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 */
|
|
20
|
|
21 //lets hope gcc will remove the unused vars ...(gcc 3.2.2 seems to do it ...)
|
|
22 //Note, the last line is there to kill these ugly unused var warnings
|
|
23 #define LOAD_COMMON(x, y)\
|
|
24 uint32_t * const score_map= s->me.score_map;\
|
|
25 const int stride= s->linesize;\
|
|
26 const int uvstride= s->uvlinesize;\
|
|
27 const int time_pp= s->pp_time;\
|
|
28 const int time_pb= s->pb_time;\
|
|
29 uint8_t * const src_y= s->new_picture.data[0] + ((y) * stride) + (x);\
|
|
30 uint8_t * const src_u= s->new_picture.data[1] + (((y)>>1) * uvstride) + ((x)>>1);\
|
|
31 uint8_t * const src_v= s->new_picture.data[2] + (((y)>>1) * uvstride) + ((x)>>1);\
|
|
32 uint8_t * const ref_y= ref_picture->data[0] + ((y) * stride) + (x);\
|
|
33 uint8_t * const ref_u= ref_picture->data[1] + (((y)>>1) * uvstride) + ((x)>>1);\
|
|
34 uint8_t * const ref_v= ref_picture->data[2] + (((y)>>1) * uvstride) + ((x)>>1);\
|
|
35 uint8_t * const ref2_y= s->next_picture.data[0] + ((y) * stride) + (x);\
|
|
36 op_pixels_func (*hpel_put)[4];\
|
|
37 op_pixels_func (*hpel_avg)[4]= &s->dsp.avg_pixels_tab[size];\
|
|
38 op_pixels_func (*chroma_hpel_put)[4];\
|
|
39 qpel_mc_func (*qpel_put)[16];\
|
|
40 qpel_mc_func (*qpel_avg)[16]= &s->dsp.avg_qpel_pixels_tab[size];\
|
|
41 const __attribute__((unused)) int unu= time_pp + time_pb + (int)src_u + (int)src_v + (int)ref_u + (int)ref_v\
|
1013
|
42 + (int)ref2_y + (int)hpel_avg + (int)qpel_avg + (int)score_map;\
|
936
|
43 if(s->no_rounding /*FIXME b_type*/){\
|
|
44 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];\
|
|
45 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];\
|
|
46 qpel_put= &s->dsp.put_no_rnd_qpel_pixels_tab[size];\
|
|
47 }else{\
|
|
48 hpel_put=& s->dsp.put_pixels_tab[size];\
|
|
49 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];\
|
|
50 qpel_put= &s->dsp.put_qpel_pixels_tab[size];\
|
|
51 }
|
|
52
|
|
53
|
|
54 #ifdef CMP_HPEL
|
|
55
|
|
56 #define CHECK_HALF_MV(dx, dy, x, y)\
|
|
57 {\
|
|
58 const int hx= 2*(x)+(dx);\
|
|
59 const int hy= 2*(y)+(dy);\
|
|
60 CMP_HPEL(d, dx, dy, x, y, size);\
|
|
61 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
|
|
62 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
|
|
63 }
|
|
64
|
|
65 #if 0
|
|
66 static int RENAME(hpel_motion_search)(MpegEncContext * s,
|
|
67 int *mx_ptr, int *my_ptr, int dmin,
|
|
68 int xmin, int ymin, int xmax, int ymax,
|
|
69 int pred_x, int pred_y, Picture *ref_picture,
|
948
|
70 int n, int size, uint16_t * const mv_penalty)
|
936
|
71 {
|
|
72 const int xx = 16 * s->mb_x + 8*(n&1);
|
|
73 const int yy = 16 * s->mb_y + 8*(n>>1);
|
|
74 const int mx = *mx_ptr;
|
|
75 const int my = *my_ptr;
|
948
|
76 const int penalty_factor= s->me.sub_penalty_factor;
|
936
|
77
|
|
78 LOAD_COMMON(xx, yy);
|
|
79
|
|
80 // INIT;
|
|
81 //FIXME factorize
|
|
82 me_cmp_func cmp, chroma_cmp, cmp_sub, chroma_cmp_sub;
|
|
83
|
|
84 if(s->no_rounding /*FIXME b_type*/){
|
|
85 hpel_put= &s->dsp.put_no_rnd_pixels_tab[size];
|
|
86 chroma_hpel_put= &s->dsp.put_no_rnd_pixels_tab[size+1];
|
|
87 }else{
|
|
88 hpel_put=& s->dsp.put_pixels_tab[size];
|
|
89 chroma_hpel_put= &s->dsp.put_pixels_tab[size+1];
|
|
90 }
|
|
91 cmp= s->dsp.me_cmp[size];
|
|
92 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
93 cmp_sub= s->dsp.me_sub_cmp[size];
|
|
94 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
|
|
95
|
|
96 if(s->me.skip){ //FIXME somehow move up (benchmark)
|
|
97 *mx_ptr = 0;
|
|
98 *my_ptr = 0;
|
|
99 return dmin;
|
|
100 }
|
|
101
|
|
102 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
|
|
103 CMP_HPEL(dmin, 0, 0, mx, my, size);
|
|
104 if(mx || my)
|
|
105 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
|
|
106 }
|
|
107
|
|
108 if (mx > xmin && mx < xmax &&
|
|
109 my > ymin && my < ymax) {
|
|
110 int bx=2*mx, by=2*my;
|
|
111 int d= dmin;
|
|
112
|
|
113 CHECK_HALF_MV(1, 1, mx-1, my-1)
|
|
114 CHECK_HALF_MV(0, 1, mx , my-1)
|
|
115 CHECK_HALF_MV(1, 1, mx , my-1)
|
|
116 CHECK_HALF_MV(1, 0, mx-1, my )
|
|
117 CHECK_HALF_MV(1, 0, mx , my )
|
|
118 CHECK_HALF_MV(1, 1, mx-1, my )
|
|
119 CHECK_HALF_MV(0, 1, mx , my )
|
|
120 CHECK_HALF_MV(1, 1, mx , my )
|
|
121
|
948
|
122 assert(bx >= xmin*2 || bx <= xmax*2 || by >= ymin*2 || by <= ymax*2);
|
936
|
123
|
|
124 *mx_ptr = bx;
|
|
125 *my_ptr = by;
|
|
126 }else{
|
|
127 *mx_ptr =2*mx;
|
|
128 *my_ptr =2*my;
|
|
129 }
|
|
130
|
|
131 return dmin;
|
|
132 }
|
|
133
|
|
134 #else
|
|
135 static int RENAME(hpel_motion_search)(MpegEncContext * s,
|
|
136 int *mx_ptr, int *my_ptr, int dmin,
|
|
137 int xmin, int ymin, int xmax, int ymax,
|
|
138 int pred_x, int pred_y, Picture *ref_picture,
|
|
139 int n, int size, uint16_t * const mv_penalty)
|
|
140 {
|
|
141 const int xx = 16 * s->mb_x + 8*(n&1);
|
|
142 const int yy = 16 * s->mb_y + 8*(n>>1);
|
|
143 const int mx = *mx_ptr;
|
|
144 const int my = *my_ptr;
|
|
145 const int penalty_factor= s->me.sub_penalty_factor;
|
|
146 me_cmp_func cmp_sub, chroma_cmp_sub;
|
1013
|
147 int bx=2*mx, by=2*my;
|
936
|
148
|
|
149 LOAD_COMMON(xx, yy);
|
|
150
|
|
151 //FIXME factorize
|
|
152
|
|
153 cmp_sub= s->dsp.me_sub_cmp[size];
|
|
154 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
|
|
155
|
|
156 if(s->me.skip){ //FIXME move out of hpel?
