motion_est_template.c
Go to the documentation of this file.
1 /*
2  * Motion estimation
3  * Copyright (c) 2002-2004 Michael Niedermayer
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 
27 //Let us hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...)
28 #define LOAD_COMMON\
29  uint32_t av_unused * const score_map= c->score_map;\
30  const int av_unused xmin= c->xmin;\
31  const int av_unused ymin= c->ymin;\
32  const int av_unused xmax= c->xmax;\
33  const int av_unused ymax= c->ymax;\
34  uint8_t *mv_penalty= c->current_mv_penalty;\
35  const int pred_x= c->pred_x;\
36  const int pred_y= c->pred_y;\
37 
38 #define CHECK_HALF_MV(dx, dy, x, y)\
39 {\
40  const int hx= 2*(x)+(dx);\
41  const int hy= 2*(y)+(dy);\
42  d= cmp_hpel(s, x, y, dx, dy, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);\
43  d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
44  COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
45 }
46 
48  int *mx_ptr, int *my_ptr, int dmin,
49  int src_index, int ref_index,
50  int size, int h)
51 {
52  MotionEstContext * const c= &s->me;
53  const int mx = *mx_ptr;
54  const int my = *my_ptr;
55  const int penalty_factor= c->sub_penalty_factor;
56  me_cmp_func cmp_sub, chroma_cmp_sub;
57  int bx=2*mx, by=2*my;
58 
60  int flags= c->sub_flags;
61 
62  //FIXME factorize
63 
64  cmp_sub= s->dsp.me_sub_cmp[size];
65  chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
66 
67  if(c->skip){ //FIXME move out of hpel?
68  *mx_ptr = 0;
69  *my_ptr = 0;
70  return dmin;
71  }
72 
73  if(c->avctx->me_cmp != c->avctx->me_sub_cmp){
74  dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
75  if(mx || my || size>0)
76  dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
77  }
78 
79  if (mx > xmin && mx < xmax &&
80  my > ymin && my < ymax) {
81  int d= dmin;
82  const int index= (my<<ME_MAP_SHIFT) + mx;
83  const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
84  + (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*c->penalty_factor;
85  const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)]
86  + (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*c->penalty_factor;
87  const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)]
88  + (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*c->penalty_factor;
89  const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
90  + (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*c->penalty_factor;
91 
92  unsigned key;
93  unsigned map_generation= c->map_generation;
94 #ifndef NDEBUG
95  uint32_t *map= c->map;
96 #endif
97  key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
98  assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
99  key= ((my+1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
100  assert(map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
101  key= ((my)<<ME_MAP_MV_BITS) + (mx+1) + map_generation;
102  assert(map[(index+1)&(ME_MAP_SIZE-1)] == key);
103  key= ((my)<<ME_MAP_MV_BITS) + (mx-1) + map_generation;
104  assert(map[(index-1)&(ME_MAP_SIZE-1)] == key);
105  if(t<=b){
106  CHECK_HALF_MV(0, 1, mx ,my-1)
107  if(l<=r){
108  CHECK_HALF_MV(1, 1, mx-1, my-1)
109  if(t+r<=b+l){
110  CHECK_HALF_MV(1, 1, mx , my-1)
111  }else{
112  CHECK_HALF_MV(1, 1, mx-1, my )
113  }
114  CHECK_HALF_MV(1, 0, mx-1, my )
115  }else{
116  CHECK_HALF_MV(1, 1, mx , my-1)
117  if(t+l<=b+r){
118  CHECK_HALF_MV(1, 1, mx-1, my-1)
119  }else{
120  CHECK_HALF_MV(1, 1, mx , my )
121  }
122  CHECK_HALF_MV(1, 0, mx , my )
123  }
124  }else{
125  if(l<=r){
126  if(t+l<=b+r){
127  CHECK_HALF_MV(1, 1, mx-1, my-1)
128  }else{
129  CHECK_HALF_MV(1, 1, mx , my )
130  }
131  CHECK_HALF_MV(1, 0, mx-1, my)
132  CHECK_HALF_MV(1, 1, mx-1, my)
133  }else{
134  if(t+r<=b+l){
135  CHECK_HALF_MV(1, 1, mx , my-1)
136  }else{
137  CHECK_HALF_MV(1, 1, mx-1, my)
138  }
139  CHECK_HALF_MV(1, 0, mx , my)
140  CHECK_HALF_MV(1, 1, mx , my)
141  }
142  CHECK_HALF_MV(0, 1, mx , my)
143  }
144  assert(bx >= xmin*2 && bx <= xmax*2 && by >= ymin*2 && by <= ymax*2);
145  }
146 
147  *mx_ptr = bx;
148  *my_ptr = by;
149 
150  return dmin;
151 }
152 
154  int *mx_ptr, int *my_ptr, int dmin,
155  int src_index, int ref_index,
156  int size, int h)
157 {
158  (*mx_ptr)<<=1;
159  (*my_ptr)<<=1;
160  return dmin;
161 }
162 
