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 Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
28 #include "libavutil/imgutils.h"
29 #include "internal.h"
30 #include "cabac.h"
31 #include "cabac_functions.h"
32 #include "dsputil.h"
33 #include "avcodec.h"
34 #include "mpegvideo.h"
35 #include "h264.h"
36 #include "h264data.h"
37 #include "h264_mvpred.h"
38 #include "golomb.h"
39 #include "mathops.h"
40 #include "rectangle.h"
41 #include "thread.h"
42 #include "vdpau_internal.h"
43 #include "libavutil/avassert.h"
44 
45 //#undef NDEBUG
46 #include <assert.h>
47 
48 static const uint8_t rem6[QP_MAX_NUM+1]={
49 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
50 };
51 
52 static const uint8_t div6[QP_MAX_NUM+1]={
53 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9,10,10,10,10,
54 };
55 
62 };
63 
69  MpegEncContext * const s = &h->s;
70  static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
71  static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
72  int i;
73 
74  if(!(h->top_samples_available&0x8000)){
75  for(i=0; i<4; i++){
76  int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
77  if(status<0){
78  av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
79  return -1;
80  } else if(status){
81  h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
82  }
83  }
84  }
85 
86  if((h->left_samples_available&0x8888)!=0x8888){
87  static const int mask[4]={0x8000,0x2000,0x80,0x20};
88  for(i=0; i<4; i++){
89  if(!(h->left_samples_available&mask[i])){
90  int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
91  if(status<0){
92  av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
93  return -1;
94  } else if(status){
95  h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
96  }
97  }
98  }
99  }
100 
101  return 0;
102 } //FIXME cleanup like ff_h264_check_intra_pred_mode
103 
108 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma){
109  MpegEncContext * const s = &h->s;
110  static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
111  static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
112 
113  if(mode > 6U) {
114  av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
115  return -1;
116  }
117 
118  if(!(h->top_samples_available&0x8000)){
119  mode= top[ mode ];
120  if(mode<0){
121  av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
122  return -1;
123  }
124  }
125 
126  if((h->left_samples_available&0x8080) != 0x8080){
127  mode= left[ mode ];
128  if(is_chroma && (h->left_samples_available&0x8080)){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
129  mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
130  }
131  if(mode<0){
132  av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
133  return -1;
134  }
135  }
136 
137  return mode;
138 }
139 
140 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length){
141  int i, si, di;
142  uint8_t *dst;
143  int bufidx;
144 
145 // src[0]&0x80; //forbidden bit
146  h->nal_ref_idc= src[0]>>5;
147  h->nal_unit_type= src[0]&0x1F;
148 
149  src++; length--;
150 
151 #if HAVE_FAST_UNALIGNED
152 # if HAVE_FAST_64BIT
153 # define RS 7
154  for(i=0; i+1<length; i+=9){
155  if(!((~AV_RN64A(src+i) & (AV_RN64A(src+i) - 0x0100010001000101ULL)) & 0x8000800080008080ULL))
156 # else
157 # define RS 3
158  for(i=0; i+1<length; i+=5){
159  if(!((~AV_RN32A(src+i) & (AV_RN32A(src+i) - 0x01000101U)) & 0x80008080U))
160 # endif
161  continue;
162  if(i>0 && !src[i]) i--;
163  while(src[i]) i++;
164 #else
165 # define RS 0
166  for(i=0; i+1<length; i+=2){
167  if(src[i]) continue;
168  if(i>0 && src[i-1]==0) i--;
169 #endif
170  if(i+2<length && src[i+1]==0 && src[i+2]<=3){
171  if(src[i+2]!=3){
172  /* startcode, so we must be past the end */
173  length=i;
174  }
175  break;
176  }
177  i-= RS;
178  }
179 
180  if(i>=length-1){ //no escaped 0
181  *dst_length= length;
182  *consumed= length+1; //+1 for the header
183  return src;
184  }
185 
186  bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0; // use second escape buffer for inter data
188  dst= h->rbsp_buffer[bufidx];
189 
190  if (dst == NULL){
191  return NULL;
192  }
193 
194 //printf("decoding esc\n");
195  memcpy(dst, src, i);
196  si=di=i;
197  while(si+2<length){
198  //remove escapes (very rare 1:2^22)
199  if(src[si+2]>3){
200  dst[di++]= src[si++];
201  dst[di++]= src[si++];
202  }else if(src[si]==0 && src[si+1]==0){
203  if(src[si+2]==3){ //escape
204  dst[di++]= 0;
205  dst[di++]= 0;
206  si+=3;
207  continue;
208  }else //next start code
209  goto nsc;
210  }
211 
212  dst[di++]= src[si++];
213  }
214  while(si<length)
215  dst[di++]= src[si++];
216 nsc:
217 
218  memset(dst+di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
219 
220  *dst_length= di;
221  *consumed= si + 1;//+1 for the header
222 //FIXME store exact number of bits in the getbitcontext (it is needed for decoding)
223  return dst;
224 }
225 
230 static int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src){
231  int v= *src;
232  int r;
233 
234  tprintf(h->s.avctx, "rbsp trailing %X\n", v);
235 
236  for(r=1; r<9; r++){
237  if(v&1) return r;
238  v>>=1;
239  }
240  return 0;
241 }
242 
243 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n, int height,
244  int y_offset, int list){
245  int raw_my= h->mv_cache[list][ scan8[n] ][1];
246  int filter_height= (raw_my&3) ? 2 : 0;
247  int full_my= (raw_my>>2) + y_offset;
248  int top = full_my - filter_height, bottom = full_my + height + filter_height;
249 
250  return FFMAX(abs(top), bottom);
251 }
252 
253 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n, int height,
254  int y_offset, int list0, int list1, int *nrefs){
255  MpegEncContext * const s = &h->s;
256  int my;
257 
258  y_offset += 16*(s->mb_y >> MB_FIELD);
259 
260  if(list0){
261  int ref_n = h->ref_cache[0][ scan8[n] ];
262  Picture *ref= &h->ref_list[0][ref_n];
263 
264  // Error resilience puts the current picture in the ref list.
265  // Don't try to wait on these as it will cause a deadlock.
266  // Fields can wait on each other, though.
267  if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
268  (ref->f.reference & 3) != s->picture_structure) {
269  my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
270  if (refs[0][ref_n] < 0) nrefs[0] += 1;
271  refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
272  }
273  }
274 
275  if(list1){
276  int ref_n = h->ref_cache[1][ scan8[n] ];
277  Picture *ref= &h->ref_list[1][ref_n];
278 
279  if (ref->f.thread_opaque != s->current_picture.f.thread_opaque ||
280  (ref->f.reference & 3) != s->picture_structure) {
281  my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
282  if (refs[1][ref_n] < 0) nrefs[1] += 1;
283  refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
284  }
285  }
286 }
287 
294  MpegEncContext * const s = &h->s;
295  const int mb_xy= h->mb_xy;
296  const int mb_type = s->current_picture.f.mb_type[mb_xy];
297  int refs[2][48];
298  int nrefs[2] = {0};
299  int ref, list;
300 
301  memset(refs, -1, sizeof(refs));
302 
303  if(IS_16X16(mb_type)){
304  get_lowest_part_y(h, refs, 0, 16, 0,
305  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
306  }else if(IS_16X8(mb_type)){
307  get_lowest_part_y(h, refs, 0, 8, 0,
308  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
309  get_lowest_part_y(h, refs, 8, 8, 8,
310  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
311  }else if(IS_8X16(mb_type)){
312  get_lowest_part_y(h, refs, 0, 16, 0,
313  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
314  get_lowest_part_y(h, refs, 4, 16, 0,
315  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
316  }else{
317  int i;
318 
319  assert(IS_8X8(mb_type));
320 
321  for(i=0; i<4; i++){
322  const int sub_mb_type= h->sub_mb_type[i];
323  const int n= 4*i;
324  int y_offset= (i&2)<<2;
325 
326  if(IS_SUB_8X8(sub_mb_type)){
327  get_lowest_part_y(h, refs, n , 8, y_offset,
328  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
329  }else if(IS_SUB_8X4(sub_mb_type)){
330  get_lowest_part_y(h, refs, n , 4, y_offset,
331  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
332  get_lowest_part_y(h, refs, n+2, 4, y_offset+4,
333  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
334  }else if(IS_SUB_4X8(sub_mb_type)){
335  get_lowest_part_y(h, refs, n , 8, y_offset,
336  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
337  get_lowest_part_y(h, refs, n+1, 8, y_offset,
338  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
339  }else{
340  int j;
341  assert(IS_SUB_4X4(sub_mb_type));
342  for(j=0; j<4; j++){
343  int sub_y_offset= y_offset + 2*(j&2);
344  get_lowest_part_y(h, refs, n+j, 4, sub_y_offset,
345  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1), nrefs);
346  }
347  }
348  }
349  }
350 
351  for(list=h->list_count-1; list>=0; list--){
352  for(ref=0; ref<48 && nrefs[list]; ref++){
353  int row = refs[list][ref];
354  if(row >= 0){
355  Picture *ref_pic = &h->ref_list[list][ref];
356  int ref_field = ref_pic->f.reference - 1;
357  int ref_field_picture = ref_pic->field_picture;
358  int pic_height = 16*s->mb_height >> ref_field_picture;
359 
360  row <<= MB_MBAFF;
361  nrefs[list]--;
362 
363  if(!FIELD_PICTURE && ref_field_picture){ // frame referencing two fields
364  ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) - !(row&1), pic_height-1), 1);
365  ff_thread_await_progress((AVFrame*)ref_pic, FFMIN((row >> 1) , pic_height-1), 0);
366  }else if(FIELD_PICTURE && !ref_field_picture){ // field referencing one field of a frame
367  ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row*2 + ref_field , pic_height-1), 0);
368  }else if(FIELD_PICTURE){
369  ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), ref_field);
370  }else{
371  ff_thread_await_progress((AVFrame*)ref_pic, FFMIN(row, pic_height-1), 0);
372  }
373  }
374  }
375  }
376 }
377 
378 #if 0
379 
383 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
384 // const int qmul= dequant_coeff[qp][0];
385  int i;
386  int temp[16]; //FIXME check if this is a good idea
387  static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
388  static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
389 
390  for(i=0; i<4; i++){
391  const int offset= y_offset[i];
392  const int z0= block[offset+stride*0] + block[offset+stride*4];
393  const int z1= block[offset+stride*0] - block[offset+stride*4];
394  const int z2= block[offset+stride*1] - block[offset+stride*5];
395  const int z3= block[offset+stride*1] + block[offset+stride*5];
396 
397  temp[4*i+0]= z0+z3;
398  temp[4*i+1]= z1+z2;
399  temp[4*i+2]= z1-z2;
400  temp[4*i+3]= z0-z3;
401  }
402 
403  for(i=0; i<4; i++){
404  const int offset= x_offset[i];
405  const int z0= temp[4*0+i] + temp[4*2+i];
406  const int z1= temp[4*0+i] - temp[4*2+i];
407  const int z2= temp[4*1+i] - temp[4*3+i];
408  const int z3= temp[4*1+i] + temp[4*3+i];
409 
410  block[stride*0 +offset]= (z0 + z3)>>1;
411  block[stride*2 +offset]= (z1 + z2)>>1;
412  block[stride*8 +offset]= (z1 - z2)>>1;
413  block[stride*10+offset]= (z0 - z3)>>1;
414  }
415 }
416 #endif
417 
418 #undef xStride
419 #undef stride
420 
421 #if 0
422 static void chroma_dc_dct_c(DCTELEM *block){
423  const int stride= 16*2;
424  const int xStride= 16;
425  int a,b,c,d,e;
426 
427  a= block[stride*0 + xStride*0];
428  b= block[stride*0 + xStride*1];
429  c= block[stride*1 + xStride*0];
430  d= block[stride*1 + xStride*1];
431 
432  e= a-b;
433  a= a+b;
434  b= c-d;
435  c= c+d;
436 
437  block[stride*0 + xStride*0]= (a+c);
438  block[stride*0 + xStride*1]= (e+b);
439  block[stride*1 + xStride*0]= (a-c);
440  block[stride*1 + xStride*1]= (e-b);
441 }
442 #endif
443 
444 static av_always_inline void
445 mc_dir_part(H264Context *h, Picture *pic, int n, int square,
446  int height, int delta, int list,
447  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
448  int src_x_offset, int src_y_offset,
449  qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
450  int pixel_shift, int chroma_idc)
451 {
452  MpegEncContext * const s = &h->s;
453  const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
454  int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
455  const int luma_xy= (mx&3) + ((my&3)<<2);
456  int offset = ((mx>>2) << pixel_shift) + (my>>2)*h->mb_linesize;
457  uint8_t * src_y = pic->f.data[0] + offset;
458  uint8_t * src_cb, * src_cr;
459  int extra_width= h->emu_edge_width;
460  int extra_height= h->emu_edge_height;
461  int emu=0;
462  const int full_mx= mx>>2;
463  const int full_my= my>>2;
464  const int pic_width = 16*s->mb_width;
465  const int pic_height = 16*s->mb_height >> MB_FIELD;
466  int ysh;
467 
468  if(mx&7) extra_width -= 3;
469  if(my&7) extra_height -= 3;
470 
471  if( full_mx < 0-extra_width
472  || full_my < 0-extra_height
473  || full_mx + 16/*FIXME*/ > pic_width + extra_width
474  || full_my + 16/*FIXME*/ > pic_height + extra_height){
475  s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
476  16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
477  src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
478  emu=1;
479  }
480 
481  qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
482  if(!square){
483  qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
484  }
485 
486  if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
487 
488  if(chroma_idc == 3 /* yuv444 */){
489  src_cb = pic->f.data[1] + offset;
490  if(emu){
491  s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
492  16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
493  src_cb= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
494  }
495  qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); //FIXME try variable height perhaps?
496  if(!square){
497  qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
498  }
499 
500  src_cr = pic->f.data[2] + offset;
501  if(emu){
502  s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
503  16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
504  src_cr= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
505  }
506  qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); //FIXME try variable height perhaps?
