interlaced motion estimation
[libav.git] / libavcodec / motion_est_template.c
CommitLineData
1457ab52
MN
1/*
2 * Motion estimation
3 * Copyright (c) 2002 Michael Niedermayer
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 */
983e3246
MN
20
21/**
22 * @file motion_est_template.c
23 * Motion estimation template.
24 */
bb198e19 25//FIXME ref2_y next_pic?
1457ab52
MN
26//lets hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...)
27//Note, the last line is there to kill these ugly unused var warnings
bb198e19 28#define LOAD_COMMON\
1457ab52 29 uint32_t * const score_map= s->me.score_map;\
1457ab52
MN
30 const int time_pp= s->pp_time;\
31 const int time_pb= s->pb_time;\
bb198e19
MN
32 const int xmin= s->me.xmin;\
33 const int ymin= s->me.ymin;\
34 const int xmax= s->me.xmax;\
35 const int ymax= s->me.ymax;\
36 uint8_t * const src_y= src_data[0];\
37 uint8_t * const src_u= src_data[1];\
38 uint8_t * const src_v= src_data[2];\
39 uint8_t * const ref_y= ref_data[0];\
40 uint8_t * const ref_u= ref_data[1];\
41 uint8_t * const ref_v= ref_data[2];\
1457ab52
MN
42 op_pixels_func (*hpel_put)[4];\
43 op_pixels_func (*hpel_avg)[4]= &s->dsp.avg_pixels_tab[size];\
44 op_pixels_func (*chroma_hpel_put)[4];\
45 qpel_mc_func (*qpel_put)[16];\
46 qpel_mc_func (*qpel_avg)[16]= &s->dsp.avg_qpel_pixels_tab[size];\
3db320ea 47 const __attribute__((unused)) int unu= time_pp + time_pb + (size_t)src_u + (size_t)src_v + (size_t)ref_u + (size_t)ref_v\
bb198e19
MN
48 + (size_t)hpel_avg + (size_t)qpel_avg + (size_t)score_map\
49 + xmin + xmax + ymin + ymax;\
1457ab52
MN
50 if(s->no_rounding /*FIXME b_type*/){\
51 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];\
52 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];\
53 qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];\
54 }else{\
55 hpel_put=& s->dsp.put_pixels_tab[size];\
56 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];\
57 qpel_put= &s->dsp.put_qpel_pixels_tab[size];\
58 }
59
60
61#ifdef CMP_HPEL
62
63#define CHECK_HALF_MV(dx, dy, x, y)\
64{\
65 const int hx= 2*(x)+(dx);\
66 const int hy= 2*(y)+(dy);\
67 CMP_HPEL(d, dx, dy, x, y, size);\
68 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
69 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
70}
71
72#if 0
73static int RENAME(hpel_motion_search)(MpegEncContext * s,
74 int *mx_ptr, int *my_ptr, int dmin,
bb198e19
MN
75 int pred_x, int pred_y, uint8_t *ref_data[3],
76 int size, uint8_t * const mv_penalty)
1457ab52 77{
1457ab52
MN
78 const int xx = 16 * s->mb_x + 8*(n&1);
79 const int yy = 16 * s->mb_y + 8*(n>>1);
80 const int mx = *mx_ptr;
81 const int my = *my_ptr;
b07a5980 82 const int penalty_factor= s->me.sub_penalty_factor;
1457ab52 83
bb198e19 84 LOAD_COMMON
1457ab52
MN
85
86 // INIT;
87 //FIXME factorize
88 me_cmp_func cmp, chroma_cmp, cmp_sub, chroma_cmp_sub;
89
90 if(s->no_rounding /*FIXME b_type*/){
91 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
92 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
93 }else{
94 hpel_put=& s->dsp.put_pixels_tab[size];
95 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
96 }
97 cmp= s->dsp.me_cmp[size];
98 chroma_cmp= s->dsp.me_cmp[size+1];
99 cmp_sub= s->dsp.me_sub_cmp[size];
100 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
101
102 if(s->me.skip){ //FIXME somehow move up (benchmark)
103 *mx_ptr = 0;
104 *my_ptr = 0;
105 return dmin;
106 }
107
108 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
109 CMP_HPEL(dmin, 0, 0, mx, my, size);
110 if(mx || my)
111 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
112 }
113
114 if (mx > xmin && mx < xmax &&
115 my > ymin && my < ymax) {
116 int bx=2*mx, by=2*my;
117 int d= dmin;
118
119 CHECK_HALF_MV(1, 1, mx-1, my-1)
120 CHECK_HALF_MV(0, 1, mx , my-1)
121 CHECK_HALF_MV(1, 1, mx , my-1)
122 CHECK_HALF_MV(1, 0, mx-1, my )
123 CHECK_HALF_MV(1, 0, mx , my )
124 CHECK_HALF_MV(1, 1, mx-1, my )
125 CHECK_HALF_MV(0, 1, mx , my )
126 CHECK_HALF_MV(1, 1, mx , my )
127
b07a5980 128 assert(bx >= xmin*2 || bx <= xmax*2 || by >= ymin*2 || by <= ymax*2);
1457ab52
MN
129
130 *mx_ptr = bx;
131 *my_ptr = by;
132 }else{
133 *mx_ptr =2*mx;
134 *my_ptr =2*my;
135 }
136
137 return dmin;
138}
139
140#else
141static int RENAME(hpel_motion_search)(MpegEncContext * s,
142 int *mx_ptr, int *my_ptr, int dmin,
bb198e19
MN
143 int pred_x, int pred_y, uint8_t *src_data[3],
144 uint8_t *ref_data[3], int stride, int uvstride,
145 int size, int h, uint8_t * const mv_penalty)
1457ab52 146{
1457ab52
MN
147 const int mx = *mx_ptr;
148 const int my = *my_ptr;
149 const int penalty_factor= s->me.sub_penalty_factor;
150 me_cmp_func cmp_sub, chroma_cmp_sub;
67725183 151 int bx=2*mx, by=2*my;
1457ab52 152
bb198e19 153 LOAD_COMMON
1457ab52
MN
154
155 //FIXME factorize
156
157 cmp_sub= s->dsp.me_sub_cmp[size];
158 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
159
160 if(s->me.skip){ //FIXME move out of hpel?