163 inline int ff_get_mb_score(MpegEncContext * s, int mx, int my, int src_index,
164  int ref_index, int size, int h, int add_rate)
165 {
166 // const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp;
167  MotionEstContext * const c= &s->me;
168  const int penalty_factor= c->mb_penalty_factor;
169  const int flags= c->mb_flags;
170  const int qpel= flags & FLAG_QPEL;
171  const int mask= 1+2*qpel;
172  me_cmp_func cmp_sub, chroma_cmp_sub;
173  int d;
174 
176 
177  //FIXME factorize
178 
179  cmp_sub= s->dsp.mb_cmp[size];
180  chroma_cmp_sub= s->dsp.mb_cmp[size+1];
181 
182 // assert(!c->skip);
183 // assert(c->avctx->me_sub_cmp != c->avctx->mb_cmp);
184 
185  d= cmp(s, mx>>(qpel+1), my>>(qpel+1), mx&mask, my&mask, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
186  //FIXME check cbp before adding penalty for (0,0) vector
187  if(add_rate && (mx || my || size>0))
188  d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
189 
190  return d;
191 }
192 
193 #define CHECK_QUARTER_MV(dx, dy, x, y)\
194 {\
195  const int hx= 4*(x)+(dx);\
196  const int hy= 4*(y)+(dy);\
197  d= cmp_qpel(s, x, y, dx, dy, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
198  d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
199  COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
200 }
201 
203  int *mx_ptr, int *my_ptr, int dmin,
204  int src_index, int ref_index,
205  int size, int h)
206 {
207  MotionEstContext * const c= &s->me;
208  const int mx = *mx_ptr;
209  const int my = *my_ptr;
210  const int penalty_factor= c->sub_penalty_factor;
211  const unsigned map_generation = c->map_generation;
212  const int subpel_quality= c->avctx->me_subpel_quality;
213  uint32_t *map= c->map;
214  me_cmp_func cmpf, chroma_cmpf;
215  me_cmp_func cmp_sub, chroma_cmp_sub;
216 
218  int flags= c->sub_flags;
219 
220  cmpf= s->dsp.me_cmp[size];
221  chroma_cmpf= s->dsp.me_cmp[size+1]; //factorize FIXME
222  //FIXME factorize
223 
224  cmp_sub= s->dsp.me_sub_cmp[size];
225  chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
226 
227  if(c->skip){ //FIXME somehow move up (benchmark)
228  *mx_ptr = 0;
229  *my_ptr = 0;
230  return dmin;
231  }
232 
233  if(c->avctx->me_cmp != c->avctx->me_sub_cmp){
234  dmin= cmp(s, mx, my, 0, 0, size, h, ref_index, src_index, cmp_sub, chroma_cmp_sub, flags);
235  if(mx || my || size>0)
236  dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor;
237  }
238 
239  if (mx > xmin && mx < xmax &&
240  my > ymin && my < ymax) {
241  int bx=4*mx, by=4*my;
242  int d= dmin;
243  int i, nx, ny;
244  const int index= (my<<ME_MAP_SHIFT) + mx;
245  const int t= score_map[(index-(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
246  const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
247  const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
248  const int b= score_map[(index+(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
249  const int c= score_map[(index )&(ME_MAP_SIZE-1)];
250  int best[8];
251  int best_pos[8][2];
252 
253  memset(best, 64, sizeof(int)*8);
254  if(s->me.dia_size>=2){
255  const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
256  const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
257  const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
258  const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
259 
260  for(ny= -3; ny <= 3; ny++){
261  for(nx= -3; nx <= 3; nx++){
262  //FIXME this could overflow (unlikely though)
263  const int64_t t2= nx*nx*(tr + tl - 2*t) + 4*nx*(tr-tl) + 32*t;
264  const int64_t c2= nx*nx*( r + l - 2*c) + 4*nx*( r- l) + 32*c;
265  const int64_t b2= nx*nx*(br + bl - 2*b) + 4*nx*(br-bl) + 32*b;
266  int score= (ny*ny*(b2 + t2 - 2*c2) + 4*ny*(b2 - t2) + 32*c2 + 512)>>10;
267  int i;
268 
269  if((nx&3)==0 && (ny&3)==0) continue;
270 
271  score += (mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
272 
273 // if(nx&1) score-=1024*c->penalty_factor;
274 // if(ny&1) score-=1024*c->penalty_factor;
275 
276  for(i=0; i<8; i++){
277  if(score < best[i]){
278  memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
279  memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
280  best[i]= score;
281  best_pos[i][0]= nx + 4*mx;
282  best_pos[i][1]= ny + 4*my;
283  break;
284  }
285  }
286  }
287  }
288  }else{
289  int tl;