|
|
157 *mx_ptr = 0;
|
|
158 *my_ptr = 0;
|
|
159 return dmin;
|
|
160 }
|
|
161
|
|
162 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
|
|
163 CMP_HPEL(dmin, 0, 0, mx, my, size);
|
1011
|
164 if(mx || my || size>0)
|
936
|
165 dmin += (mv_penalty[2*mx - pred_x] + mv_penalty[2*my - pred_y])*penalty_factor;
|
|
166 }
|
|
167
|
|
168 if (mx > xmin && mx < xmax &&
|
|
169 my > ymin && my < ymax) {
|
|
170 int d= dmin;
|
|
171 const int index= (my<<ME_MAP_SHIFT) + mx;
|
|
172 const int t= score_map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
|
948
|
173 + (mv_penalty[bx - pred_x] + mv_penalty[by-2 - pred_y])*s->me.penalty_factor;
|
936
|
174 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)]
|
948
|
175 + (mv_penalty[bx-2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor;
|
936
|
176 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)]
|
948
|
177 + (mv_penalty[bx+2 - pred_x] + mv_penalty[by - pred_y])*s->me.penalty_factor;
|
936
|
178 const int b= score_map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)]
|
948
|
179 + (mv_penalty[bx - pred_x] + mv_penalty[by+2 - pred_y])*s->me.penalty_factor;
|
|
180
|
1013
|
181 #if 1
|
948
|
182 int key;
|
|
183 int map_generation= s->me.map_generation;
|
|
184 uint32_t *map= s->me.map;
|
|
185 key= ((my-1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
|
|
186 assert(map[(index-(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
|
|
187 key= ((my+1)<<ME_MAP_MV_BITS) + (mx) + map_generation;
|
|
188 assert(map[(index+(1<<ME_MAP_SHIFT))&(ME_MAP_SIZE-1)] == key);
|
|
189 key= ((my)<<ME_MAP_MV_BITS) + (mx+1) + map_generation;
|
|
190 assert(map[(index+1)&(ME_MAP_SIZE-1)] == key);
|
|
191 key= ((my)<<ME_MAP_MV_BITS) + (mx-1) + map_generation;
|
|
192 assert(map[(index-1)&(ME_MAP_SIZE-1)] == key);
|
|
193 #endif
|
936
|
194 if(t<=b){
|
|
195 CHECK_HALF_MV(0, 1, mx ,my-1)
|
|
196 if(l<=r){
|
|
197 CHECK_HALF_MV(1, 1, mx-1, my-1)
|
|
198 if(t+r<=b+l){
|
|
199 CHECK_HALF_MV(1, 1, mx , my-1)
|
|
200 }else{
|
|
201 CHECK_HALF_MV(1, 1, mx-1, my )
|
|
202 }
|
|
203 CHECK_HALF_MV(1, 0, mx-1, my )
|
|
204 }else{
|
|
205 CHECK_HALF_MV(1, 1, mx , my-1)
|
|
206 if(t+l<=b+r){
|
|
207 CHECK_HALF_MV(1, 1, mx-1, my-1)
|
|
208 }else{
|
|
209 CHECK_HALF_MV(1, 1, mx , my )
|
|
210 }
|
|
211 CHECK_HALF_MV(1, 0, mx , my )
|
|
212 }
|
|
213 }else{
|
|
214 if(l<=r){
|
|
215 if(t+l<=b+r){
|
|
216 CHECK_HALF_MV(1, 1, mx-1, my-1)
|
|
217 }else{
|
|
218 CHECK_HALF_MV(1, 1, mx , my )
|
|
219 }
|
|
220 CHECK_HALF_MV(1, 0, mx-1, my)
|
|
221 CHECK_HALF_MV(1, 1, mx-1, my)
|
|
222 }else{
|
|
223 if(t+r<=b+l){
|
|
224 CHECK_HALF_MV(1, 1, mx , my-1)
|
|
225 }else{
|
|
226 CHECK_HALF_MV(1, 1, mx-1, my)
|
|
227 }
|
|
228 CHECK_HALF_MV(1, 0, mx , my)
|
|
229 CHECK_HALF_MV(1, 1, mx , my)
|
|
230 }
|
|
231 CHECK_HALF_MV(0, 1, mx , my)
|
|
232 }
|
|
233 assert(bx >= xmin*2 && bx <= xmax*2 && by >= ymin*2 && by <= ymax*2);
|
|
234 }
|
|
235
|
1013
|
236 *mx_ptr = bx;
|
|
237 *my_ptr = by;
|
|
238
|
936
|
239 return dmin;
|
|
240 }
|
|
241 #endif
|
|
242
|
1013
|
243 static int RENAME(hpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, Picture *ref_picture,
|
|
244 uint16_t * const mv_penalty)
|
|
245 {
|
|
246 // const int check_luma= s->dsp.me_sub_cmp != s->dsp.mb_cmp;
|
|
247 const int size= 0;
|
|
248 const int xx = 16 * s->mb_x;
|
|
249 const int yy = 16 * s->mb_y;
|
|
250 const int penalty_factor= s->me.mb_penalty_factor;
|
|
251 const int xmin= -256*256, ymin= -256*256, xmax= 256*256, ymax= 256*256; //assume that the caller checked these
|
|
252 const __attribute__((unused)) int unu2= xmin + xmax +ymin + ymax; //no unused warning shit
|
|
253 me_cmp_func cmp_sub, chroma_cmp_sub;
|
|
254 int d;
|
|
255
|
|
256 LOAD_COMMON(xx, yy);
|
|
257
|
|
258 //FIXME factorize
|
|
259
|
|
260 cmp_sub= s->dsp.mb_cmp[size];
|
|
261 chroma_cmp_sub= s->dsp.mb_cmp[size+1];
|
|
262
|
|
263 assert(!s->me.skip);
|
|
264 assert(s->avctx->me_sub_cmp != s->avctx->mb_cmp);
|
|
265
|
|
266 CMP_HPEL(d, mx&1, my&1, mx>>1, my>>1, size);
|
|
267 //FIXME check cbp before adding penalty for (0,0) vector
|
|
268 if(mx || my || size>0)
|
|
269 d += (mv_penalty[mx - pred_x] + mv_penalty[my - pred_y])*penalty_factor;
|
|
270
|
|
271 return d;
|
|
272 }
|
|
273
|
936
|
274 #endif /* CMP_HPEL */
|
|
275
|
1013
|
276
|
|
277
|
936
|
278 #ifdef CMP_QPEL
|
|
279
|
|
280 #define CHECK_QUARTER_MV(dx, dy, x, y)\
|
|
281 {\
|
|
282 const int hx= 4*(x)+(dx);\
|
|
283 const int hy= 4*(y)+(dy);\