507  if(!square){
508  qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
509  }
510  return;
511  }
512 
513  ysh = 3 - (chroma_idc == 2 /* yuv422 */);
514  if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
515  // chroma offset when predicting from a field of opposite parity
516  my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
517  emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
518  }
519 
520  src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
521  src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
522 
523  if(emu){
525  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
526  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
527  src_cb= s->edge_emu_buffer;
528  }
529  chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
530  mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
531 
532  if(emu){
534  9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
535  pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
536  src_cr= s->edge_emu_buffer;
537  }
538  chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
539  mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
540 }
541 
542 static av_always_inline void
543 mc_part_std(H264Context *h, int n, int square, int height, int delta,
544  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
545  int x_offset, int y_offset,
546  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
547  qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
548  int list0, int list1, int pixel_shift, int chroma_idc)
549 {
550  MpegEncContext * const s = &h->s;
551  qpel_mc_func *qpix_op= qpix_put;
552  h264_chroma_mc_func chroma_op= chroma_put;
553 
554  dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
555  if (chroma_idc == 3 /* yuv444 */) {
556  dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
557  dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
558  } else if (chroma_idc == 2 /* yuv422 */) {
559  dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
560  dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
561  } else /* yuv420 */ {
562  dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
563  dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
564  }
565  x_offset += 8*s->mb_x;
566  y_offset += 8*(s->mb_y >> MB_FIELD);
567 
568  if(list0){
569  Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
570  mc_dir_part(h, ref, n, square, height, delta, 0,
571  dest_y, dest_cb, dest_cr, x_offset, y_offset,
572  qpix_op, chroma_op, pixel_shift, chroma_idc);
573 
574  qpix_op= qpix_avg;
575  chroma_op= chroma_avg;
576  }
577 
578  if(list1){
579  Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
580  mc_dir_part(h, ref, n, square, height, delta, 1,
581  dest_y, dest_cb, dest_cr, x_offset, y_offset,
582  qpix_op, chroma_op, pixel_shift, chroma_idc);
583  }
584 }
585 
586 static av_always_inline void
587 mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
588  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
589  int x_offset, int y_offset,
590  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
591  h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
592  h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
593  int list0, int list1, int pixel_shift, int chroma_idc){
594  MpegEncContext * const s = &h->s;
595  int chroma_height;
596 
597  dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
598  if (chroma_idc == 3 /* yuv444 */) {
599  chroma_height = height;
600  chroma_weight_avg = luma_weight_avg;
601  chroma_weight_op = luma_weight_op;
602  dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
603  dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
604  } else if (chroma_idc == 2 /* yuv422 */) {
605  chroma_height = height;
606  dest_cb += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
607  dest_cr += ( x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
608  } else /* yuv420 */ {
609  chroma_height = height >> 1;
610  dest_cb += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
611  dest_cr += ( x_offset << pixel_shift) + y_offset*h->mb_uvlinesize;
612  }
613  x_offset += 8*s->mb_x;
614  y_offset += 8*(s->mb_y >> MB_FIELD);
615 
616  if(list0 && list1){
617  /* don't optimize for luma-only case, since B-frames usually
618  * use implicit weights => chroma too. */
619  uint8_t *tmp_cb = s->obmc_scratchpad;
620  uint8_t *tmp_cr = s->obmc_scratchpad + (16 << pixel_shift);
621  uint8_t *tmp_y = s->obmc_scratchpad + 16*h->mb_uvlinesize;
622  int refn0 = h->ref_cache[0][ scan8[n] ];
623  int refn1 = h->ref_cache[1][ scan8[n] ];
624 
625  mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
626  dest_y, dest_cb, dest_cr,
627  x_offset, y_offset, qpix_put, chroma_put,
628  pixel_shift, chroma_idc);
629  mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
630  tmp_y, tmp_cb, tmp_cr,
631  x_offset, y_offset, qpix_put, chroma_put,
632  pixel_shift, chroma_idc);
633 
634  if(h->use_weight == 2){
635  int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
636  int weight1 = 64 - weight0;
637  luma_weight_avg( dest_y, tmp_y, h-> mb_linesize,
638  height, 5, weight0, weight1, 0);
639  chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
640  chroma_height, 5, weight0, weight1, 0);
641  chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
642  chroma_height, 5, weight0, weight1, 0);
643  }else{
644  luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
645  h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
646  h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
647  chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
648  h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
649  h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
650  chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
651  h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
652  h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
653  }
654  }else{
655  int list = list1 ? 1 : 0;
656  int refn = h->ref_cache[list][ scan8[n] ];
657  Picture *ref= &h->ref_list[list][refn];
658  mc_dir_part(h, ref, n, square, height, delta, list,
659  dest_y, dest_cb, dest_cr, x_offset, y_offset,
660  qpix_put, chroma_put, pixel_shift, chroma_idc);
661 
662  luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
663  h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
664  if(h->use_weight_chroma){
665  chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
666  h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
667  chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
668  h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
669  }
670  }
671 }
672 
673 static av_always_inline void
674 mc_part(H264Context *h, int n, int square, int height, int delta,
675  uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
676  int x_offset, int y_offset,
677  qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
678  qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
679  h264_weight_func *weight_op, h264_biweight_func *weight_avg,
680  int list0, int list1, int pixel_shift, int chroma_idc)
681 {
682  if((h->use_weight==2 && list0 && list1
683  && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
684  || h->use_weight==1)
685  mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
686  x_offset, y_offset, qpix_put, chroma_put,
687  weight_op[0], weight_op[1], weight_avg[0],
688  weight_avg[1], list0, list1, pixel_shift, chroma_idc);
689  else
690  mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
691  x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
692  chroma_avg, list0, list1, pixel_shift, chroma_idc);
693 }
694 
695 static av_always_inline void
696 prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
697 {
698  /* fetch pixels for estimated mv 4 macroblocks ahead
699  * optimized for 64byte cache lines */
700  MpegEncContext * const s = &h->s;
701  const int refn = h->ref_cache[list][scan8[0]];
702  if(refn >= 0){
703  const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
704  const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
705  uint8_t **src = h->ref_list[list][refn].f.data;
706  int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
707  s->dsp.prefetch(src[0]+off, s->linesize, 4);
708  if (chroma_idc == 3 /* yuv444 */) {
709  s->dsp.prefetch(src[1]+off, s->linesize, 4);
710  s->dsp.prefetch(src[2]+off, s->linesize, 4);
711  }else{
712  off= ((mx>>1) << pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize + (64 << pixel_shift);
713  s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
714  }
715  }
716 }
717 
718 static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
719  qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
720  qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
721  h264_weight_func *weight_op, h264_biweight_func *weight_avg,
722  int pixel_shift, int chroma_idc)
723 {
724  MpegEncContext * const s = &h->s;
725  const int mb_xy= h->mb_xy;
726  const int mb_type = s->current_picture.f.mb_type[mb_xy];
727 
728  assert(IS_INTER(mb_type));
729 
731  await_references(h);
732  prefetch_motion(h, 0, pixel_shift, chroma_idc);
733 
734  if(IS_16X16(mb_type)){
735  mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
736  qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
737  weight_op, weight_avg,
738  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
739  pixel_shift, chroma_idc);
740  }else if(IS_16X8(mb_type)){
741  mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
742  qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
743  weight_op, weight_avg,
744  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
745  pixel_shift, chroma_idc);
746  mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
747  qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
748  weight_op, weight_avg,
749  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
750  pixel_shift, chroma_idc);
751  }else if(IS_8X16(mb_type)){
752  mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
753  qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
754  &weight_op[1], &weight_avg[1],
755  IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
756  pixel_shift, chroma_idc);
757  mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
758  qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
759  &weight_op[1], &weight_avg[1],
760  IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
761  pixel_shift, chroma_idc);
762  }else{
763  int i;
764 
765  assert(IS_8X8(mb_type));
766 
767  for(i=0; i<4; i++){
768  const int sub_mb_type= h->sub_mb_type[i];
769  const int n= 4*i;
770  int x_offset= (i&1)<<2;
771  int y_offset= (i&2)<<1;
772 
773  if(IS_SUB_8X8(sub_mb_type)){
774  mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
775  qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
776  &weight_op[1], &weight_avg[1],
777  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
778  pixel_shift, chroma_idc);
779  }else if(IS_SUB_8X4(sub_mb_type)){
780  mc_part(h, n , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
781  qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
782  &weight_op[1], &weight_avg[1],
783  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
784  pixel_shift, chroma_idc);
785  mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
786  qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
787  &weight_op[1], &weight_avg[1],
788  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
789  pixel_shift, chroma_idc);
790  }else if(IS_SUB_4X8(sub_mb_type)){
791  mc_part(h, n , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
792  qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
793  &weight_op[2], &weight_avg[2],
794  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
795  pixel_shift, chroma_idc);
796  mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
797  qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
798  &weight_op[2], &weight_avg[2],
799  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
800  pixel_shift, chroma_idc);
801  }else{
802  int j;
803  assert(IS_SUB_4X4(sub_mb_type));
804  for(j=0; j<4; j++){
805  int sub_x_offset= x_offset + 2*(j&1);
806  int sub_y_offset= y_offset + (j&2);
807  mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
808  qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
809  &weight_op[2], &weight_avg[2],
810  IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
811  pixel_shift, chroma_idc);
812  }
813  }
814  }
815  }
816 
817  prefetch_motion(h, 1, pixel_shift, chroma_idc);
818 }
819 
820 static av_always_inline void
821 hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
822  qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
823  qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
824  h264_weight_func *weight_op, h264_biweight_func *weight_avg,
825  int pixel_shift)
826 {
827  hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
828  qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
829 }
830 
831 static av_always_inline void
832 hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
833  qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
834  qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
835  h264_weight_func *weight_op, h264_biweight_func *weight_avg,
836  int pixel_shift)
837 {
838  hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
839  qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
840 }
841 
842 static void free_tables(H264Context *h, int free_rbsp){
843  int i;
844  H264Context *hx;
845 
848  av_freep(&h->cbp_table);
849  av_freep(&h->mvd_table[0]);
850  av_freep(&h->mvd_table[1]);
851  av_freep(&h->direct_table);
854  h->slice_table= NULL;
855  av_freep(&h->list_counts);
856 
857  av_freep(&h->mb2b_xy);
858  av_freep(&h->mb2br_xy);
859 
860  for(i = 0; i < MAX_THREADS; i++) {
861  hx = h->thread_context[i];
862  if(!hx) continue;
863  av_freep(&hx->top_borders[1]);
864  av_freep(&hx->top_borders[0]);
865  av_freep(&hx->s.obmc_scratchpad);
866  if (free_rbsp){
867  av_freep(&hx->rbsp_buffer[1]);
868  av_freep(&hx->rbsp_buffer[0]);
869  hx->rbsp_buffer_size[0] = 0;
870  hx->rbsp_buffer_size[1] = 0;
871  }
872  if (i) av_freep(&h->thread_context[i]);
873  }
874 }
875 
877  int i,j,q,x;
878  const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
879 
880  for(i=0; i<6; i++ ){
881  h->dequant8_coeff[i] = h->dequant8_buffer[i];
882  for(j=0; j<i; j++){
883  if(!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i], 64*sizeof(uint8_t))){
884  h->dequant8_coeff[i] = h->dequant8_buffer[j];
885  break;
886  }
887  }
888  if(j<i)
889  continue;
890 
891  for(q=0; q<max_qp+1; q++){
892  int shift = div6[q];
893  int idx = rem6[q];
894  for(x=0; x<64; x++)
895  h->dequant8_coeff[i][q][(x>>3)|((x&7)<<3)] =
896  ((uint32_t)dequant8_coeff_init[idx][ dequant8_coeff_init_scan[((x>>1)&12) | (x&3)] ] *
897  h->pps.scaling_matrix8[i][x]) << shift;
898  }
899  }
900 }
901 
903  int i,j,q,x;
904  const int max_qp = 51 + 6*(h->sps.bit_depth_luma-8);
905  for(i=0; i<6; i++ ){
906  h->dequant4_coeff[i] = h->dequant4_buffer[i];
907  for(j=0; j<i; j++){
908  if(!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i], 16*sizeof(uint8_t))){
909  h->dequant4_coeff[i] = h->dequant4_buffer[j];
910  break;
911  }
912  }
913  if(j<i)
914  continue;
915 
916  for(q=0; q<max_qp+1; q++){
917  int shift = div6[q] + 2;
918  int idx = rem6[q];
919  for(x=0; x<16; x++)
920  h->dequant4_coeff[i][q][(x>>2)|((x<<2)&0xF)] =
921  ((uint32_t)dequant4_coeff_init[idx][(x&1) + ((x>>2)&1)] *
922  h->pps.scaling_matrix4[i][x]) << shift;
923  }
924  }
925 }
926 
928  int i,x;
930  if(h->pps.transform_8x8_mode)
932  if(h->sps.transform_bypass){
933  for(i=0; i<6; i++)
934  for(x=0; x<16; x++)
935  h->dequant4_coeff[i][0][x] = 1<<6;
936  if(h->pps.transform_8x8_mode)
937  for(i=0; i<6; i++)
938  for(x=0; x<64; x++)
939  h->dequant8_coeff[i][0][x] = 1<<6;
940  }
941 }
942 
943 
945  MpegEncContext * const s = &h->s;
946  const int big_mb_num= s->mb_stride * (s->mb_height+1);
947  const int row_mb_num= 2*s->mb_stride*s->avctx->thread_count;
948  int x,y;
949 
950  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode, row_mb_num * 8 * sizeof(uint8_t), fail)
951 
952  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count , big_mb_num * 48 * sizeof(uint8_t), fail)
953  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base , (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base), fail)
954  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table, big_mb_num * sizeof(uint16_t), fail)
955 
956  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t), fail)
957  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0], 16*row_mb_num * sizeof(uint8_t), fail);
958  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1], 16*row_mb_num * sizeof(uint8_t), fail);
959  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table, 4*big_mb_num * sizeof(uint8_t) , fail);
960  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts, big_mb_num * sizeof(uint8_t), fail)
961 
962  memset(h->slice_table_base, -1, (big_mb_num+s->mb_stride) * sizeof(*h->slice_table_base));
963  h->slice_table= h->slice_table_base + s->mb_stride*2 + 1;
964 
965  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy , big_mb_num * sizeof(uint32_t), fail);
966  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy , big_mb_num * sizeof(uint32_t), fail);
967  for(y=0; y<s->mb_height; y++){
968  for(x=0; x<s->mb_width; x++){
969  const int mb_xy= x + y*s->mb_stride;
970  const int b_xy = 4*x + 4*y*h->b_stride;
971 
972  h->mb2b_xy [mb_xy]= b_xy;
973  h->mb2br_xy[mb_xy]= 8*(FMO ? mb_xy : (mb_xy % (2*s->mb_stride)));
974  }
975  }
976 
977  s->obmc_scratchpad = NULL;
978 
979  if(!h->dequant4_coeff[0])
981 
982  return 0;
983 fail:
984  free_tables(h, 1);
985  return -1;
986 }
987 
991 static void clone_tables(H264Context *dst, H264Context *src, int i){
992  MpegEncContext * const s = &src->s;
993  dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i*8*2*s->mb_stride;
994  dst->non_zero_count = src->non_zero_count;
995  dst->slice_table = src->slice_table;
996  dst->cbp_table = src->cbp_table;
997  dst->mb2b_xy = src->mb2b_xy;
998  dst->mb2br_xy = src->mb2br_xy;
1000  dst->mvd_table[0] = src->mvd_table[0] + i*8*2*s->mb_stride;
1001  dst->mvd_table[1] = src->mvd_table[1] + i*8*2*s->mb_stride;
1002  dst->direct_table = src->direct_table;
1003  dst->list_counts = src->list_counts;
1004 
1005  dst->s.obmc_scratchpad = NULL;
1007 }
1008 
1013 static int context_init(H264Context *h){
1014  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1015  FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1], h->s.mb_width * 16*3 * sizeof(uint8_t)*2, fail)
1016 
1017  h->ref_cache[0][scan8[5 ]+1] = h->ref_cache[0][scan8[7 ]+1] = h->ref_cache[0][scan8[13]+1] =
1018  h->ref_cache[1][scan8[5 ]+1] = h->ref_cache[1][scan8[7 ]+1] = h->ref_cache[1][scan8[13]+1] = PART_NOT_AVAILABLE;
1019 
1020  return 0;
1021 fail:
1022  return -1; // free_tables will clean up for us
1023 }
1024 
1025 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
1026 
1028  MpegEncContext * const s = &h->s;
1029 
1030  s->width = s->avctx->width;
1031  s->height = s->avctx->height;
1032  s->codec_id= s->avctx->codec->id;
1033 
1034  ff_h264dsp_init(&h->h264dsp, 8, 1);
1035  ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
1036 
1037  h->dequant_coeff_pps= -1;
1038  s->unrestricted_mv=1;
1039 
1040  dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
1041 
1042  memset(h->pps.scaling_matrix4, 16, 6*16*sizeof(uint8_t));
1043  memset(h->pps.scaling_matrix8, 16, 2*64*sizeof(uint8_t));
1044 }
1045 
1047 {
1048  AVCodecContext *avctx = h->s.avctx;
1049 
1050  if(avctx->extradata[0] == 1){
1051  int i, cnt, nalsize;
1052  unsigned char *p = avctx->extradata;
1053 
1054  h->is_avc = 1;
1055 
1056  if(avctx->extradata_size < 7) {
1057  av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1058  return -1;
1059  }
1060  /* sps and pps in the avcC always have length coded with 2 bytes,
1061  so put a fake nal_length_size = 2 while parsing them */
1062  h->nal_length_size = 2;
1063  // Decode sps from avcC
1064  cnt = *(p+5) & 0x1f; // Number of sps
1065  p += 6;
1066  for (i = 0; i < cnt; i++) {
1067  nalsize = AV_RB16(p) + 2;
1068  if (p - avctx->extradata + nalsize > avctx->extradata_size)
1069  return -1;
1070  if(decode_nal_units(h, p, nalsize) < 0) {
1071  av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
1072  return -1;
1073  }
1074  p += nalsize;
1075  }
1076  // Decode pps from avcC
1077  cnt = *(p++); // Number of pps
1078  for (i = 0; i < cnt; i++) {
1079  nalsize = AV_RB16(p) + 2;
1080  if (p - avctx->extradata + nalsize > avctx->extradata_size)
1081  return -1;
1082  if (decode_nal_units(h, p, nalsize) < 0) {
1083  av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
1084  return -1;
1085  }
1086  p += nalsize;
1087  }
1088  // Now store right nal length size, that will be use to parse all other nals
1089  h->nal_length_size = (avctx->extradata[4] & 0x03) + 1;
1090  } else {
1091  h->is_avc = 0;
1092  if(decode_nal_units(h, avctx->extradata, avctx->extradata_size) < 0)
1093  return -1;
1094  }
1095  return 0;
1096 }
1097 
1099  H264Context *h= avctx->priv_data;
1100  MpegEncContext * const s = &h->s;
1101  int i;
1102 
1104 
1105  s->avctx = avctx;
1106  common_init(h);
1107 
1108  s->out_format = FMT_H264;
1109  s->workaround_bugs= avctx->workaround_bugs;
1110 
1111  // set defaults
1112 // s->decode_mb= ff_h263_decode_mb;
1113  s->quarter_sample = 1;
1114  if(!avctx->has_b_frames)
1115  s->low_delay= 1;
1116 
1118 
1120 
1121  h->pixel_shift = 0;
1122  h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1123 
1124  h->thread_context[0] = h;
1125  h->outputed_poc = h->next_outputed_poc = INT_MIN;
1126  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1127  h->last_pocs[i] = INT_MIN;
1128  h->prev_poc_msb= 1<<16;
1129  h->x264_build = -1;
1130  ff_h264_reset_sei(h);
1131  if(avctx->codec_id == CODEC_ID_H264){
1132  if(avctx->ticks_per_frame == 1){
1133  s->avctx->time_base.den *=2;
1134  }
1135  avctx->ticks_per_frame = 2;
1136  }
1137 
1138  if(avctx->extradata_size > 0 && avctx->extradata &&
1140  return -1;
1141 
1144  s->low_delay = 0;
1145  }
1146 
1147  return 0;
1148 }
1149 
1150 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b)+(size))))
1151 static void copy_picture_range(Picture **to, Picture **from, int count, MpegEncContext *new_base, MpegEncContext *old_base)
1152 {
1153  int i;
1154 
1155  for (i=0; i<count; i++){
1156  assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1157  IN_RANGE(from[i], old_base->picture, sizeof(Picture) * old_base->picture_count) ||
1158  !from[i]));
1159  to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1160  }
1161 }
1162 
1163 static void copy_parameter_set(void **to, void **from, int count, int size)
1164 {
1165  int i;
1166 
1167  for (i=0; i<count; i++){
1168  if (to[i] && !from[i]) av_freep(&to[i]);
1169  else if (from[i] && !to[i]) to[i] = av_malloc(size);
1170 
1171  if (from[i]) memcpy(to[i], from[i], size);
1172  }
1173 }
1174 
1176  H264Context *h= avctx->priv_data;
1177 
1178  if (!avctx->internal->is_copy)
1179  return 0;
1180  memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1181  memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1182 
1183  return 0;
1184 }
1185 
1186 #define copy_fields(to, from, start_field, end_field) memcpy(&to->start_field, &from->start_field, (char*)&to->end_field - (char*)&to->start_field)
1188  H264Context *h= dst->priv_data, *h1= src->priv_data;
1189  MpegEncContext * const s = &h->s, * const s1 = &h1->s;
1190  int inited = s->context_initialized, err;
1191  int i;
1192 
1193  if(dst == src || !s1->context_initialized) return 0;
1194 
1195  err = ff_mpeg_update_thread_context(dst, src);
1196  if(err) return err;
1197 
1198  //FIXME handle width/height changing
1199  if(!inited){
1200  for(i = 0; i < MAX_SPS_COUNT; i++)
1201  av_freep(h->sps_buffers + i);
1202 
1203  for(i = 0; i < MAX_PPS_COUNT; i++)
1204  av_freep(h->pps_buffers + i);
1205 
1206  memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
1207  memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1208  memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1209  if (ff_h264_alloc_tables(h) < 0) {
1210  av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1211  return AVERROR(ENOMEM);
1212  }
1213  context_init(h);
1214 
1215  for(i=0; i<2; i++){
1216  h->rbsp_buffer[i] = NULL;
1217  h->rbsp_buffer_size[i] = 0;
1218  }
1219 
1220  h->thread_context[0] = h;
1221 
1222  // frame_start may not be called for the next thread (if it's decoding a bottom field)
1223  // so this has to be allocated here
1224  h->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
1225 
1226  s->dsp.clear_blocks(h->mb);
1227  s->dsp.clear_blocks(h->mb+(24*16<<h->pixel_shift));
1228  }
1229 
1230  //extradata/NAL handling
1231  h->is_avc = h1->is_avc;
1232 
1233  //SPS/PPS
1234  copy_parameter_set((void**)h->sps_buffers, (void**)h1->sps_buffers, MAX_SPS_COUNT, sizeof(SPS));
1235  h->sps = h1->sps;
1236  copy_parameter_set((void**)h->pps_buffers, (void**)h1->pps_buffers, MAX_PPS_COUNT, sizeof(PPS));
1237  h->pps = h1->pps;
1238 
1239  //Dequantization matrices
1240  //FIXME these are big - can they be only copied when PPS changes?
