Mercurial > libavcodec.hg
annotate motion_est_template.c @ 1433:46d3fa8501cd libavcodec
ASV2 decoder
author | michaelni |
---|---|
date | Mon, 01 Sep 2003 08:10:01 +0000 |
parents | a7a9df478e46 |
children | dea5b2946999 |
rev | line source |
---|---|
936 | 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 */ | |
1106 | 20 |
21 /** | |
22 * @file motion_est_template.c | |
23 * Motion estimation template. | |
24 */ | |
936 | 25 |
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 | |
28 #define LOAD_COMMON(x, y)\ | |
29 uint32_t * const score_map= s->me.score_map;\ | |
30 const int stride= s->linesize;\ | |
31 const int uvstride= s->uvlinesize;\ | |
32 const int time_pp= s->pp_time;\ | |
33 const int time_pb= s->pb_time;\ | |
34 uint8_t * const src_y= s->new_picture.data[0] + ((y) * stride) + (x);\ | |
35 uint8_t * const src_u= s->new_picture.data[1] + (((y)>>1) * uvstride) + ((x)>>1);\ | |
36 uint8_t * const src_v= s->new_picture.data[2] + (((y)>>1) * uvstride) + ((x)>>1);\ | |
37 uint8_t * const ref_y= ref_picture->data[0] + ((y) * stride) + (x);\ | |
38 uint8_t * const ref_u= ref_picture->data[1] + (((y)>>1) * uvstride) + ((x)>>1);\ | |
39 uint8_t * const ref_v= ref_picture->data[2] + (((y)>>1) * uvstride) + ((x)>>1);\ | |
40 uint8_t * const ref2_y= s->next_picture.data[0] + ((y) * stride) + (x);\ | |
41 op_pixels_func (*hpel_put)[4];\ | |
42 op_pixels_func (*hpel_avg)[4]= &s->dsp.avg_pixels_tab[size];\ | |
43 op_pixels_func (*chroma_hpel_put)[4];\ | |
44 qpel_mc_func (*qpel_put)[16];\ | |
45 qpel_mc_func (*qpel_avg)[16]= &s->dsp.avg_qpel_pixels_tab[size];\ | |
1266 | 46 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\ |
47 + (size_t)ref2_y + (size_t)hpel_avg + (size_t)qpel_avg + (size_t)score_map;\ | |
936 | 48 if(s->no_rounding /*FIXME b_type*/){\ |
49 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];\ | |
50 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];\ | |
51 qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];\ | |
52 }else{\ | |
53 hpel_put=& s->dsp.put_pixels_tab[size];\ | |
54 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];\ | |
55 qpel_put= &s->dsp.put_qpel_pixels_tab[size];\ | |
56 } | |
57 | |
58 | |
59 #ifdef CMP_HPEL | |
60 | |
61 #define CHECK_HALF_MV(dx, dy, x, y)\ | |
62 {\ | |
63 const int hx= 2*(x)+(dx);\ | |
64 const int hy= 2*(y)+(dy);\ | |
65 CMP_HPEL(d, dx, dy, x, y, size);\ | |
66 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\ | |
67 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\ | |
68 } | |
69 | |
70 #if 0 | |
71 static int RENAME(hpel_motion_search)(MpegEncContext * s, | |
72 int *mx_ptr, int *my_ptr, int dmin, | |
73 int xmin, int ymin, int xmax, int ymax, | |
74 int pred_x, int pred_y, Picture *ref_picture, | |
1162 | 75 int n, int size, uint8_t * const mv_penalty) |
936 | 76 { |
77 const int xx = 16 * s->mb_x + 8*(n&1); | |
78 const int yy = 16 * s->mb_y + 8*(n>>1); | |
79 const int mx = *mx_ptr; | |
80 const int my = *my_ptr; | |
948 | 81 const int penalty_factor= s->me.sub_penalty_factor; |
936 | 82 |
83 LOAD_COMMON(xx, yy); | |
84 | |
85 // INIT; | |
86 //FIXME factorize | |
87 me_cmp_func cmp, chroma_cmp, cmp_sub, chroma_cmp_sub; | |
88 | |
89 if(s->no_rounding /*FIXME b_type*/){ | |
90 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size]; | |
91 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1]; | |
92 }else{ | |
93 hpel_put=& s->dsp.put_pixels_tab[size]; | |
94 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1]; | |
95 } | |
96 cmp= s->dsp.me_cmp[size]; | |
97 chroma_cmp= s->dsp.me_cmp[size+1]; | |
98 cmp_sub= s->dsp.me_sub_cmp[size]; | |
99 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1]; | |
100 | |
101 if(s->me.skip){ //FIXME somehow move up (benchmark) | |
102 *mx_ptr = 0; | |
103 *my_ptr = 0; | |
104 return dmin; | |
105 } | |
106 | |
107 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){ | |
108 CMP_HPEL(dmin, 0, 0, mx, my, size); | |
109 if(mx || my) | |
110 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor; | |
111 } | |
112 | |
113 if (mx > xmin && mx < xmax && | |
114 my > ymin && my < ymax) { | |
115 int bx=2*mx, by=2*my; | |
116 int d= dmin; | |
117 | |
118 CHECK_HALF_MV(1, 1, mx-1, my-1) | |
119 CHECK_HALF_MV(0, 1, mx , my-1) | |
120 CHECK_HALF_MV(1, 1, mx , my-1) | |
121 CHECK_HALF_MV(1, 0, mx-1, my ) | |
122 CHECK_HALF_MV(1, 0, mx , my ) | |
123 CHECK_HALF_MV(1, 1, mx-1, my ) | |
124 CHECK_HALF_MV(0, 1, mx , my ) | |
125 CHECK_HALF_MV(1, 1, mx , my ) | |
126 | |
948 | 127 assert(bx >= xmin*2 || bx <= xmax*2 || by >= ymin*2 || by <= ymax*2); |
936 | 128 |
129 *mx_ptr = bx; | |
130 *my_ptr = by; | |
131 }else{ | |
132 *mx_ptr =2*mx; | |
133 *my_ptr =2*my; | |