290  //FIXME this could overflow (unlikely though)
291  const int cx = 4*(r - l);
292  const int cx2= r + l - 2*c;
293  const int cy = 4*(b - t);
294  const int cy2= b + t - 2*c;
295  int cxy;
296 
297  if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME
298  tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
299  }else{
300  tl= cmp(s, mx-1, my-1, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);//FIXME wrong if chroma me is different
301  }
302 
303  cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c;
304 
305  assert(16*cx2 + 4*cx + 32*c == 32*r);
306  assert(16*cx2 - 4*cx + 32*c == 32*l);
307  assert(16*cy2 + 4*cy + 32*c == 32*b);
308  assert(16*cy2 - 4*cy + 32*c == 32*t);
309  assert(16*cxy + 16*cy2 + 16*cx2 - 4*cy - 4*cx + 32*c == 32*tl);
310 
311  for(ny= -3; ny <= 3; ny++){
312  for(nx= -3; nx <= 3; nx++){
313  //FIXME this could overflow (unlikely though)
314  int score= ny*nx*cxy + nx*nx*cx2 + ny*ny*cy2 + nx*cx + ny*cy + 32*c; //FIXME factor
315  int i;
316 
317  if((nx&3)==0 && (ny&3)==0) continue;
318 
319  score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
320 // if(nx&1) score-=32*c->penalty_factor;
321  // if(ny&1) score-=32*c->penalty_factor;
322 
323  for(i=0; i<8; i++){
324  if(score < best[i]){
325  memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
326  memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
327  best[i]= score;
328  best_pos[i][0]= nx + 4*mx;
329  best_pos[i][1]= ny + 4*my;
330  break;
331  }
332  }
333  }
334  }
335  }
336  for(i=0; i<subpel_quality; i++){
337  nx= best_pos[i][0];
338  ny= best_pos[i][1];
339  CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2)
340  }
341 
342  assert(bx >= xmin*4 && bx <= xmax*4 && by >= ymin*4 && by <= ymax*4);
343 
344  *mx_ptr = bx;
345  *my_ptr = by;
346  }else{
347  *mx_ptr =4*mx;
348  *my_ptr =4*my;
349  }
350 
351  return dmin;
352 }
353 
354 
355 #define CHECK_MV(x,y)\
356 {\
357  const unsigned key = ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
358  const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
359  assert((x) >= xmin);\
360  assert((x) <= xmax);\
361  assert((y) >= ymin);\
362  assert((y) <= ymax);\
363 /*printf("check_mv %d %d\n", x, y);*/\
364  if(map[index]!=key){\
365  d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
366  map[index]= key;\
367  score_map[index]= d;\
368  d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
369 /*printf("score:%d\n", d);*/\
370  COPY3_IF_LT(dmin, d, best[0], x, best[1], y)\
371  }\
372 }
373 
374 #define CHECK_CLIPPED_MV(ax,ay)\
375 {\
376  const int Lx= ax;\
377  const int Ly= ay;\
378  const int Lx2= FFMAX(xmin, FFMIN(Lx, xmax));\
379  const int Ly2= FFMAX(ymin, FFMIN(Ly, ymax));\
380  CHECK_MV(Lx2, Ly2)\
381 }
382 
383 #define CHECK_MV_DIR(x,y,new_dir)\
384 {\
385  const unsigned key = ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
386  const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
387 /*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\
388  if(map[index]!=key){\
389  d= cmp(s, x, y, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
390  map[index]= key;\
391  score_map[index]= d;\
392  d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
393 /*printf("score:%d\n", d);*/\
394  if(d<dmin){\
395  best[0]=x;\
396  best[1]=y;\
397  dmin=d;\
398  next_dir= new_dir;\
399  }\
400  }\
401 }
402 
403 #define check(x,y,S,v)\
404 if( (x)<(xmin<<(S)) ) printf("%d %d %d %d %d xmin" #v, xmin, (x), (y), s->mb_x, s->mb_y);\
405 if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, s->mb_y);\
406 if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\
407 if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\
408 
409 #define LOAD_COMMON2\
410  uint32_t *map= c->map;\
411  const int qpel= flags&FLAG_QPEL;\
412  const int shift= 1+qpel;\
413 
414 static av_always_inline int small_diamond_search(MpegEncContext * s, int *best, int dmin,
415  int src_index, int ref_index, int const penalty_factor,
416  int size, int h, int flags)
417 {
418  MotionEstContext * const c= &s->me;
419  me_cmp_func cmpf, chroma_cmpf;
420  int next_dir=-1;
423  unsigned map_generation = c->map_generation;
424 
425  cmpf= s->dsp.me_cmp[size];
426  chroma_cmpf= s->dsp.me_cmp[size+1];
427 
428  { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