1241  copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1242 
1243  for(i=0; i<6; i++)
1244  h->dequant4_coeff[i] = h->dequant4_buffer[0] + (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1245 
1246  for(i=0; i<6; i++)
1247  h->dequant8_coeff[i] = h->dequant8_buffer[0] + (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1248 
1249  h->dequant_coeff_pps = h1->dequant_coeff_pps;
1250 
1251  //POC timing
1252  copy_fields(h, h1, poc_lsb, redundant_pic_count);
1253 
1254  //reference lists
1255  copy_fields(h, h1, ref_count, list_count);
1256  copy_fields(h, h1, ref_list, intra_gb);
1257  copy_fields(h, h1, short_ref, cabac_init_idc);
1258 
1259  copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1260  copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1261  copy_picture_range(h->delayed_pic, h1->delayed_pic, MAX_DELAYED_PIC_COUNT+2, s, s1);
1262 
1263  h->last_slice_type = h1->last_slice_type;
1264 
1265  if(!s->current_picture_ptr) return 0;
1266 
1267  if(!s->dropable) {
1269  h->prev_poc_msb = h->poc_msb;
1270  h->prev_poc_lsb = h->poc_lsb;
1271  }
1273  h->prev_frame_num = h->frame_num;
1275 
1276  return err;
1277 }
1278 
1280  MpegEncContext * const s = &h->s;
1281  int i;
1282  const int pixel_shift = h->pixel_shift;
1283 
1284  if(MPV_frame_start(s, s->avctx) < 0)
1285  return -1;
1286  ff_er_frame_start(s);
1287  /*
1288  * MPV_frame_start uses pict_type to derive key_frame.
1289  * This is incorrect for H.264; IDR markings must be used.
1290  * Zero here; IDR markings per slice in frame or fields are ORed in later.
1291  * See decode_nal_units().
1292  */
1295 
1296  assert(s->linesize && s->uvlinesize);
1297 
1298  for(i=0; i<16; i++){
1299  h->block_offset[i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
1300  h->block_offset[48+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
1301  }
1302  for(i=0; i<16; i++){
1303  h->block_offset[16+i]=
1304  h->block_offset[32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1305  h->block_offset[48+16+i]=
1306  h->block_offset[48+32+i]= (4*((scan8[i] - scan8[0])&7) << pixel_shift) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
1307  }
1308 
1309  /* can't be in alloc_tables because linesize isn't known there.
1310  * FIXME: redo bipred weight to not require extra buffer? */
1311  for(i = 0; i < s->slice_context_count; i++)
1312  if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
1314 
1315  /* some macroblocks can be accessed before they're available in case of lost slices, mbaff or threading*/
1316  memset(h->slice_table, -1, (s->mb_height*s->mb_stride-1) * sizeof(*h->slice_table));
1317 
1318 // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.f.reference /*|| h->contains_intra*/ || 1;
1319 
1320  // We mark the current picture as non-reference after allocating it, so
1321  // that if we break out due to an error it can be released automatically
1322  // in the next MPV_frame_start().
1323  // SVQ3 as well as most other codecs have only last/next/current and thus
1324  // get released even with set reference, besides SVQ3 and others do not
1325  // mark frames as reference later "naturally".
1326  if(s->codec_id != CODEC_ID_SVQ3)
1328 
1330  s->current_picture_ptr->field_poc[1]= INT_MAX;
1331 
1332  h->next_output_pic = NULL;
1333 
1334  assert(s->current_picture_ptr->long_ref==0);
1335 
1336  return 0;
1337 }
1338 
1347 static void decode_postinit(H264Context *h, int setup_finished){
1348  MpegEncContext * const s = &h->s;
1349  Picture *out = s->current_picture_ptr;
1350  Picture *cur = s->current_picture_ptr;
1351  int i, pics, out_of_order, out_idx;
1352  int invalid = 0, cnt = 0;
1353 
1356 
1357  if (h->next_output_pic) return;
1358 
1359  if (cur->field_poc[0]==INT_MAX || cur->field_poc[1]==INT_MAX) {
1360  //FIXME: if we have two PAFF fields in one packet, we can't start the next thread here.
1361  //If we have one field per packet, we can. The check in decode_nal_units() is not good enough
1362  //to find this yet, so we assume the worst for now.
1363  //if (setup_finished)
1364  // ff_thread_finish_setup(s->avctx);
1365  return;
1366  }
1367 
1368  cur->f.interlaced_frame = 0;
1369  cur->f.repeat_pict = 0;
1370 
1371  /* Signal interlacing information externally. */
1372  /* Prioritize picture timing SEI information over used decoding process if it exists. */
1373 
1375  switch (h->sei_pic_struct)
1376  {
1377  case SEI_PIC_STRUCT_FRAME:
1378  break;
1381  cur->f.interlaced_frame = 1;
1382  break;
1386  cur->f.interlaced_frame = 1;
1387  else
1388  // try to flag soft telecine progressive
1390  break;
1393  // Signal the possibility of telecined film externally (pic_struct 5,6)
1394  // From these hints, let the applications decide if they apply deinterlacing.
1395  cur->f.repeat_pict = 1;
1396  break;
1398  // Force progressive here, as doubling interlaced frame is a bad idea.
1399  cur->f.repeat_pict = 2;
1400  break;
1402  cur->f.repeat_pict = 4;
1403  break;
1404  }
1405 
1407  cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1408  }else{
1409  /* Derive interlacing flag from used decoding process. */
1411  }
1413 
1414  if (cur->field_poc[0] != cur->field_poc[1]){
1415  /* Derive top_field_first from field pocs. */
1416  cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1417  }else{
1418  if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1419  /* Use picture timing SEI information. Even if it is a information of a past frame, better than nothing. */
1422  cur->f.top_field_first = 1;
1423  else
1424  cur->f.top_field_first = 0;
1425  }else{
1426  /* Most likely progressive */
1427  cur->f.top_field_first = 0;
1428  }
1429  }
1430 
1431  //FIXME do something with unavailable reference frames
1432 
1433  /* Sort B-frames into display order */
1434 
1438  s->low_delay = 0;
1439  }
1440 
1444  s->low_delay= 0;
1445  }
1446 
1447  pics = 0;
1448  while(h->delayed_pic[pics]) pics++;
1449 
1450  assert(pics <= MAX_DELAYED_PIC_COUNT);
1451 
1452  h->delayed_pic[pics++] = cur;
1453  if (cur->f.reference == 0)
1454  cur->f.reference = DELAYED_PIC_REF;
1455 
1456  /* Frame reordering. This code takes pictures from coding order and sorts
1457  * them by their incremental POC value into display order. It supports POC
1458  * gaps, MMCO reset codes and random resets.
1459  * A "display group" can start either with a IDR frame (f.key_frame = 1),
1460  * and/or can be closed down with a MMCO reset code. In sequences where
1461  * there is no delay, we can't detect that (since the frame was already
1462  * output to the user), so we also set h->mmco_reset to detect the MMCO
1463  * reset code.
1464  * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
1465  * we increase the delay between input and output. All frames affected by
1466  * the lag (e.g. those that should have been output before another frame
1467  * that we already returned to the user) will be dropped. This is a bug
1468  * that we will fix later. */
1469  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
1470  cnt += out->poc < h->last_pocs[i];
1471  invalid += out->poc == INT_MIN;
1472  }
1473  if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
1474  h->mmco_reset = 2;
1475  if (pics > 1)
1476  h->delayed_pic[pics - 2]->mmco_reset = 2;
1477  }
1478  if (h->mmco_reset || cur->f.key_frame) {
1479  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1480  h->last_pocs[i] = INT_MIN;
1481  cnt = 0;
1482  invalid = MAX_DELAYED_PIC_COUNT;
1483  }
1484  out = h->delayed_pic[0];
1485  out_idx = 0;
1486  for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
1487  !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
1488  {
1489  if(h->delayed_pic[i]->poc < out->poc){
1490  out = h->delayed_pic[i];
1491  out_idx = i;
1492  }
1493  }
1494  if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
1495  h->next_outputed_poc = INT_MIN;
1496  out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
1497 
1499  { }
1500  else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
1501  s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
1502  if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
1503  s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
1504  }
1505  s->low_delay = 0;
1506  } else if (s->low_delay &&
1507  ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
1508  cur->f.pict_type == AV_PICTURE_TYPE_B)) {
1509  s->low_delay = 0;
1510  s->avctx->has_b_frames++;
1511  }
1512 
1513  if(pics > s->avctx->has_b_frames){
1514  out->f.reference &= ~DELAYED_PIC_REF;
1515  out->owner2 = s; // for frame threading, the owner must be the second field's thread
1516  // or else the first thread can release the picture and reuse it unsafely
1517  for(i=out_idx; h->delayed_pic[i]; i++)
1518  h->delayed_pic[i] = h->delayed_pic[i+1];
1519  }
1520  memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
1521  h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
1522  if(!out_of_order && pics > s->avctx->has_b_frames){
1523  h->next_output_pic = out;
1524  if (out->mmco_reset) {
1525  if (out_idx > 0) {
1526  h->next_outputed_poc = out->poc;
1527  h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
1528  } else {
1529  h->next_outputed_poc = INT_MIN;
1530  }
1531  } else {
1532  if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
1533  h->next_outputed_poc = INT_MIN;
1534  } else {
1535  h->next_outputed_poc = out->poc;
1536  }
1537  }
1538  h->mmco_reset = 0;
1539  }else{
1540  av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
1541  }
1542 
1543  if (setup_finished)
1545 }
1546 
1547 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1548  uint8_t *src_cb, uint8_t *src_cr,
1549  int linesize, int uvlinesize, int simple)
1550 {
1551  MpegEncContext * const s = &h->s;
1552  uint8_t *top_border;
1553  int top_idx = 1;
1554  const int pixel_shift = h->pixel_shift;
1555  int chroma444 = CHROMA444;
1556  int chroma422 = CHROMA422;
1557 
1558  src_y -= linesize;
1559  src_cb -= uvlinesize;
1560  src_cr -= uvlinesize;
1561 
1562  if(!simple && FRAME_MBAFF){
1563  if(s->mb_y&1){
1564  if(!MB_MBAFF){
1565  top_border = h->top_borders[0][s->mb_x];
1566  AV_COPY128(top_border, src_y + 15*linesize);
1567  if (pixel_shift)
1568  AV_COPY128(top_border+16, src_y+15*linesize+16);
1569  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1570  if(chroma444){
1571  if (pixel_shift){
1572  AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1573  AV_COPY128(top_border+48, src_cb + 15*uvlinesize+16);
1574  AV_COPY128(top_border+64, src_cr + 15*uvlinesize);
1575  AV_COPY128(top_border+80, src_cr + 15*uvlinesize+16);
1576  } else {
1577  AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
1578  AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
1579  }
1580  } else if(chroma422) {
1581  if (pixel_shift) {
1582  AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
1583  AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
1584  } else {
1585  AV_COPY64(top_border+16, src_cb + 15*uvlinesize);
1586  AV_COPY64(top_border+24, src_cr + 15*uvlinesize);
1587  }
1588  } else {
1589  if (pixel_shift) {
1590  AV_COPY128(top_border+32, src_cb+7*uvlinesize);
1591  AV_COPY128(top_border+48, src_cr+7*uvlinesize);
1592  } else {
1593  AV_COPY64(top_border+16, src_cb+7*uvlinesize);
1594  AV_COPY64(top_border+24, src_cr+7*uvlinesize);
1595  }
1596  }
1597  }
1598  }
1599  }else if(MB_MBAFF){
1600  top_idx = 0;
1601  }else
1602  return;
1603  }
1604 
1605  top_border = h->top_borders[top_idx][s->mb_x];
1606  // There are two lines saved, the line above the the top macroblock of a pair,
1607  // and the line above the bottom macroblock
1608  AV_COPY128(top_border, src_y + 16*linesize);
1609  if (pixel_shift)
1610  AV_COPY128(top_border+16, src_y+16*linesize+16);
1611 
1612  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1613  if(chroma444){
1614  if (pixel_shift){
1615  AV_COPY128(top_border+32, src_cb + 16*linesize);
1616  AV_COPY128(top_border+48, src_cb + 16*linesize+16);
1617  AV_COPY128(top_border+64, src_cr + 16*linesize);
1618  AV_COPY128(top_border+80, src_cr + 16*linesize+16);
1619  } else {
1620  AV_COPY128(top_border+16, src_cb + 16*linesize);
1621  AV_COPY128(top_border+32, src_cr + 16*linesize);
1622  }
1623  } else if(chroma422) {
1624  if (pixel_shift) {
1625  AV_COPY128(top_border+32, src_cb+16*uvlinesize);
1626  AV_COPY128(top_border+48, src_cr+16*uvlinesize);
1627  } else {
1628  AV_COPY64(top_border+16, src_cb+16*uvlinesize);
1629  AV_COPY64(top_border+24, src_cr+16*uvlinesize);
1630  }
1631  } else {
1632  if (pixel_shift) {
1633  AV_COPY128(top_border+32, src_cb+8*uvlinesize);
1634  AV_COPY128(top_border+48, src_cr+8*uvlinesize);
1635  } else {
1636  AV_COPY64(top_border+16, src_cb+8*uvlinesize);
1637  AV_COPY64(top_border+24, src_cr+8*uvlinesize);
1638  }
1639  }
1640  }
1641 }
1642 
1643 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1644  uint8_t *src_cb, uint8_t *src_cr,
1645  int linesize, int uvlinesize,
1646  int xchg, int chroma444,
1647  int simple, int pixel_shift){
1648  MpegEncContext * const s = &h->s;
1649  int deblock_topleft;
1650  int deblock_top;
1651  int top_idx = 1;
1652  uint8_t *top_border_m1;
1653  uint8_t *top_border;
1654 
1655  if(!simple && FRAME_MBAFF){
1656  if(s->mb_y&1){
1657  if(!MB_MBAFF)
1658  return;
1659  }else{
1660  top_idx = MB_MBAFF ? 0 : 1;
1661  }
1662  }
1663 
1664  if(h->deblocking_filter == 2) {
1665  deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1666  deblock_top = h->top_type;
1667  } else {
1668  deblock_topleft = (s->mb_x > 0);
1669  deblock_top = (s->mb_y > !!MB_FIELD);
1670  }
1671 
1672  src_y -= linesize + 1 + pixel_shift;
1673  src_cb -= uvlinesize + 1 + pixel_shift;
1674  src_cr -= uvlinesize + 1 + pixel_shift;
1675 
1676  top_border_m1 = h->top_borders[top_idx][s->mb_x-1];
1677  top_border = h->top_borders[top_idx][s->mb_x];
1678 
1679 #define XCHG(a,b,xchg)\
1680  if (pixel_shift) {\
1681  if (xchg) {\
1682  AV_SWAP64(b+0,a+0);\
1683  AV_SWAP64(b+8,a+8);\
1684  } else {\
1685  AV_COPY128(b,a); \
1686  }\
1687  } else \
1688 if (xchg) AV_SWAP64(b,a);\
1689 else AV_COPY64(b,a);
1690 
1691  if(deblock_top){
1692  if(deblock_topleft){
1693  XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
1694  }
1695  XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1696  XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1697  if(s->mb_x+1 < s->mb_width){
1698  XCHG(h->top_borders[top_idx][s->mb_x+1], src_y + (17 << pixel_shift), 1);
1699  }
1700  }
1701  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1702  if(chroma444){
1703  if(deblock_topleft){
1704  XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1705  XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1706  }