134 } | |
135 | |
136 return dmin; | |
137 } | |
138 | |
139 #else | |
140 static int RENAME(hpel_motion_search)(MpegEncContext * s, | |
141 int *mx_ptr, int *my_ptr, int dmin, | |
142 int xmin, int ymin, int xmax, int ymax, | |
143 int pred_x, int pred_y, Picture *ref_picture, | |
1162 | 144 int n, int size, uint8_t * const mv_penalty) |
936 | 145 { |
146 const int xx = 16 * s->mb_x + 8*(n&1); | |
147 const int yy = 16 * s->mb_y + 8*(n>>1); | |
148 const int mx = *mx_ptr; | |
149 const int my = *my_ptr; | |
150 const int penalty_factor= s->me.sub_penalty_factor; | |
151 me_cmp_func cmp_sub, chroma_cmp_sub; | |
1013 | 152 int bx=2*mx, by=2*my; |
936 | 153 |
154 LOAD_COMMON(xx, yy); | |
155 | |
156 //FIXME factorize | |
157 | |
158 cmp_sub= s->dsp.me_sub_cmp[size]; | |
159 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1]; | |
160 | |
161 if(s->me.skip){ //FIXME move out of hpel? | |
162 *mx_ptr = 0; | |
163 *my_ptr = 0; | |
164 return dmin; | |
165 } | |
166 | |
167 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){ | |
168 CMP_HPEL(dmin, 0, 0, mx, my, size); | |
1011 | 169 if(mx || my || size>0) |
936 | 170 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor; |
171 } | |
172 | |
173 if (mx > xmin && mx < xmax && | |
174 my > ymin && my < ymax) { | |
175 int d= dmin; | |
176 const int index= (my<<ME_MAP_SHIFT) + mx; | |
177 const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] | |
948 | 178 + (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*s->me.penalty_factor; |
936 | 179 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)] |
948 | 180 + (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor; |
936 | 181 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)] |
948 | 182 + (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor; |
936 | 183 const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] |
948 | 184 + (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*s->me.penalty_factor; |
185 | |
1013 | 186 #if 1 |
948 | 187 int key; |
188 int map_generation= s->me.map_generation; | |
1419 | 189 #ifndef NDEBUG |
948 | 190 uint32_t *map= s->me.map; |
1419 | 191 #endif |
948 | 192 key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation; |
193 assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key); | |
194 key= ((my+1)<<ME_MAP_MV_BITS) + (mx) + map_generation; | |
195 assert(map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key); | |
196 key= ((my)<<ME_MAP_MV_BITS) + (mx+1) + map_generation; | |
197 assert(map[(index+1)&(ME_MAP_SIZE-1)] == key); | |
198 key= ((my)<<ME_MAP_MV_BITS) + (mx-1) + map_generation; | |
199 assert(map[(index-1)&(ME_MAP_SIZE-1)] == key); | |
200 #endif | |
936 | 201 if(t<=b){ |
202 CHECK_HALF_MV(0, 1, mx ,my-1) | |
203 if(l<=r){ | |
204 CHECK_HALF_MV(1, 1, mx-1, my-1) | |
205 if(t+r<=b+l){ | |
206 CHECK_HALF_MV(1, 1, mx , my-1) | |
207 }else{ | |
208 CHECK_HALF_MV(1, 1, mx-1, my ) | |
209 } | |
210 CHECK_HALF_MV(1, 0, mx-1, my ) | |
211 }else{ | |
212 CHECK_HALF_MV(1, 1, mx , my-1) | |
213 if(t+l<=b+r){ | |
214 CHECK_HALF_MV(1, 1, mx-1, my-1) | |
215 }else{ | |
216 CHECK_HALF_MV(1, 1, mx , my ) | |
217 } | |
218 CHECK_HALF_MV(1, 0, mx , my ) | |
219 } | |
220 }else{ | |
221 if(l<=r){ | |
222 if(t+l<=b+r){ | |
223 CHECK_HALF_MV(1, 1, mx-1, my-1) | |
224 }else{ | |
225 CHECK_HALF_MV(1, 1, mx , my ) | |
226 } | |
227 CHECK_HALF_MV(1, 0, mx-1, my) | |
228 CHECK_HALF_MV(1, 1, mx-1, my) | |
229 }else{ | |
230 if(t+r<=b+l){ | |
231 CHECK_HALF_MV(1, 1, mx , my-1) | |
232 }else{ | |
233 CHECK_HALF_MV(1, 1, mx-1, my) | |
234 } | |
235 CHECK_HALF_MV(1, 0, mx , my) | |
236 CHECK_HALF_MV(1, 1, mx , my) | |
237 } | |
238 CHECK_HALF_MV(0, 1, mx , my) | |
239 } | |
240 assert(bx >= xmin*2 && bx <= xmax*2 && by >= ymin*2 && by <= ymax*2); | |
241 } | |
242 | |
1013 | 243 *mx_ptr = bx; |
244 *my_ptr = by; | |
245 | |
936 | 246 return dmin; |
247 } | |
248 #endif | |
249 | |
1013 | 250 static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, Picture *ref_picture, |
1162 | 251 uint8_t * const mv_penalty) |
1013 | 252 { |
253 // const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp; | |
254 const int size= 0; | |
255 const int xx = 16 * s->mb_x; | |
256 const int yy = 16 * s->mb_y; | |
257 const int penalty_factor= s->me.mb_penalty_factor; | |
258 const int xmin= -256*256, ymin= -256*256, xmax= 256*256, ymax= 256*256; //assume that the caller checked these | |
259 const __attribute__((unused)) int unu2= xmin + xmax +ymin + ymax; //no unused warning shit | |
260 me_cmp_func cmp_sub, chroma_cmp_sub; | |
261 int d; | |
262 | |
263 LOAD_COMMON(xx, yy); | |
264 | |
265 //FIXME factorize | |
266 | |
267 cmp_sub= s->dsp.mb_cmp[size]; | |