161 *mx_ptr = 0;
162 *my_ptr = 0;
163 return dmin;
164 }
165
166 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
167 CMP_HPEL(dmin, 0, 0, mx, my, size);
1d0eab1d 168 if(mx || my || size>0)
1457ab52
MN
169 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
170 }
171
172 if (mx > xmin && mx < xmax &&
173 my > ymin && my < ymax) {
1457ab52
MN
174 int d= dmin;
175 const int index= (my<<ME_MAP_SHIFT) + mx;
176 const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
b07a5980 177 + (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*s->me.penalty_factor;
1457ab52 178 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)]
b07a5980 179 + (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor;
1457ab52 180 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)]
b07a5980 181 + (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor;
1457ab52 182 const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
b07a5980
MN
183 + (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*s->me.penalty_factor;
184
67725183 185#if 1
b07a5980
MN
186 int key;
187 int map_generation= s->me.map_generation;
802f454e 188#ifndef NDEBUG
b07a5980 189 uint32_t *map= s->me.map;
802f454e 190#endif
b07a5980
MN
191 key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
192 assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
193 key= ((my+1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
194 assert(map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
195 key= ((my)<<ME_MAP_MV_BITS) + (mx+1) + map_generation;
196 assert(map[(index+1)&(ME_MAP_SIZE-1)] == key);
197 key= ((my)<<ME_MAP_MV_BITS) + (mx-1) + map_generation;
198 assert(map[(index-1)&(ME_MAP_SIZE-1)] == key);
199#endif
1457ab52
MN
200 if(t<=b){
201 CHECK_HALF_MV(0, 1, mx ,my-1)
202 if(l<=r){
203 CHECK_HALF_MV(1, 1, mx-1, my-1)
204 if(t+r<=b+l){
205 CHECK_HALF_MV(1, 1, mx , my-1)
206 }else{
207 CHECK_HALF_MV(1, 1, mx-1, my )
208 }
209 CHECK_HALF_MV(1, 0, mx-1, my )
210 }else{
211 CHECK_HALF_MV(1, 1, mx , my-1)
212 if(t+l<=b+r){
213 CHECK_HALF_MV(1, 1, mx-1, my-1)
214 }else{
215 CHECK_HALF_MV(1, 1, mx , my )
216 }
217 CHECK_HALF_MV(1, 0, mx , my )
218 }
219 }else{
220 if(l<=r){
221 if(t+l<=b+r){
222 CHECK_HALF_MV(1, 1, mx-1, my-1)
223 }else{
224 CHECK_HALF_MV(1, 1, mx , my )
225 }
226 CHECK_HALF_MV(1, 0, mx-1, my)
227 CHECK_HALF_MV(1, 1, mx-1, my)
228 }else{
229 if(t+r<=b+l){
230 CHECK_HALF_MV(1, 1, mx , my-1)
231 }else{
232 CHECK_HALF_MV(1, 1, mx-1, my)
233 }
234 CHECK_HALF_MV(1, 0, mx , my)
235 CHECK_HALF_MV(1, 1, mx , my)
236 }
237 CHECK_HALF_MV(0, 1, mx , my)
238 }
239 assert(bx >= xmin*2 && bx <= xmax*2 && by >= ymin*2 && by <= ymax*2);
1457ab52
MN
240 }
241
67725183
MN
242 *mx_ptr = bx;
243 *my_ptr = by;
244
1457ab52
MN
245 return dmin;
246}
247#endif
248
bb198e19
MN
249static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, uint8_t *src_data[3],
250 uint8_t *ref_data[3], int stride, int uvstride,
30952237 251 uint8_t * const mv_penalty)
67725183
MN
252{
253// const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp;
254 const int size= 0;
bb198e19 255 const int h= 16;
67725183 256 const int penalty_factor= s->me.mb_penalty_factor;
67725183
MN
257 me_cmp_func cmp_sub, chroma_cmp_sub;
258 int d;
259
bb198e19 260 LOAD_COMMON
67725183
MN
261
262 //FIXME factorize
263
264 cmp_sub= s->dsp.mb_cmp[size];
265 chroma_cmp_sub= s->dsp.mb_cmp[size+1];
266
267 assert(!s->me.skip);
268 assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
269
270 CMP_HPEL(d, mx&1, my&1, mx>>1, my>>1, size);
271 //FIXME check cbp before adding penalty for (0,0) vector
272 if(mx || my || size>0)
273 d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
274
275 return d;
276}
277
1457ab52
MN
278#endif /* CMP_HPEL */
279
67725183
MN
280
281
1457ab52
MN
282#ifdef CMP_QPEL
283
284#define CHECK_QUARTER_MV(dx, dy, x, y)\
285{\
286 const int hx= 4*(x)+(dx);\
287 const int hy= 4*(y)+(dy);\
288 CMP_QPEL(d, dx, dy, x, y, size);\
289 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
290 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
291}
292
293static int RENAME(qpel_motion_search)(MpegEncContext * s,
294 int *mx_ptr, int *my_ptr, int dmin,
bb198e19
MN
295 int pred_x, int pred_y, uint8_t *src_data[3],
296 uint8_t *ref_data[3], int stride, int uvstride,
297 int size, int h, uint8_t * const mv_penalty)
1457ab52 298{
1457ab52
MN
299 const int mx = *mx_ptr;
300 const int my = *my_ptr;
301 const int penalty_factor= s->me.sub_penalty_factor;
302 const int map_generation= s->me.map_generation;
826f429a 303 const int subpel_quality= s->avctx->me_subpel_quality;