1707  XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1708  XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1709  XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1710  XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1711  if(s->mb_x+1 < s->mb_width){
1712  XCHG(h->top_borders[top_idx][s->mb_x+1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1713  XCHG(h->top_borders[top_idx][s->mb_x+1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1714  }
1715  } else {
1716  if(deblock_top){
1717  if(deblock_topleft){
1718  XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1719  XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1720  }
1721  XCHG(top_border + (16 << pixel_shift), src_cb+1+pixel_shift, 1);
1722  XCHG(top_border + (24 << pixel_shift), src_cr+1+pixel_shift, 1);
1723  }
1724  }
1725  }
1726 }
1727 
1728 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth, int index) {
1729  if (high_bit_depth) {
1730  return AV_RN32A(((int32_t*)mb) + index);
1731  } else
1732  return AV_RN16A(mb + index);
1733 }
1734 
1735 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth, int index, int value) {
1736  if (high_bit_depth) {
1737  AV_WN32A(((int32_t*)mb) + index, value);
1738  } else
1739  AV_WN16A(mb + index, value);
1740 }
1741 
1742 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1743  int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1744 {
1745  MpegEncContext * const s = &h->s;
1746  void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1747  void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1748  int i;
1749  int qscale = p == 0 ? s->qscale : h->chroma_qp[p-1];
1750  block_offset += 16*p;
1751  if(IS_INTRA4x4(mb_type)){
1752  if(simple || !s->encoding){
1753  if(IS_8x8DCT(mb_type)){
1754  if(transform_bypass){
1755  idct_dc_add =
1756  idct_add = s->dsp.add_pixels8;
1757  }else{
1758  idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1760  }
1761  for(i=0; i<16; i+=4){
1762  uint8_t * const ptr= dest_y + block_offset[i];
1763  const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1764  if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1765  h->hpc.pred8x8l_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1766  }else{
1767  const int nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1768  h->hpc.pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
1769  (h->topright_samples_available<<i)&0x4000, linesize);
1770  if(nnz){
1771  if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1772  idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1773  else
1774  idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1775  }
1776  }
1777  }
1778  }else{
1779  if(transform_bypass){
1780  idct_dc_add =
1781  idct_add = s->dsp.add_pixels4;
1782  }else{
1783  idct_dc_add = h->h264dsp.h264_idct_dc_add;
1785  }
1786  for(i=0; i<16; i++){
1787  uint8_t * const ptr= dest_y + block_offset[i];
1788  const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
1789 
1790  if(transform_bypass && h->sps.profile_idc==244 && dir<=1){
1791  h->hpc.pred4x4_add[dir](ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1792  }else{
1793  uint8_t *topright;
1794  int nnz, tr;
1795  uint64_t tr_high;
1796  if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
1797  const int topright_avail= (h->topright_samples_available<<i)&0x8000;
1798  assert(s->mb_y || linesize <= block_offset[i]);
1799  if(!topright_avail){
1800  if (pixel_shift) {
1801  tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
1802  topright= (uint8_t*) &tr_high;
1803  } else {
1804  tr= ptr[3 - linesize]*0x01010101u;
1805  topright= (uint8_t*) &tr;
1806  }
1807  }else
1808  topright= ptr + (4 << pixel_shift) - linesize;
1809  }else
1810  topright= NULL;
1811 
1812  h->hpc.pred4x4[ dir ](ptr, topright, linesize);
1813  nnz = h->non_zero_count_cache[ scan8[i+p*16] ];
1814  if(nnz){
1815  if(is_h264){
1816  if(nnz == 1 && dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1817  idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1818  else
1819  idct_add (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
1820  } else if (CONFIG_SVQ3_DECODER)
1821  ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
1822  }
1823  }
1824  }
1825  }
1826  }
1827  }else{
1828  h->hpc.pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
1829  if(is_h264){
1831  if(!transform_bypass)
1832  h->h264dsp.h264_luma_dc_dequant_idct(h->mb+(p*256 << pixel_shift), h->mb_luma_dc[p], h->dequant4_coeff[p][qscale][0]);
1833  else{
1834  static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
1835  8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
1836  for(i = 0; i < 16; i++)
1837  dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
1838  }
1839  }
1840  } else if (CONFIG_SVQ3_DECODER)
1841  ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
1842  }
1843 }
1844 
1845 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type, int is_h264, int simple, int transform_bypass,
1846  int pixel_shift, int *block_offset, int linesize, uint8_t *dest_y, int p)
1847 {
1848  MpegEncContext * const s = &h->s;
1849  void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1850  int i;
1851  block_offset += 16*p;
1852  if(!IS_INTRA4x4(mb_type)){
1853  if(is_h264){
1854  if(IS_INTRA16x16(mb_type)){
1855  if(transform_bypass){
1857  h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize);
1858  }else{
1859  for(i=0; i<16; i++){
1860  if(h->non_zero_count_cache[ scan8[i+p*16] ] || dctcoef_get(h->mb, pixel_shift, i*16+p*256))
1861  s->dsp.add_pixels4(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1862  }
1863  }
1864  }else{
1865  h->h264dsp.h264_idct_add16intra(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1866  }
1867  }else if(h->cbp&15){
1868  if(transform_bypass){
1869  const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1870  idct_add= IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4;
1871  for(i=0; i<16; i+=di){
1872  if(h->non_zero_count_cache[ scan8[i+p*16] ]){
1873  idct_add(dest_y + block_offset[i], h->mb + (i*16+p*256 << pixel_shift), linesize);
1874  }
1875  }
1876  }else{
1877  if(IS_8x8DCT(mb_type)){
1878  h->h264dsp.h264_idct8_add4(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1879  }else{
1880  h->h264dsp.h264_idct_add16(dest_y, block_offset, h->mb + (p*256 << pixel_shift), linesize, h->non_zero_count_cache+p*5*8);
1881  }
1882  }
1883  }
1884  } else if (CONFIG_SVQ3_DECODER) {
1885  for(i=0; i<16; i++){
1886  if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
1887  uint8_t * const ptr= dest_y + block_offset[i];
1888  ff_svq3_add_idct_c(ptr, h->mb + i*16 + p*256, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1889  }
1890  }
1891  }
1892  }
1893 }
1894 
1895 static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
1896 {
1897  MpegEncContext * const s = &h->s;
1898  const int mb_x= s->mb_x;
1899  const int mb_y= s->mb_y;
1900  const int mb_xy= h->mb_xy;
1901  const int mb_type = s->current_picture.f.mb_type[mb_xy];
1902  uint8_t *dest_y, *dest_cb, *dest_cr;
1903  int linesize, uvlinesize /*dct_offset*/;
1904  int i, j;
1905  int *block_offset = &h->block_offset[0];
1906  const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
1907  /* is_h264 should always be true if SVQ3 is disabled. */
1908  const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
1909  void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1910  const int block_h = 16 >> s->chroma_y_shift;
1911  const int chroma422 = CHROMA422;
1912 
1913  dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
1914  dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1915  dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
1916 
1917  s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
1918  s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
1919 
1920  h->list_counts[mb_xy]= h->list_count;
1921 
1922  if (!simple && MB_FIELD) {
1923  linesize = h->mb_linesize = s->linesize * 2;
1924  uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
1925  block_offset = &h->block_offset[48];
1926  if(mb_y&1){ //FIXME move out of this function?
1927  dest_y -= s->linesize*15;
1928  dest_cb-= s->uvlinesize * (block_h - 1);
1929  dest_cr-= s->uvlinesize * (block_h - 1);
1930  }
1931  if(FRAME_MBAFF) {
1932  int list;
1933  for(list=0; list<h->list_count; list++){
1934  if(!USES_LIST(mb_type, list))
1935  continue;
1936  if(IS_16X16(mb_type)){
1937  int8_t *ref = &h->ref_cache[list][scan8[0]];
1938  fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
1939  }else{
1940  for(i=0; i<16; i+=4){
1941  int ref = h->ref_cache[list][scan8[i]];
1942  if(ref >= 0)
1943  fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
1944  }
1945  }
1946  }
1947  }
1948  } else {
1949  linesize = h->mb_linesize = s->linesize;
1950  uvlinesize = h->mb_uvlinesize = s->uvlinesize;
1951 // dct_offset = s->linesize * 16;
1952  }
1953 
1954  if (!simple && IS_INTRA_PCM(mb_type)) {
1955  if (pixel_shift) {
1956  const int bit_depth = h->sps.bit_depth_luma;
1957  int j;
1958  GetBitContext gb;
1959  init_get_bits(&gb, (uint8_t*)h->mb, 384*bit_depth);
1960 
1961  for (i = 0; i < 16; i++) {
1962  uint16_t *tmp_y = (uint16_t*)(dest_y + i*linesize);
1963  for (j = 0; j < 16; j++)
1964  tmp_y[j] = get_bits(&gb, bit_depth);
1965  }
1966  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1967  if (!h->sps.chroma_format_idc) {
1968  for (i = 0; i < block_h; i++) {
1969  uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1970  for (j = 0; j < 8; j++) {
1971  tmp_cb[j] = 1 << (bit_depth - 1);
1972  }
1973  }
1974  for (i = 0; i < block_h; i++) {
1975  uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1976  for (j = 0; j < 8; j++) {
1977  tmp_cr[j] = 1 << (bit_depth - 1);
1978  }
1979  }
1980  } else {
1981  for (i = 0; i < block_h; i++) {
1982  uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
1983  for (j = 0; j < 8; j++)
1984  tmp_cb[j] = get_bits(&gb, bit_depth);
1985  }
1986  for (i = 0; i < block_h; i++) {
1987  uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
1988  for (j = 0; j < 8; j++)
1989  tmp_cr[j] = get_bits(&gb, bit_depth);
1990  }
1991  }
1992  }
1993  } else {
1994  for (i=0; i<16; i++) {
1995  memcpy(dest_y + i* linesize, h->mb + i*8, 16);
1996  }
1997  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
1998  if (!h->sps.chroma_format_idc) {
1999  for (i = 0; i < block_h; i++) {
2000  memset(dest_cb + i*uvlinesize, 128, 8);
2001  memset(dest_cr + i*uvlinesize, 128, 8);
2002  }
2003  } else {
2004  for (i = 0; i < block_h; i++) {
2005  memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4, 8);
2006  memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4, 8);
2007  }
2008  }
2009  }
2010  }
2011  } else {
2012  if(IS_INTRA(mb_type)){
2013  if(h->deblocking_filter)
2014  xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1, 0, simple, pixel_shift);
2015 
2016  if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
2017  h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2018  h->hpc.pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2019  }
2020 
2021  hl_decode_mb_predict_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2022 
2023  if(h->deblocking_filter)
2024  xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
2025  }else if(is_h264){
2026  if (chroma422) {
2027  hl_motion_422(h, dest_y, dest_cb, dest_cr,
2032  pixel_shift);
2033  } else {
2034  hl_motion_420(h, dest_y, dest_cb, dest_cr,
2039  pixel_shift);
2040  }
2041  }
2042 
2043  hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
2044 
2045  if((simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) && (h->cbp&0x30)){
2046  uint8_t *dest[2] = {dest_cb, dest_cr};
2047  if(transform_bypass){
2048  if(IS_INTRA(mb_type) && h->sps.profile_idc==244 && (h->chroma_pred_mode==VERT_PRED8x8 || h->chroma_pred_mode==HOR_PRED8x8)){
2049  h->hpc.pred8x8_add[h->chroma_pred_mode](dest[0], block_offset + 16, h->mb + (16*16*1 << pixel_shift), uvlinesize);
2050  h->hpc.pred8x8_add[h->chroma_pred_mode](dest[1], block_offset + 32, h->mb + (16*16*2 << pixel_shift), uvlinesize);
2051  }else{
2052  idct_add = s->dsp.add_pixels4;
2053  for(j=1; j<3; j++){
2054  for(i=j*16; i<j*16+4; i++){
2055  if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2056  idct_add (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
2057  }
2058  if (chroma422) {
2059  for(i=j*16+4; i<j*16+8; i++){
2060  if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
2061  idct_add (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
2062  }
2063  }
2064  }
2065  }
2066  }else{
2067  if(is_h264){
2068  int qp[2];
2069  if (chroma422) {
2070  qp[0] = h->chroma_qp[0] + 3;
2071  qp[1] = h->chroma_qp[1] + 3;
2072  } else {
2073  qp[0] = h->chroma_qp[0];
2074  qp[1] = h->chroma_qp[1];
2075  }
2077  h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
2079  h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
2080  h->h264dsp.h264_idct_add8(dest, block_offset,
2081  h->mb, uvlinesize,
2083  } else if (CONFIG_SVQ3_DECODER) {
2084  h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
2085  h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
2086  for(j=1; j<3; j++){
2087  for(i=j*16; i<j*16+4; i++){
2088  if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2089  uint8_t * const ptr= dest[j-1] + block_offset[i];
2090  ff_svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);
2091  }
2092  }
2093  }
2094  }
2095  }
2096  }
2097  }
2098  if(h->cbp || IS_INTRA(mb_type))
2099  {
2100  s->dsp.clear_blocks(h->mb);
2101  s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2102  }
2103 }
2104 
2105 static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simple, int pixel_shift){
2106  MpegEncContext * const s = &h->s;
2107  const int mb_x= s->mb_x;
2108  const int mb_y= s->mb_y;
2109  const int mb_xy= h->mb_xy;
2110  const int mb_type = s->current_picture.f.mb_type[mb_xy];
2111  uint8_t *dest[3];
2112  int linesize;
2113  int i, j, p;
2114  int *block_offset = &h->block_offset[0];
2115  const int transform_bypass = !simple && (s->qscale == 0 && h->sps.transform_bypass);
2116  const int plane_count = (simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)) ? 3 : 1;
2117 
2118  for (p = 0; p < plane_count; p++)
2119  {
2120  dest[p] = s->current_picture.f.data[p] + ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
2121  s->dsp.prefetch(dest[p] + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
2122  }
2123 
2124  h->list_counts[mb_xy]= h->list_count;
2125 
2126  if (!simple && MB_FIELD) {
2127  linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize * 2;
2128  block_offset = &h->block_offset[48];
2129  if(mb_y&1) //FIXME move out of this function?