268 chroma_cmp_sub= s->dsp.mb_cmp[size+1]; | |
269 | |
270 assert(!s->me.skip); | |
271 assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp); | |
272 | |
273 CMP_HPEL(d, mx&1, my&1, mx>>1, my>>1, size); | |
274 //FIXME check cbp before adding penalty for (0,0) vector | |
275 if(mx || my || size>0) | |
276 d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor; | |
277 | |
278 return d; | |
279 } | |
280 | |
936 | 281 #endif /* CMP_HPEL */ |
282 | |
1013 | 283 |
284 | |
936 | 285 #ifdef CMP_QPEL |
286 | |
287 #define CHECK_QUARTER_MV(dx, dy, x, y)\ | |
288 {\ | |
289 const int hx= 4*(x)+(dx);\ | |
290 const int hy= 4*(y)+(dy);\ | |
291 CMP_QPEL(d, dx, dy, x, y, size);\ | |
292 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\ | |
293 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\ | |
294 } | |
295 | |
296 static int RENAME(qpel_motion_search)(MpegEncContext * s, | |
297 int *mx_ptr, int *my_ptr, int dmin, | |
298 int xmin, int ymin, int xmax, int ymax, | |
299 int pred_x, int pred_y, Picture *ref_picture, | |
1162 | 300 int n, int size, uint8_t * const mv_penalty) |
936 | 301 { |
302 const int xx = 16 * s->mb_x + 8*(n&1); | |
303 const int yy = 16 * s->mb_y + 8*(n>>1); | |
304 const int mx = *mx_ptr; | |
305 const int my = *my_ptr; | |
306 const int penalty_factor= s->me.sub_penalty_factor; | |
307 const int map_generation= s->me.map_generation; | |
954 | 308 const int subpel_quality= s->avctx->me_subpel_quality; |
936 | 309 uint32_t *map= s->me.map; |
310 me_cmp_func cmp, chroma_cmp; | |
311 me_cmp_func cmp_sub, chroma_cmp_sub; | |
312 | |
313 LOAD_COMMON(xx, yy); | |
314 | |
315 cmp= s->dsp.me_cmp[size]; | |
316 chroma_cmp= s->dsp.me_cmp[size+1]; //factorize FIXME | |
317 //FIXME factorize | |
318 | |
319 cmp_sub= s->dsp.me_sub_cmp[size]; | |
320 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1]; | |
321 | |
322 if(s->me.skip){ //FIXME somehow move up (benchmark) | |
323 *mx_ptr = 0; | |
324 *my_ptr = 0; | |
325 return dmin; | |
326 } | |
327 | |
328 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){ | |
329 CMP_QPEL(dmin, 0, 0, mx, my, size); | |
1011 | 330 if(mx || my || size>0) |
936 | 331 dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor; |
332 } | |
333 | |
334 if (mx > xmin && mx < xmax && | |
335 my > ymin && my < ymax) { | |
336 int bx=4*mx, by=4*my; | |
337 int d= dmin; | |
338 int i, nx, ny; | |
339 const int index= (my<<ME_MAP_SHIFT) + mx; | |
340 const int t= score_map[(index-(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)]; | |
341 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)]; | |
342 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)]; | |
343 const int b= score_map[(index+(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)]; | |
344 const int c= score_map[(index )&(ME_MAP_SIZE-1)]; | |
345 int best[8]; | |
346 int best_pos[8][2]; | |
347 | |
348 memset(best, 64, sizeof(int)*8); | |
349 #if 1 | |
954 | 350 if(s->me.dia_size>=2){ |
936 | 351 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; |
352 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; | |
353 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)]; | |
354 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)]; | |
355 | |
356 for(ny= -3; ny <= 3; ny++){ | |
357 for(nx= -3; nx <= 3; nx++){ | |
358 const int t2= nx*nx*(tr + tl - 2*t) + 4*nx*(tr-tl) + 32*t; | |
359 const int c2= nx*nx*( r + l - 2*c) + 4*nx*( r- l) + 32*c; | |
360 const int b2= nx*nx*(br + bl - 2*b) + 4*nx*(br-bl) + 32*b; | |
361 int score= ny*ny*(b2 + t2 - 2*c2) + 4*ny*(b2 - t2) + 32*c2; | |
362 int i; | |
363 | |
364 if((nx&3)==0 && (ny&3)==0) continue; | |
365 | |
366 score += 1024*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor; | |
367 | |
368 // if(nx&1) score-=1024*s->me.penalty_factor; | |
369 // if(ny&1) score-=1024*s->me.penalty_factor; | |
370 | |
371 for(i=0; i<8; i++){ | |
372 if(score < best[i]){ | |
373 memmove(&best[i+1], &best[i], sizeof(int)*(7-i)); | |
374 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i)); | |
375 best[i]= score; | |
376 best_pos[i][0]= nx + 4*mx; | |
377 best_pos[i][1]= ny + 4*my; | |
378 break; | |
379 } | |
380 } | |
381 } | |
382 } | |
383 }else{ | |
384 int tl; | |
385 const int cx = 4*(r - l); | |
386 const int cx2= r + l - 2*c; | |
387 const int cy = 4*(b - t); | |
388 const int cy2= b + t - 2*c; | |
389 int cxy; | |
390 | |
391 if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME | |
392 tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; | |
393 }else{ | |
394 CMP(tl, mx-1, my-1, size); //FIXME wrong if chroma me is different | |
395 } | |
396 | |
397 cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c; | |