429  const unsigned key = (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
430  const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1);
431  if(map[index]!=key){ //this will be executed only very rarey
432  score_map[index]= cmp(s, best[0], best[1], 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
433  map[index]= key;
434  }
435  }
436 
437  for(;;){
438  int d;
439  const int dir= next_dir;
440  const int x= best[0];
441  const int y= best[1];
442  next_dir=-1;
443 
444 //printf("%d", dir);
445  if(dir!=2 && x>xmin) CHECK_MV_DIR(x-1, y , 0)
446  if(dir!=3 && y>ymin) CHECK_MV_DIR(x , y-1, 1)
447  if(dir!=0 && x<xmax) CHECK_MV_DIR(x+1, y , 2)
448  if(dir!=1 && y<ymax) CHECK_MV_DIR(x , y+1, 3)
449 
450  if(next_dir==-1){
451  return dmin;
452  }
453  }
454 }
455 
456 static int funny_diamond_search(MpegEncContext * s, int *best, int dmin,
457  int src_index, int ref_index, int const penalty_factor,
458  int size, int h, int flags)
459 {
460  MotionEstContext * const c= &s->me;
461  me_cmp_func cmpf, chroma_cmpf;
462  int dia_size;
465  unsigned map_generation = c->map_generation;
466 
467  cmpf= s->dsp.me_cmp[size];
468  chroma_cmpf= s->dsp.me_cmp[size+1];
469 
470  for(dia_size=1; dia_size<=4; dia_size++){
471  int dir;
472  const int x= best[0];
473  const int y= best[1];
474 
475  if(dia_size&(dia_size-1)) continue;
476 
477  if( x + dia_size > xmax
478  || x - dia_size < xmin
479  || y + dia_size > ymax
480  || y - dia_size < ymin)
481  continue;
482 
483  for(dir= 0; dir<dia_size; dir+=2){
484  int d;
485 
486  CHECK_MV(x + dir , y + dia_size - dir);
487  CHECK_MV(x + dia_size - dir, y - dir );
488  CHECK_MV(x - dir , y - dia_size + dir);
489  CHECK_MV(x - dia_size + dir, y + dir );
490  }
491 
492  if(x!=best[0] || y!=best[1])
493  dia_size=0;
494  }
495  return dmin;
496 }
497 
498 static int hex_search(MpegEncContext * s, int *best, int dmin,
499  int src_index, int ref_index, int const penalty_factor,
500  int size, int h, int flags, int dia_size)
501 {
502  MotionEstContext * const c= &s->me;
503  me_cmp_func cmpf, chroma_cmpf;
506  unsigned map_generation = c->map_generation;
507  int x,y,d;
508  const int dec= dia_size & (dia_size-1);
509 
510  cmpf= s->dsp.me_cmp[size];
511  chroma_cmpf= s->dsp.me_cmp[size+1];
512 
513  for(;dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
514  do{
515  x= best[0];
516  y= best[1];
517 
518  CHECK_CLIPPED_MV(x -dia_size , y);
519  CHECK_CLIPPED_MV(x+ dia_size , y);
520  CHECK_CLIPPED_MV(x+( dia_size>>1), y+dia_size);
521  CHECK_CLIPPED_MV(x+( dia_size>>1), y-dia_size);
522  if(dia_size>1){
523  CHECK_CLIPPED_MV(x+(-dia_size>>1), y+dia_size);
524  CHECK_CLIPPED_MV(x+(-dia_size>>1), y-dia_size);
525  }
526  }while(best[0] != x || best[1] != y);
527  }
528 
529  return dmin;
530 }
531 
532 static int l2s_dia_search(MpegEncContext * s, int *best, int dmin,
533  int src_index, int ref_index, int const penalty_factor,
534  int size, int h, int flags)
535 {
536  MotionEstContext * const c= &s->me;
537  me_cmp_func cmpf, chroma_cmpf;
540  unsigned map_generation = c->map_generation;
541  int x,y,i,d;
542  int dia_size= c->dia_size&0xFF;
543  const int dec= dia_size & (dia_size-1);
544  static const int hex[8][2]={{-2, 0}, {-1,-1}, { 0,-2}, { 1,-1},
545  { 2, 0}, { 1, 1}, { 0, 2}, {-1, 1}};
546 
547  cmpf= s->dsp.me_cmp[size];
548  chroma_cmpf= s->dsp.me_cmp[size+1];
549 
550  for(; dia_size; dia_size= dec ? dia_size-1 : dia_size>>1){
551  do{
552  x= best[0];
553  y= best[1];
554  for(i=0; i<8; i++){
555  CHECK_CLIPPED_MV(x+hex[i][0]*dia_size, y+hex[i][1]*dia_size);
556  }
557  }while(best[0] != x || best[1] != y);
558  }
559 
560  x= best[0];
561  y= best[1];
562  CHECK_CLIPPED_MV(x+1, y);
563  CHECK_CLIPPED_MV(x, y+1);
564  CHECK_CLIPPED_MV(x-1, y);
565  CHECK_CLIPPED_MV(x, y-1);
566 
567  return dmin;
568 }
569 
570 static int umh_search(MpegEncContext * s, int *best, int dmin,
571  int src_index, int ref_index, int const penalty_factor,
572  int size, int h, int flags)
573 {
574  MotionEstContext * const c= &s->me;
575  me_cmp_func cmpf, chroma_cmpf;
578  unsigned map_generation = c->map_generation;
579  int x,y,x2,y2, i, j, d;
580  const int dia_size= c->dia_size&0xFE;
581  static const int hex[16][2]={{-4,-2}, {-4,-1}, {-4, 0}, {-4, 1}, {-4, 2},