2130  for (p = 0; p < 3; p++)
2131  dest[p] -= s->linesize*15;
2132  if(FRAME_MBAFF) {
2133  int list;
2134  for(list=0; list<h->list_count; list++){
2135  if(!USES_LIST(mb_type, list))
2136  continue;
2137  if(IS_16X16(mb_type)){
2138  int8_t *ref = &h->ref_cache[list][scan8[0]];
2139  fill_rectangle(ref, 4, 4, 8, (16+*ref)^(s->mb_y&1), 1);
2140  }else{
2141  for(i=0; i<16; i+=4){
2142  int ref = h->ref_cache[list][scan8[i]];
2143  if(ref >= 0)
2144  fill_rectangle(&h->ref_cache[list][scan8[i]], 2, 2, 8, (16+ref)^(s->mb_y&1), 1);
2145  }
2146  }
2147  }
2148  }
2149  } else {
2150  linesize = h->mb_linesize = h->mb_uvlinesize = s->linesize;
2151  }
2152 
2153  if (!simple && IS_INTRA_PCM(mb_type)) {
2154  if (pixel_shift) {
2155  const int bit_depth = h->sps.bit_depth_luma;
2156  GetBitContext gb;
2157  init_get_bits(&gb, (uint8_t*)h->mb, 768*bit_depth);
2158 
2159  for (p = 0; p < plane_count; p++) {
2160  for (i = 0; i < 16; i++) {
2161  uint16_t *tmp = (uint16_t*)(dest[p] + i*linesize);
2162  for (j = 0; j < 16; j++)
2163  tmp[j] = get_bits(&gb, bit_depth);
2164  }
2165  }
2166  } else {
2167  for (p = 0; p < plane_count; p++) {
2168  for (i = 0; i < 16; i++) {
2169  memcpy(dest[p] + i*linesize, h->mb + p*128 + i*8, 16);
2170  }
2171  }
2172  }
2173  } else {
2174  if(IS_INTRA(mb_type)){
2175  if(h->deblocking_filter)
2176  xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 1, 1, simple, pixel_shift);
2177 
2178  for (p = 0; p < plane_count; p++)
2179  hl_decode_mb_predict_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2180 
2181  if(h->deblocking_filter)
2182  xchg_mb_border(h, dest[0], dest[1], dest[2], linesize, linesize, 0, 1, simple, pixel_shift);
2183  }else{
2184  hl_motion(h, dest[0], dest[1], dest[2],
2188  h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
2189  }
2190 
2191  for (p = 0; p < plane_count; p++)
2192  hl_decode_mb_idct_luma(h, mb_type, 1, simple, transform_bypass, pixel_shift, block_offset, linesize, dest[p], p);
2193  }
2194  if(h->cbp || IS_INTRA(mb_type))
2195  {
2196  s->dsp.clear_blocks(h->mb);
2197  s->dsp.clear_blocks(h->mb+(24*16<<pixel_shift));
2198  }
2199 }
2200 
2204 #define hl_decode_mb_simple(sh, bits) \
2205 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
2206  hl_decode_mb_internal(h, 1, sh); \
2207 }
2209 hl_decode_mb_simple(1, 16)
2210 
2214 static void av_noinline hl_decode_mb_complex(H264Context *h){
2215  hl_decode_mb_internal(h, 0, h->pixel_shift);
2216 }
2217 
2220 }
2221 
2223  hl_decode_mb_444_internal(h, 1, 0);
2224 }
2225 
2227  MpegEncContext * const s = &h->s;
2228  const int mb_xy= h->mb_xy;
2229  const int mb_type = s->current_picture.f.mb_type[mb_xy];
2230  int is_complex = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
2231 
2232  if (CHROMA444) {
2233  if(is_complex || h->pixel_shift)
2235  else
2237  } else if (is_complex) {
2238  hl_decode_mb_complex(h);
2239  } else if (h->pixel_shift) {
2240  hl_decode_mb_simple_16(h);
2241  } else
2242  hl_decode_mb_simple_8(h);
2243 }
2244 
2246  MpegEncContext * const s = &h->s;
2247  int list, i;
2248  int luma_def, chroma_def;
2249 
2250  h->use_weight= 0;
2251  h->use_weight_chroma= 0;
2253  if(h->sps.chroma_format_idc)
2255  luma_def = 1<<h->luma_log2_weight_denom;
2256  chroma_def = 1<<h->chroma_log2_weight_denom;
2257 
2258  for(list=0; list<2; list++){
2259  h->luma_weight_flag[list] = 0;
2260  h->chroma_weight_flag[list] = 0;
2261  for(i=0; i<h->ref_count[list]; i++){
2262  int luma_weight_flag, chroma_weight_flag;
2263 
2264  luma_weight_flag= get_bits1(&s->gb);
2265  if(luma_weight_flag){
2266  h->luma_weight[i][list][0]= get_se_golomb(&s->gb);
2267  h->luma_weight[i][list][1]= get_se_golomb(&s->gb);
2268  if( h->luma_weight[i][list][0] != luma_def
2269  || h->luma_weight[i][list][1] != 0) {
2270  h->use_weight= 1;
2271  h->luma_weight_flag[list]= 1;
2272  }
2273  }else{
2274  h->luma_weight[i][list][0]= luma_def;
2275  h->luma_weight[i][list][1]= 0;
2276  }
2277 
2278  if(h->sps.chroma_format_idc){
2279  chroma_weight_flag= get_bits1(&s->gb);
2280  if(chroma_weight_flag){
2281  int j;
2282  for(j=0; j<2; j++){
2283  h->chroma_weight[i][list][j][0]= get_se_golomb(&s->gb);
2284  h->chroma_weight[i][list][j][1]= get_se_golomb(&s->gb);
2285  if( h->chroma_weight[i][list][j][0] != chroma_def
2286  || h->chroma_weight[i][list][j][1] != 0) {
2287  h->use_weight_chroma= 1;
2288  h->chroma_weight_flag[list]= 1;
2289  }
2290  }
2291  }else{
2292  int j;
2293  for(j=0; j<2; j++){
2294  h->chroma_weight[i][list][j][0]= chroma_def;
2295  h->chroma_weight[i][list][j][1]= 0;
2296  }
2297  }
2298  }
2299  }
2300  if(h->slice_type_nos != AV_PICTURE_TYPE_B) break;
2301  }
2303  return 0;
2304 }
2305 
2311 static void implicit_weight_table(H264Context *h, int field){
2312  MpegEncContext * const s = &h->s;
2313  int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2314 
2315  for (i = 0; i < 2; i++) {
2316  h->luma_weight_flag[i] = 0;
2317  h->chroma_weight_flag[i] = 0;
2318  }
2319 
2320  if(field < 0){
2321  if (s->picture_structure == PICT_FRAME) {
2322  cur_poc = s->current_picture_ptr->poc;
2323  } else {
2324  cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2325  }
2326  if( h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
2327  && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
2328  h->use_weight= 0;
2329  h->use_weight_chroma= 0;
2330  return;
2331  }
2332  ref_start= 0;
2333  ref_count0= h->ref_count[0];
2334  ref_count1= h->ref_count[1];
2335  }else{
2336  cur_poc = s->current_picture_ptr->field_poc[field];
2337  ref_start= 16;
2338  ref_count0= 16+2*h->ref_count[0];
2339  ref_count1= 16+2*h->ref_count[1];
2340  }
2341 
2342  h->use_weight= 2;
2343  h->use_weight_chroma= 2;
2344  h->luma_log2_weight_denom= 5;
2346 
2347  for(ref0=ref_start; ref0 < ref_count0; ref0++){
2348  int poc0 = h->ref_list[0][ref0].poc;
2349  for(ref1=ref_start; ref1 < ref_count1; ref1++){
2350  int w = 32;
2351  if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2352  int poc1 = h->ref_list[1][ref1].poc;
2353  int td = av_clip(poc1 - poc0, -128, 127);
2354  if(td){
2355  int tb = av_clip(cur_poc - poc0, -128, 127);
2356  int tx = (16384 + (FFABS(td) >> 1)) / td;
2357  int dist_scale_factor = (tb*tx + 32) >> 8;
2358  if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
2359  w = 64 - dist_scale_factor;
2360  }
2361  }
2362  if(field<0){
2363  h->implicit_weight[ref0][ref1][0]=
2364  h->implicit_weight[ref0][ref1][1]= w;
2365  }else{
2366  h->implicit_weight[ref0][ref1][field]=w;
2367  }
2368  }
2369  }
2370 }
2371 
2375 static void idr(H264Context *h){
2377  h->prev_frame_num= 0;
2378  h->prev_frame_num_offset= 0;
2379  h->prev_poc_msb=
2380  h->prev_poc_lsb= 0;
2381 }
2382 
2383 /* forget old pics after a seek */
2384 static void flush_dpb(AVCodecContext *avctx){
2385  H264Context *h= avctx->priv_data;
2386  int i;
2387  for(i=0; i<MAX_DELAYED_PIC_COUNT; i++) {
2388  if(h->delayed_pic[i])
2389  h->delayed_pic[i]->f.reference = 0;
2390  h->delayed_pic[i]= NULL;
2391  }
2392  for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2393  h->last_pocs[i] = INT_MIN;
2394  h->outputed_poc=h->next_outputed_poc= INT_MIN;
2395  h->prev_interlaced_frame = 1;
2396  idr(h);
2397  if(h->s.current_picture_ptr)
2398  h->s.current_picture_ptr->f.reference = 0;
2399  h->s.first_field= 0;
2400  ff_h264_reset_sei(h);
2401  ff_mpeg_flush(avctx);
2402 }
2403 
2404 static int init_poc(H264Context *h){
2405  MpegEncContext * const s = &h->s;
2406  const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2407  int field_poc[2];
2408  Picture *cur = s->current_picture_ptr;
2409 
2411  if(h->frame_num < h->prev_frame_num)
2412  h->frame_num_offset += max_frame_num;
2413 
2414  if(h->sps.poc_type==0){
2415  const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2416 
2417  if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2418  h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2419  else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2420  h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2421  else
2422  h->poc_msb = h->prev_poc_msb;
2423 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2424  field_poc[0] =
2425  field_poc[1] = h->poc_msb + h->poc_lsb;
2426  if(s->picture_structure == PICT_FRAME)
2427  field_poc[1] += h->delta_poc_bottom;
2428  }else if(h->sps.poc_type==1){
2429  int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2430  int i;
2431 
2432  if(h->sps.poc_cycle_length != 0)
2433  abs_frame_num = h->frame_num_offset + h->frame_num;
2434  else
2435  abs_frame_num = 0;
2436 
2437  if(h->nal_ref_idc==0 && abs_frame_num > 0)
2438  abs_frame_num--;
2439 
2440  expected_delta_per_poc_cycle = 0;
2441  for(i=0; i < h->sps.poc_cycle_length; i++)
2442  expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2443 
2444  if(abs_frame_num > 0){
2445  int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2446  int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2447 
2448  expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2449  for(i = 0; i <= frame_num_in_poc_cycle; i++)
2450  expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2451  } else
2452  expectedpoc = 0;
2453 
2454  if(h->nal_ref_idc == 0)
2455  expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2456 
2457  field_poc[0] = expectedpoc + h->delta_poc[0];
2458  field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2459 
2460  if(s->picture_structure == PICT_FRAME)
2461  field_poc[1] += h->delta_poc[1];
2462  }else{
2463  int poc= 2*(h->frame_num_offset + h->frame_num);
2464 
2465  if(!h->nal_ref_idc)
2466  poc--;
2467 
2468  field_poc[0]= poc;
2469  field_poc[1]= poc;
2470  }
2471 
2473  s->current_picture_ptr->field_poc[0]= field_poc[0];
2475  s->current_picture_ptr->field_poc[1]= field_poc[1];
2476  cur->poc= FFMIN(cur->field_poc[0], cur->field_poc[1]);
2477 
2478  return 0;
2479 }
2480 
2481 
2486  int i;
2487  for(i=0; i<16; i++){
2488 #define T(x) (x>>2) | ((x<<2) & 0xF)
2489  h->zigzag_scan[i] = T(zigzag_scan[i]);
2490  h-> field_scan[i] = T( field_scan[i]);
2491 #undef T
2492  }
2493  for(i=0; i<64; i++){
2494 #define T(x) (x>>3) | ((x&7)<<3)
2495  h->zigzag_scan8x8[i] = T(ff_zigzag_direct[i]);
2497  h->field_scan8x8[i] = T(field_scan8x8[i]);
2499 #undef T
2500  }
2501  if(h->sps.transform_bypass){ //FIXME same ugly
2508  }else{
2509  h->zigzag_scan_q0 = h->zigzag_scan;
2512  h->field_scan_q0 = h->field_scan;
2515  }
2516 }
2517 
2518 static int field_end(H264Context *h, int in_setup){
2519  MpegEncContext * const s = &h->s;
2520  AVCodecContext * const avctx= s->avctx;
2521  int err = 0;
2522  s->mb_y= 0;
2523 
2524  if (!in_setup && !s->dropable)
2527 
2530 
2531  if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
2532  if(!s->dropable) {
2534  h->prev_poc_msb= h->poc_msb;
2535  h->prev_poc_lsb= h->poc_lsb;
2536  }
2538  h->prev_frame_num= h->frame_num;
2540  }
2541 
2542  if (avctx->hwaccel) {
2543  if (avctx->hwaccel->end_frame(avctx) < 0)
2544  av_log(avctx, AV_LOG_ERROR, "hardware accelerator failed to decode picture\n");
2545  }
2546 
2549 
2550  /*
2551  * FIXME: Error handling code does not seem to support interlaced
2552  * when slices span multiple rows
2553  * The ff_er_add_slice calls don't work right for bottom
2554  * fields; they cause massive erroneous error concealing
2555  * Error marking covers both fields (top and bottom).
2556  * This causes a mismatched s->error_count
2557  * and a bad error table. Further, the error count goes to
2558  * INT_MAX when called for bottom field, because mb_y is
2559  * past end by one (callers fault) and resync_mb_y != 0
2560  * causes problems for the first MB line, too.
2561  */
2562  if (!FIELD_PICTURE)
2563  ff_er_frame_end(s);
2564 
2565  MPV_frame_end(s);
2566 
2567  h->current_slice=0;
2568 
2569  return err;
2570 }
2571 
2575 static void clone_slice(H264Context *dst, H264Context *src)
2576 {
2577  memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2579  dst->s.current_picture = src->s.current_picture;
2580  dst->s.linesize = src->s.linesize;
2581  dst->s.uvlinesize = src->s.uvlinesize;
2582  dst->s.first_field = src->s.first_field;
2583 
2584  dst->prev_poc_msb = src->prev_poc_msb;
2585  dst->prev_poc_lsb = src->prev_poc_lsb;
2587  dst->prev_frame_num = src->prev_frame_num;
2588  dst->short_ref_count = src->short_ref_count;
2589 
2590  memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
2591  memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
2592  memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2593  memcpy(dst->ref_list, src->ref_list, sizeof(dst->ref_list));
2594 
2595  memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
2596  memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
2597 }
2598 
2607 {
2608  int profile = sps->profile_idc;
2609 
2610  switch(sps->profile_idc) {
2612  // constraint_set1_flag set to 1
2613  profile |= (sps->constraint_set_flags & 1<<1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2614  break;
2618  // constraint_set3_flag set to 1
2619  profile |= (sps->constraint_set_flags & 1<<3) ? FF_PROFILE_H264_INTRA : 0;
2620  break;
2621  }
2622 
2623  return profile;
2624 }
2625 
2636  MpegEncContext * const s = &h->s;
2637  MpegEncContext * const s0 = &h0->s;
2638  unsigned int first_mb_in_slice;
2639  unsigned int pps_id;
2640  int num_ref_idx_active_override_flag;
2641  unsigned int slice_type, tmp, i, j;
2642  int default_ref_list_done = 0;
2643  int last_pic_structure, last_pic_dropable;
2644 
2645  /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2646  if((s->avctx->flags2 & CODEC_FLAG2_FAST) && !h->nal_ref_idc && !h->pixel_shift){
2649  }else{
2652  }
2653 
2654  first_mb_in_slice= get_ue_golomb(&s->gb);
2655 
2656  if(first_mb_in_slice == 0){ //FIXME better field boundary detection
2657  if(h0->current_slice && FIELD_PICTURE){
2658  field_end(h, 1);
2659  }
2660 
2661  h0->current_slice = 0;
2662  if (!s0->first_field) {
2663  if (s->current_picture_ptr && !s->dropable &&
2664  s->current_picture_ptr->owner2 == s) {
2667  }
2669  }
2670  }
2671 
2672  slice_type= get_ue_golomb_31(&s->gb);
2673  if(slice_type > 9){
2674  av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2675  return -1;
2676  }
2677  if(slice_type > 4){
2678  slice_type -= 5;
2679  h->slice_type_fixed=1;
2680  }else
2681  h->slice_type_fixed=0;
2682 
2683  slice_type= golomb_to_pict_type[ slice_type ];
2684  if (slice_type == AV_PICTURE_TYPE_I
2685  || (h0->current_slice != 0 && slice_type == h0->last_slice_type) ) {
2686  default_ref_list_done = 1;
2687  }
2688  h->slice_type= slice_type;
2689  h->slice_type_nos= slice_type & 3;
2690 
2691  if (h->nal_unit_type == NAL_IDR_SLICE &&
2693  av_log(h->s.avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
2694  return AVERROR_INVALIDDATA;
2695  }
2696 
2697  // to make a few old functions happy, it's wrong though
2698  s->pict_type = h->slice_type;
2699 
2700  pps_id= get_ue_golomb(&s->gb);
2701  if(pps_id>=MAX_PPS_COUNT){
2702  av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2703  return -1;
2704  }
2705  if(!h0->pps_buffers[pps_id]) {
2706  av_log(h->s.avctx, AV_LOG_ERROR, "non-existing PPS %u referenced\n", pps_id);
2707  return -1;
2708  }
2709  h->pps= *h0->pps_buffers[pps_id];
2710 
2711  if(!h0->sps_buffers[h->pps.sps_id]) {
2712  av_log(h->s.avctx, AV_LOG_ERROR, "non-existing SPS %u referenced\n", h->pps.sps_id);
2713  return -1;
2714  }
2715  h->sps = *h0->sps_buffers[h->pps.sps_id];
2716 
2717  s->avctx->profile = ff_h264_get_profile(&h->sps);
2718  s->avctx->level = h->sps.level_idc;
2719  s->avctx->refs = h->sps.ref_frame_count;
2720 
2721  s->mb_width= h->sps.mb_width;
2722  s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2723 
2724  h->b_stride= s->mb_width*4;
2725 
2726  s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2727 
2728  s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2729  if(h->sps.frame_mbs_only_flag)
2730  s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2731  else
2732  s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
2733 
2734  if (FFALIGN(s->avctx->width, 16) == s->width &&
2735  FFALIGN(s->avctx->height, 16) == s->height) {
2736  s->width = s->avctx->width;
2737  s->height = s->avctx->height;
2738  }
2739 
2740  if (s->context_initialized
2741  && ( s->width != s->avctx->width || s->height != s->avctx->height
2742  || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio))) {
2743  if(h != h0 || (HAVE_THREADS && h->s.avctx->active_thread_type & FF_THREAD_FRAME)) {
2744  av_log_missing_feature(s->avctx, "Width/height changing with threads is", 0);
2745  return AVERROR_PATCHWELCOME; // width / height changed during parallelized decoding
2746  }
2747  free_tables(h, 0);
2748  flush_dpb(s->avctx);
2749  MPV_common_end(s);
2750  }
2751  if (!s->context_initialized) {
2752  if (h != h0) {
2753  av_log(h->s.avctx, AV_LOG_ERROR, "Cannot (re-)initialize context during parallel decoding.\n");
2754  return -1;
2755  }
2756 
2758  s->avctx->sample_aspect_ratio= h->sps.sar;
2760 
2765  s->avctx->color_trc = h->sps.color_trc;
2766  s->avctx->colorspace = h->sps.colorspace;
2767  }
2768  }
2769 
2771  int64_t den= h->sps.time_scale;
2772  if(h->x264_build < 44U)
2773  den *= 2;
2775  h->sps.num_units_in_tick, den, 1<<30);
2776  }
2777 
2778  switch (h->sps.bit_depth_luma) {
2779  case 9 :
2780  if (CHROMA444) {
2781  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2782  s->avctx->pix_fmt = PIX_FMT_GBRP9;
2783  } else
2785  } else if (CHROMA422)
2787  else
2789  break;
2790  case 10 :
2791  if (CHROMA444) {
2792  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2794  } else
2796  } else if (CHROMA422)
2798  else
2800  break;
2801  case 8:
2802  if (CHROMA444){
2803  if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2804  s->avctx->pix_fmt = PIX_FMT_GBRP;
2805  } else
2807  } else if (CHROMA422) {
2809  }else{
2810  s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2811  s->avctx->codec->pix_fmts ?