|
|
284 CMP_QPEL(d, dx, dy, x, y, size);\
|
|
285 d += (mv_penalty[hx - pred_x] + mv_penalty[hy - pred_y])*penalty_factor;\
|
|
286 COPY3_IF_LT(dmin, d, bx, hx, by, hy)\
|
|
287 }
|
|
288
|
|
289 static int RENAME(qpel_motion_search)(MpegEncContext * s,
|
|
290 int *mx_ptr, int *my_ptr, int dmin,
|
|
291 int xmin, int ymin, int xmax, int ymax,
|
|
292 int pred_x, int pred_y, Picture *ref_picture,
|
|
293 int n, int size, uint16_t * const mv_penalty)
|
|
294 {
|
|
295 const int xx = 16 * s->mb_x + 8*(n&1);
|
|
296 const int yy = 16 * s->mb_y + 8*(n>>1);
|
|
297 const int mx = *mx_ptr;
|
|
298 const int my = *my_ptr;
|
|
299 const int penalty_factor= s->me.sub_penalty_factor;
|
|
300 const int map_generation= s->me.map_generation;
|
954
|
301 const int subpel_quality= s->avctx->me_subpel_quality;
|
936
|
302 uint32_t *map= s->me.map;
|
|
303 me_cmp_func cmp, chroma_cmp;
|
|
304 me_cmp_func cmp_sub, chroma_cmp_sub;
|
|
305
|
|
306 LOAD_COMMON(xx, yy);
|
|
307
|
|
308 cmp= s->dsp.me_cmp[size];
|
|
309 chroma_cmp= s->dsp.me_cmp[size+1]; //factorize FIXME
|
|
310 //FIXME factorize
|
|
311
|
|
312 cmp_sub= s->dsp.me_sub_cmp[size];
|
|
313 chroma_cmp_sub= s->dsp.me_sub_cmp[size+1];
|
|
314
|
|
315 if(s->me.skip){ //FIXME somehow move up (benchmark)
|
|
316 *mx_ptr = 0;
|
|
317 *my_ptr = 0;
|
|
318 return dmin;
|
|
319 }
|
|
320
|
|
321 if(s->avctx->me_cmp != s->avctx->me_sub_cmp){
|
|
322 CMP_QPEL(dmin, 0, 0, mx, my, size);
|
1011
|
323 if(mx || my || size>0)
|
936
|
324 dmin += (mv_penalty[4*mx - pred_x] + mv_penalty[4*my - pred_y])*penalty_factor;
|
|
325 }
|
|
326
|
|
327 if (mx > xmin && mx < xmax &&
|
|
328 my > ymin && my < ymax) {
|
|
329 int bx=4*mx, by=4*my;
|
|
330 int d= dmin;
|
|
331 int i, nx, ny;
|
|
332 const int index= (my<<ME_MAP_SHIFT) + mx;
|
|
333 const int t= score_map[(index-(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
|
|
334 const int l= score_map[(index- 1 )&(ME_MAP_SIZE-1)];
|
|
335 const int r= score_map[(index+ 1 )&(ME_MAP_SIZE-1)];
|
|
336 const int b= score_map[(index+(1<<ME_MAP_SHIFT) )&(ME_MAP_SIZE-1)];
|
|
337 const int c= score_map[(index )&(ME_MAP_SIZE-1)];
|
|
338 int best[8];
|
|
339 int best_pos[8][2];
|
|
340
|
|
341 memset(best, 64, sizeof(int)*8);
|
|
342 #if 1
|
954
|
343 if(s->me.dia_size>=2){
|
936
|
344 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
|
|
345 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
|
|
346 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
|
|
347 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
|
|
348
|
|
349 for(ny= -3; ny <= 3; ny++){
|
|
350 for(nx= -3; nx <= 3; nx++){
|
|
351 const int t2= nx*nx*(tr + tl - 2*t) + 4*nx*(tr-tl) + 32*t;
|
|
352 const int c2= nx*nx*( r + l - 2*c) + 4*nx*( r- l) + 32*c;
|
|
353 const int b2= nx*nx*(br + bl - 2*b) + 4*nx*(br-bl) + 32*b;
|
|
354 int score= ny*ny*(b2 + t2 - 2*c2) + 4*ny*(b2 - t2) + 32*c2;
|
|
355 int i;
|
|
356
|
|
357 if((nx&3)==0 && (ny&3)==0) continue;
|
|
358
|
|
359 score += 1024*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
|
|
360
|
|
361 // if(nx&1) score-=1024*s->me.penalty_factor;
|
|
362 // if(ny&1) score-=1024*s->me.penalty_factor;
|
|
363
|
|
364 for(i=0; i<8; i++){
|
|
365 if(score < best[i]){
|
|
366 memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
|
|
367 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
|
|
368 best[i]= score;
|
|
369 best_pos[i][0]= nx + 4*mx;
|
|
370 best_pos[i][1]= ny + 4*my;
|
|
371 break;
|
|
372 }
|
|
373 }
|
|
374 }
|
|
375 }
|
|
376 }else{
|
|
377 int tl;
|
|
378 const int cx = 4*(r - l);
|
|
379 const int cx2= r + l - 2*c;
|
|
380 const int cy = 4*(b - t);
|
|
381 const int cy2= b + t - 2*c;
|
|
382 int cxy;
|
|
383
|
|
384 if(map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)] == (my<<ME_MAP_MV_BITS) + mx + map_generation && 0){ //FIXME
|
|
385 tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
|
|
386 }else{
|
|
387 CMP(tl, mx-1, my-1, size); //FIXME wrong if chroma me is different
|
|
388 }
|
|
389
|
|
390 cxy= 2*tl + (cx + cy)/4 - (cx2 + cy2) - 2*c;
|
|
391
|
|
392 assert(16*cx2 + 4*cx + 32*c == 32*r);
|
|
393 assert(16*cx2 - 4*cx + 32*c == 32*l);
|
|
394 assert(16*cy2 + 4*cy + 32*c == 32*b);
|
|
395 assert(16*cy2 - 4*cy + 32*c == 32*t);
|
|
396 assert(16*cxy + 16*cy2 + 16*cx2 - 4*cy - 4*cx + 32*c == 32*tl);
|
|
397
|
|
398 for(ny= -3; ny <= 3; ny++){
|
|
399 for(nx= -3; nx <= 3; nx++){
|
|
400 int score= ny*nx*cxy + nx*nx*cx2 + ny*ny*cy2 + nx*cx + ny*cy + 32*c; //FIXME factor