1457ab52
MN
304 uint32_t *map= s->me.map;
305 me_cmp_func cmp, chroma_cmp;
306 me_cmp_func cmp_sub, chroma_cmp_sub;
307
bb198e19 308 LOAD_COMMON
1457ab52
MN
309
310 cmp= s->dsp.me_cmp[size];
311 chroma_cmp= s->dsp.me_cmp[size+1]; //factorize FIXME
312 //FIXME factorize
313
314 cmp_sub= s->dsp.me_sub_cmp[size];
315 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
316
317 if(s->me.skip){ //FIXME somehow move up (benchmark)
318 *mx_ptr = 0;
319 *my_ptr = 0;
320 return dmin;
321 }
322
323 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
324 CMP_QPEL(dmin, 0, 0, mx, my, size);
1d0eab1d 325 if(mx || my || size>0)
1457ab52
MN
326 dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor;
327 }
328
329 if (mx > xmin && mx < xmax &&
330 my > ymin && my < ymax) {
331 int bx=4*mx, by=4*my;
332 int d= dmin;
333 int i, nx, ny;
334 const int index= (my<<ME_MAP_SHIFT) + mx;
335 const int t= score_map[(index-(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
336 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
337 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
338 const int b= score_map[(index+(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
339 const int c= score_map[(index )&(ME_MAP_SIZE-1)];
340 int best[8];
341 int best_pos[8][2];
342
343 memset(best, 64, sizeof(int)*8);
344#if 1
826f429a 345 if(s->me.dia_size>=2){
1457ab52
MN
346 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
347 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
348 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
349 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
350
351 for(ny= -3; ny <= 3; ny++){
352 for(nx= -3; nx <= 3; nx++){
353 const int t2= nx*nx*(tr + tl - 2*t) + 4*nx*(tr-tl) + 32*t;
354 const int c2= nx*nx*( r + l - 2*c) + 4*nx*( r- l) + 32*c;
355 const int b2= nx*nx*(br + bl - 2*b) + 4*nx*(br-bl) + 32*b;
356 int score= ny*ny*(b2 + t2 - 2*c2) + 4*ny*(b2 - t2) + 32*c2;
357 int i;
358
359 if((nx&3)==0 && (ny&3)==0) continue;
360
361 score += 1024*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
362
363// if(nx&1) score-=1024*s->me.penalty_factor;
364// if(ny&1) score-=1024*s->me.penalty_factor;
365
366 for(i=0; i<8; i++){
367 if(score < best[i]){
368 memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
369 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
370 best[i]= score;
371 best_pos[i][0]= nx + 4*mx;
372 best_pos[i][1]= ny + 4*my;
373 break;
374 }
375 }
376 }
377 }
378 }else{
379 int tl;
380 const int cx = 4*(r - l);
381 const int cx2= r + l - 2*c;
382 const int cy = 4*(b - t);
383 const int cy2= b + t - 2*c;
384 int cxy;
385
386 if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME
387 tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
388 }else{
389 CMP(tl, mx-1, my-1, size); //FIXME wrong if chroma me is different
390 }
391
392 cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c;
393
394 assert(16*cx2 + 4*cx + 32*c == 32*r);
395 assert(16*cx2 - 4*cx + 32*c == 32*l);
396 assert(16*cy2 + 4*cy + 32*c == 32*b);
397 assert(16*cy2 - 4*cy + 32*c == 32*t);
398 assert(16*cxy + 16*cy2 + 16*cx2 - 4*cy - 4*cx + 32*c == 32*tl);
399
400 for(ny= -3; ny <= 3; ny++){
401 for(nx= -3; nx <= 3; nx++){
402 int score= ny*nx*cxy + nx*nx*cx2 + ny*ny*cy2 + nx*cx + ny*cy + 32*c; //FIXME factor
403 int i;
404
405 if((nx&3)==0 && (ny&3)==0) continue;
406
407 score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
408// if(nx&1) score-=32*s->me.penalty_factor;
409 // if(ny&1) score-=32*s->me.penalty_factor;
410
411 for(i=0; i<8; i++){
412 if(score < best[i]){
413 memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
414 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
415 best[i]= score;
416 best_pos[i][0]= nx + 4*mx;
417 best_pos[i][1]= ny + 4*my;
418 break;
419 }
420 }
421 }
422 }
423 }
826f429a 424 for(i=0; i<subpel_quality; i++){
1457ab52
MN
425 nx= best_pos[i][0];
426 ny= best_pos[i][1];
427 CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2)
428 }
826f429a 429
1457ab52 430#if 0
826f429a
MN
431 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
432 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
433 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
434 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
435// if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){
436 if(tl<br){
437
438// nx= FFMAX(4*mx - bx, bx - 4*mx);
439// ny= FFMAX(4*my - by, by - 4*my);
1457ab52 440
826f429a
MN
441 static int stats[7][7], count;
442 count++;
443 stats[4*mx - bx + 3][4*my - by + 3]++;