582  { 4,-2}, { 4,-1}, { 4, 0}, { 4, 1}, { 4, 2},
583  {-2, 3}, { 0, 4}, { 2, 3},
584  {-2,-3}, { 0,-4}, { 2,-3},};
585 
586  cmpf= s->dsp.me_cmp[size];
587  chroma_cmpf= s->dsp.me_cmp[size+1];
588 
589  x= best[0];
590  y= best[1];
591  for(x2=FFMAX(x-dia_size+1, xmin); x2<=FFMIN(x+dia_size-1,xmax); x2+=2){
592  CHECK_MV(x2, y);
593  }
594  for(y2=FFMAX(y-dia_size/2+1, ymin); y2<=FFMIN(y+dia_size/2-1,ymax); y2+=2){
595  CHECK_MV(x, y2);
596  }
597 
598  x= best[0];
599  y= best[1];
600  for(y2=FFMAX(y-2, ymin); y2<=FFMIN(y+2,ymax); y2++){
601  for(x2=FFMAX(x-2, xmin); x2<=FFMIN(x+2,xmax); x2++){
602  CHECK_MV(x2, y2);
603  }
604  }
605 
606 //FIXME prevent the CLIP stuff
607 
608  for(j=1; j<=dia_size/4; j++){
609  for(i=0; i<16; i++){
610  CHECK_CLIPPED_MV(x+hex[i][0]*j, y+hex[i][1]*j);
611  }
612  }
613 
614  return hex_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags, 2);
615 }
616 
617 static int full_search(MpegEncContext * s, int *best, int dmin,
618  int src_index, int ref_index, int const penalty_factor,
619  int size, int h, int flags)
620 {
621  MotionEstContext * const c= &s->me;
622  me_cmp_func cmpf, chroma_cmpf;
625  unsigned map_generation = c->map_generation;
626  int x,y, d;
627  const int dia_size= c->dia_size&0xFF;
628 
629  cmpf= s->dsp.me_cmp[size];
630  chroma_cmpf= s->dsp.me_cmp[size+1];
631 
632  for(y=FFMAX(-dia_size, ymin); y<=FFMIN(dia_size,ymax); y++){
633  for(x=FFMAX(-dia_size, xmin); x<=FFMIN(dia_size,xmax); x++){
634  CHECK_MV(x, y);
635  }
636  }
637 
638  x= best[0];
639  y= best[1];
640  d= dmin;
641  CHECK_CLIPPED_MV(x , y);
642  CHECK_CLIPPED_MV(x+1, y);
643  CHECK_CLIPPED_MV(x, y+1);
644  CHECK_CLIPPED_MV(x-1, y);
645  CHECK_CLIPPED_MV(x, y-1);
646  best[0]= x;
647  best[1]= y;
648 
649  return d;
650 }
651 
652 #define SAB_CHECK_MV(ax,ay)\
653 {\
654  const unsigned key = ((ay)<<ME_MAP_MV_BITS) + (ax) + map_generation;\
655  const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\
656 /*printf("sab check %d %d\n", ax, ay);*/\
657  if(map[index]!=key){\
658  d= cmp(s, ax, ay, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);\
659  map[index]= key;\
660  score_map[index]= d;\
661  d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\
662 /*printf("score: %d\n", d);*/\
663  if(d < minima[minima_count-1].height){\
664  int j=0;\
665  \
666  while(d >= minima[j].height) j++;\
667 \
668  memmove(&minima [j+1], &minima [j], (minima_count - j - 1)*sizeof(Minima));\
669 \
670  minima[j].checked= 0;\
671  minima[j].height= d;\
672  minima[j].x= ax;\
673  minima[j].y= ay;\
674  \
675  i=-1;\
676  continue;\
677  }\
678  }\
679 }
680 
681 #define MAX_SAB_SIZE ME_MAP_SIZE
682 static int sab_diamond_search(MpegEncContext * s, int *best, int dmin,
683  int src_index, int ref_index, int const penalty_factor,
684  int size, int h, int flags)
685 {
686  MotionEstContext * const c= &s->me;
687  me_cmp_func cmpf, chroma_cmpf;
688  Minima minima[MAX_SAB_SIZE];
689  const int minima_count= FFABS(c->dia_size);
690  int i, j;
693  unsigned map_generation = c->map_generation;
694 
695  cmpf= s->dsp.me_cmp[size];
696  chroma_cmpf= s->dsp.me_cmp[size+1];
697 
698  /*Note j<MAX_SAB_SIZE is needed if MAX_SAB_SIZE < ME_MAP_SIZE as j can
699  become larger due to MVs overflowing their ME_MAP_MV_BITS bits space in map
700  */
701  for(j=i=0; i<ME_MAP_SIZE && j<MAX_SAB_SIZE; i++){
702  uint32_t key= map[i];
703 
704  key += (1<<(ME_MAP_MV_BITS-1)) + (1<<(2*ME_MAP_MV_BITS-1));
705 
706  if((key&((-1)<<(2*ME_MAP_MV_BITS))) != map_generation) continue;
707 
708  minima[j].height= score_map[i];
709  minima[j].x= key & ((1<<ME_MAP_MV_BITS)-1); key>>=ME_MAP_MV_BITS;
710  minima[j].y= key & ((1<<ME_MAP_MV_BITS)-1);
711  minima[j].x-= (1<<(ME_MAP_MV_BITS-1));
712  minima[j].y-= (1<<(ME_MAP_MV_BITS-1));
713 
714  // all entries in map should be in range except if the mv overflows their ME_MAP_MV_BITS bits space
715  if( minima[j].x > xmax || minima[j].x < xmin
716  || minima[j].y > ymax || minima[j].y < ymin)
717  continue;
718 
719  minima[j].checked=0;
720  if(minima[j].x || minima[j].y)
721  minima[j].height+= (mv_penalty[((minima[j].x)<<shift)-pred_x] + mv_penalty[((minima[j].y)<<shift)-pred_y])*penalty_factor;
722 
723  j++;
724  }
725 
726  qsort(minima, j, sizeof(Minima), minima_cmp);
