h264_direct.c
Go to the documentation of this file.
1 /*
2  * H.26L/H.264/AVC/JVT/14496-10/... direct mb/block decoding
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 "internal.h"
29 #include "dsputil.h"
30 #include "avcodec.h"
31 #include "mpegvideo.h"
32 #include "h264.h"
33 #include "rectangle.h"
34 #include "thread.h"
35 
36 //#undef NDEBUG
37 #include <assert.h>
38 
39 
40 static int get_scale_factor(H264Context * const h, int poc, int poc1, int i){
41  int poc0 = h->ref_list[0][i].poc;
42  int td = av_clip(poc1 - poc0, -128, 127);
43  if(td == 0 || h->ref_list[0][i].long_ref){
44  return 256;
45  }else{
46  int tb = av_clip(poc - poc0, -128, 127);
47  int tx = (16384 + (FFABS(td) >> 1)) / td;
48  return av_clip((tb*tx + 32) >> 6, -1024, 1023);
49  }
50 }
51 
53  MpegEncContext * const s = &h->s;
55  const int poc1 = h->ref_list[1][0].poc;
56  int i, field;
57  for(field=0; field<2; field++){
58  const int poc = h->s.current_picture_ptr->field_poc[field];
59  const int poc1 = h->ref_list[1][0].field_poc[field];
60  for(i=0; i < 2*h->ref_count[0]; i++)
61  h->dist_scale_factor_field[field][i^field] = get_scale_factor(h, poc, poc1, i+16);
62  }
63 
64  for(i=0; i<h->ref_count[0]; i++){
65  h->dist_scale_factor[i] = get_scale_factor(h, poc, poc1, i);
66  }
67 }
68 
69 static void fill_colmap(H264Context *h, int map[2][16+32], int list, int field, int colfield, int mbafi){
70  MpegEncContext * const s = &h->s;
71  Picture * const ref1 = &h->ref_list[1][0];
72  int j, old_ref, rfield;
73  int start= mbafi ? 16 : 0;
74  int end = mbafi ? 16+2*h->ref_count[0] : h->ref_count[0];
75  int interl= mbafi || s->picture_structure != PICT_FRAME;
76 
77  /* bogus; fills in for missing frames */
78  memset(map[list], 0, sizeof(map[list]));
79 
80  for(rfield=0; rfield<2; rfield++){
81  for(old_ref=0; old_ref<ref1->ref_count[colfield][list]; old_ref++){
82  int poc = ref1->ref_poc[colfield][list][old_ref];
83 
84  if (!interl)
85  poc |= 3;
86  else if( interl && (poc&3) == 3) //FIXME store all MBAFF references so this isnt needed
87  poc= (poc&~3) + rfield + 1;
88 
89  for(j=start; j<end; j++){
90  if (4 * h->ref_list[0][j].frame_num + (h->ref_list[0][j].f.reference & 3) == poc) {
91  int cur_ref= mbafi ? (j-16)^field : j;
92  map[list][2*old_ref + (rfield^field) + 16] = cur_ref;
93  if(rfield == field || !interl)
94  map[list][old_ref] = cur_ref;
95  break;
96  }
97  }
98  }
99  }
100 }
101 
103  MpegEncContext * const s = &h->s;
104  Picture * const ref1 = &h->ref_list[1][0];
105  Picture * const cur = s->current_picture_ptr;
106  int list, j, field;
107  int sidx= (s->picture_structure&1)^1;
108  int ref1sidx = (ref1->f.reference&1)^1;
109 
110  for(list=0; list<2; list++){
111  cur->ref_count[sidx][list] = h->ref_count[list];
112  for(j=0; j<h->ref_count[list]; j++)
113  cur->ref_poc[sidx][list][j] = 4 * h->ref_list[list][j].frame_num + (h->ref_list[list][j].f.reference & 3);
114  }
115 
116  if(s->picture_structure == PICT_FRAME){
117  memcpy(cur->ref_count[1], cur->ref_count[0], sizeof(cur->ref_count[0]));
118  memcpy(cur->ref_poc [1], cur->ref_poc [0], sizeof(cur->ref_poc [0]));
119  }
120 
121  cur->mbaff= FRAME_MBAFF;
122 
123  h->col_fieldoff= 0;
124  if(s->picture_structure == PICT_FRAME){
125  int cur_poc = s->current_picture_ptr->poc;
126  int *col_poc = h->ref_list[1]->field_poc;
127  h->col_parity= (FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc));
128  ref1sidx=sidx= h->col_parity;
129  } else if (!(s->picture_structure & h->ref_list[1][0].f.reference) && !h->ref_list[1][0].mbaff) { // FL -> FL & differ parity
130  h->col_fieldoff = 2 * h->ref_list[1][0].f.reference - 3;
131  }
132 
134  return;
135 
136  for(list=0; list<2; list++){
137  fill_colmap(h, h->map_col_to_list0, list, sidx, ref1sidx, 0);
138  if(FRAME_MBAFF)
139  for(field=0; field<2; field++)
140  fill_colmap(h, h->map_col_to_list0_field[field], list, field, field, 1);
141  }
142 }
143 
144 static void await_reference_mb_row(H264Context * const h, Picture *ref, int mb_y)
145 {
146  int ref_field = ref->f.reference - 1;