|
|
401 int i;
|
|
402
|
|
403 if((nx&3)==0 && (ny&3)==0) continue;
|
|
404
|
|
405 score += 32*(mv_penalty[4*mx + nx - pred_x] + mv_penalty[4*my + ny - pred_y])*penalty_factor;
|
|
406 // if(nx&1) score-=32*s->me.penalty_factor;
|
|
407 // if(ny&1) score-=32*s->me.penalty_factor;
|
|
408
|
|
409 for(i=0; i<8; i++){
|
|
410 if(score < best[i]){
|
|
411 memmove(&best[i+1], &best[i], sizeof(int)*(7-i));
|
|
412 memmove(&best_pos[i+1][0], &best_pos[i][0], sizeof(int)*2*(7-i));
|
|
413 best[i]= score;
|
|
414 best_pos[i][0]= nx + 4*mx;
|
|
415 best_pos[i][1]= ny + 4*my;
|
|
416 break;
|
|
417 }
|
|
418 }
|
|
419 }
|
|
420 }
|
|
421 }
|
954
|
422 for(i=0; i<subpel_quality; i++){
|
936
|
423 nx= best_pos[i][0];
|
|
424 ny= best_pos[i][1];
|
|
425 CHECK_QUARTER_MV(nx&3, ny&3, nx>>2, ny>>2)
|
|
426 }
|
954
|
427
|
936
|
428 #if 0
|
954
|
429 const int tl= score_map[(index-(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
|
|
430 const int bl= score_map[(index+(1<<ME_MAP_SHIFT)-1)&(ME_MAP_SIZE-1)];
|
|
431 const int tr= score_map[(index-(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
|
|
432 const int br= score_map[(index+(1<<ME_MAP_SHIFT)+1)&(ME_MAP_SIZE-1)];
|
|
433 // if(l < r && l < t && l < b && l < tl && l < bl && l < tr && l < br && bl < tl){
|
|
434 if(tl<br){
|
|
435
|
|
436 // nx= FFMAX(4*mx - bx, bx - 4*mx);
|
|
437 // ny= FFMAX(4*my - by, by - 4*my);
|
936
|
438
|
954
|
439 static int stats[7][7], count;
|
|
440 count++;
|
|
441 stats[4*mx - bx + 3][4*my - by + 3]++;
|
|
442 if(256*256*256*64 % count ==0){
|
|
443 for(i=0; i<49; i++){
|
|
444 if((i%7)==0) printf("\n");
|
936
|
445 printf("%6d ", stats[0][i]);
|
|
446 }
|
|
447 printf("\n");
|
|
448 }
|
954
|
449 }
|
936
|
450 #endif
|
|
451 #else
|
|
452
|
|
453 CHECK_QUARTER_MV(2, 2, mx-1, my-1)
|
|
454 CHECK_QUARTER_MV(0, 2, mx , my-1)
|
|
455 CHECK_QUARTER_MV(2, 2, mx , my-1)
|
|
456 CHECK_QUARTER_MV(2, 0, mx , my )
|
|
457 CHECK_QUARTER_MV(2, 2, mx , my )
|
|
458 CHECK_QUARTER_MV(0, 2, mx , my )
|
|
459 CHECK_QUARTER_MV(2, 2, mx-1, my )
|
|
460 CHECK_QUARTER_MV(2, 0, mx-1, my )
|
|
461
|
|
462 nx= bx;
|
|
463 ny= by;
|
|
464
|
|
465 for(i=0; i<8; i++){
|
|
466 int ox[8]= {0, 1, 1, 1, 0,-1,-1,-1};
|
|
467 int oy[8]= {1, 1, 0,-1,-1,-1, 0, 1};
|
|
468 CHECK_QUARTER_MV((nx + ox[i])&3, (ny + oy[i])&3, (nx + ox[i])>>2, (ny + oy[i])>>2)
|
|
469 }
|
|
470 #endif
|
|
471 #if 0
|
|
472 //outer ring
|
|
473 CHECK_QUARTER_MV(1, 3, mx-1, my-1)
|
|
474 CHECK_QUARTER_MV(1, 2, mx-1, my-1)
|
|
475 CHECK_QUARTER_MV(1, 1, mx-1, my-1)
|
|
476 CHECK_QUARTER_MV(2, 1, mx-1, my-1)
|
|
477 CHECK_QUARTER_MV(3, 1, mx-1, my-1)
|
|
478 CHECK_QUARTER_MV(0, 1, mx , my-1)
|
|
479 CHECK_QUARTER_MV(1, 1, mx , my-1)
|
|
480 CHECK_QUARTER_MV(2, 1, mx , my-1)
|
|
481 CHECK_QUARTER_MV(3, 1, mx , my-1)
|
|
482 CHECK_QUARTER_MV(3, 2, mx , my-1)
|
|
483 CHECK_QUARTER_MV(3, 3, mx , my-1)
|
|
484 CHECK_QUARTER_MV(3, 0, mx , my )
|
|
485 CHECK_QUARTER_MV(3, 1, mx , my )
|
|
486 CHECK_QUARTER_MV(3, 2, mx , my )
|
|
487 CHECK_QUARTER_MV(3, 3, mx , my )
|
|
488 CHECK_QUARTER_MV(2, 3, mx , my )
|
|
489 CHECK_QUARTER_MV(1, 3, mx , my )
|
|
490 CHECK_QUARTER_MV(0, 3, mx , my )
|
|
491 CHECK_QUARTER_MV(3, 3, mx-1, my )
|
|
492 CHECK_QUARTER_MV(2, 3, mx-1, my )
|
|
493 CHECK_QUARTER_MV(1, 3, mx-1, my )
|
|
494 CHECK_QUARTER_MV(1, 2, mx-1, my )
|
|
495 CHECK_QUARTER_MV(1, 1, mx-1, my )
|
|
496 CHECK_QUARTER_MV(1, 0, mx-1, my )
|
|
497 #endif
|
|
498 assert(bx >= xmin*4 && bx <= xmax*4 && by >= ymin*4 && by <= ymax*4);
|
|
499
|
|
500 *mx_ptr = bx;
|
|
501 *my_ptr = by;
|
|
502 }else{
|
|
503 *mx_ptr =4*mx;
|
|
504 *my_ptr =4*my;
|
|
505 }
|
|
506
|
|
507 return dmin;
|
|
508 }
|
|
509
|
1013
|
510 static int RENAME(qpel_get_mb_score)(MpegEncContext * s, int mx, int my, int pred_x, int pred_y, Picture *ref_picture,
|
|
511 uint16_t * const mv_penalty)
|
|
512 {
|
|
513 const int size= 0;
|
|
514 const int xx = 16 * s->mb_x;
|
|
515 const int yy = 16 * s->mb_y;
|
|
516 const int penalty_factor= s->me.mb_penalty_factor;
|
|
517 const int xmin= -256*256, ymin= -256*256, xmax= 256*256, ymax= 256*256; //assume that the caller checked these
|
|
518 const __attribute__((unused)) int unu2= xmin + xmax +ymin + ymax; //no unused warning shit
|
|
519 me_cmp_func cmp_sub, chroma_cmp_sub;
|
|
520 int d;
|
|
521
|
|
522 LOAD_COMMON(xx, yy);
|
|
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
|
936
|
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);\