398 | |
399 assert(16*cx2 + 4*cx + 32*c == 32*r); | |
400 assert(16*cx2 - 4*cx + 32*c == 32*l); | |
401 assert(16*cy2 + 4*cy + 32*c == 32*b); | |
402 assert(16*cy2 - 4*cy + 32*c == 32*t); | |
403 assert(16*cxy + 16*cy2 + 16*cx2 - 4*cy - 4*cx + 32*c == 32*tl); | |
404 | |
405 for(ny= -3; ny <= 3; ny++){ | |
406 for(nx= -3; nx <= 3; nx++){ | |
407 int score= ny*nx*cxy + nx*nx*cx2 + ny*ny*cy2 + nx*cx + ny*cy + 32*c; //FIXME factor | |
408 int i; | |
409 | |
410 if((nx&3)==0 && (ny&3)==0) continue; | |
411 | |
412 score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor; | |
413 // if(nx&1) score-=32*s->me.penalty_factor; | |
414 // if(ny&1) score-=32*s->me.penalty_factor; | |
415 | |
416 for(i=0; i<8; i++){ | |
417 if(score < best[i]){ | |
418 memmove(&best[i+1], &best[i], sizeof(int)*(7-i)); | |
419 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i)); | |
420 best[i]= score; | |
421 best_pos[i][0]= nx + 4*mx; | |
422 best_pos[i][1]= ny + 4*my; | |
423 break; | |
424 } | |
425 } | |
426 } | |
427 } | |
428 } | |
954 | 429 for(i=0; i<subpel_quality; i++){ |
936 | 430 nx= best_pos[i][0]; |
431 ny= best_pos[i][1]; | |
432 CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2) | |
433 } | |
954 | 434 |
936 | 435 #if 0 |
954 | 436 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; |
437 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)]; | |
438 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)]; | |
439 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)]; | |
440 // if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){ | |
441 if(tl<br){ | |
442 | |
443 // nx= FFMAX(4*mx - bx, bx - 4*mx); | |
444 // ny= FFMAX(4*my - by, by - 4*my); | |
936 | 445 |
954 | 446 static int stats[7][7], count; |
447 count++; | |
448 stats[4*mx - bx + 3][4*my - by + 3]++; | |
449 if(256*256*256*64 % count ==0){ | |
450 for(i=0; i<49; i++){ | |
451 if((i%7)==0) printf("\n"); | |
936 | 452 printf("%6d ", stats[0][i]); |
453 } | |
454 printf("\n"); | |
455 } | |
954 | 456 } |
936 | 457 #endif |
458 #else | |
459 | |
460 CHECK_QUARTER_MV(2, 2, mx-1, my-1) | |
461 CHECK_QUARTER_MV(0, 2, mx , my-1) | |
462 CHECK_QUARTER_MV(2, 2, mx , my-1) | |
463 CHECK_QUARTER_MV(2, 0, mx , my ) | |
464 CHECK_QUARTER_MV(2, 2, mx , my ) | |
465 CHECK_QUARTER_MV(0, 2, mx , my ) | |
466 CHECK_QUARTER_MV(2, 2, mx-1, my ) | |
467 CHECK_QUARTER_MV(2, 0, mx-1, my ) | |
468 | |
469 nx= bx; | |
470 ny= by; | |
471 | |
472 for(i=0; i<8; i++){ | |
473 int ox[8]= {0, 1, 1, 1, 0,-1,-1,-1}; | |
474 int oy[8]= {1, 1, 0,-1,-1,-1, 0, 1}; | |
475 CHECK_QUARTER_MV((nx + ox[i])&3, (ny + oy[i])&3, (nx + ox[i])>>2, (ny + oy[i])>>2) | |
476 } | |
477 #endif | |
478 #if 0 | |
479 //outer ring | |
480 CHECK_QUARTER_MV(1, 3, mx-1, my-1) | |
481 CHECK_QUARTER_MV(1, 2, mx-1, my-1) | |
482 CHECK_QUARTER_MV(1, 1, mx-1, my-1) | |
483 CHECK_QUARTER_MV(2, 1, mx-1, my-1) | |
484 CHECK_QUARTER_MV(3, 1, mx-1, my-1) | |
485 CHECK_QUARTER_MV(0, 1, mx , my-1) | |
486 CHECK_QUARTER_MV(1, 1, mx , my-1) | |
487 CHECK_QUARTER_MV(2, 1, mx , my-1) | |
488 CHECK_QUARTER_MV(3, 1, mx , my-1) | |
489 CHECK_QUARTER_MV(3, 2, mx , my-1) | |
490 CHECK_QUARTER_MV(3, 3, mx , my-1) | |
491 CHECK_QUARTER_MV(3, 0, mx , my ) | |
492 CHECK_QUARTER_MV(3, 1, mx , my ) | |
493 CHECK_QUARTER_MV(3, 2, mx , my ) | |
494 CHECK_QUARTER_MV(3, 3, mx , my ) | |
495 CHECK_QUARTER_MV(2, 3, mx , my ) | |
496 CHECK_QUARTER_MV(1, 3, mx , my ) | |
497 CHECK_QUARTER_MV(0, 3, mx , my ) | |
498 CHECK_QUARTER_MV(3, 3, mx-1, my ) | |
499 CHECK_QUARTER_MV(2, 3, mx-1, my ) | |
500 CHECK_QUARTER_MV(1, 3, mx-1, my ) | |
501 CHECK_QUARTER_MV(1, 2, mx-1, my ) | |
502 CHECK_QUARTER_MV(1, 1, mx-1, my ) | |
503 CHECK_QUARTER_MV(1, 0, mx-1, my ) | |
504 #endif | |
505 assert(bx >= xmin*4 && bx <= xmax*4 && by >= ymin*4 && by <= ymax*4); | |
506 | |
507 *mx_ptr = bx; | |
508 *my_ptr = by; | |
509 }else{ | |
510 *mx_ptr =4*mx; | |
511 *my_ptr =4*my; | |
512 } | |
513 | |
514 return dmin; | |
515 } | |
516 | |
1013 | 517 static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, Picture *ref_picture, |
1162 | 518 uint8_t * const mv_penalty) |
1013 | 519 { |
520 const int size= 0; | |
521 const int xx = 16 * s->mb_x; | |
522 const int yy = 16 * s->mb_y; | |
523 const int penalty_factor= s->me.mb_penalty_factor; | |
524 const int xmin= -256*256, ymin= -256*256, xmax= 256*256, ymax= 256*256; //assume that the caller checked these | |
525 const __attribute__((unused)) int unu2= xmin + xmax +ymin + ymax; //no unused warning shit | |
526 me_cmp_func cmp_sub, chroma_cmp_sub; | |
527 int d; | |
528 | |
529 LOAD_COMMON(xx, yy); | |
530 | |
531 //FIXME factorize | |
532 | |
533 cmp_sub= s->dsp.mb_cmp[size]; | |