147  int ref_field_picture = ref->field_picture;
148  int ref_height = 16*h->s.mb_height >> ref_field_picture;
149 
151  return;
152 
153  //FIXME it can be safe to access mb stuff
154  //even if pixels aren't deblocked yet
155 
156  ff_thread_await_progress((AVFrame*)ref, FFMIN(16*mb_y >> ref_field_picture, ref_height-1),
157  ref_field_picture && ref_field);
158 }
159 
160 static void pred_spatial_direct_motion(H264Context * const h, int *mb_type){
161  MpegEncContext * const s = &h->s;
162  int b8_stride = 2;
163  int b4_stride = h->b_stride;
164  int mb_xy = h->mb_xy, mb_y = s->mb_y;
165  int mb_type_col[2];
166  const int16_t (*l1mv0)[2], (*l1mv1)[2];
167  const int8_t *l1ref0, *l1ref1;
168  const int is_b8x8 = IS_8X8(*mb_type);
169  unsigned int sub_mb_type= MB_TYPE_L0L1;
170  int i8, i4;
171  int ref[2];
172  int mv[2];
173  int list;
174 
175  assert(h->ref_list[1][0].f.reference & 3);
176 
177  await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
178 
179 #define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
180 
181 
182  /* ref = min(neighbors) */
183  for(list=0; list<2; list++){
184  int left_ref = h->ref_cache[list][scan8[0] - 1];
185  int top_ref = h->ref_cache[list][scan8[0] - 8];
186  int refc = h->ref_cache[list][scan8[0] - 8 + 4];
187  const int16_t *C= h->mv_cache[list][ scan8[0] - 8 + 4];
188  if(refc == PART_NOT_AVAILABLE){
189  refc = h->ref_cache[list][scan8[0] - 8 - 1];
190  C = h-> mv_cache[list][scan8[0] - 8 - 1];
191  }
192  ref[list] = FFMIN3((unsigned)left_ref, (unsigned)top_ref, (unsigned)refc);
193  if(ref[list] >= 0){
194  //this is just pred_motion() but with the cases removed that cannot happen for direct blocks
195  const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
196  const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
197 
198  int match_count= (left_ref==ref[list]) + (top_ref==ref[list]) + (refc==ref[list]);
199  if(match_count > 1){ //most common
200  mv[list]= pack16to32(mid_pred(A[0], B[0], C[0]),
201  mid_pred(A[1], B[1], C[1]) );
202  }else {
203  assert(match_count==1);
204  if(left_ref==ref[list]){
205  mv[list]= AV_RN32A(A);
206  }else if(top_ref==ref[list]){
207  mv[list]= AV_RN32A(B);
208  }else{
209  mv[list]= AV_RN32A(C);
210  }
211  }
212  }else{
213  int mask= ~(MB_TYPE_L0 << (2*list));
214  mv[list] = 0;
215  ref[list] = -1;
216  if(!is_b8x8)
217  *mb_type &= mask;
218  sub_mb_type &= mask;
219  }
220  }
221  if(ref[0] < 0 && ref[1] < 0){
222  ref[0] = ref[1] = 0;
223  if(!is_b8x8)
224  *mb_type |= MB_TYPE_L0L1;
225  sub_mb_type |= MB_TYPE_L0L1;
226  }
227 
228  if(!(is_b8x8|mv[0]|mv[1])){
229  fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
230  fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
231  fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
232  fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
234  return;
235  }
236 
237  if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
238  if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
239  mb_y = (s->mb_y&~1) + h->col_parity;
240  mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
241  b8_stride = 0;
242  }else{
243  mb_y += h->col_fieldoff;
244  mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
245  }
246  goto single_col;
247  }else{ // AFL/AFR/FR/FL -> AFR/FR
248  if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
249  mb_y = s->mb_y&~1;
250  mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
251  mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
252  mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
253  b8_stride = 2+4*s->mb_stride;
254  b4_stride *= 6;
255  if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
256  mb_type_col[0] &= ~MB_TYPE_INTERLACED;
257  mb_type_col[1] &= ~MB_TYPE_INTERLACED;
258  }
259 
260  sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
261  if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
262  && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
263  && !is_b8x8){
264  *mb_type |= MB_TYPE_16x8 |MB_TYPE_DIRECT2; /* B_16x8 */
265  }else{
266  *mb_type |= MB_TYPE_8x8;
267  }