444 if(256*256*256*64 % count ==0){
445 for(i=0; i<49; i++){
446 if((i%7)==0) printf("\n");
1457ab52
MN
447 printf("%6d ", stats[0][i]);
448 }
449 printf("\n");
450 }
826f429a 451 }
1457ab52
MN
452#endif
453#else
454
455 CHECK_QUARTER_MV(2, 2, mx-1, my-1)
456 CHECK_QUARTER_MV(0, 2, mx , my-1)
457 CHECK_QUARTER_MV(2, 2, mx , my-1)
458 CHECK_QUARTER_MV(2, 0, mx , my )
459 CHECK_QUARTER_MV(2, 2, mx , my )
460 CHECK_QUARTER_MV(0, 2, mx , my )
461 CHECK_QUARTER_MV(2, 2, mx-1, my )
462 CHECK_QUARTER_MV(2, 0, mx-1, my )
463
464 nx= bx;
465 ny= by;
466
467 for(i=0; i<8; i++){
468 int ox[8]= {0, 1, 1, 1, 0,-1,-1,-1};
469 int oy[8]= {1, 1, 0,-1,-1,-1, 0, 1};
470 CHECK_QUARTER_MV((nx + ox[i])&3, (ny + oy[i])&3, (nx + ox[i])>>2, (ny + oy[i])>>2)
471 }
472#endif
473#if 0
474 //outer ring
475 CHECK_QUARTER_MV(1, 3, mx-1, my-1)
476 CHECK_QUARTER_MV(1, 2, mx-1, my-1)
477 CHECK_QUARTER_MV(1, 1, mx-1, my-1)
478 CHECK_QUARTER_MV(2, 1, mx-1, my-1)
479 CHECK_QUARTER_MV(3, 1, mx-1, my-1)
480 CHECK_QUARTER_MV(0, 1, mx , my-1)
481 CHECK_QUARTER_MV(1, 1, mx , my-1)
482 CHECK_QUARTER_MV(2, 1, mx , my-1)
483 CHECK_QUARTER_MV(3, 1, mx , my-1)
484 CHECK_QUARTER_MV(3, 2, mx , my-1)
485 CHECK_QUARTER_MV(3, 3, mx , my-1)
486 CHECK_QUARTER_MV(3, 0, mx , my )
487 CHECK_QUARTER_MV(3, 1, mx , my )
488 CHECK_QUARTER_MV(3, 2, mx , my )
489 CHECK_QUARTER_MV(3, 3, mx , my )
490 CHECK_QUARTER_MV(2, 3, mx , my )
491 CHECK_QUARTER_MV(1, 3, mx , my )
492 CHECK_QUARTER_MV(0, 3, mx , my )
493 CHECK_QUARTER_MV(3, 3, mx-1, my )
494 CHECK_QUARTER_MV(2, 3, mx-1, my )
495 CHECK_QUARTER_MV(1, 3, mx-1, my )
496 CHECK_QUARTER_MV(1, 2, mx-1, my )
497 CHECK_QUARTER_MV(1, 1, mx-1, my )
498 CHECK_QUARTER_MV(1, 0, mx-1, my )
499#endif
500 assert(bx >= xmin*4 && bx <= xmax*4 && by >= ymin*4 && by <= ymax*4);
501
502 *mx_ptr = bx;
503 *my_ptr = by;
504 }else{
505 *mx_ptr =4*mx;
506 *my_ptr =4*my;
507 }
508
509 return dmin;
510}
511
bb198e19
MN
512static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, uint8_t *src_data[3],
513 uint8_t *ref_data[3], int stride, int uvstride,
30952237 514 uint8_t * const mv_penalty)
67725183
MN
515{
516 const int size= 0;
bb198e19 517 const int h= 16;
67725183 518 const int penalty_factor= s->me.mb_penalty_factor;
67725183
MN
519 me_cmp_func cmp_sub, chroma_cmp_sub;
520 int d;
521
bb198e19 522 LOAD_COMMON
67725183
MN
523
524 //FIXME factorize
525
526 cmp_sub= s->dsp.mb_cmp[size];
527 chroma_cmp_sub= s->dsp.mb_cmp[size+1];
528
529 assert(!s->me.skip);
530 assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
531
532 CMP_QPEL(d, mx&3, my&3, mx>>2, my>>2, size);
533 //FIXME check cbp before adding penalty for (0,0) vector
534 if(mx || my || size>0)
535 d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
536
537 return d;
538}
539
540
1457ab52
MN
541#endif /* CMP_QPEL */
542
543#define CHECK_MV(x,y)\
544{\
545 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
546 const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
b07a5980 547/*printf("check_mv %d %d\n", x, y);*/\
1457ab52
MN
548 if(map[index]!=key){\
549 CMP(d, x, y, size);\
550 map[index]= key;\
551 score_map[index]= d;\
552 d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
b07a5980 553/*printf("score:%d\n", d);*/\
1457ab52
MN
554 COPY3_IF_LT(dmin, d, best[0], x, best[1], y)\
555 }\
556}
557
b07a5980
MN
558#define CHECK_CLIPED_MV(ax,ay)\
559{\
560 const int x= FFMAX(xmin, FFMIN(ax, xmax));\
561 const int y= FFMAX(ymin, FFMIN(ay, ymax));\
562 CHECK_MV(x, y)\
563}
564
1457ab52
MN
565#define CHECK_MV_DIR(x,y,new_dir)\
566{\
567 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\
568 const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\
b07a5980 569/*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\
1457ab52
MN
570 if(map[index]!=key){\
571 CMP(d, x, y, size);\
572 map[index]= key;\
573 score_map[index]= d;\
574 d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\
b07a5980 575/*printf("score:%d\n", d);*/\
1457ab52
MN
576 if(d<dmin){\
577 best[0]=x;\
578 best[1]=y;\
579 dmin=d;\
580 next_dir= new_dir;\
581 }\
582 }\
583}
584
585#define check(x,y,S,v)\
586if( (x)<(xmin<<(S)) ) printf("%d %d %d %d %d xmin" #v, xmin, (x), (y), s->mb_x, s->mb_y);\
587if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, s->mb_y);\
588if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\
589if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\
590
591
592static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, int dmin,
bb198e19
MN
593 uint8_t *src_data[3],
594 uint8_t *ref_data[3], int stride, int uvstride,