534 chroma_cmp_sub= s->dsp.mb_cmp[size+1]; | |
535 | |
536 assert(!s->me.skip); | |
537 assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp); | |
538 | |
539 CMP_QPEL(d, mx&3, my&3, mx>>2, my>>2, size); | |
540 //FIXME check cbp before adding penalty for (0,0) vector | |
541 if(mx || my || size>0) | |
542 d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor; | |
543 | |
544 return d; | |
545 } | |
546 | |
547 | |
936 | 548 #endif /* CMP_QPEL */ |
549 | |
550 #define CHECK_MV(x,y)\ | |
551 {\ | |
552 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\ | |
553 const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\ | |
948 | 554 /*printf("check_mv %d %d\n", x, y);*/\ |
936 | 555 if(map[index]!=key){\ |
556 CMP(d, x, y, size);\ | |
557 map[index]= key;\ | |
558 score_map[index]= d;\ | |
559 d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\ | |
948 | 560 /*printf("score:%d\n", d);*/\ |
936 | 561 COPY3_IF_LT(dmin, d, best[0], x, best[1], y)\ |
562 }\ | |
563 } | |
564 | |
948 | 565 #define CHECK_CLIPED_MV(ax,ay)\ |
566 {\ | |
567 const int x= FFMAX(xmin, FFMIN(ax, xmax));\ | |
568 const int y= FFMAX(ymin, FFMIN(ay, ymax));\ | |
569 CHECK_MV(x, y)\ | |
570 } | |
571 | |
936 | 572 #define CHECK_MV_DIR(x,y,new_dir)\ |
573 {\ | |
574 const int key= ((y)<<ME_MAP_MV_BITS) + (x) + map_generation;\ | |
575 const int index= (((y)<<ME_MAP_SHIFT) + (x))&(ME_MAP_SIZE-1);\ | |
948 | 576 /*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\ |
936 | 577 if(map[index]!=key){\ |
578 CMP(d, x, y, size);\ | |
579 map[index]= key;\ | |
580 score_map[index]= d;\ | |
581 d += (mv_penalty[((x)<<shift)-pred_x] + mv_penalty[((y)<<shift)-pred_y])*penalty_factor;\ | |
948 | 582 /*printf("score:%d\n", d);*/\ |
936 | 583 if(d<dmin){\ |
584 best[0]=x;\ | |
585 best[1]=y;\ | |
586 dmin=d;\ | |
587 next_dir= new_dir;\ | |
588 }\ | |
589 }\ | |
590 } | |
591 | |
592 #define check(x,y,S,v)\ | |
593 if( (x)<(xmin<<(S)) ) printf("%d %d %d %d %d xmin" #v, xmin, (x), (y), s->mb_x, s->mb_y);\ | |
594 if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, s->mb_y);\ | |
595 if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\ | |
596 if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\ | |
597 | |
598 | |
599 static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, int dmin, | |
600 Picture *ref_picture, | |
601 int const pred_x, int const pred_y, int const penalty_factor, | |
602 int const xmin, int const ymin, int const xmax, int const ymax, int const shift, | |
1162 | 603 uint32_t *map, int map_generation, int size, uint8_t * const mv_penalty |
936 | 604 ) |
605 { | |
606 me_cmp_func cmp, chroma_cmp; | |
607 int next_dir=-1; | |
608 LOAD_COMMON(s->mb_x*16, s->mb_y*16); | |
609 | |
610 cmp= s->dsp.me_cmp[size]; | |
611 chroma_cmp= s->dsp.me_cmp[size+1]; | |
612 | |
948 | 613 { /* ensure that the best point is in the MAP as h/qpel refinement needs it */ |
614 const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation; | |
615 const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1); | |
616 if(map[index]!=key){ //this will be executed only very rarey | |
617 CMP(score_map[index], best[0], best[1], size); | |
618 map[index]= key; | |
619 } | |
620 } | |
621 | |
936 | 622 for(;;){ |
623 int d; | |
624 const int dir= next_dir; | |
625 const int x= best[0]; | |
626 const int y= best[1]; | |
627 next_dir=-1; | |
628 | |
629 //printf("%d", dir); | |
630 if(dir!=2 && x>xmin) CHECK_MV_DIR(x-1, y , 0) | |
631 if(dir!=3 && y>ymin) CHECK_MV_DIR(x , y-1, 1) | |
632 if(dir!=0 && x<xmax) CHECK_MV_DIR(x+1, y , 2) | |
633 if(dir!=1 && y<ymax) CHECK_MV_DIR(x , y+1, 3) | |
634 | |
635 if(next_dir==-1){ | |
636 return dmin; | |
637 } | |
638 } | |
639 } | |
640 | |
948 | 641 static inline int RENAME(funny_diamond_search)(MpegEncContext * s, int *best, int dmin, |
642 Picture *ref_picture, | |
643 int const pred_x, int const pred_y, int const penalty_factor, | |
644 int const xmin, int const ymin, int const xmax, int const ymax, int const shift, | |
1162 | 645 uint32_t *map, int map_generation, int size, uint8_t * const mv_penalty |
948 | 646 ) |
647 { | |
648 me_cmp_func cmp, chroma_cmp; | |
649 int dia_size; | |
650 LOAD_COMMON(s->mb_x*16, s->mb_y*16); | |
651 | |
652 cmp= s->dsp.me_cmp[size]; | |
653 chroma_cmp= s->dsp.me_cmp[size+1]; | |
654 | |
655 for(dia_size=1; dia_size<=4; dia_size++){ | |
656 int dir; | |
657 const int x= best[0]; | |
658 const int y= best[1]; | |
659 | |
660 if(dia_size&(dia_size-1)) continue; | |
661 | |
662 if( x + dia_size > xmax | |
663 || x - dia_size < xmin | |
664 || y + dia_size > ymax | |
665 || y - dia_size < ymin) | |
666 continue; | |
667 | |
668 for(dir= 0; dir<dia_size; dir+=2){ | |