2812  s->avctx->codec->pix_fmts :
2816  }
2817  break;
2818  default:
2820  "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2821  return AVERROR_INVALIDDATA;
2822  }
2823 
2825 
2826  if (MPV_common_init(s) < 0) {
2827  av_log(h->s.avctx, AV_LOG_ERROR, "MPV_common_init() failed.\n");
2828  return -1;
2829  }
2830  s->first_field = 0;
2831  h->prev_interlaced_frame = 1;
2832 
2833  init_scan_tables(h);
2834  if (ff_h264_alloc_tables(h) < 0) {
2835  av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
2836  return AVERROR(ENOMEM);
2837  }
2838 
2840  if (context_init(h) < 0) {
2841  av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2842  return -1;
2843  }
2844  } else {
2845  for(i = 1; i < s->slice_context_count; i++) {
2846  H264Context *c;
2847  c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2848  memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2849  memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2850  c->h264dsp = h->h264dsp;
2851  c->sps = h->sps;
2852  c->pps = h->pps;
2853  c->pixel_shift = h->pixel_shift;
2854  init_scan_tables(c);
2855  clone_tables(c, h, i);
2856  }
2857 
2858  for(i = 0; i < s->slice_context_count; i++)
2859  if (context_init(h->thread_context[i]) < 0) {
2860  av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2861  return -1;
2862  }
2863  }
2864  }
2865 
2866  if(h == h0 && h->dequant_coeff_pps != pps_id){
2867  h->dequant_coeff_pps = pps_id;
2869  }
2870 
2872 
2873  h->mb_mbaff = 0;
2874  h->mb_aff_frame = 0;
2875  last_pic_structure = s0->picture_structure;
2876  last_pic_dropable = s0->dropable;
2877  s->dropable = h->nal_ref_idc == 0;
2878  if(h->sps.frame_mbs_only_flag){
2880  }else{
2881  if(get_bits1(&s->gb)) { //field_pic_flag
2882  s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2883  } else {
2885  h->mb_aff_frame = h->sps.mb_aff;
2886  }
2887  }
2889 
2890  if (h0->current_slice != 0) {
2891  if (last_pic_structure != s->picture_structure ||
2892  last_pic_dropable != s->dropable) {
2893  av_log(h->s.avctx, AV_LOG_ERROR,
2894  "Changing field mode (%d -> %d) between slices is not allowed\n",
2895  last_pic_structure, s->picture_structure);
2896  s->picture_structure = last_pic_structure;
2897  s->dropable = last_pic_dropable;
2898  return AVERROR_INVALIDDATA;
2899  } else if (!s->current_picture_ptr) {
2901  "unset current_picture_ptr on %d. slice\n",
2902  h0->current_slice + 1);
2903  return AVERROR_INVALIDDATA;
2904  }
2905  } else {
2906  /* Shorten frame num gaps so we don't have to allocate reference
2907  * frames just to throw them away */
2908  if (h->frame_num != h->prev_frame_num) {
2909  int unwrap_prev_frame_num = h->prev_frame_num;
2910  int max_frame_num = 1 << h->sps.log2_max_frame_num;
2911 
2912  if (unwrap_prev_frame_num > h->frame_num) unwrap_prev_frame_num -= max_frame_num;
2913 
2914  if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2915  unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2916  if (unwrap_prev_frame_num < 0)
2917  unwrap_prev_frame_num += max_frame_num;
2918 
2919  h->prev_frame_num = unwrap_prev_frame_num;
2920  }
2921  }
2922 
2923  /* See if we have a decoded first field looking for a pair...
2924  * Here, we're using that to see if we should mark previously
2925  * decode frames as "finished".
2926  * We have to do that before the "dummy" in-between frame allocation,
2927  * since that can modify s->current_picture_ptr. */
2928  if (s0->first_field) {
2929  assert(s0->current_picture_ptr);
2930  assert(s0->current_picture_ptr->f.data[0]);
2932 
2933  /* Mark old field/frame as completed */
2934  if (!last_pic_dropable && s0->current_picture_ptr->owner2 == s0) {
2936  last_pic_structure == PICT_BOTTOM_FIELD);
2937  }
2938 
2939  /* figure out if we have a complementary field pair */
2940  if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2941  /* Previous field is unmatched. Don't display it, but let it
2942  * remain for reference if marked as such. */
2943  if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2945  last_pic_structure == PICT_TOP_FIELD);
2946  }
2947  } else {
2948  if (s0->current_picture_ptr->frame_num != h->frame_num) {
2949  /* This and previous field were reference, but had
2950  * different frame_nums. Consider this field first in
2951  * pair. Throw away previous field except for reference
2952  * purposes. */
2953  if (!last_pic_dropable && last_pic_structure != PICT_FRAME) {
2955  last_pic_structure == PICT_TOP_FIELD);
2956  }
2957  } else {
2958  /* Second field in complementary pair */
2959  if (!((last_pic_structure == PICT_TOP_FIELD &&
2961  (last_pic_structure == PICT_BOTTOM_FIELD &&
2964  "Invalid field mode combination %d/%d\n",
2965  last_pic_structure, s->picture_structure);
2966  s->picture_structure = last_pic_structure;
2967  s->dropable = last_pic_dropable;
2968  return AVERROR_INVALIDDATA;
2969  } else if (last_pic_dropable != s->dropable) {
2971  "Cannot combine reference and non-reference fields in the same frame\n");
2973  s->picture_structure = last_pic_structure;
2974  s->dropable = last_pic_dropable;
2975  return AVERROR_INVALIDDATA;
2976  }
2977 
2978  /* Take ownership of this buffer. Note that if another thread owned
2979  * the first field of this buffer, we're not operating on that pointer,
2980  * so the original thread is still responsible for reporting progress
2981  * on that first field (or if that was us, we just did that above).
2982  * By taking ownership, we assign responsibility to ourselves to
2983  * report progress on the second field. */
2984  s0->current_picture_ptr->owner2 = s0;
2985  }
2986  }
2987  }
2988 
2989  while (h->frame_num != h->prev_frame_num &&
2990  h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2991  Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2992  av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n", h->frame_num, h->prev_frame_num);
2993  if (ff_h264_frame_start(h) < 0)
2994  return -1;
2995  h->prev_frame_num++;
3001  if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
3003  return AVERROR_INVALIDDATA;
3004  /* Error concealment: if a ref is missing, copy the previous ref in its place.
3005  * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
3006  * about there being no actual duplicates.
3007  * FIXME: this doesn't copy padding for out-of-frame motion vectors. Given we're
3008  * concealing a lost frame, this probably isn't noticeable by comparison, but it should
3009  * be fixed. */
3010  if (h->short_ref_count) {
3011  if (prev) {
3012  av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
3013  (const uint8_t**)prev->f.data, prev->f.linesize,
3014  s->avctx->pix_fmt, s->mb_width*16, s->mb_height*16);
3015  h->short_ref[0]->poc = prev->poc+2;
3016  }
3017  h->short_ref[0]->frame_num = h->prev_frame_num;
3018  }
3019  }
3020 
3021  /* See if we have a decoded first field looking for a pair...
3022  * We're using that to see whether to continue decoding in that
3023  * frame, or to allocate a new one. */
3024  if (s0->first_field) {
3025  assert(s0->current_picture_ptr);
3026  assert(s0->current_picture_ptr->f.data[0]);
3028 
3029  /* figure out if we have a complementary field pair */
3030  if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3031  /*
3032  * Previous field is unmatched. Don't display it, but let it
3033  * remain for reference if marked as such.
3034  */
3035  s0->current_picture_ptr = NULL;
3036  s0->first_field = FIELD_PICTURE;
3037 
3038  } else {
3039  if (s0->current_picture_ptr->frame_num != h->frame_num) {
3040  /* This and the previous field had different frame_nums.
3041  * Consider this field first in pair. Throw away previous
3042  * one except for reference purposes. */
3043  s0->first_field = 1;
3044  s0->current_picture_ptr = NULL;
3045 
3046  } else {
3047  /* Second field in complementary pair */
3048  s0->first_field = 0;
3049  }
3050  }
3051 
3052  } else {
3053  /* Frame or first field in a potentially complementary pair */
3054  assert(!s0->current_picture_ptr);
3055  s0->first_field = FIELD_PICTURE;
3056  }
3057 
3058  if(!FIELD_PICTURE || s0->first_field) {
3059  if (ff_h264_frame_start(h) < 0) {
3060  s0->first_field = 0;
3061  return -1;
3062  }
3063  } else {
3065  }
3066  }
3067  if(h != h0)
3068  clone_slice(h, h0);
3069 
3070  s->current_picture_ptr->frame_num= h->frame_num; //FIXME frame_num cleanup
3071 
3072  assert(s->mb_num == s->mb_width * s->mb_height);
3073  if(first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3074  first_mb_in_slice >= s->mb_num){
3075  av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3076  return -1;
3077  }
3078  s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3079  s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3081  s->resync_mb_y = s->mb_y = s->mb_y + 1;
3082  assert(s->mb_y < s->mb_height);
3083 
3084  if(s->picture_structure==PICT_FRAME){
3085  h->curr_pic_num= h->frame_num;
3086  h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3087  }else{
3088  h->curr_pic_num= 2*h->frame_num + 1;
3089  h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3090  }
3091 
3092  if(h->nal_unit_type == NAL_IDR_SLICE){
3093  get_ue_golomb(&s->gb); /* idr_pic_id */
3094  }
3095 
3096  if(h->sps.poc_type==0){
3097  h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3098 
3101  }
3102  }
3103 
3105  h->delta_poc[0]= get_se_golomb(&s->gb);
3106 
3108  h->delta_poc[1]= get_se_golomb(&s->gb);
3109  }
3110 
3111  init_poc(h);
3112 
3115  }
3116 
3117  //set defaults, might be overridden a few lines later
3118  h->ref_count[0]= h->pps.ref_count[0];
3119  h->ref_count[1]= h->pps.ref_count[1];
3120 
3122  int max_refs = s->picture_structure == PICT_FRAME ? 16 : 32;
3123 
3126  }
3127  num_ref_idx_active_override_flag= get_bits1(&s->gb);
3128 
3129  if(num_ref_idx_active_override_flag){
3130  h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3131  if (h->ref_count[0] < 1)
3132  return AVERROR_INVALIDDATA;
3133  if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3134  h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3135  if (h->ref_count[1] < 1)
3136  return AVERROR_INVALIDDATA;
3137  }
3138  }
3139 
3140  if (h->ref_count[0] > max_refs || h->ref_count[1] > max_refs) {
3141  av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3142  h->ref_count[0] = h->ref_count[1] = 1;
3143  return AVERROR_INVALIDDATA;
3144  }
3145 
3147  h->list_count= 2;
3148  else
3149  h->list_count= 1;
3150  }else
3151  h->list_count= 0;
3152 
3153  if(!default_ref_list_done){
3155  }
3156 
3158  h->ref_count[1]= h->ref_count[0]= 0;
3159  return -1;
3160  }
3161 
3163  s->last_picture_ptr= &h->ref_list[0][0];
3165  }
3167  s->next_picture_ptr= &h->ref_list[1][0];
3169  }
3170 
3173  pred_weight_table(h);
3175  implicit_weight_table(h, -1);
3176  }else {
3177  h->use_weight = 0;
3178  for (i = 0; i < 2; i++) {
3179  h->luma_weight_flag[i] = 0;
3180  h->chroma_weight_flag[i] = 0;
3181  }
3182  }
3183 
3184  if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 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  }
3224  get_se_golomb(&s->gb); /* slice_qs_delta */
3225  }
3226 
3227  h->deblocking_filter = 1;
3228  h->slice_alpha_c0_offset = 52;
3229  h->slice_beta_offset = 52;
3231  tmp= get_ue_golomb_31(&s->gb);
3232  if(tmp > 2){
3233  av_log(s->avctx, AV_LOG_ERROR, "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){
3245  av_log(s->avctx, AV_LOG_ERROR, "deblocking filter parameters %d %d out of range\n", h->slice_alpha_c0_offset, h->slice_beta_offset);
3246  return -1;
3247  }
3248  }
3249  }
3250 
3254  ||(s->avctx->skip_loop_filter >= AVDISCARD_NONREF && h->nal_ref_idc == 0))
3255  h->deblocking_filter= 0;
3256 
3257  if(h->deblocking_filter == 1 && h0->max_contexts > 1) {
3258  if(s->avctx->flags2 & CODEC_FLAG2_FAST) {
3259  /* Cheat slightly for speed:
3260  Do not bother to deblock across slices. */
3261  h->deblocking_filter = 2;
3262  } else {
3263  h0->max_contexts = 1;
3264  if(!h0->single_decode_warning) {
3265  av_log(s->avctx, AV_LOG_INFO, "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3266  h0->single_decode_warning = 1;
3267  }
3268  if (h != h0) {
3269  av_log(h->s.avctx, AV_LOG_ERROR, "Deblocking switched inside frame.\n");
3270  return 1;
3271  }
3272  }
3273  }
3276  + 6 * (h->sps.bit_depth_luma - 8);
3277 
3278 #if 0 //FMO
3279  if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3280  slice_group_change_cycle= get_bits(&s->gb, ?);
3281 #endif
3282 
3283  h0->last_slice_type = slice_type;
3284  h->slice_num = ++h0->current_slice;
3285  if(h->slice_num >= MAX_SLICES){
3286  av_log(s->avctx, AV_LOG_ERROR, "Too many slices, increase MAX_SLICES and recompile\n");
3287  }
3288 
3289  for(j=0; j<2; j++){
3290  int id_list[16];
3291  int *ref2frm= h->ref2frm[h->slice_num&(MAX_SLICES-1)][j];
3292  for(i=0; i<16; i++){
3293  id_list[i]= 60;
3294  if (h->ref_list[j][i].f.data[0]) {
3295  int k;
3296  uint8_t *base = h->ref_list[j][i].f.base[0];
3297  for(k=0; k<h->short_ref_count; k++)
3298  if (h->short_ref[k]->f.base[0] == base) {
3299  id_list[i]= k;
3300  break;
3301  }
3302  for(k=0; k<h->long_ref_count; k++)
3303  if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3304  id_list[i]= h->short_ref_count + k;
3305  break;
3306  }
3307  }
3308  }
3309 
3310  ref2frm[0]=
3311  ref2frm[1]= -1;
3312  for(i=0; i<16; i++)
3313  ref2frm[i+2]= 4*id_list[i]
3314  + (h->ref_list[j][i].f.reference & 3);
3315  ref2frm[18+0]=
3316  ref2frm[18+1]= -1;
3317  for(i=16; i<48; i++)
3318  ref2frm[i+4]= 4*id_list[(i-16)>>1]
3319  + (h->ref_list[j][i].f.reference & 3);
3320  }
3321 
3322  //FIXME: fix draw_edges+PAFF+frame threads
3325 
3326  if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3327  av_log(h->s.avctx, AV_LOG_DEBUG, "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",
3328  h->slice_num,
3329  (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3330  first_mb_in_slice,
3331  av_get_picture_type_char(h->slice_type), h->slice_type_fixed ? " fix" : "", h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3332  pps_id, h->frame_num,
3334  h->ref_count[0], h->ref_count[1],
3335  s->qscale,
3337  h->use_weight,
3338  h->use_weight==1 && h->use_weight_chroma ? "c" : "",
3339  h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : ""
3340  );
3341  }
3342 
3343  return 0;
3344 }
3345 
3347 {
3348  switch (h->slice_type) {
3349  case AV_PICTURE_TYPE_P: return 0;
3350  case AV_PICTURE_TYPE_B: return 1;
3351  case AV_PICTURE_TYPE_I: return 2;
3352  case AV_PICTURE_TYPE_SP: return 3;
3353  case AV_PICTURE_TYPE_SI: return 4;
3354  default: return -1;
3355  }
3356 }
3357 
3358 static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncContext * const s, int mb_type, int top_xy,
3359  int left_xy[LEFT_MBS], int top_type, int left_type[LEFT_MBS], int mb_xy, int list)
3360 {
3361  int b_stride = h->b_stride;
3362  int16_t (*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3363  int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3364  if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
3365  if(USES_LIST(top_type, list)){
3366  const int b_xy= h->mb2b_xy[top_xy] + 3*b_stride;
3367  const int b8_xy= 4*top_xy + 2;
3368  int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3369  AV_COPY128(mv_dst - 1*8, s->current_picture.f.motion_val[list][b_xy + 0]);
3370  ref_cache[0 - 1*8]=
3371  ref_cache[1 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3372  ref_cache[2 - 1*8]=
3373  ref_cache[3 - 1*8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3374  }else{
3375  AV_ZERO128(mv_dst - 1*8);
3376  AV_WN32A(&ref_cache[0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3377  }
3378 
3379  if(!IS_INTERLACED(mb_type^left_type[LTOP])){
3380  if(USES_LIST(left_type[LTOP], list)){
3381  const int b_xy= h->mb2b_xy[left_xy[LTOP]] + 3;
3382  const int b8_xy= 4*left_xy[LTOP] + 1;
3383  int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[LTOP]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3384  AV_COPY32(mv_dst - 1 + 0, s->current_picture.f.motion_val[list][b_xy + b_stride*0]);