727 
728  for(; j<minima_count; j++){
729  minima[j].height=256*256*256*64;
730  minima[j].checked=0;
731  minima[j].x= minima[j].y=0;
732  }
733 
734  for(i=0; i<minima_count; i++){
735  const int x= minima[i].x;
736  const int y= minima[i].y;
737  int d;
738 
739  if(minima[i].checked) continue;
740 
741  if( x >= xmax || x <= xmin
742  || y >= ymax || y <= ymin)
743  continue;
744 
745  SAB_CHECK_MV(x-1, y)
746  SAB_CHECK_MV(x+1, y)
747  SAB_CHECK_MV(x , y-1)
748  SAB_CHECK_MV(x , y+1)
749 
750  minima[i].checked= 1;
751  }
752 
753  best[0]= minima[0].x;
754  best[1]= minima[0].y;
755  dmin= minima[0].height;
756 
757  if( best[0] < xmax && best[0] > xmin
758  && best[1] < ymax && best[1] > ymin){
759  int d;
760  //ensure that the refernece samples for hpel refinement are in the map
761  CHECK_MV(best[0]-1, best[1])
762  CHECK_MV(best[0]+1, best[1])
763  CHECK_MV(best[0], best[1]-1)
764  CHECK_MV(best[0], best[1]+1)
765  }
766  return dmin;
767 }
768 
769 static int var_diamond_search(MpegEncContext * s, int *best, int dmin,
770  int src_index, int ref_index, int const penalty_factor,
771  int size, int h, int flags)
772 {
773  MotionEstContext * const c= &s->me;
774  me_cmp_func cmpf, chroma_cmpf;
775  int dia_size;
778  unsigned map_generation = c->map_generation;
779 
780  cmpf= s->dsp.me_cmp[size];
781  chroma_cmpf= s->dsp.me_cmp[size+1];
782 
783  for(dia_size=1; dia_size<=c->dia_size; dia_size++){
784  int dir, start, end;
785  const int x= best[0];
786  const int y= best[1];
787 
788  start= FFMAX(0, y + dia_size - ymax);
789  end = FFMIN(dia_size, xmax - x + 1);
790  for(dir= start; dir<end; dir++){
791  int d;
792 
793 //check(x + dir,y + dia_size - dir,0, a0)
794  CHECK_MV(x + dir , y + dia_size - dir);
795  }
796 
797  start= FFMAX(0, x + dia_size - xmax);
798  end = FFMIN(dia_size, y - ymin + 1);
799  for(dir= start; dir<end; dir++){
800  int d;
801 
802 //check(x + dia_size - dir, y - dir,0, a1)
803  CHECK_MV(x + dia_size - dir, y - dir );
804  }
805 
806  start= FFMAX(0, -y + dia_size + ymin );
807  end = FFMIN(dia_size, x - xmin + 1);
808  for(dir= start; dir<end; dir++){
809  int d;
810 
811 //check(x - dir,y - dia_size + dir,0, a2)
812  CHECK_MV(x - dir , y - dia_size + dir);
813  }
814 
815  start= FFMAX(0, -x + dia_size + xmin );
816  end = FFMIN(dia_size, ymax - y + 1);
817  for(dir= start; dir<end; dir++){
818  int d;
819 
820 //check(x - dia_size + dir, y + dir,0, a3)
821  CHECK_MV(x - dia_size + dir, y + dir );
822  }
823 
824  if(x!=best[0] || y!=best[1])
825  dia_size=0;
826  }
827  return dmin;
828 }
829 
830 static av_always_inline int diamond_search(MpegEncContext * s, int *best, int dmin,
831  int src_index, int ref_index, int const penalty_factor,
832  int size, int h, int flags){
833  MotionEstContext * const c= &s->me;
834  if(c->dia_size==-1)
835  return funny_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
836  else if(c->dia_size<-1)
837  return sab_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
838  else if(c->dia_size<2)
839  return small_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
840  else if(c->dia_size>1024)
841  return full_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
842  else if(c->dia_size>768)
843  return umh_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
844  else if(c->dia_size>512)
845  return hex_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags, c->dia_size&0xFF);
846  else if(c->dia_size>256)
847  return l2s_dia_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
848  else
849  return var_diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
850 }
851 
858 static av_always_inline int epzs_motion_search_internal(MpegEncContext * s, int *mx_ptr, int *my_ptr,
859  int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
860  int ref_mv_scale, int flags, int size, int h)
861 {
862  MotionEstContext * const c= &s->me;
863  int best[2]={0, 0};
867  int d;
868  int dmin;
870  unsigned map_generation;
871  int penalty_factor;
872  const int ref_mv_stride= s->mb_stride; //pass as arg FIXME
873  const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; //add to last_mv beforepassing FIXME
874  me_cmp_func cmpf, chroma_cmpf;
875 
878 
879  if(c->pre_pass){
880  penalty_factor= c->pre_penalty_factor;