1457ab52 595 int const pred_x, int const pred_y, int const penalty_factor,
bb198e19
MN
596 int const shift,
597 uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
1457ab52
MN
598 )
599{
600 me_cmp_func cmp, chroma_cmp;
601 int next_dir=-1;
bb198e19 602 LOAD_COMMON
1457ab52
MN
603
604 cmp= s->dsp.me_cmp[size];
605 chroma_cmp= s->dsp.me_cmp[size+1];
606
b07a5980
MN
607 { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
608 const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
609 const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1);
610 if(map[index]!=key){ //this will be executed only very rarey
611 CMP(score_map[index], best[0], best[1], size);
612 map[index]= key;
613 }
614 }
615
1457ab52
MN
616 for(;;){
617 int d;
618 const int dir= next_dir;
619 const int x= best[0];
620 const int y= best[1];
621 next_dir=-1;
622
623//printf("%d", dir);
624 if(dir!=2 && x>xmin) CHECK_MV_DIR(x-1, y , 0)
625 if(dir!=3 && y>ymin) CHECK_MV_DIR(x , y-1, 1)
626 if(dir!=0 && x<xmax) CHECK_MV_DIR(x+1, y , 2)
627 if(dir!=1 && y<ymax) CHECK_MV_DIR(x , y+1, 3)
628
629 if(next_dir==-1){
630 return dmin;
631 }
632 }
633}
634
b07a5980 635static inline int RENAME(funny_diamond_search)(MpegEncContext * s, int *best, int dmin,
bb198e19
MN
636 uint8_t *src_data[3],
637 uint8_t *ref_data[3], int stride, int uvstride,
b07a5980 638 int const pred_x, int const pred_y, int const penalty_factor,
bb198e19
MN
639 int const shift,
640 uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
b07a5980
MN
641 )
642{
643 me_cmp_func cmp, chroma_cmp;
644 int dia_size;
bb198e19 645 LOAD_COMMON
b07a5980
MN
646
647 cmp= s->dsp.me_cmp[size];
648 chroma_cmp= s->dsp.me_cmp[size+1];
649
650 for(dia_size=1; dia_size<=4; dia_size++){
651 int dir;
652 const int x= best[0];
653 const int y= best[1];
654
655 if(dia_size&(dia_size-1)) continue;
656
657 if( x + dia_size > xmax
658 || x - dia_size < xmin
659 || y + dia_size > ymax
660 || y - dia_size < ymin)
661 continue;
662
663 for(dir= 0; dir<dia_size; dir+=2){
664 int d;
665
666 CHECK_MV(x + dir , y + dia_size - dir);
667 CHECK_MV(x + dia_size - dir, y - dir );
668 CHECK_MV(x - dir , y - dia_size + dir);
669 CHECK_MV(x - dia_size + dir, y + dir );
670 }
671
672 if(x!=best[0] || y!=best[1])
673 dia_size=0;
674#if 0
675{
676int dx, dy, i;
677static int stats[8*8];
678dx= ABS(x-best[0]);
679dy= ABS(y-best[1]);
680if(dy>dx){
681 dx^=dy; dy^=dx; dx^=dy;
682}
683stats[dy*8 + dx] ++;
684if(256*256*256*64 % (stats[0]+1)==0){
685 for(i=0; i<64; i++){
686 if((i&7)==0) printf("\n");
687 printf("%8d ", stats[i]);
688 }
689 printf("\n");
690}
691}
692#endif
693 }
694 return dmin;
695}
696
697#define SAB_CHECK_MV(ax,ay)\
698{\
699 const int key= ((ay)<<ME_MAP_MV_BITS) + (ax) + map_generation;\
700 const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\
701/*printf("sab check %d %d\n", ax, ay);*/\
702 if(map[index]!=key){\
703 CMP(d, ax, ay, size);\
704 map[index]= key;\
705 score_map[index]= d;\
706 d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\
707/*printf("score: %d\n", d);*/\
708 if(d < minima[minima_count-1].height){\
709 int j=0;\
710 \
711 while(d >= minima[j].height) j++;\
712\
713 memmove(&minima [j+1], &minima [j], (minima_count - j - 1)*sizeof(Minima));\
714\
715 minima[j].checked= 0;\
716 minima[j].height= d;\
717 minima[j].x= ax;\
718 minima[j].y= ay;\
719 \
720 i=-1;\
721 continue;\
722 }\
723 }\
724}
725
726#define MAX_SAB_SIZE 16
727static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int dmin,
bb198e19
MN
728 uint8_t *src_data[3],
729 uint8_t *ref_data[3], int stride, int uvstride,
b07a5980 730 int const pred_x, int const pred_y, int const penalty_factor,
bb198e19
MN
731 int const shift,
732 uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
b07a5980
MN
733 )
734{
735 me_cmp_func cmp, chroma_cmp;
736 Minima minima[MAX_SAB_SIZE];
826f429a 737 const int minima_count= ABS(s->me.dia_size);
b07a5980 738 int i, j;
bb198e19 739 LOAD_COMMON
b07a5980
MN
740
741 cmp= s->dsp.me_cmp[size];
742 chroma_cmp= s->dsp.me_cmp[size+1];
743
744 for(j=i=0; i<ME_MAP_SIZE; i++){
745 uint32_t key= map[i];
746
747 key += (1<<(ME_MAP_MV_BITS-1)) + (1<<(2*ME_MAP_MV_BITS-1));
748
749 if((key&((-1)<<(2*ME_MAP_MV_BITS))) != map_generation) continue;
750
751 assert(j<MAX_SAB_SIZE); //max j = number of predictors
752
753 minima[j].height= score_map[i];
754 minima[j].x= key & ((1<<ME_MAP_MV_BITS)-1); key>>=ME_MAP_MV_BITS;
755 minima[j].y= key & ((1<<ME_MAP_MV_BITS)-1);
756 minima[j].x-= (1<<(ME_MAP_MV_BITS-1));
757 minima[j].y-= (1<<(ME_MAP_MV_BITS-1));
758 minima[j].checked=0;