268  }else{ // AFR/FR -> AFR/FR
269 single_col:
270  mb_type_col[0] =
271  mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
272 
273  sub_mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
274  if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
275  *mb_type |= MB_TYPE_16x16|MB_TYPE_DIRECT2; /* B_16x16 */
276  }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
277  *mb_type |= MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
278  }else{
280  /* FIXME save sub mb types from previous frames (or derive from MVs)
281  * so we know exactly what block size to use */
282  sub_mb_type += (MB_TYPE_8x8-MB_TYPE_16x16); /* B_SUB_4x4 */
283  }
284  *mb_type |= MB_TYPE_8x8;
285  }
286  }
287  }
288 
289  await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
290 
291  l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
292  l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
293  l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
294  l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
295  if(!b8_stride){
296  if(s->mb_y&1){
297  l1ref0 += 2;
298  l1ref1 += 2;
299  l1mv0 += 2*b4_stride;
300  l1mv1 += 2*b4_stride;
301  }
302  }
303 
304 
305  if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
306  int n=0;
307  for(i8=0; i8<4; i8++){
308  int x8 = i8&1;
309  int y8 = i8>>1;
310  int xy8 = x8+y8*b8_stride;
311  int xy4 = 3*x8+y8*b4_stride;
312  int a,b;
313 
314  if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
315  continue;
316  h->sub_mb_type[i8] = sub_mb_type;
317 
318  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
319  fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
320  if(!IS_INTRA(mb_type_col[y8]) && !h->ref_list[1][0].long_ref
321  && ( (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
322  || (l1ref0[xy8] < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
323  a=b=0;
324  if(ref[0] > 0)
325  a= mv[0];
326  if(ref[1] > 0)
327  b= mv[1];
328  n++;
329  }else{
330  a= mv[0];
331  b= mv[1];
332  }
333  fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
334  fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
335  }
336  if(!is_b8x8 && !(n&3))
338  }else if(IS_16X16(*mb_type)){
339  int a,b;
340 
341  fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
342  fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
343  if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref
344  && ( (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
345  || (l1ref0[0] < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
346  && h->x264_build>33U))){
347  a=b=0;
348  if(ref[0] > 0)
349  a= mv[0];
350  if(ref[1] > 0)
351  b= mv[1];
352  }else{
353  a= mv[0];
354  b= mv[1];
355  }
356  fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
357  fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
358  }else{
359  int n=0;
360  for(i8=0; i8<4; i8++){
361  const int x8 = i8&1;
362  const int y8 = i8>>1;
363 
364  if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
365  continue;
366  h->sub_mb_type[i8] = sub_mb_type;
367 
368  fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, mv[0], 4);
369  fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, mv[1], 4);
370  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
371  fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
372 
373  assert(b8_stride==2);
374  /* col_zero_flag */
375  if(!IS_INTRA(mb_type_col[0]) && !h->ref_list[1][0].long_ref && ( l1ref0[i8] == 0
376  || (l1ref0[i8] < 0 && l1ref1[i8] == 0
377  && h->x264_build>33U))){
378  const int16_t (*l1mv)[2]= l1ref0[i8] == 0 ? l1mv0 : l1mv1;
379  if(IS_SUB_8X8(sub_mb_type)){
380  const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
381  if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
382  if(ref[0] == 0)
383  fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
384  if(ref[1] == 0)
385  fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
386  n+=4;
387  }
388  }else{
389  int m=0;
390  for(i4=0; i4<4; i4++){
391  const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
392  if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