669 int d; | |
670 | |
671 CHECK_MV(x + dir , y + dia_size - dir); | |
672 CHECK_MV(x + dia_size - dir, y - dir ); | |
673 CHECK_MV(x - dir , y - dia_size + dir); | |
674 CHECK_MV(x - dia_size + dir, y + dir ); | |
675 } | |
676 | |
677 if(x!=best[0] || y!=best[1]) | |
678 dia_size=0; | |
679 #if 0 | |
680 { | |
681 int dx, dy, i; | |
682 static int stats[8*8]; | |
683 dx= ABS(x-best[0]); | |
684 dy= ABS(y-best[1]); | |
685 if(dy>dx){ | |
686 dx^=dy; dy^=dx; dx^=dy; | |
687 } | |
688 stats[dy*8 + dx] ++; | |
689 if(256*256*256*64 % (stats[0]+1)==0){ | |
690 for(i=0; i<64; i++){ | |
691 if((i&7)==0) printf("\n"); | |
692 printf("%8d ", stats[i]); | |
693 } | |
694 printf("\n"); | |
695 } | |
696 } | |
697 #endif | |
698 } | |
699 return dmin; | |
700 } | |
701 | |
702 #define SAB_CHECK_MV(ax,ay)\ | |
703 {\ | |
704 const int key= ((ay)<<ME_MAP_MV_BITS) + (ax) + map_generation;\ | |
705 const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\ | |
706 /*printf("sab check %d %d\n", ax, ay);*/\ | |
707 if(map[index]!=key){\ | |
708 CMP(d, ax, ay, size);\ | |
709 map[index]= key;\ | |
710 score_map[index]= d;\ | |
711 d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\ | |
712 /*printf("score: %d\n", d);*/\ | |
713 if(d < minima[minima_count-1].height){\ | |
714 int j=0;\ | |
715 \ | |
716 while(d >= minima[j].height) j++;\ | |
717 \ | |
718 memmove(&minima [j+1], &minima [j], (minima_count - j - 1)*sizeof(Minima));\ | |
719 \ | |
720 minima[j].checked= 0;\ | |
721 minima[j].height= d;\ | |
722 minima[j].x= ax;\ | |
723 minima[j].y= ay;\ | |
724 \ | |
725 i=-1;\ | |
726 continue;\ | |
727 }\ | |
728 }\ | |
729 } | |
730 | |
731 #define MAX_SAB_SIZE 16 | |
732 static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int dmin, | |
733 Picture *ref_picture, | |
734 int const pred_x, int const pred_y, int const penalty_factor, | |
735 int const xmin, int const ymin, int const xmax, int const ymax, int const shift, | |
1162 | 736 uint32_t *map, int map_generation, int size, uint8_t * const mv_penalty |
948 | 737 ) |
738 { | |
739 me_cmp_func cmp, chroma_cmp; | |
740 Minima minima[MAX_SAB_SIZE]; | |
954 | 741 const int minima_count= ABS(s->me.dia_size); |
948 | 742 int i, j; |
743 LOAD_COMMON(s->mb_x*16, s->mb_y*16); | |
744 | |
745 cmp= s->dsp.me_cmp[size]; | |
746 chroma_cmp= s->dsp.me_cmp[size+1]; | |
747 | |
748 for(j=i=0; i<ME_MAP_SIZE; i++){ | |
749 uint32_t key= map[i]; | |
750 | |
751 key += (1<<(ME_MAP_MV_BITS-1)) + (1<<(2*ME_MAP_MV_BITS-1)); | |
752 | |
753 if((key&((-1)<<(2*ME_MAP_MV_BITS))) != map_generation) continue; | |
754 | |
755 assert(j<MAX_SAB_SIZE); //max j = number of predictors | |
756 | |
757 minima[j].height= score_map[i]; | |
758 minima[j].x= key & ((1<<ME_MAP_MV_BITS)-1); key>>=ME_MAP_MV_BITS; | |
759 minima[j].y= key & ((1<<ME_MAP_MV_BITS)-1); | |
760 minima[j].x-= (1<<(ME_MAP_MV_BITS-1)); | |
761 minima[j].y-= (1<<(ME_MAP_MV_BITS-1)); | |
762 minima[j].checked=0; | |
763 if(minima[j].x || minima[j].y) | |
764 minima[j].height+= (mv_penalty[((minima[j].x)<<shift)-pred_x] + mv_penalty[((minima[j].y)<<shift)-pred_y])*penalty_factor; | |
765 | |
766 j++; | |
767 } | |
768 | |
769 qsort(minima, j, sizeof(Minima), minima_cmp); | |
770 | |
771 for(; j<minima_count; j++){ | |
772 minima[j].height=256*256*256*64; | |
773 minima[j].checked=0; | |
774 minima[j].x= minima[j].y=0; | |
775 } | |
776 | |
777 for(i=0; i<minima_count; i++){ | |
778 const int x= minima[i].x; | |
779 const int y= minima[i].y; | |
780 int d; | |
781 | |
782 if(minima[i].checked) continue; | |
783 | |
784 if( x >= xmax || x <= xmin | |
785 || y >= ymax || y <= ymin) | |
786 continue; | |
787 | |
788 SAB_CHECK_MV(x-1, y) | |
789 SAB_CHECK_MV(x+1, y) | |
790 SAB_CHECK_MV(x , y-1) | |
791 SAB_CHECK_MV(x , y+1) | |
792 | |
793 minima[i].checked= 1; | |
794 } | |
795 | |
796 best[0]= minima[0].x; | |
797 best[1]= minima[0].y; | |
798 dmin= minima[0].height; | |
799 | |
800 if( best[0] < xmax && best[0] > xmin | |
801 && best[1] < ymax && best[1] > ymin){ | |
802 int d; | |
803 //ensure that the refernece samples for hpel refinement are in the map | |
804 CHECK_MV(best[0]-1, best[1]) | |
805 CHECK_MV(best[0]+1, best[1]) | |
806 CHECK_MV(best[0], best[1]-1) | |
807 CHECK_MV(best[0], best[1]+1) | |
808 } | |
809 return dmin; | |
810 } | |
811 | |
936 | 812 static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int dmin, |
813 Picture *ref_picture, | |
814 int const pred_x, int const pred_y, int const penalty_factor, | |
815 int const xmin, int const ymin, int const xmax, int const ymax, int const shift, | |
1162 | 816 uint32_t *map, int map_generation, int size, uint8_t * const mv_penalty |