759 if(minima[j].x || minima[j].y)
760 minima[j].height+= (mv_penalty[((minima[j].x)<<shift)-pred_x] + mv_penalty[((minima[j].y)<<shift)-pred_y])*penalty_factor;
761
762 j++;
763 }
764
765 qsort(minima, j, sizeof(Minima), minima_cmp);
766
767 for(; j<minima_count; j++){
768 minima[j].height=256*256*256*64;
769 minima[j].checked=0;
770 minima[j].x= minima[j].y=0;
771 }
772
773 for(i=0; i<minima_count; i++){
774 const int x= minima[i].x;
775 const int y= minima[i].y;
776 int d;
777
778 if(minima[i].checked) continue;
779
780 if( x >= xmax || x <= xmin
781 || y >= ymax || y <= ymin)
782 continue;
783
784 SAB_CHECK_MV(x-1, y)
785 SAB_CHECK_MV(x+1, y)
786 SAB_CHECK_MV(x , y-1)
787 SAB_CHECK_MV(x , y+1)
788
789 minima[i].checked= 1;
790 }
791
792 best[0]= minima[0].x;
793 best[1]= minima[0].y;
794 dmin= minima[0].height;
795
796 if( best[0] < xmax && best[0] > xmin
797 && best[1] < ymax && best[1] > ymin){
798 int d;
799 //ensure that the refernece samples for hpel refinement are in the map
800 CHECK_MV(best[0]-1, best[1])
801 CHECK_MV(best[0]+1, best[1])
802 CHECK_MV(best[0], best[1]-1)
803 CHECK_MV(best[0], best[1]+1)
804 }
805 return dmin;
806}
807
1457ab52 808static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int dmin,
bb198e19
MN
809 uint8_t *src_data[3],
810 uint8_t *ref_data[3], int stride, int uvstride,
1457ab52 811 int const pred_x, int const pred_y, int const penalty_factor,
bb198e19
MN
812 int const shift,
813 uint32_t *map, int map_generation, int size, int h, uint8_t * const mv_penalty
1457ab52
MN
814 )
815{
816 me_cmp_func cmp, chroma_cmp;
b07a5980 817 int dia_size;
bb198e19 818 LOAD_COMMON
1457ab52
MN
819
820 cmp= s->dsp.me_cmp[size];
821 chroma_cmp= s->dsp.me_cmp[size+1];
822
826f429a 823 for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){
1457ab52
MN
824 int dir, start, end;
825 const int x= best[0];
826 const int y= best[1];
827
828 start= FFMAX(0, y + dia_size - ymax);
b07a5980 829 end = FFMIN(dia_size, xmax - x + 1);
1457ab52
MN
830 for(dir= start; dir<end; dir++){
831 int d;
832
833//check(x + dir,y + dia_size - dir,0, a0)
834 CHECK_MV(x + dir , y + dia_size - dir);
835 }
836
837 start= FFMAX(0, x + dia_size - xmax);
b07a5980 838 end = FFMIN(dia_size, y - ymin + 1);
1457ab52
MN
839 for(dir= start; dir<end; dir++){
840 int d;
841
842//check(x + dia_size - dir, y - dir,0, a1)
843 CHECK_MV(x + dia_size - dir, y - dir );
844 }
845
846 start= FFMAX(0, -y + dia_size + ymin );
b07a5980 847 end = FFMIN(dia_size, x - xmin + 1);
1457ab52
MN
848 for(dir= start; dir<end; dir++){
849 int d;
850
851//check(x - dir,y - dia_size + dir,0, a2)
852 CHECK_MV(x - dir , y - dia_size + dir);
853 }
854
855 start= FFMAX(0, -x + dia_size + xmin );
b07a5980 856 end = FFMIN(dia_size, ymax - y + 1);
1457ab52
MN
857 for(dir= start; dir<end; dir++){
858 int d;
859
860//check(x - dia_size + dir, y + dir,0, a3)
861 CHECK_MV(x - dia_size + dir, y + dir );
862 }
863
864 if(x!=best[0] || y!=best[1])
865 dia_size=0;
b07a5980
MN
866#if 0
867{
868int dx, dy, i;
869static int stats[8*8];
870dx= ABS(x-best[0]);
871dy= ABS(y-best[1]);
872stats[dy*8 + dx] ++;
873if(256*256*256*64 % (stats[0]+1)==0){
874 for(i=0; i<64; i++){
875 if((i&7)==0) printf("\n");
876 printf("%6d ", stats[i]);
877 }
878 printf("\n");
879}
880}
881#endif
1457ab52
MN
882 }
883 return dmin;
884}
885
bb198e19 886static int RENAME(epzs_motion_search)(MpegEncContext * s,
1457ab52 887 int *mx_ptr, int *my_ptr,
bb198e19
MN
888 int P[10][2], int pred_x, int pred_y, uint8_t *src_data[3],
889 uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2],
30952237 890 int ref_mv_scale, uint8_t * const mv_penalty)
1457ab52
MN
891{
892 int best[2]={0, 0};
893 int d, dmin;
894 const int shift= 1+s->quarter_sample;
895 uint32_t *map= s->me.map;
896 int map_generation;
897 const int penalty_factor= s->me.penalty_factor;
898 const int size=0;
bb198e19
MN
899 const int h=16;
900 const int ref_mv_stride= s->mb_stride; //pass as arg FIXME
901 const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; //add to last_mv beforepassing FIXME
1457ab52 902 me_cmp_func cmp, chroma_cmp;
bb198e19 903 LOAD_COMMON
1457ab52
MN
904
905 cmp= s->dsp.me_cmp[size];
906 chroma_cmp= s->dsp.me_cmp[size+1];
907
908 map_generation= update_map_generation(s);
909
910 CMP(dmin, 0, 0, size);
911 map[0]= map_generation;
912 score_map[0]= dmin;
913
914 /* first line */
f931ff7b 915 if (s->mb_y == 0) {
1457ab52 916 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
b07a5980
MN
917 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