881  cmpf= s->dsp.me_pre_cmp[size];
882  chroma_cmpf= s->dsp.me_pre_cmp[size+1];
883  }else{
884  penalty_factor= c->penalty_factor;
885  cmpf= s->dsp.me_cmp[size];
886  chroma_cmpf= s->dsp.me_cmp[size+1];
887  }
888 
889  map_generation= update_map_generation(c);
890 
891  assert(cmpf);
892  dmin= cmp(s, 0, 0, 0, 0, size, h, ref_index, src_index, cmpf, chroma_cmpf, flags);
893  map[0]= map_generation;
894  score_map[0]= dmin;
895 
896  //FIXME precalc first term below?
897  if((s->pict_type == AV_PICTURE_TYPE_B && !(c->flags & FLAG_DIRECT)) || s->flags&CODEC_FLAG_MV0)
898  dmin += (mv_penalty[pred_x] + mv_penalty[pred_y])*penalty_factor;
899 
900  /* first line */
901  if (s->first_slice_line) {
902  CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
903  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
904  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
905  }else{
906  if(dmin<((h*h*s->avctx->mv0_threshold)>>8)
907  && ( P_LEFT[0] |P_LEFT[1]
908  |P_TOP[0] |P_TOP[1]
909  |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){
910  *mx_ptr= 0;
911  *my_ptr= 0;
912  c->skip=1;
913  return dmin;
914  }
915  CHECK_MV( P_MEDIAN[0] >>shift , P_MEDIAN[1] >>shift)
916  CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift) , (P_MEDIAN[1]>>shift)-1)
917  CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift) , (P_MEDIAN[1]>>shift)+1)
918  CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)-1, (P_MEDIAN[1]>>shift) )
919  CHECK_CLIPPED_MV((P_MEDIAN[0]>>shift)+1, (P_MEDIAN[1]>>shift) )
920  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
921  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
922  CHECK_MV(P_LEFT[0] >>shift, P_LEFT[1] >>shift)
923  CHECK_MV(P_TOP[0] >>shift, P_TOP[1] >>shift)
924  CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
925  }
926  if(dmin>h*h*4){
927  if(c->pre_pass){
928  CHECK_CLIPPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16,
929  (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16)
930  if(!s->first_slice_line)
931  CHECK_CLIPPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
932  (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
933  }else{
934  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
935  (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
936  if(s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line
937  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
938  (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
939  }
940  }
941 
942  if(c->avctx->last_predictor_count){
943  const int count= c->avctx->last_predictor_count;
944  const int xstart= FFMAX(0, s->mb_x - count);
945  const int ystart= FFMAX(0, s->mb_y - count);
946  const int xend= FFMIN(s->mb_width , s->mb_x + count + 1);
947  const int yend= FFMIN(s->mb_height, s->mb_y + count + 1);
948  int mb_y;
949 
950  for(mb_y=ystart; mb_y<yend; mb_y++){
951  int mb_x;
952  for(mb_x=xstart; mb_x<xend; mb_x++){
953  const int xy= mb_x + 1 + (mb_y + 1)*ref_mv_stride;
954  int mx= (last_mv[xy][0]*ref_mv_scale + (1<<15))>>16;
955  int my= (last_mv[xy][1]*ref_mv_scale + (1<<15))>>16;
956 
957  if(mx>xmax || mx<xmin || my>ymax || my<ymin) continue;
958  CHECK_MV(mx,my)
959  }
960  }
961  }
962 
963 //check(best[0],best[1],0, b0)
964  dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
965 
966 //check(best[0],best[1],0, b1)
967  *mx_ptr= best[0];
968  *my_ptr= best[1];
969 
970 // printf("%d %d %d \n", best[0], best[1], dmin);
971  return dmin;
972 }
973 
974 //this function is dedicated to the braindamaged gcc
975 inline int ff_epzs_motion_search(MpegEncContext * s, int *mx_ptr, int *my_ptr,
976  int P[10][2], int src_index, int ref_index, int16_t (*last_mv)[2],
977  int ref_mv_scale, int size, int h)
978 {
979  MotionEstContext * const c= &s->me;
980 //FIXME convert other functions in the same way if faster
981  if(c->flags==0 && h==16 && size==0){
982  return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, 0, 0, 16);
983 // case FLAG_QPEL:
984 // return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, FLAG_QPEL);
985  }else{
986  return epzs_motion_search_internal(s, mx_ptr, my_ptr, P, src_index, ref_index, last_mv, ref_mv_scale, c->flags, size, h);
987  }
988 }
989 
991  int *mx_ptr, int *my_ptr, int P[10][2],