3385  AV_COPY32(mv_dst - 1 + 8, s->current_picture.f.motion_val[list][b_xy + b_stride*1]);
3386  AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride*2]);
3387  AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride*3]);
3388  ref_cache[-1 + 0]=
3389  ref_cache[-1 + 8]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*0]];
3390  ref_cache[-1 + 16]=
3391  ref_cache[-1 + 24]= ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2*1]];
3392  }else{
3393  AV_ZERO32(mv_dst - 1 + 0);
3394  AV_ZERO32(mv_dst - 1 + 8);
3395  AV_ZERO32(mv_dst - 1 +16);
3396  AV_ZERO32(mv_dst - 1 +24);
3397  ref_cache[-1 + 0]=
3398  ref_cache[-1 + 8]=
3399  ref_cache[-1 + 16]=
3400  ref_cache[-1 + 24]= LIST_NOT_USED;
3401  }
3402  }
3403  }
3404 
3405  if(!USES_LIST(mb_type, list)){
3406  fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0,0), 4);
3407  AV_WN32A(&ref_cache[0*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3408  AV_WN32A(&ref_cache[1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3409  AV_WN32A(&ref_cache[2*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3410  AV_WN32A(&ref_cache[3*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
3411  return;
3412  }
3413 
3414  {
3415  int8_t *ref = &s->current_picture.f.ref_index[list][4*mb_xy];
3416  int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
3417  uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101;
3418  uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]],ref2frm[list][ref[3]])&0x00FF00FF)*0x0101;
3419  AV_WN32A(&ref_cache[0*8], ref01);
3420  AV_WN32A(&ref_cache[1*8], ref01);
3421  AV_WN32A(&ref_cache[2*8], ref23);
3422  AV_WN32A(&ref_cache[3*8], ref23);
3423  }
3424 
3425  {
3426  int16_t (*mv_src)[2] = &s->current_picture.f.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
3427  AV_COPY128(mv_dst + 8*0, mv_src + 0*b_stride);
3428  AV_COPY128(mv_dst + 8*1, mv_src + 1*b_stride);
3429  AV_COPY128(mv_dst + 8*2, mv_src + 2*b_stride);
3430  AV_COPY128(mv_dst + 8*3, mv_src + 3*b_stride);
3431  }
3432 }
3433 
3438 static int fill_filter_caches(H264Context *h, int mb_type){
3439  MpegEncContext * const s = &h->s;
3440  const int mb_xy= h->mb_xy;
3441  int top_xy, left_xy[LEFT_MBS];
3442  int top_type, left_type[LEFT_MBS];
3443  uint8_t *nnz;
3444  uint8_t *nnz_cache;
3445 
3446  top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3447 
3448  /* Wow, what a mess, why didn't they simplify the interlacing & intra
3449  * stuff, I can't imagine that these complex rules are worth it. */
3450 
3451  left_xy[LBOT] = left_xy[LTOP] = mb_xy-1;
3452  if(FRAME_MBAFF){
3453  const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3454  const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3455  if(s->mb_y&1){
3456  if (left_mb_field_flag != curr_mb_field_flag) {
3457  left_xy[LTOP] -= s->mb_stride;
3458  }
3459  }else{
3460  if(curr_mb_field_flag){
3461  top_xy += s->mb_stride & (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3462  }
3463  if (left_mb_field_flag != curr_mb_field_flag) {
3464  left_xy[LBOT] += s->mb_stride;
3465  }
3466  }
3467  }
3468 
3469  h->top_mb_xy = top_xy;
3470  h->left_mb_xy[LTOP] = left_xy[LTOP];
3471  h->left_mb_xy[LBOT] = left_xy[LBOT];
3472  {
3473  //for sufficiently low qp, filtering wouldn't do anything
3474  //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
3475  int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
3476  int qp = s->current_picture.f.qscale_table[mb_xy];
3477  if(qp <= qp_thresh
3478  && (left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh)
3479  && (top_xy < 0 || ((qp + s->current_picture.f.qscale_table[top_xy ] + 1) >> 1) <= qp_thresh)) {
3480  if(!FRAME_MBAFF)
3481  return 1;
3482  if ((left_xy[LTOP] < 0 || ((qp + s->current_picture.f.qscale_table[left_xy[LBOT] ] + 1) >> 1) <= qp_thresh) &&
3483  (top_xy < s->mb_stride || ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3484  return 1;
3485  }
3486  }
3487 
3488  top_type = s->current_picture.f.mb_type[top_xy];
3489  left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3490  left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3491  if(h->deblocking_filter == 2){
3492  if(h->slice_table[top_xy ] != h->slice_num) top_type= 0;
3493  if(h->slice_table[left_xy[LBOT]] != h->slice_num) left_type[LTOP]= left_type[LBOT]= 0;
3494  }else{
3495  if(h->slice_table[top_xy ] == 0xFFFF) top_type= 0;
3496  if(h->slice_table[left_xy[LBOT]] == 0xFFFF) left_type[LTOP]= left_type[LBOT] =0;
3497  }
3498  h->top_type = top_type;
3499  h->left_type[LTOP]= left_type[LTOP];
3500  h->left_type[LBOT]= left_type[LBOT];
3501 
3502  if(IS_INTRA(mb_type))
3503  return 0;
3504 
3505  fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 0);
3506  if(h->list_count == 2)
3507  fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy, top_type, left_type, mb_xy, 1);
3508 
3509  nnz = h->non_zero_count[mb_xy];
3510  nnz_cache = h->non_zero_count_cache;
3511  AV_COPY32(&nnz_cache[4+8*1], &nnz[ 0]);
3512  AV_COPY32(&nnz_cache[4+8*2], &nnz[ 4]);
3513  AV_COPY32(&nnz_cache[4+8*3], &nnz[ 8]);
3514  AV_COPY32(&nnz_cache[4+8*4], &nnz[12]);
3515  h->cbp= h->cbp_table[mb_xy];
3516 
3517  if(top_type){
3518  nnz = h->non_zero_count[top_xy];
3519  AV_COPY32(&nnz_cache[4+8*0], &nnz[3*4]);
3520  }
3521 
3522  if(left_type[LTOP]){
3523  nnz = h->non_zero_count[left_xy[LTOP]];
3524  nnz_cache[3+8*1]= nnz[3+0*4];
3525  nnz_cache[3+8*2]= nnz[3+1*4];
3526  nnz_cache[3+8*3]= nnz[3+2*4];
3527  nnz_cache[3+8*4]= nnz[3+3*4];
3528  }
3529 
3530  // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
3531  if(!CABAC && h->pps.transform_8x8_mode){
3532  if(IS_8x8DCT(top_type)){
3533  nnz_cache[4+8*0]=
3534  nnz_cache[5+8*0]= (h->cbp_table[top_xy] & 0x4000) >> 12;
3535  nnz_cache[6+8*0]=
3536  nnz_cache[7+8*0]= (h->cbp_table[top_xy] & 0x8000) >> 12;
3537  }
3538  if(IS_8x8DCT(left_type[LTOP])){
3539  nnz_cache[3+8*1]=
3540  nnz_cache[3+8*2]= (h->cbp_table[left_xy[LTOP]]&0x2000) >> 12; //FIXME check MBAFF
3541  }
3542  if(IS_8x8DCT(left_type[LBOT])){
3543  nnz_cache[3+8*3]=
3544  nnz_cache[3+8*4]= (h->cbp_table[left_xy[LBOT]]&0x8000) >> 12; //FIXME check MBAFF
3545  }
3546 
3547  if(IS_8x8DCT(mb_type)){
3548  nnz_cache[scan8[0 ]]= nnz_cache[scan8[1 ]]=
3549  nnz_cache[scan8[2 ]]= nnz_cache[scan8[3 ]]= (h->cbp & 0x1000) >> 12;
3550 
3551  nnz_cache[scan8[0+ 4]]= nnz_cache[scan8[1+ 4]]=
3552  nnz_cache[scan8[2+ 4]]= nnz_cache[scan8[3+ 4]]= (h->cbp & 0x2000) >> 12;
3553 
3554  nnz_cache[scan8[0+ 8]]= nnz_cache[scan8[1+ 8]]=
3555  nnz_cache[scan8[2+ 8]]= nnz_cache[scan8[3+ 8]]= (h->cbp & 0x4000) >> 12;
3556 
3557  nnz_cache[scan8[0+12]]= nnz_cache[scan8[1+12]]=
3558  nnz_cache[scan8[2+12]]= nnz_cache[scan8[3+12]]= (h->cbp & 0x8000) >> 12;
3559  }
3560  }
3561 
3562  return 0;
3563 }
3564 
3565 static void loop_filter(H264Context *h, int start_x, int end_x){
3566  MpegEncContext * const s = &h->s;
3567  uint8_t *dest_y, *dest_cb, *dest_cr;
3568  int linesize, uvlinesize, mb_x, mb_y;
3569  const int end_mb_y= s->mb_y + FRAME_MBAFF;
3570  const int old_slice_type= h->slice_type;
3571  const int pixel_shift = h->pixel_shift;
3572  const int block_h = 16 >> s->chroma_y_shift;
3573 
3574  if(h->deblocking_filter) {
3575  for(mb_x= start_x; mb_x<end_x; mb_x++){
3576  for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
3577  int mb_xy, mb_type;
3578  mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
3579  h->slice_num= h->slice_table[mb_xy];
3580  mb_type = s->current_picture.f.mb_type[mb_xy];
3581  h->list_count= h->list_counts[mb_xy];
3582 
3583  if(FRAME_MBAFF)
3584  h->mb_mbaff = h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3585 
3586  s->mb_x= mb_x;
3587  s->mb_y= mb_y;
3588  dest_y = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize ) * 16;
3589  dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3590  dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
3591  //FIXME simplify above
3592 
3593  if (MB_FIELD) {
3594  linesize = h->mb_linesize = s->linesize * 2;
3595  uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3596  if(mb_y&1){ //FIXME move out of this function?
3597  dest_y -= s->linesize*15;
3598  dest_cb-= s->uvlinesize * (block_h - 1);
3599  dest_cr-= s->uvlinesize * (block_h - 1);
3600  }
3601  } else {
3602  linesize = h->mb_linesize = s->linesize;
3603  uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3604  }
3605  backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3606  if(fill_filter_caches(h, mb_type))
3607  continue;
3608  h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3609  h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3610 
3611  if (FRAME_MBAFF) {
3612  ff_h264_filter_mb (h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3613  } else {
3614  ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3615  }
3616  }
3617  }
3618  }
3619  h->slice_type= old_slice_type;
3620  s->mb_x= end_x;
3621  s->mb_y= end_mb_y - FRAME_MBAFF;
3622  h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3623  h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3624 }
3625 
3627  MpegEncContext * const s = &h->s;
3628  const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3629  int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
3630  ? s->current_picture.f.mb_type[mb_xy - 1]
3631  : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
3632  ? s->current_picture.f.mb_type[mb_xy - s->mb_stride]
3633  : 0;
3634  h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3635 }
3636 
3641  MpegEncContext * const s = &h->s;
3642  int top = 16*(s->mb_y >> FIELD_PICTURE);
3643  int height = 16 << FRAME_MBAFF;
3644  int deblock_border = (16 + 4) << FRAME_MBAFF;
3645  int pic_height = 16*s->mb_height >> FIELD_PICTURE;
3646 
3647  if (h->deblocking_filter) {
3648  if((top + height) >= pic_height)
3649  height += deblock_border;
3650 
3651  top -= deblock_border;
3652  }
3653 
3654  if (top >= pic_height || (top + height) < h->emu_edge_height)
3655  return;
3656 
3657  height = FFMIN(height, pic_height - top);
3658  if (top < h->emu_edge_height) {
3659  height = top+height;
3660  top = 0;
3661  }
3662 
3663  ff_draw_horiz_band(s, top, height);
3664 
3665  if (s->dropable) return;
3666 
3667  ff_thread_report_progress((AVFrame*)s->current_picture_ptr, top + height - 1,
3669 }
3670 
3671 static int decode_slice(struct AVCodecContext *avctx, void *arg){
3672  H264Context *h = *(void**)arg;
3673  MpegEncContext * const s = &h->s;
3674  const int part_mask= s->partitioned_frame ? (ER_AC_END|ER_AC_ERROR) : 0x7F;
3675  int lf_x_start = s->mb_x;
3676 
3677  s->mb_skip_run= -1;
3678 
3680  (CONFIG_GRAY && (s->flags&CODEC_FLAG_GRAY));
3681 
3682  if( h->pps.cabac ) {
3683  /* realign */
3684  align_get_bits( &s->gb );
3685 
3686  /* init cabac */
3689  s->gb.buffer + get_bits_count(&s->gb)/8,
3690  (get_bits_left(&s->gb) + 7)/8);
3691 
3693 
3694  for(;;){
3695 //START_TIMER
3696  int ret = ff_h264_decode_mb_cabac(h);
3697  int eos;
3698 //STOP_TIMER("decode_mb_cabac")
3699 
3700  if(ret>=0) ff_h264_hl_decode_mb(h);
3701 
3702  if( ret >= 0 && FRAME_MBAFF ) { //FIXME optimal? or let mb_decode decode 16x32 ?
3703  s->mb_y++;
3704 
3705  ret = ff_h264_decode_mb_cabac(h);
3706 
3707  if(ret>=0) ff_h264_hl_decode_mb(h);
3708  s->mb_y--;
3709  }
3710  eos = get_cabac_terminate( &h->cabac );
3711 
3713  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3714  if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
3715  return 0;
3716  }
3717  if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3718  av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
3719  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3720  return -1;
3721  }
3722 
3723  if( ++s->mb_x >= s->mb_width ) {
3724  loop_filter(h, lf_x_start, s->mb_x);
3725  s->mb_x = lf_x_start = 0;
3726  decode_finish_row(h);
3727  ++s->mb_y;
3729  ++s->mb_y;
3730  if(FRAME_MBAFF && s->mb_y < s->mb_height)
3732  }
3733  }
3734 
3735  if( eos || s->mb_y >= s->mb_height ) {
3736  tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3737  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3738  if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3739  return 0;
3740  }
3741  }
3742 
3743  } else {
3744  for(;;){
3745  int ret = ff_h264_decode_mb_cavlc(h);
3746 
3747  if(ret>=0) ff_h264_hl_decode_mb(h);
3748 
3749  if(ret>=0 && FRAME_MBAFF){ //FIXME optimal? or let mb_decode decode 16x32 ?
3750  s->mb_y++;
3751  ret = ff_h264_decode_mb_cavlc(h);
3752 
3753  if(ret>=0) ff_h264_hl_decode_mb(h);
3754  s->mb_y--;
3755  }
3756 
3757  if(ret<0){
3758  av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3759  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3760  return -1;
3761  }
3762 
3763  if(++s->mb_x >= s->mb_width){
3764  loop_filter(h, lf_x_start, s->mb_x);
3765  s->mb_x = lf_x_start = 0;
3766  decode_finish_row(h);
3767  ++s->mb_y;
3769  ++s->mb_y;
3770  if(FRAME_MBAFF && s->mb_y < s->mb_height)
3772  }
3773  if(s->mb_y >= s->mb_height){
3774  tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3775 
3776  if (get_bits_left(&s->gb) == 0) {
3777  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3778 
3779  return 0;
3780  } else {
3782  s->mb_x - 1, s->mb_y,
3783  ER_MB_END & part_mask);
3784  return -1;
3785  }
3786  }
3787  }
3788 
3789  if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0){
3790  tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3791  if (get_bits_left(&s->gb) == 0) {
3792  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
3793  if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
3794 
3795  return 0;
3796  }else{
3797  ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
3798 
3799  return -1;
3800  }
3801  }
3802  }
3803  }
3804 }
3805 
3812 static int execute_decode_slices(H264Context *h, int context_count){
3813  MpegEncContext * const s = &h->s;
3814  AVCodecContext * const avctx= s->avctx;
3815  H264Context *hx;
3816  int i;
3817 
3819  return 0;
3820  if(context_count == 1) {
3821  return decode_slice(avctx, &h);
3822  } else {
3823  for(i = 1; i < context_count; i++) {
3824  hx = h->thread_context[i];
3825  hx->s.err_recognition = avctx->err_recognition;
3826  hx->s.error_count = 0;
3827  }
3828 
3829  avctx->execute(avctx, decode_slice,
3830  h->thread_context, NULL, context_count, sizeof(void*));
3831 
3832  /* pull back stuff from slices to master context */
3833  hx = h->thread_context[context_count - 1];
3834  s->mb_x = hx->s.mb_x;
3835  s->mb_y = hx->s.mb_y;
3836  s->dropable = hx->s.dropable;
3838  for(i = 1; i < context_count; i++)
3839  h->s.error_count += h->thread_context[i]->s.error_count;
3840  }
3841 
3842  return 0;
3843 }
3844 
3845 
3846 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
3847  MpegEncContext * const s = &h->s;
3848  AVCodecContext * const avctx= s->avctx;
3849  H264Context *hx;
3850  int buf_index;
3851  int context_count;
3852  int next_avc;
3853  int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3854  int nals_needed=0;
3855  int nal_index;
3856 
3858  if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
3859  h->current_slice = 0;
3860  if (!s->first_field)
3862  ff_h264_reset_sei(h);
3863  }
3864 
3865  for(;pass <= 1;pass++){
3866  buf_index = 0;
3867  context_count = 0;
3868  next_avc = h->is_avc ? 0 : buf_size;
3869  nal_index = 0;
3870  for(;;){
3871  int consumed;
3872  int dst_length;
3873  int bit_length;
3874  const uint8_t *ptr;
3875  int i, nalsize = 0;
3876  int err;
3877 
3878  if(buf_index >= next_avc) {
3879  if (buf_index >= buf_size - h->nal_length_size) break;
3880  nalsize = 0;
3881  for(i = 0; i < h->nal_length_size; i++)
3882  nalsize = (nalsize << 8) | buf[buf_index++];
3883  if(nalsize <= 0 || nalsize > buf_size - buf_index){
3884  av_log(h->s.avctx, AV_LOG_ERROR, "AVC: nal size %d\n", nalsize);
3885  break;
3886  }
3887  next_avc= buf_index + nalsize;
3888  } else {
3889  // start code prefix search
3890  for(; buf_index + 3 < next_avc; buf_index++){
3891  // This should always succeed in the first iteration.