918 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1457ab52
MN
919 }else{
920 if(dmin<256 && ( P_LEFT[0] |P_LEFT[1]
921 |P_TOP[0] |P_TOP[1]
b07a5980 922 |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){
1457ab52
MN
923 *mx_ptr= 0;
924 *my_ptr= 0;
925 s->me.skip=1;
926 return dmin;
927 }
928 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
929 if(dmin>256*2){
b07a5980
MN
930 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
931 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1457ab52
MN
932 CHECK_MV(P_LEFT[0] >>shift, P_LEFT[1] >>shift)
933 CHECK_MV(P_TOP[0] >>shift, P_TOP[1] >>shift)
934 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
935 }
936 }
937 if(dmin>256*4){
f931ff7b
MN
938 if(s->me.pre_pass){
939 CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16,
940 (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16)
941 CHECK_CLIPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
942 (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
943 }else{
944 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
945 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
946 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
947 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
948 }
1457ab52 949 }
b07a5980
MN
950
951 if(s->avctx->last_predictor_count){
952 const int count= s->avctx->last_predictor_count;
953 const int xstart= FFMAX(0, s->mb_x - count);
954 const int ystart= FFMAX(0, s->mb_y - count);
955 const int xend= FFMIN(s->mb_width , s->mb_x + count + 1);
956 const int yend= FFMIN(s->mb_height, s->mb_y + count + 1);
957 int mb_y;
958
959 for(mb_y=ystart; mb_y<yend; mb_y++){
960 int mb_x;
961 for(mb_x=xstart; mb_x<xend; mb_x++){
962 const int xy= mb_x + 1 + (mb_y + 1)*ref_mv_stride;
963 int mx= (last_mv[xy][0]*ref_mv_scale + (1<<15))>>16;
964 int my= (last_mv[xy][1]*ref_mv_scale + (1<<15))>>16;
965
966 if(mx>xmax || mx<xmin || my>ymax || my<ymin) continue;
967 CHECK_MV(mx,my)
1457ab52
MN
968 }
969 }
970 }
b07a5980 971
1457ab52 972//check(best[0],best[1],0, b0)
826f429a 973 if(s->me.dia_size==-1)
bb198e19
MN
974 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
975 pred_x, pred_y, penalty_factor,
976 shift, map, map_generation, size, h, mv_penalty);
826f429a 977 else if(s->me.dia_size<-1)
bb198e19
MN
978 dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
979 pred_x, pred_y, penalty_factor,
980 shift, map, map_generation, size, h, mv_penalty);
826f429a 981 else if(s->me.dia_size<2)
bb198e19
MN
982 dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
983 pred_x, pred_y, penalty_factor,
984 shift, map, map_generation, size, h, mv_penalty);
1457ab52 985 else
bb198e19
MN
986 dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
987 pred_x, pred_y, penalty_factor,
988 shift, map, map_generation, size, h, mv_penalty);
1457ab52
MN
989
990//check(best[0],best[1],0, b1)
991 *mx_ptr= best[0];
992 *my_ptr= best[1];
993
994// printf("%d %d %d \n", best[0], best[1], dmin);
995 return dmin;
996}
997
998#ifndef CMP_DIRECT /* no 4mv search needed in direct mode */
bb198e19 999static int RENAME(epzs_motion_search4)(MpegEncContext * s,
1457ab52
MN
1000 int *mx_ptr, int *my_ptr,
1001 int P[10][2], int pred_x, int pred_y,
bb198e19
MN
1002 uint8_t *src_data[3],
1003 uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2],
30952237 1004 int ref_mv_scale, uint8_t * const mv_penalty)
1457ab52
MN
1005{
1006 int best[2]={0, 0};
1007 int d, dmin;
1008 const int shift= 1+s->quarter_sample;
1009 uint32_t *map= s->me.map;
1010 int map_generation;
1011 const int penalty_factor= s->me.penalty_factor;
1012 const int size=1;
bb198e19 1013 const int h=8;
7bc9090a
MN
1014 const int ref_mv_stride= s->mb_stride;
1015 const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
1457ab52 1016 me_cmp_func cmp, chroma_cmp;
bb198e19 1017 LOAD_COMMON
1457ab52
MN
1018
1019 cmp= s->dsp.me_cmp[size];
1020 chroma_cmp= s->dsp.me_cmp[size+1];
1021
1022 map_generation= update_map_generation(s);
1023
1024 dmin = 1000000;
1025//printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
1026 /* first line */
bb198e19 1027 if (s->mb_y == 0/* && block<2*/) {
1457ab52 1028 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
b07a5980
MN
1029 CHECK_CLIPED_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)
1457ab52
MN
1031 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1032 }else{
1033 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1034 //FIXME try some early stop
1035 if(dmin>64*2){
1036 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