|
948
|
547 /*printf("check_mv %d %d\n", x, y);*/\
|
936
|
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;\
|
948
|
553 /*printf("score:%d\n", d);*/\
|
936
|
554 COPY3_IF_LT(dmin, d, best[0], x, best[1], y)\
|
|
555 }\
|
|
556 }
|
|
557
|
948
|
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
|
936
|
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);\
|
948
|
569 /*printf("check_mv_dir %d %d %d\n", x, y, new_dir);*/\
|
936
|
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;\
|
948
|
575 /*printf("score:%d\n", d);*/\
|
936
|
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)\
|
|
586 if( (x)<(xmin<<(S)) ) printf("%d %d %d %d %d xmin" #v, xmin, (x), (y), s->mb_x, s->mb_y);\
|
|
587 if( (x)>(xmax<<(S)) ) printf("%d %d %d %d %d xmax" #v, xmax, (x), (y), s->mb_x, s->mb_y);\
|
|
588 if( (y)<(ymin<<(S)) ) printf("%d %d %d %d %d ymin" #v, ymin, (x), (y), s->mb_x, s->mb_y);\
|
|
589 if( (y)>(ymax<<(S)) ) printf("%d %d %d %d %d ymax" #v, ymax, (x), (y), s->mb_x, s->mb_y);\
|
|
590
|
|
591
|
|
592 static inline int RENAME(small_diamond_search)(MpegEncContext * s, int *best, int dmin,
|
|
593 Picture *ref_picture,
|
|
594 int const pred_x, int const pred_y, int const penalty_factor,
|
|
595 int const xmin, int const ymin, int const xmax, int const ymax, int const shift,
|
|
596 uint32_t *map, int map_generation, int size, uint16_t * const mv_penalty
|
|
597 )
|
|
598 {
|
|
599 me_cmp_func cmp, chroma_cmp;
|
|
600 int next_dir=-1;
|
|
601 LOAD_COMMON(s->mb_x*16, s->mb_y*16);
|
|
602
|
|
603 cmp= s->dsp.me_cmp[size];
|
|
604 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
605
|
948
|
606 { /* ensure that the best point is in the MAP as h/qpel refinement needs it */
|
|
607 const int key= (best[1]<<ME_MAP_MV_BITS) + best[0] + map_generation;
|
|
608 const int index= ((best[1]<<ME_MAP_SHIFT) + best[0])&(ME_MAP_SIZE-1);
|
|
609 if(map[index]!=key){ //this will be executed only very rarey
|
|
610 CMP(score_map[index], best[0], best[1], size);
|
|
611 map[index]= key;
|
|
612 }
|
|
613 }
|
|
614
|
936
|
615 for(;;){
|
|
616 int d;
|
|
617 const int dir= next_dir;
|
|
618 const int x= best[0];
|
|
619 const int y= best[1];
|
|
620 next_dir=-1;
|
|
621
|
|
622 //printf("%d", dir);
|
|
623 if(dir!=2 && x>xmin) CHECK_MV_DIR(x-1, y , 0)
|
|
624 if(dir!=3 && y>ymin) CHECK_MV_DIR(x , y-1, 1)
|
|
625 if(dir!=0 && x<xmax) CHECK_MV_DIR(x+1, y , 2)
|
|
626 if(dir!=1 && y<ymax) CHECK_MV_DIR(x , y+1, 3)
|
|
627
|
|
628 if(next_dir==-1){
|
|
629 return dmin;
|
|
630 }
|
|
631 }
|
|
632 }
|
|
633
|
948
|
634 static inline int RENAME(funny_diamond_search)(MpegEncContext * s, int *best, int dmin,
|
|
635 Picture *ref_picture,
|
|
636 int const pred_x, int const pred_y, int const penalty_factor,
|
|
637 int const xmin, int const ymin, int const xmax, int const ymax, int const shift,
|
|
638 uint32_t *map, int map_generation, int size, uint16_t * const mv_penalty
|
|
639 )
|
|
640 {
|
|
641 me_cmp_func cmp, chroma_cmp;
|
|
642 int dia_size;
|
|
643 LOAD_COMMON(s->mb_x*16, s->mb_y*16);
|
|
644
|
|
645 cmp= s->dsp.me_cmp[size];
|
|
646 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
647
|
|
648 for(dia_size=1; dia_size<=4; dia_size++){
|
|
649 int dir;
|
|
650 const int x= best[0];
|
|
651 const int y= best[1];
|
|
652
|
|
653 if(dia_size&(dia_size-1)) continue;
|
|
654
|
|
655 if( x + dia_size > xmax
|
|
656 || x - dia_size < xmin
|
|
657 || y + dia_size > ymax
|
|
658 || y - dia_size < ymin)
|
|
659 continue;
|
|
660
|
|
661 for(dir= 0; dir<dia_size; dir+=2){
|
|
662 int d;
|
|
663
|
|
664 CHECK_MV(x + dir , y + dia_size - dir);
|
|
665 CHECK_MV(x + dia_size - dir, y - dir );
|
|
666 CHECK_MV(x - dir , y - dia_size + dir);
|
|
667 CHECK_MV(x - dia_size + dir, y + dir );
|
|
668 }
|
|
669
|
|
670 if(x!=best[0] || y!=best[1])
|
|
671 dia_size=0;
|
|
672 #if 0
|
|
673 {
|
|
674 int dx, dy, i;
|
|
675 static int stats[8*8];
|
|
676 dx= ABS(x-best[0]);
|
|
677 dy= ABS(y-best[1]);
|
|
678 if(dy>dx){
|
|
679 dx^=dy; dy^=dx; dx^=dy;
|
|
680 }
|
|
681 stats[dy*8 + dx] ++;
|
|
682 if(256*256*256*64 % (stats[0]+1)==0){
|
|
683 for(i=0; i<64; i++){
|
|
684 if((i&7)==0) printf("\n");
|
|
685 printf("%8d ", stats[i]);
|
|
686 }
|
|
687 printf("\n");
|
|
688 }
|
|
689 }
|
|
690 #endif
|
|
691 }
|
|
692 return dmin;
|
|
693 }
|
|
694
|
|
695 #define SAB_CHECK_MV(ax,ay)\
|
|
696 {\
|
|
697 const int key= ((ay)<<ME_MAP_MV_BITS) + (ax) + map_generation;\
|
|