3892  if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
3893  break;
3894  }
3895 
3896 
3897  if (buf_index + 3 >= buf_size) {
3898  buf_index = buf_size;
3899  break;
3900  }
3901 
3902  buf_index+=3;
3903  if(buf_index >= next_avc) continue;
3904  }
3905 
3906  hx = h->thread_context[context_count];
3907 
3908  ptr= ff_h264_decode_nal(hx, buf + buf_index, &dst_length, &consumed, next_avc - buf_index);
3909  if (ptr == NULL || dst_length < 0) {
3910  buf_index = -1;
3911  goto end;
3912  }
3913  i= buf_index + consumed;
3914  if((s->workaround_bugs & FF_BUG_AUTODETECT) && i+3<next_avc &&
3915  buf[i]==0x00 && buf[i+1]==0x00 && buf[i+2]==0x01 && buf[i+3]==0xE0)
3917 
3918  if(!(s->workaround_bugs & FF_BUG_TRUNCATED)){
3919  while(ptr[dst_length - 1] == 0 && dst_length > 0)
3920  dst_length--;
3921  }
3922  bit_length= !dst_length ? 0 : (8*dst_length - ff_h264_decode_rbsp_trailing(h, ptr + dst_length - 1));
3923 
3924  if(s->avctx->debug&FF_DEBUG_STARTCODE){
3925  av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", hx->nal_unit_type, buf_index, buf_size, dst_length);
3926  }
3927 
3928  if (h->is_avc && (nalsize != consumed) && nalsize){
3929  av_log(h->s.avctx, AV_LOG_DEBUG, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
3930  }
3931 
3932  buf_index += consumed;
3933  nal_index++;
3934 
3935  if(pass == 0) {
3936  // packets can sometimes contain multiple PPS/SPS
3937  // e.g. two PAFF field pictures in one packet, or a demuxer which splits NALs strangely
3938  // if so, when frame threading we can't start the next thread until we've read all of them
3939  switch (hx->nal_unit_type) {
3940  case NAL_SPS:
3941  case NAL_PPS:
3942  nals_needed = nal_index;
3943  break;
3944  case NAL_IDR_SLICE:
3945  case NAL_SLICE:
3946  init_get_bits(&hx->s.gb, ptr, bit_length);
3947  if (!get_ue_golomb(&hx->s.gb))
3948  nals_needed = nal_index;
3949  }
3950  continue;
3951  }
3952 
3953  //FIXME do not discard SEI id
3954  if(avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3955  continue;
3956 
3957  again:
3958  err = 0;
3959  switch(hx->nal_unit_type){
3960  case NAL_IDR_SLICE:
3961  if (h->nal_unit_type != NAL_IDR_SLICE) {
3962  av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
3963  buf_index = -1;
3964  goto end;
3965  }
3966  idr(h); // FIXME ensure we don't lose some frames if there is reordering
3967  case NAL_SLICE:
3968  init_get_bits(&hx->s.gb, ptr, bit_length);
3969  hx->intra_gb_ptr=
3970  hx->inter_gb_ptr= &hx->s.gb;
3971  hx->s.data_partitioning = 0;
3972 
3973  if((err = decode_slice_header(hx, h)))
3974  break;
3975 
3977  (hx->nal_unit_type == NAL_IDR_SLICE) ||
3978  (h->sei_recovery_frame_cnt >= 0);
3979 
3980  if (h->current_slice == 1) {
3981  if(!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3982  decode_postinit(h, nal_index >= nals_needed);
3983  }
3984 
3985  if (s->avctx->hwaccel && s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3986  return -1;
3989  }
3990 
3991  if(hx->redundant_pic_count==0
3992  && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
3995  && avctx->skip_frame < AVDISCARD_ALL){
3996  if(avctx->hwaccel) {
3997  if (avctx->hwaccel->decode_slice(avctx, &buf[buf_index - consumed], consumed) < 0)
3998  return -1;
3999  }else
4001  static const uint8_t start_code[] = {0x00, 0x00, 0x01};
4002  ff_vdpau_add_data_chunk(s, start_code, sizeof(start_code));
4003  ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed], consumed );
4004  }else
4005  context_count++;
4006  }
4007  break;
4008  case NAL_DPA:
4009  init_get_bits(&hx->s.gb, ptr, bit_length);
4010  hx->intra_gb_ptr=
4011  hx->inter_gb_ptr= NULL;
4012 
4013  if ((err = decode_slice_header(hx, h)) < 0)
4014  break;
4015 
4016  hx->s.data_partitioning = 1;
4017 
4018  break;
4019  case NAL_DPB:
4020  init_get_bits(&hx->intra_gb, ptr, bit_length);
4021  hx->intra_gb_ptr= &hx->intra_gb;
4022  break;
4023  case NAL_DPC:
4024  init_get_bits(&hx->inter_gb, ptr, bit_length);
4025  hx->inter_gb_ptr= &hx->inter_gb;
4026 
4027  if(hx->redundant_pic_count==0 && hx->intra_gb_ptr && hx->s.data_partitioning
4028  && s->current_picture_ptr
4029  && s->context_initialized
4030  && (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc)
4033  && avctx->skip_frame < AVDISCARD_ALL)
4034  context_count++;
4035  break;
4036  case NAL_SEI:
4037  init_get_bits(&s->gb, ptr, bit_length);
4038  ff_h264_decode_sei(h);
4039  break;
4040  case NAL_SPS:
4041  init_get_bits(&s->gb, ptr, bit_length);
4042  if (ff_h264_decode_seq_parameter_set(h) < 0 &&
4043  h->is_avc && (nalsize != consumed) && nalsize) {
4044  av_log(h->s.avctx, AV_LOG_DEBUG, "SPS decoding failure, "
4045  "try parsing the coomplete NAL\n");
4046  init_get_bits(&s->gb, buf + buf_index + 1 - consumed,
4047  8 * (nalsize - 1));
4049  }
4050 
4051  if (s->flags & CODEC_FLAG_LOW_DELAY ||
4053  !h->sps.num_reorder_frames)) {
4054  if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
4055  av_log(avctx, AV_LOG_WARNING, "Delayed frames seen "
4056  "reenabling low delay requires a codec "
4057  "flush.\n");
4058  else
4059  s->low_delay = 1;
4060  }
4061 
4062  if(avctx->has_b_frames < 2)
4063  avctx->has_b_frames= !s->low_delay;
4064 
4065  if (h->sps.bit_depth_luma != h->sps.bit_depth_chroma) {
4067  "Different bit depth between chroma and luma", 1);
4068  return AVERROR_PATCHWELCOME;
4069  }
4070 
4071  if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
4073  if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
4076  h->pixel_shift = h->sps.bit_depth_luma > 8;
4077 
4080  s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
4081  dsputil_init(&s->dsp, s->avctx);
4082  } else {
4083  av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
4084  buf_index = -1;
4085  goto end;
4086  }
4087  }
4088  break;
4089  case NAL_PPS:
4090  init_get_bits(&s->gb, ptr, bit_length);
4091 
4092  ff_h264_decode_picture_parameter_set(h, bit_length);
4093 
4094  break;
4095  case NAL_AUD:
4096  case NAL_END_SEQUENCE:
4097  case NAL_END_STREAM:
4098  case NAL_FILLER_DATA:
4099  case NAL_SPS_EXT:
4100  case NAL_AUXILIARY_SLICE:
4101  break;
4102  default:
4103  av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n", hx->nal_unit_type, bit_length);
4104  }
4105 
4106  if(context_count == h->max_contexts) {
4107  execute_decode_slices(h, context_count);
4108  context_count = 0;
4109  }
4110 
4111  if (err < 0)
4112  av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4113  else if(err == 1) {
4114  /* Slice could not be decoded in parallel mode, copy down
4115  * NAL unit stuff to context 0 and restart. Note that
4116  * rbsp_buffer is not transferred, but since we no longer
4117  * run in parallel mode this should not be an issue. */
4118  h->nal_unit_type = hx->nal_unit_type;
4119  h->nal_ref_idc = hx->nal_ref_idc;
4120  hx = h;
4121  goto again;
4122  }
4123  }
4124  }
4125  if(context_count)
4126  execute_decode_slices(h, context_count);
4127 
4128 end:
4129  /* clean up */
4130  if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4131  !s->dropable) {
4134  }
4135 
4136  return buf_index;
4137 }
4138 
4142 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4143  if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
4144  if(pos+10>buf_size) pos=buf_size; // oops ;)
4145 
4146  return pos;
4147 }
4148 
4149 static int decode_frame(AVCodecContext *avctx,
4150  void *data, int *data_size,
4151  AVPacket *avpkt)
4152 {
4153  const uint8_t *buf = avpkt->data;
4154  int buf_size = avpkt->size;
4155  H264Context *h = avctx->priv_data;
4156  MpegEncContext *s = &h->s;
4157  AVFrame *pict = data;
4158  int buf_index = 0;
4159 
4160  s->flags= avctx->flags;
4161  s->flags2= avctx->flags2;
4162 
4163  /* end of stream, output what is still in the buffers */
4164  out:
4165  if (buf_size == 0) {
4166  Picture *out;
4167  int i, out_idx;
4168 
4170 
4171 //FIXME factorize this with the output code below
4172  out = h->delayed_pic[0];
4173  out_idx = 0;
4174  for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
4175  if(h->delayed_pic[i]->poc < out->poc){
4176  out = h->delayed_pic[i];
4177  out_idx = i;
4178  }
4179 
4180  for(i=out_idx; h->delayed_pic[i]; i++)
4181  h->delayed_pic[i] = h->delayed_pic[i+1];
4182 
4183  if(out){
4184  *data_size = sizeof(AVFrame);
4185  *pict= *(AVFrame*)out;
4186  }
4187 
4188  return buf_index;
4189  }
4190 
4191  buf_index=decode_nal_units(h, buf, buf_size);
4192  if(buf_index < 0)
4193  return -1;
4194 
4196  buf_size = 0;
4197  goto out;
4198  }
4199 
4200  if(!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr){
4201  if (avctx->skip_frame >= AVDISCARD_NONREF)
4202  return 0;
4203  av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4204  return -1;
4205  }
4206 
4207  if(!(s->flags2 & CODEC_FLAG2_CHUNKS) || (s->mb_y >= s->mb_height && s->mb_height)){
4208 
4209  if(s->flags2 & CODEC_FLAG2_CHUNKS) decode_postinit(h, 1);
4210 
4211  field_end(h, 0);
4212 
4213  if (!h->next_output_pic) {
4214  /* Wait for second field. */
4215  *data_size = 0;
4216 
4217  } else {
4218  *data_size = sizeof(AVFrame);
4219  *pict = *(AVFrame*)h->next_output_pic;
4220  }
4221  }
4222 
4223  assert(pict->data[0] || !*data_size);
4224  ff_print_debug_info(s, pict);
4225 //printf("out %d\n", (int)pict->data[0]);
4226 
4227  return get_consumed_bytes(s, buf_index, buf_size);
4228 }
4229 #if 0
4230 static inline void fill_mb_avail(H264Context *h){
4231  MpegEncContext * const s = &h->s;
4232  const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4233 
4234  if(s->mb_y){
4235  h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4236  h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
4237  h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4238  }else{
4239  h->mb_avail[0]=
4240  h->mb_avail[1]=
4241  h->mb_avail[2]= 0;
4242  }
4243  h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4244  h->mb_avail[4]= 1; //FIXME move out
4245  h->mb_avail[5]= 0; //FIXME move out
4246 }
4247 #endif
4248 
4249 #ifdef TEST
4250 #undef printf
4251 #undef random
4252 #define COUNT 8000
4253 #define SIZE (COUNT*40)
4254 int main(void){
4255  int i;
4256  uint8_t temp[SIZE];
4257  PutBitContext pb;
4258  GetBitContext gb;
4259  DSPContext dsp;
4260  AVCodecContext avctx;
4261 
4262  avctx.av_class = avcodec_get_class();
4263  dsputil_init(&dsp, &avctx);
4264 
4265  init_put_bits(&pb, temp, SIZE);
4266  printf("testing unsigned exp golomb\n");
4267  for(i=0; i<COUNT; i++){
4268  START_TIMER
4269  set_ue_golomb(&pb, i);
4270  STOP_TIMER("set_ue_golomb");
4271  }
4272  flush_put_bits(&pb);
4273 
4274  init_get_bits(&gb, temp, 8*SIZE);
4275  for(i=0; i<COUNT; i++){
4276  int j, s = show_bits(&gb, 24);
4277 
4278  START_TIMER
4279  j= get_ue_golomb(&gb);
4280  if(j != i){
4281  printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4282 // return -1;
4283  }
4284  STOP_TIMER("get_ue_golomb");
4285  }
4286 
4287 
4288  init_put_bits(&pb, temp, SIZE);
4289  printf("testing signed exp golomb\n");
4290  for(i=0; i<COUNT; i++){
4291  START_TIMER
4292  set_se_golomb(&pb, i - COUNT/2);
4293  STOP_TIMER("set_se_golomb");
4294  }
4295  flush_put_bits(&pb);
4296 
4297  init_get_bits(&gb, temp, 8*SIZE);
4298  for(i=0; i<COUNT; i++){
4299  int j, s = show_bits(&gb, 24);
4300 
4301  START_TIMER
4302  j= get_se_golomb(&gb);
4303  if(j != i - COUNT/2){
4304  printf("mismatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4305 // return -1;
4306  }
4307  STOP_TIMER("get_se_golomb");
4308  }
4309 
4310  printf("Testing RBSP\n");
4311 
4312 
4313  return 0;
4314 }
4315 #endif /* TEST */
4316 
4317 
4319 {
4320  int i;
4321 
4322  free_tables(h, 1); //FIXME cleanup init stuff perhaps
4323 
4324  for(i = 0; i < MAX_SPS_COUNT; i++)
4325  av_freep(h->sps_buffers + i);
4326 
4327  for(i = 0; i < MAX_PPS_COUNT; i++)
4328  av_freep(h->pps_buffers + i);
4329 }
4330 
4332 {
4333  H264Context *h = avctx->priv_data;
4334  MpegEncContext *s = &h->s;
4335 
4337 
4338  MPV_common_end(s);
4339 
4340 // memset(h, 0, sizeof(H264Context));
4341 
4342  return 0;
4343 }
4344 
4345 static const AVProfile profiles[] = {
4346  { FF_PROFILE_H264_BASELINE, "Baseline" },
4347  { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline" },
4348  { FF_PROFILE_H264_MAIN, "Main" },
4349  { FF_PROFILE_H264_EXTENDED, "Extended" },
4350  { FF_PROFILE_H264_HIGH, "High" },
4351  { FF_PROFILE_H264_HIGH_10, "High 10" },
4352  { FF_PROFILE_H264_HIGH_10_INTRA, "High 10 Intra" },
4353  { FF_PROFILE_H264_HIGH_422, "High 4:2:2" },
4354  { FF_PROFILE_H264_HIGH_422_INTRA, "High 4:2:2 Intra" },
4355  { FF_PROFILE_H264_HIGH_444, "High 4:4:4" },
4356  { FF_PROFILE_H264_HIGH_444_PREDICTIVE, "High 4:4:4 Predictive" },
4357  { FF_PROFILE_H264_HIGH_444_INTRA, "High 4:4:4 Intra" },
4358  { FF_PROFILE_H264_CAVLC_444, "CAVLC 4:4:4" },
4359  { FF_PROFILE_UNKNOWN },
4360 };
4361 
4363  .name = "h264",
4364  .type = AVMEDIA_TYPE_VIDEO,
4365  .id = CODEC_ID_H264,
4366  .priv_data_size = sizeof(H264Context),
4369  .decode = decode_frame,
4370  .capabilities = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
4372  .flush= flush_dpb,
4373  .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4374  .init_thread_copy = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4375  .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4376  .profiles = NULL_IF_CONFIG_SMALL(profiles),
4377 };
4378 
4379 #if CONFIG_H264_VDPAU_DECODER
4380 AVCodec ff_h264_vdpau_decoder = {
4381  .name = "h264_vdpau",
4382  .type = AVMEDIA_TYPE_VIDEO,
4383  .id = CODEC_ID_H264,
4384  .priv_data_size = sizeof(H264Context),
4387  .decode = decode_frame,
4389  .flush= flush_dpb,
4390  .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4391  .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},
4392  .profiles = NULL_IF_CONFIG_SMALL(profiles),
4393 };
4394 #endif