1037 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1038 CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
1039 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
b07a5980
MN
1040 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1041 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1457ab52
MN
1042 }
1043 }
1044 if(dmin>64*4){
b07a5980
MN
1045 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
1046 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
1047 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
1048 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
1457ab52
MN
1049 }
1050
826f429a 1051 if(s->me.dia_size==-1)
bb198e19
MN
1052 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1053 pred_x, pred_y, penalty_factor,
1054 shift, map, map_generation, size, h, mv_penalty);
826f429a 1055 else if(s->me.dia_size<-1)
bb198e19
MN
1056 dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1057 pred_x, pred_y, penalty_factor,
1058 shift, map, map_generation, size, h, mv_penalty);
826f429a 1059 else if(s->me.dia_size<2)
bb198e19
MN
1060 dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1061 pred_x, pred_y, penalty_factor,
1062 shift, map, map_generation, size, h, mv_penalty);
1457ab52 1063 else
bb198e19
MN
1064 dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1065 pred_x, pred_y, penalty_factor,
1066 shift, map, map_generation, size, h, mv_penalty);
1067
1068
1069 *mx_ptr= best[0];
1070 *my_ptr= best[1];
1071
1072// printf("%d %d %d \n", best[0], best[1], dmin);
1073 return dmin;
1074}
1075
1076//try to merge with above FIXME (needs PSNR test)
1077static int RENAME(epzs_motion_search2)(MpegEncContext * s,
1078 int *mx_ptr, int *my_ptr,
1079 int P[10][2], int pred_x, int pred_y,
1080 uint8_t *src_data[3],
1081 uint8_t *ref_data[3], int stride, int uvstride, int16_t (*last_mv)[2],
1082 int ref_mv_scale, uint8_t * const mv_penalty)
1083{
1084 int best[2]={0, 0};
1085 int d, dmin;
1086 const int shift= 1+s->quarter_sample;
1087 uint32_t *map= s->me.map;
1088 int map_generation;
1089 const int penalty_factor= s->me.penalty_factor;
1090 const int size=0; //FIXME pass as arg
1091 const int h=8;
1092 const int ref_mv_stride= s->mb_stride;
1093 const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride;
1094 me_cmp_func cmp, chroma_cmp;
1095 LOAD_COMMON
1096
1097 cmp= s->dsp.me_cmp[size];
1098 chroma_cmp= s->dsp.me_cmp[size+1];
1099
1100 map_generation= update_map_generation(s);
1101
1102 dmin = 1000000;
1103//printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
1104 /* first line */
1105 if (s->mb_y == 0) {
1106 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1107 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1108 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1109 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1110 }else{
1111 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
1112 //FIXME try some early stop
1113 if(dmin>64*2){
1114 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
1115 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
1116 CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
1117 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
1118 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
1119 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
1120 }
1121 }
1122 if(dmin>64*4){
1123 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
1124 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
1125 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
1126 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
1127 }
1128
1129 if(s->me.dia_size==-1)
1130 dmin= RENAME(funny_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1131 pred_x, pred_y, penalty_factor,
1132 shift, map, map_generation, size, h, mv_penalty);
1133 else if(s->me.dia_size<-1)
1134 dmin= RENAME(sab_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1135 pred_x, pred_y, penalty_factor,
1136 shift, map, map_generation, size, h, mv_penalty);
1137 else if(s->me.dia_size<2)
1138 dmin= RENAME(small_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1139 pred_x, pred_y, penalty_factor,
1140 shift, map, map_generation, size, h, mv_penalty);
1141 else
1142 dmin= RENAME(var_diamond_search)(s, best, dmin, src_data, ref_data, stride, uvstride,
1143 pred_x, pred_y, penalty_factor,
1144 shift, map, map_generation, size, h, mv_penalty);
1145
b07a5980 1146
1457ab52
MN
1147 *mx_ptr= best[0];
1148 *my_ptr= best[1];
1149
1150// printf("%d %d %d \n", best[0], best[1], dmin);
1151 return dmin;
1152}
1153#endif /* !CMP_DIRECT */