992  int src_index, int ref_index, int16_t (*last_mv)[2],
993  int ref_mv_scale)
994 {
995  MotionEstContext * const c= &s->me;
996  int best[2]={0, 0};
997  int d, dmin;
998  unsigned map_generation;
999  const int penalty_factor= c->penalty_factor;
1000  const int size=1;
1001  const int h=8;
1002  const int ref_mv_stride= s->mb_stride;
1003  const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
1004  me_cmp_func cmpf, chroma_cmpf;
1005  LOAD_COMMON
1006  int flags= c->flags;
1007  LOAD_COMMON2
1008 
1009  cmpf= s->dsp.me_cmp[size];
1010  chroma_cmpf= s->dsp.me_cmp[size+1];
1011 
1012  map_generation= update_map_generation(c);
1013 
1014  dmin = 1000000;
1015 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
1016  /* first line */
1017  if (s->first_slice_line) {
1018  CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1019  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1020  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1021  CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1022  }else{
1023  CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1024  //FIXME try some early stop
1025  CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
1026  CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1027  CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
1028  CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
1029  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1030  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1031  }
1032  if(dmin>64*4){
1033  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
1034  (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
1035  if(s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line
1036  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
1037  (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
1038  }
1039 
1040  dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
1041 
1042  *mx_ptr= best[0];
1043  *my_ptr= best[1];
1044 
1045 // printf("%d %d %d \n", best[0], best[1], dmin);
1046  return dmin;
1047 }
1048 
1049 //try to merge with above FIXME (needs PSNR test)
1051  int *mx_ptr, int *my_ptr, int P[10][2],
1052  int src_index, int ref_index, int16_t (*last_mv)[2],
1053  int ref_mv_scale)
1054 {
1055  MotionEstContext * const c= &s->me;
1056  int best[2]={0, 0};
1057  int d, dmin;
1058  unsigned map_generation;
1059  const int penalty_factor= c->penalty_factor;
1060  const int size=0; //FIXME pass as arg
1061  const int h=8;
1062  const int ref_mv_stride= s->mb_stride;
1063  const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
1064  me_cmp_func cmpf, chroma_cmpf;
1065  LOAD_COMMON
1066  int flags= c->flags;
1067  LOAD_COMMON2
1068 
1069  cmpf= s->dsp.me_cmp[size];
1070  chroma_cmpf= s->dsp.me_cmp[size+1];
1071 
1072  map_generation= update_map_generation(c);
1073 
1074  dmin = 1000000;
1075 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
1076  /* first line */
1077  if (s->first_slice_line) {
1078  CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1079  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1080  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1081  CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1082  }else{
1083  CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1084  //FIXME try some early stop
1085  CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
1086  CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1087  CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
1088  CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
1089  CHECK_CLIPPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1090  (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1091  }
1092  if(dmin>64*4){
1093  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
1094  (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
1095  if(s->mb_y+1<s->end_mb_y) //FIXME replace at least with last_slice_line
1096  CHECK_CLIPPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
1097  (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
1098  }
1099 
1100  dmin= diamond_search(s, best, dmin, src_index, ref_index, penalty_factor, size, h, flags);
1101 
1102  *mx_ptr= best[0];
1103  *my_ptr= best[1];
1104 
1105 // printf("%d %d %d \n", best[0], best[1], dmin);
1106  return dmin;
1107 }