393  if(ref[0] == 0)
394  AV_ZERO32(h->mv_cache[0][scan8[i8*4+i4]]);
395  if(ref[1] == 0)
396  AV_ZERO32(h->mv_cache[1][scan8[i8*4+i4]]);
397  m++;
398  }
399  }
400  if(!(m&3))
402  n+=m;
403  }
404  }
405  }
406  if(!is_b8x8 && !(n&15))
408  }
409 }
410 
411 static void pred_temp_direct_motion(H264Context * const h, int *mb_type){
412  MpegEncContext * const s = &h->s;
413  int b8_stride = 2;
414  int b4_stride = h->b_stride;
415  int mb_xy = h->mb_xy, mb_y = s->mb_y;
416  int mb_type_col[2];
417  const int16_t (*l1mv0)[2], (*l1mv1)[2];
418  const int8_t *l1ref0, *l1ref1;
419  const int is_b8x8 = IS_8X8(*mb_type);
420  unsigned int sub_mb_type;
421  int i8, i4;
422 
423  assert(h->ref_list[1][0].f.reference & 3);
424 
425  await_reference_mb_row(h, &h->ref_list[1][0], s->mb_y + !!IS_INTERLACED(*mb_type));
426 
427  if (IS_INTERLACED(h->ref_list[1][0].f.mb_type[mb_xy])) { // AFL/AFR/FR/FL -> AFL/FL
428  if (!IS_INTERLACED(*mb_type)) { // AFR/FR -> AFL/FL
429  mb_y = (s->mb_y&~1) + h->col_parity;
430  mb_xy= s->mb_x + ((s->mb_y&~1) + h->col_parity)*s->mb_stride;
431  b8_stride = 0;
432  }else{
433  mb_y += h->col_fieldoff;
434  mb_xy += s->mb_stride*h->col_fieldoff; // non zero for FL -> FL & differ parity
435  }
436  goto single_col;
437  }else{ // AFL/AFR/FR/FL -> AFR/FR
438  if(IS_INTERLACED(*mb_type)){ // AFL /FL -> AFR/FR
439  mb_y = s->mb_y&~1;
440  mb_xy= s->mb_x + (s->mb_y&~1)*s->mb_stride;
441  mb_type_col[0] = h->ref_list[1][0].f.mb_type[mb_xy];
442  mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy + s->mb_stride];
443  b8_stride = 2+4*s->mb_stride;
444  b4_stride *= 6;
445  if (IS_INTERLACED(mb_type_col[0]) != IS_INTERLACED(mb_type_col[1])) {
446  mb_type_col[0] &= ~MB_TYPE_INTERLACED;
447  mb_type_col[1] &= ~MB_TYPE_INTERLACED;
448  }
449 
450  sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
451 
452  if( (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)
453  && (mb_type_col[1] & MB_TYPE_16x16_OR_INTRA)
454  && !is_b8x8){
455  *mb_type |= MB_TYPE_16x8 |MB_TYPE_L0L1|MB_TYPE_DIRECT2; /* B_16x8 */
456  }else{
457  *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
458  }
459  }else{ // AFR/FR -> AFR/FR
460 single_col:
461  mb_type_col[0] =
462  mb_type_col[1] = h->ref_list[1][0].f.mb_type[mb_xy];
463 
464  sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
465  if(!is_b8x8 && (mb_type_col[0] & MB_TYPE_16x16_OR_INTRA)){
466  *mb_type |= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
467  }else if(!is_b8x8 && (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16))){
468  *mb_type |= MB_TYPE_L0L1|MB_TYPE_DIRECT2 | (mb_type_col[0] & (MB_TYPE_16x8|MB_TYPE_8x16));
469  }else{
471  /* FIXME save sub mb types from previous frames (or derive from MVs)
472  * so we know exactly what block size to use */
473  sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
474  }
475  *mb_type |= MB_TYPE_8x8|MB_TYPE_L0L1;
476  }
477  }
478  }
479 
480  await_reference_mb_row(h, &h->ref_list[1][0], mb_y);
481 
482  l1mv0 = &h->ref_list[1][0].f.motion_val[0][h->mb2b_xy [mb_xy]];
483  l1mv1 = &h->ref_list[1][0].f.motion_val[1][h->mb2b_xy [mb_xy]];
484  l1ref0 = &h->ref_list[1][0].f.ref_index [0][4 * mb_xy];
485  l1ref1 = &h->ref_list[1][0].f.ref_index [1][4 * mb_xy];
486  if(!b8_stride){
487  if(s->mb_y&1){
488  l1ref0 += 2;
489  l1ref1 += 2;
490  l1mv0 += 2*b4_stride;
491  l1mv1 += 2*b4_stride;
492  }
493  }
494 
495  {
496  const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
497  const int *dist_scale_factor = h->dist_scale_factor;
498  int ref_offset;
499 
500  if(FRAME_MBAFF && IS_INTERLACED(*mb_type)){
501  map_col_to_list0[0] = h->map_col_to_list0_field[s->mb_y&1][0];
502  map_col_to_list0[1] = h->map_col_to_list0_field[s->mb_y&1][1];
503  dist_scale_factor =h->dist_scale_factor_field[s->mb_y&1];
504  }
505  ref_offset = (h->ref_list[1][0].mbaff<<4) & (mb_type_col[0]>>3); //if(h->ref_list[1][0].mbaff && IS_INTERLACED(mb_type_col[0])) ref_offset=16 else 0
506 
507  if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col[0])){
508  int y_shift = 2*!IS_INTERLACED(*mb_type);
