbsp_buffer is not transferred, but since we no longer
4264  * run in parallel mode this should not be an issue. */
4265  h->nal_unit_type = hx->nal_unit_type;
4266  h->nal_ref_idc = hx->nal_ref_idc;
4267  hx = h;
4268  goto again;
4269  }
4270  }
4271  }
4272  if (context_count)
4273  execute_decode_slices(h, context_count);
4274 
4275 end:
4276  /* clean up */
4277  if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4278  !s->droppable) {
4281  }
4282 
4283  return buf_index;
4284 }
4285 
4286 /**
4287  * Return the number of bytes consumed for building the current frame.
4288  */
4289 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4290 {
4291  if (pos == 0)
4292  pos = 1; // avoid infinite loops (i doubt that is needed but ...)
4293  if (pos + 10 > buf_size)
4294  pos = buf_size; // oops ;)
4295 
4296  return pos;
4297 }
4298 
4299 static int decode_frame(AVCodecContext *avctx, void *data,
4300  int *got_frame, AVPacket *avpkt)
4301 {
4302  const uint8_t *buf = avpkt->data;
4303  int buf_size = avpkt->size;
4304  H264Context *h = avctx->priv_data;
4305  MpegEncContext *s = &h->s;
4306  AVFrame *pict = data;
4307  int buf_index = 0;
4308  Picture *out;
4309  int i, out_idx;
4310 
4311  s->flags = avctx->flags;
4312  s->flags2 = avctx->flags2;
4313 
4314  /* end of stream, output what is still in the buffers */
4315  if (buf_size == 0) {
4316  out:
4317 
4319 
4320  // FIXME factorize this with the output code below
4321  out = h->delayed_pic[0];
4322  out_idx = 0;
4323  for (i = 1;
4324  h->delayed_pic[i] &&
4325  !h->delayed_pic[i]->f.key_frame &&
4326  !h->delayed_pic[i]->mmco_reset;
4327  i++)
4328  if (h->delayed_pic[i]->poc < out->poc) {
4329  out = h->delayed_pic[i];
4330  out_idx = i;
4331  }
4332 
4333  for (i = out_idx; h->delayed_pic[i]; i++)
4334  h->delayed_pic[i] = h->delayed_pic[i + 1];
4335 
4336  if (out) {
4337  *got_frame = 1;
4338  *pict = out->f;
4339  }
4340 
4341  return buf_index;
4342  }
4343  if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4344  int cnt= buf[5]&0x1f;
4345  const uint8_t *p= buf+6;
4346  while(cnt--){
4347  int nalsize= AV_RB16(p) + 2;
4348  if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4349  goto not_extra;
4350  p += nalsize;
4351  }
4352  cnt = *(p++);
4353  if(!cnt)
4354  goto not_extra;
4355  while(cnt--){
4356  int nalsize= AV_RB16(p) + 2;
4357  if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4358  goto not_extra;
4359  p += nalsize;
4360  }
4361 
4362  return ff_h264_decode_extradata(h, buf, buf_size);
4363  }
4364 not_extra:
4365 
4366  buf_index = decode_nal_units(h, buf, buf_size, 0);
4367  if (buf_index < 0)
4368  return -1;
4369 
4371  av_assert0(buf_index <= buf_size);
4372  goto out;
4373  }
4374 
4375  if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4376  if (avctx->skip_frame >= AVDISCARD_NONREF ||
4377  buf_size >= 4 && !memcmp("Q264", buf, 4))
4378  return buf_size;
4379  av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4380  return -1;
4381  }
4382 
4383  if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4384  (s->mb_y >= s->mb_height && s->mb_height)) {
4385  if (s->flags2 & CODEC_FLAG2_CHUNKS)
4386  decode_postinit(h, 1);
4387 
4388  field_end(h, 0);
4389  h->context_reinitialized = 0;
4390 
4391  /* Wait for second field. */
4392  *got_frame = 0;
4393  if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4394  *got_frame = 1;
4395  *pict = h->next_output_pic->f;
4396  }
4397  }
4398 
4399  assert(pict->data[0] || !*got_frame);
4400  ff_print_debug_info(s, pict);
4401 
4402  return get_consumed_bytes(s, buf_index, buf_size);
4403 }
4404 
4406 {
4407  int i;
4408 
4409  free_tables(h, 1); // FIXME cleanup init stuff perhaps
4410 
4411  for (i = 0; i < MAX_SPS_COUNT; i++)
4412  av_freep(h->sps_buffers + i);
4413 
4414  for (i = 0; i < MAX_PPS_COUNT; i++)
4415  av_freep(h->pps_buffers + i);
4416 }
4417 
4419 {
4420  H264Context *h = avctx->priv_data;
4421  MpegEncContext *s = &h->s;
4422 
4425 
4426  ff_MPV_common_end(s);
4427 
4428  // memset(h, 0, sizeof(H264Context));
4429 
4430  return 0;
4431 }
4432 
4433 static const AVProfile profiles[] = {
4434  { FF_PROFILE_H264_BASELINE, "Baseline" },
4435  { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4436  { FF_PROFILE_H264_MAIN, "Main" },
4437  { FF_PROFILE_H264_EXTENDED, "Extended" },
4438  { FF_PROFILE_H264_HIGH, "High" },
4439  { FF_PROFILE_H264_HIGH_10, "High 10" },
4440  { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4441  { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4442  { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4443  { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4444  { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4445  { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4446  { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4447  { FF_PROFILE_UNKNOWN },
4448 };
4449 
4450 static const AVOption h264_options[] = {
4451  {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
4452  {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
4453  {NULL}
4454 };
4455 
4456 static const AVClass h264_class = {
4457  .class_name = "H264 Decoder",
4458  .item_name = av_default_item_name,
4459  .option = h264_options,
4460  .version = LIBAVUTIL_VERSION_INT,
4461 };
4462 
4463 static const AVClass h264_vdpau_class = {
4464  .class_name = "H264 VDPAU Decoder",
4465  .item_name = av_default_item_name,
4466  .option = h264_options,
4467  .version = LIBAVUTIL_VERSION_INT,
4468 };
4469 
4471  .name = "h264",
4472  .type = AVMEDIA_TYPE_VIDEO,
4473  .id = AV_CODEC_ID_H264,
4474  .priv_data_size = sizeof(H264Context),
4477  .decode = decode_frame,
4478  .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4481  .flush = flush_dpb,
4482  .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4483  .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4484  .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4485  .profiles = NULL_IF_CONFIG_SMALL(profiles),
4486  .priv_class = &h264_class,
4487 };
4488 
4489 #if CONFIG_H264_VDPAU_DECODER
4490 AVCodec ff_h264_vdpau_decoder = {
4491  .name = "h264_vdpau",
4492  .type = AVMEDIA_TYPE_VIDEO,
4493  .id = AV_CODEC_ID_H264,
4494  .priv_data_size = sizeof(H264Context),
4497  .decode = decode_frame,
4499  .flush = flush_dpb,
4500  .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4501  .pix_fmts = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
4502  AV_PIX_FMT_NONE},
4503  .profiles = NULL_IF_CONFIG_SMALL(profiles),
4504  .priv_class = &h264_vdpau_class,
4505 };
4506 #endif