698 const int index= (((ay)<<ME_MAP_SHIFT) + (ax))&(ME_MAP_SIZE-1);\
|
|
699 /*printf("sab check %d %d\n", ax, ay);*/\
|
|
700 if(map[index]!=key){\
|
|
701 CMP(d, ax, ay, size);\
|
|
702 map[index]= key;\
|
|
703 score_map[index]= d;\
|
|
704 d += (mv_penalty[((ax)<<shift)-pred_x] + mv_penalty[((ay)<<shift)-pred_y])*penalty_factor;\
|
|
705 /*printf("score: %d\n", d);*/\
|
|
706 if(d < minima[minima_count-1].height){\
|
|
707 int j=0;\
|
|
708 \
|
|
709 while(d >= minima[j].height) j++;\
|
|
710 \
|
|
711 memmove(&minima [j+1], &minima [j], (minima_count - j - 1)*sizeof(Minima));\
|
|
712 \
|
|
713 minima[j].checked= 0;\
|
|
714 minima[j].height= d;\
|
|
715 minima[j].x= ax;\
|
|
716 minima[j].y= ay;\
|
|
717 \
|
|
718 i=-1;\
|
|
719 continue;\
|
|
720 }\
|
|
721 }\
|
|
722 }
|
|
723
|
|
724 #define MAX_SAB_SIZE 16
|
|
725 static inline int RENAME(sab_diamond_search)(MpegEncContext * s, int *best, int dmin,
|
|
726 Picture *ref_picture,
|
|
727 int const pred_x, int const pred_y, int const penalty_factor,
|
|
728 int const xmin, int const ymin, int const xmax, int const ymax, int const shift,
|
|
729 uint32_t *map, int map_generation, int size, uint16_t * const mv_penalty
|
|
730 )
|
|
731 {
|
|
732 me_cmp_func cmp, chroma_cmp;
|
|
733 Minima minima[MAX_SAB_SIZE];
|
954
|
734 const int minima_count= ABS(s->me.dia_size);
|
948
|
735 int i, j;
|
|
736 LOAD_COMMON(s->mb_x*16, s->mb_y*16);
|
|
737
|
|
738 cmp= s->dsp.me_cmp[size];
|
|
739 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
740
|
|
741 for(j=i=0; i<ME_MAP_SIZE; i++){
|
|
742 uint32_t key= map[i];
|
|
743
|
|
744 key += (1<<(ME_MAP_MV_BITS-1)) + (1<<(2*ME_MAP_MV_BITS-1));
|
|
745
|
|
746 if((key&((-1)<<(2*ME_MAP_MV_BITS))) != map_generation) continue;
|
|
747
|
|
748 assert(j<MAX_SAB_SIZE); //max j = number of predictors
|
|
749
|
|
750 minima[j].height= score_map[i];
|
|
751 minima[j].x= key & ((1<<ME_MAP_MV_BITS)-1); key>>=ME_MAP_MV_BITS;
|
|
752 minima[j].y= key & ((1<<ME_MAP_MV_BITS)-1);
|
|
753 minima[j].x-= (1<<(ME_MAP_MV_BITS-1));
|
|
754 minima[j].y-= (1<<(ME_MAP_MV_BITS-1));
|
|
755 minima[j].checked=0;
|
|
756 if(minima[j].x || minima[j].y)
|
|
757 minima[j].height+= (mv_penalty[((minima[j].x)<<shift)-pred_x] + mv_penalty[((minima[j].y)<<shift)-pred_y])*penalty_factor;
|
|
758
|
|
759 j++;
|
|
760 }
|
|
761
|
|
762 qsort(minima, j, sizeof(Minima), minima_cmp);
|
|
763
|
|
764 for(; j<minima_count; j++){
|
|
765 minima[j].height=256*256*256*64;
|
|
766 minima[j].checked=0;
|
|
767 minima[j].x= minima[j].y=0;
|
|
768 }
|
|
769
|
|
770 for(i=0; i<minima_count; i++){
|
|
771 const int x= minima[i].x;
|
|
772 const int y= minima[i].y;
|
|
773 int d;
|
|
774
|
|
775 if(minima[i].checked) continue;
|
|
776
|
|
777 if( x >= xmax || x <= xmin
|
|
778 || y >= ymax || y <= ymin)
|
|
779 continue;
|
|
780
|
|
781 SAB_CHECK_MV(x-1, y)
|
|
782 SAB_CHECK_MV(x+1, y)
|
|
783 SAB_CHECK_MV(x , y-1)
|
|
784 SAB_CHECK_MV(x , y+1)
|
|
785
|
|
786 minima[i].checked= 1;
|
|
787 }
|
|
788
|
|
789 best[0]= minima[0].x;
|
|
790 best[1]= minima[0].y;
|
|
791 dmin= minima[0].height;
|
|
792
|
|
793 if( best[0] < xmax && best[0] > xmin
|
|
794 && best[1] < ymax && best[1] > ymin){
|
|
795 int d;
|
|
796 //ensure that the refernece samples for hpel refinement are in the map
|
|
797 CHECK_MV(best[0]-1, best[1])
|
|
798 CHECK_MV(best[0]+1, best[1])
|
|
799 CHECK_MV(best[0], best[1]-1)
|
|
800 CHECK_MV(best[0], best[1]+1)
|
|
801 }
|
|
802 return dmin;
|
|
803 }
|
|
804
|
936
|
805 static inline int RENAME(var_diamond_search)(MpegEncContext * s, int *best, int dmin,
|
|
806 Picture *ref_picture,
|
|
807 int const pred_x, int const pred_y, int const penalty_factor,
|
|
808 int const xmin, int const ymin, int const xmax, int const ymax, int const shift,
|
|
809 uint32_t *map, int map_generation, int size, uint16_t * const mv_penalty
|
|
810 )
|
|
811 {
|
|
812 me_cmp_func cmp, chroma_cmp;
|
948
|
813 int dia_size;
|
936
|
814 LOAD_COMMON(s->mb_x*16, s->mb_y*16);
|
|
815
|
|
816 cmp= s->dsp.me_cmp[size];
|
|
817 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
818
|
954
|
819 for(dia_size=1; dia_size<=s->me.dia_size; dia_size++){
|
936
|
820 int dir, start, end;
|
|
821 const int x= best[0];
|
|
822 const int y= best[1];
|
|
823
|
|
824 start= FFMAX(0, y + dia_size - ymax);
|
948
|
825 end = FFMIN(dia_size, xmax - x + 1);
|
936
|
826 for(dir= start; dir<end; dir++){
|
|
827 int d;
|
|
828
|
|
829 //check(x + dir,y + dia_size - dir,0, a0)
|
|