936 | 817 ) |
818 { | |
819 me_cmp_func cmp, chroma_cmp; | |
948 | 820 int dia_size; |
936 | 821 LOAD_COMMON(s->mb_x*16, s->mb_y*16); |
822 | |
823 cmp= s->dsp.me_cmp[size]; | |
824 chroma_cmp= s->dsp.me_cmp[size+1]; | |
825 | |
954 | 826 for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){ |
936 | 827 int dir, start, end; |
828 const int x= best[0]; | |
829 const int y= best[1]; | |
830 | |
831 start= FFMAX(0, y + dia_size - ymax); | |
948 | 832 end = FFMIN(dia_size, xmax - x + 1); |
936 | 833 for(dir= start; dir<end; dir++){ |
834 int d; | |
835 | |
836 //check(x + dir,y + dia_size - dir,0, a0) | |
837 CHECK_MV(x + dir , y + dia_size - dir); | |
838 } | |
839 | |
840 start= FFMAX(0, x + dia_size - xmax); | |
948 | 841 end = FFMIN(dia_size, y - ymin + 1); |
936 | 842 for(dir= start; dir<end; dir++){ |
843 int d; | |
844 | |
845 //check(x + dia_size - dir, y - dir,0, a1) | |
846 CHECK_MV(x + dia_size - dir, y - dir ); | |
847 } | |
848 | |
849 start= FFMAX(0, -y + dia_size + ymin ); | |
948 | 850 end = FFMIN(dia_size, x - xmin + 1); |
936 | 851 for(dir= start; dir<end; dir++){ |
852 int d; | |
853 | |
854 //check(x - dir,y - dia_size + dir,0, a2) | |
855 CHECK_MV(x - dir , y - dia_size + dir); | |
856 } | |
857 | |
858 start= FFMAX(0, -x + dia_size + xmin ); | |
948 | 859 end = FFMIN(dia_size, ymax - y + 1); |
936 | 860 for(dir= start; dir<end; dir++){ |
861 int d; | |
862 | |
863 //check(x - dia_size + dir, y + dir,0, a3) | |
864 CHECK_MV(x - dia_size + dir, y + dir ); | |
865 } | |
866 | |
867 if(x!=best[0] || y!=best[1]) | |
868 dia_size=0; | |
948 | 869 #if 0 |
870 { | |
871 int dx, dy, i; | |
872 static int stats[8*8]; | |
873 dx= ABS(x-best[0]); | |
874 dy= ABS(y-best[1]); | |
875 stats[dy*8 + dx] ++; | |
876 if(256*256*256*64 % (stats[0]+1)==0){ | |
877 for(i=0; i<64; i++){ | |
878 if((i&7)==0) printf("\n"); | |
879 printf("%6d ", stats[i]); | |
880 } | |
881 printf("\n"); | |
882 } | |
883 } | |
884 #endif | |
936 | 885 } |
886 return dmin; | |
887 } | |
888 | |
889 static int RENAME(epzs_motion_search)(MpegEncContext * s, int block, | |
890 int *mx_ptr, int *my_ptr, | |
891 int P[10][2], int pred_x, int pred_y, | |
948 | 892 int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2], |
1162 | 893 int ref_mv_scale, uint8_t * const mv_penalty) |
936 | 894 { |
895 int best[2]={0, 0}; | |
896 int d, dmin; | |
897 const int shift= 1+s->quarter_sample; | |
898 uint32_t *map= s->me.map; | |
899 int map_generation; | |
900 const int penalty_factor= s->me.penalty_factor; | |
901 const int size=0; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1162
diff
changeset
|
902 const int ref_mv_stride= s->mb_stride; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1162
diff
changeset
|
903 const int ref_mv_xy= s->mb_x + s->mb_y*ref_mv_stride; |
936 | 904 me_cmp_func cmp, chroma_cmp; |
905 LOAD_COMMON(s->mb_x*16, s->mb_y*16); | |
906 | |
907 cmp= s->dsp.me_cmp[size]; | |
908 chroma_cmp= s->dsp.me_cmp[size+1]; | |
909 | |
910 map_generation= update_map_generation(s); | |
911 | |
912 CMP(dmin, 0, 0, size); | |
913 map[0]= map_generation; | |
914 score_map[0]= dmin; | |
915 | |
916 /* first line */ | |
952 | 917 if (s->mb_y == 0) { |
936 | 918 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) |
948 | 919 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, |
920 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) | |
936 | 921 }else{ |
922 if(dmin<256 && ( P_LEFT[0] |P_LEFT[1] | |
923 |P_TOP[0] |P_TOP[1] | |
948 | 924 |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){ |
936 | 925 *mx_ptr= 0; |
926 *my_ptr= 0; | |
927 s->me.skip=1; | |
928 return dmin; | |
929 } | |
930 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift) | |
931 if(dmin>256*2){ | |
948 | 932 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16, |
933 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16) | |
936 | 934 CHECK_MV(P_LEFT[0] >>shift, P_LEFT[1] >>shift) |
935 CHECK_MV(P_TOP[0] >>shift, P_TOP[1] >>shift) | |
936 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift) | |
937 } | |
938 } | |
939 if(dmin>256*4){ | |
952 | 940 if(s->me.pre_pass){ |
941 CHECK_CLIPED_MV((last_mv[ref_mv_xy-1][0]*ref_mv_scale + (1<<15))>>16, | |
942 (last_mv[ref_mv_xy-1][1]*ref_mv_scale + (1<<15))>>16) | |
943 CHECK_CLIPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | |
944 (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
945 }else{ | |
946 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16, | |
947 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16) | |
948 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16, | |
949 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16) | |
950 } | |