509  assert(h->sps.direct_8x8_inference_flag);
510 
511  for(i8=0; i8<4; i8++){
512  const int x8 = i8&1;
513  const int y8 = i8>>1;
514  int ref0, scale;
515  const int16_t (*l1mv)[2]= l1mv0;
516 
517  if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
518  continue;
519  h->sub_mb_type[i8] = sub_mb_type;
520 
521  fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
522  if(IS_INTRA(mb_type_col[y8])){
523  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
524  fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
525  fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
526  continue;
527  }
528 
529  ref0 = l1ref0[x8 + y8*b8_stride];
530  if(ref0 >= 0)
531  ref0 = map_col_to_list0[0][ref0 + ref_offset];
532  else{
533  ref0 = map_col_to_list0[1][l1ref1[x8 + y8*b8_stride] + ref_offset];
534  l1mv= l1mv1;
535  }
536  scale = dist_scale_factor[ref0];
537  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
538 
539  {
540  const int16_t *mv_col = l1mv[x8*3 + y8*b4_stride];
541  int my_col = (mv_col[1]<<y_shift)/2;
542  int mx = (scale * mv_col[0] + 128) >> 8;
543  int my = (scale * my_col + 128) >> 8;
544  fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
545  fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-my_col), 4);
546  }
547  }
548  return;
549  }
550 
551  /* one-to-one mv scaling */
552 
553  if(IS_16X16(*mb_type)){
554  int ref, mv0, mv1;
555 
556  fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
557  if(IS_INTRA(mb_type_col[0])){
558  ref=mv0=mv1=0;
559  }else{
560  const int ref0 = l1ref0[0] >= 0 ? map_col_to_list0[0][l1ref0[0] + ref_offset]
561  : map_col_to_list0[1][l1ref1[0] + ref_offset];
562  const int scale = dist_scale_factor[ref0];
563  const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
564  int mv_l0[2];
565  mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
566  mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
567  ref= ref0;
568  mv0= pack16to32(mv_l0[0],mv_l0[1]);
569  mv1= pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
570  }
571  fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref, 1);
572  fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, mv0, 4);
573  fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, mv1, 4);
574  }else{
575  for(i8=0; i8<4; i8++){
576  const int x8 = i8&1;
577  const int y8 = i8>>1;
578  int ref0, scale;
579  const int16_t (*l1mv)[2]= l1mv0;
580 
581  if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
582  continue;
583  h->sub_mb_type[i8] = sub_mb_type;
584  fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
585  if(IS_INTRA(mb_type_col[0])){
586  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
587  fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
588  fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
589  continue;
590  }
591 
592  assert(b8_stride == 2);
593  ref0 = l1ref0[i8];
594  if(ref0 >= 0)
595  ref0 = map_col_to_list0[0][ref0 + ref_offset];
596  else{
597  ref0 = map_col_to_list0[1][l1ref1[i8] + ref_offset];
598  l1mv= l1mv1;
599  }
600  scale = dist_scale_factor[ref0];
601 
602  fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
603  if(IS_SUB_8X8(sub_mb_type)){
604  const int16_t *mv_col = l1mv[x8*3 + y8*3*b4_stride];
605  int mx = (scale * mv_col[0] + 128) >> 8;
606  int my = (scale * mv_col[1] + 128) >> 8;
607  fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
608  fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
609  }else
610  for(i4=0; i4<4; i4++){
611  const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*b4_stride];
612  int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
613  mv_l0[0] = (scale * mv_col[0] + 128) >> 8;
614  mv_l0[1] = (scale * mv_col[1] + 128) >> 8;
615  AV_WN32A(h->mv_cache[1][scan8[i8*4+i4]],
616  pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]));
617  }
618  }
619  }
620  }
621 }
622 
623 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type){
624  if(h->direct_spatial_mv_pred){
625  pred_spatial_direct_motion(h, mb_type);
626  }else{
627  pred_temp_direct_motion(h, mb_type);
628  }
629 }