830 CHECK_MV(x + dir , y + dia_size - dir);
|
|
831 }
|
|
832
|
|
833 start= FFMAX(0, x + dia_size - xmax);
|
948
|
834 end = FFMIN(dia_size, y - ymin + 1);
|
936
|
835 for(dir= start; dir<end; dir++){
|
|
836 int d;
|
|
837
|
|
838 //check(x + dia_size - dir, y - dir,0, a1)
|
|
839 CHECK_MV(x + dia_size - dir, y - dir );
|
|
840 }
|
|
841
|
|
842 start= FFMAX(0, -y + dia_size + ymin );
|
948
|
843 end = FFMIN(dia_size, x - xmin + 1);
|
936
|
844 for(dir= start; dir<end; dir++){
|
|
845 int d;
|
|
846
|
|
847 //check(x - dir,y - dia_size + dir,0, a2)
|
|
848 CHECK_MV(x - dir , y - dia_size + dir);
|
|
849 }
|
|
850
|
|
851 start= FFMAX(0, -x + dia_size + xmin );
|
948
|
852 end = FFMIN(dia_size, ymax - y + 1);
|
936
|
853 for(dir= start; dir<end; dir++){
|
|
854 int d;
|
|
855
|
|
856 //check(x - dia_size + dir, y + dir,0, a3)
|
|
857 CHECK_MV(x - dia_size + dir, y + dir );
|
|
858 }
|
|
859
|
|
860 if(x!=best[0] || y!=best[1])
|
|
861 dia_size=0;
|
948
|
862 #if 0
|
|
863 {
|
|
864 int dx, dy, i;
|
|
865 static int stats[8*8];
|
|
866 dx= ABS(x-best[0]);
|
|
867 dy= ABS(y-best[1]);
|
|
868 stats[dy*8 + dx] ++;
|
|
869 if(256*256*256*64 % (stats[0]+1)==0){
|
|
870 for(i=0; i<64; i++){
|
|
871 if((i&7)==0) printf("\n");
|
|
872 printf("%6d ", stats[i]);
|
|
873 }
|
|
874 printf("\n");
|
|
875 }
|
|
876 }
|
|
877 #endif
|
936
|
878 }
|
|
879 return dmin;
|
|
880 }
|
|
881
|
|
882 static int RENAME(epzs_motion_search)(MpegEncContext * s, int block,
|
|
883 int *mx_ptr, int *my_ptr,
|
|
884 int P[10][2], int pred_x, int pred_y,
|
948
|
885 int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2],
|
|
886 int ref_mv_scale, uint16_t * const mv_penalty)
|
936
|
887 {
|
|
888 int best[2]={0, 0};
|
|
889 int d, dmin;
|
|
890 const int shift= 1+s->quarter_sample;
|
|
891 uint32_t *map= s->me.map;
|
|
892 int map_generation;
|
|
893 const int penalty_factor= s->me.penalty_factor;
|
|
894 const int size=0;
|
948
|
895 const int ref_mv_stride= s->mb_width+2;
|
|
896 const int ref_mv_xy= 1 + s->mb_x + (s->mb_y + 1)*ref_mv_stride;
|
936
|
897 me_cmp_func cmp, chroma_cmp;
|
|
898 LOAD_COMMON(s->mb_x*16, s->mb_y*16);
|
|
899
|
|
900 cmp= s->dsp.me_cmp[size];
|
|
901 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
902
|
|
903 map_generation= update_map_generation(s);
|
|
904
|
|
905 CMP(dmin, 0, 0, size);
|
|
906 map[0]= map_generation;
|
|
907 score_map[0]= dmin;
|
|
908
|
|
909 /* first line */
|
952
|
910 if (s->mb_y == 0) {
|
936
|
911 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
|
948
|
912 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
|
|
913 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
|
936
|
914 }else{
|
|
915 if(dmin<256 && ( P_LEFT[0] |P_LEFT[1]
|
|
916 |P_TOP[0] |P_TOP[1]
|
948
|
917 |P_TOPRIGHT[0]|P_TOPRIGHT[1])==0){
|
936
|
918 *mx_ptr= 0;
|
|
919 *my_ptr= 0;
|
|
920 s->me.skip=1;
|
|
921 return dmin;
|
|
922 }
|
|
923 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
|
|
924 if(dmin>256*2){
|
948
|
925 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
|
|
926 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
|
936
|
927 CHECK_MV(P_LEFT[0] >>shift, P_LEFT[1] >>shift)
|
|
928 CHECK_MV(P_TOP[0] >>shift, P_TOP[1] >>shift)
|
|
929 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
|
|
930 }
|
|
931 }
|
|
932 if(dmin>256*4){
|
952
|
933 if(s->me.pre_pass){
|
|
934 CHECK_CLIPED_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 CHECK_CLIPED_MV((last_mv[ref_mv_xy-ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
|
|
937 (last_mv[ref_mv_xy-ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
|
|
938 }else{
|
|
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 }
|
936
|
944 }
|
948
|
945
|
|
946 if(s->avctx->last_predictor_count){
|
|
947 const int count= s->avctx->last_predictor_count;
|
|
948 const int xstart= FFMAX(0, s->mb_x - count);
|
|
949 const int ystart= FFMAX(0, s->mb_y - count);
|
|
950 const int xend= FFMIN(s->mb_width , s->mb_x + count + 1);
|
|
951 const int yend= FFMIN(s->mb_height, s->mb_y + count + 1);
|
|
952 int mb_y;
|
936
|
953
|
948
|
954 for(mb_y=ystart; mb_y<yend; mb_y++){
|
|
955 int mb_x;
|
|
956 for(mb_x=xstart; mb_x<xend; mb_x++){
|
|
957 const int xy= mb_x + 1 + (mb_y + 1)*ref_mv_stride;
|
|
958 int mx= (last_mv[xy][0]*ref_mv_scale + (1<<15))>>16;
|
|
959 int my= (last_mv[xy][1]*ref_mv_scale + (1<<15))>>16;
|
|
960
|
|
961 if(mx>xmax || mx<xmin || my>ymax || my<ymin) continue;
|
|
962 CHECK_MV(mx,my)
|
936
|