936 | 951 } |
948 | 952 |
953 if(s->avctx->last_predictor_count){ | |
954 const int count= s->avctx->last_predictor_count; | |
955 const int xstart= FFMAX(0, s->mb_x - count); | |
956 const int ystart= FFMAX(0, s->mb_y - count); | |
957 const int xend= FFMIN(s->mb_width , s->mb_x + count + 1); | |
958 const int yend= FFMIN(s->mb_height, s->mb_y + count + 1); | |
959 int mb_y; | |
936 | 960 |
948 | 961 for(mb_y=ystart; mb_y<yend; mb_y++){ |
962 int mb_x; | |
963 for(mb_x=xstart; mb_x<xend; mb_x++){ | |
964 const int xy= mb_x + 1 + (mb_y + 1)*ref_mv_stride; | |
965 int mx= (last_mv[xy][0]*ref_mv_scale + (1<<15))>>16; | |
966 int my= (last_mv[xy][1]*ref_mv_scale + (1<<15))>>16; | |
967 | |
968 if(mx>xmax || mx<xmin || my>ymax || my<ymin) continue; | |
969 CHECK_MV(mx,my) | |
936 | 970 } |
971 } | |
972 } | |
948 | 973 |
936 | 974 //check(best[0],best[1],0, b0) |
954 | 975 if(s->me.dia_size==-1) |
948 | 976 dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture, |
977 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
978 shift, map, map_generation, size, mv_penalty); | |
954 | 979 else if(s->me.dia_size<-1) |
948 | 980 dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture, |
981 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
982 shift, map, map_generation, size, mv_penalty); | |
954 | 983 else if(s->me.dia_size<2) |
936 | 984 dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture, |
985 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
986 shift, map, map_generation, size, mv_penalty); | |
987 else | |
988 dmin= RENAME(var_diamond_search)(s, best, dmin, ref_picture, | |
989 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
990 shift, map, map_generation, size, mv_penalty); | |
991 | |
992 //check(best[0],best[1],0, b1) | |
993 *mx_ptr= best[0]; | |
994 *my_ptr= best[1]; | |
995 | |
996 // printf("%d %d %d \n", best[0], best[1], dmin); | |
997 return dmin; | |
998 } | |
999 | |
1000 #ifndef CMP_DIRECT /* no 4mv search needed in direct mode */ | |
1001 static int RENAME(epzs_motion_search4)(MpegEncContext * s, int block, | |
1002 int *mx_ptr, int *my_ptr, | |
1003 int P[10][2], int pred_x, int pred_y, | |
948 | 1004 int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2], |
1162 | 1005 int ref_mv_scale, uint8_t * const mv_penalty) |
936 | 1006 { |
1007 int best[2]={0, 0}; | |
1008 int d, dmin; | |
1009 const int shift= 1+s->quarter_sample; | |
1010 uint32_t *map= s->me.map; | |
1011 int map_generation; | |
1012 const int penalty_factor= s->me.penalty_factor; | |
1013 const int size=1; | |
1177
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1162
diff
changeset
|
1014 const int ref_mv_stride= s->mb_stride; |
fea03d2c4946
simplified adressing of most mb based arrays (mb_x + mb_y*s->mb_stride) now instead of mb_x + mb_y*mb_width and 1+mb_x + (1+mb_y)*(mb_width+2) and ... mixture
michaelni
parents:
1162
diff
changeset
|
1015 const int ref_mv_xy= s->mb_x + s->mb_y *ref_mv_stride; |
936 | 1016 me_cmp_func cmp, chroma_cmp; |
1017 LOAD_COMMON((s->mb_x*2 + (block&1))*8, (s->mb_y*2 + (block>>1))*8); | |
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 */ | |
952 | 1027 if (s->mb_y == 0 && block<2) { |
936 | 1028 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift) |
948 | 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) | |
936 | 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) | |
948 | 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) | |
936 | 1042 } |
1043 } | |
1044 if(dmin>64*4){ | |
948 | 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) | |
936 | 1049 } |
1050 | |
954 | 1051 if(s->me.dia_size==-1) |
948 | 1052 dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture, |
1053 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
1054 shift, map, map_generation, size, mv_penalty); | |
954 | 1055 else if(s->me.dia_size<-1) |
948 | 1056 dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture, |
1057 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
1058 shift, map, map_generation, size, mv_penalty); | |
954 | 1059 else if(s->me.dia_size<2) |
936 | 1060 dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture, |
1061 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
1062 shift, map, map_generation, size, mv_penalty); | |
1063 else | |
1064 dmin= RENAME(var_diamond_search)(s, best, dmin, ref_picture, | |
1065 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax, | |
1066 shift, map, map_generation, size, mv_penalty); | |
948 | 1067 |
936 | 1068 *mx_ptr= best[0]; |
1069 *my_ptr= best[1]; | |
1070 | |
1071 // printf("%d %d %d \n", best[0], best[1], dmin); | |
1072 return dmin; | |
1073 } | |
1074 #endif /* !CMP_DIRECT */ |