963 }
|
|
964 }
|
|
965 }
|
948
|
966
|
936
|
967 //check(best[0],best[1],0, b0)
|
954
|
968 if(s->me.dia_size==-1)
|
948
|
969 dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture,
|
|
970 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
971 shift, map, map_generation, size, mv_penalty);
|
954
|
972 else if(s->me.dia_size<-1)
|
948
|
973 dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture,
|
|
974 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
975 shift, map, map_generation, size, mv_penalty);
|
954
|
976 else if(s->me.dia_size<2)
|
936
|
977 dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture,
|
|
978 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
979 shift, map, map_generation, size, mv_penalty);
|
|
980 else
|
|
981 dmin= RENAME(var_diamond_search)(s, best, dmin, ref_picture,
|
|
982 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
983 shift, map, map_generation, size, mv_penalty);
|
|
984
|
|
985 //check(best[0],best[1],0, b1)
|
|
986 *mx_ptr= best[0];
|
|
987 *my_ptr= best[1];
|
|
988
|
|
989 // printf("%d %d %d \n", best[0], best[1], dmin);
|
|
990 return dmin;
|
|
991 }
|
|
992
|
|
993 #ifndef CMP_DIRECT /* no 4mv search needed in direct mode */
|
|
994 static int RENAME(epzs_motion_search4)(MpegEncContext * s, int block,
|
|
995 int *mx_ptr, int *my_ptr,
|
|
996 int P[10][2], int pred_x, int pred_y,
|
948
|
997 int xmin, int ymin, int xmax, int ymax, Picture *ref_picture, int16_t (*last_mv)[2],
|
|
998 int ref_mv_scale, uint16_t * const mv_penalty)
|
936
|
999 {
|
|
1000 int best[2]={0, 0};
|
|
1001 int d, dmin;
|
|
1002 const int shift= 1+s->quarter_sample;
|
|
1003 uint32_t *map= s->me.map;
|
|
1004 int map_generation;
|
|
1005 const int penalty_factor= s->me.penalty_factor;
|
|
1006 const int size=1;
|
948
|
1007 const int ref_mv_stride= s->mb_width+2;
|
|
1008 const int ref_mv_xy= 1 + s->mb_x + (s->mb_y + 1)*ref_mv_stride;
|
936
|
1009 me_cmp_func cmp, chroma_cmp;
|
|
1010 LOAD_COMMON((s->mb_x*2 + (block&1))*8, (s->mb_y*2 + (block>>1))*8);
|
|
1011
|
|
1012 cmp= s->dsp.me_cmp[size];
|
|
1013 chroma_cmp= s->dsp.me_cmp[size+1];
|
|
1014
|
|
1015 map_generation= update_map_generation(s);
|
|
1016
|
|
1017 dmin = 1000000;
|
|
1018 //printf("%d %d %d %d //",xmin, ymin, xmax, ymax);
|
|
1019 /* first line */
|
952
|
1020 if (s->mb_y == 0 && block<2) {
|
936
|
1021 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
|
948
|
1022 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
|
|
1023 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
|
936
|
1024 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
|
|
1025 }else{
|
|
1026 CHECK_MV(P_MV1[0]>>shift, P_MV1[1]>>shift)
|
|
1027 //FIXME try some early stop
|
|
1028 if(dmin>64*2){
|
|
1029 CHECK_MV(P_MEDIAN[0]>>shift, P_MEDIAN[1]>>shift)
|
|
1030 CHECK_MV(P_LEFT[0]>>shift, P_LEFT[1]>>shift)
|
|
1031 CHECK_MV(P_TOP[0]>>shift, P_TOP[1]>>shift)
|
|
1032 CHECK_MV(P_TOPRIGHT[0]>>shift, P_TOPRIGHT[1]>>shift)
|
948
|
1033 CHECK_CLIPED_MV((last_mv[ref_mv_xy][0]*ref_mv_scale + (1<<15))>>16,
|
|
1034 (last_mv[ref_mv_xy][1]*ref_mv_scale + (1<<15))>>16)
|
936
|
1035 }
|
|
1036 }
|
|
1037 if(dmin>64*4){
|
948
|
1038 CHECK_CLIPED_MV((last_mv[ref_mv_xy+1][0]*ref_mv_scale + (1<<15))>>16,
|
|
1039 (last_mv[ref_mv_xy+1][1]*ref_mv_scale + (1<<15))>>16)
|
|
1040 CHECK_CLIPED_MV((last_mv[ref_mv_xy+ref_mv_stride][0]*ref_mv_scale + (1<<15))>>16,
|
|
1041 (last_mv[ref_mv_xy+ref_mv_stride][1]*ref_mv_scale + (1<<15))>>16)
|
936
|
1042 }
|
|
1043
|
954
|
1044 if(s->me.dia_size==-1)
|
948
|
1045 dmin= RENAME(funny_diamond_search)(s, best, dmin, ref_picture,
|
|
1046 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
1047 shift, map, map_generation, size, mv_penalty);
|
954
|
1048 else if(s->me.dia_size<-1)
|
948
|
1049 dmin= RENAME(sab_diamond_search)(s, best, dmin, ref_picture,
|
|
1050 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
1051 shift, map, map_generation, size, mv_penalty);
|
954
|
1052 else if(s->me.dia_size<2)
|
936
|
1053 dmin= RENAME(small_diamond_search)(s, best, dmin, ref_picture,
|
|
1054 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
1055 shift, map, map_generation, size, mv_penalty);
|
|
1056 else
|
|
1057 dmin= RENAME(var_diamond_search)(s, best, dmin, ref_picture,
|
|
1058 pred_x, pred_y, penalty_factor, xmin, ymin, xmax, ymax,
|
|
1059 shift, map, map_generation, size, mv_penalty);
|
948
|
1060
|
936
|
1061 *mx_ptr= best[0];
|
|
1062 *my_ptr= best[1];
|
|
1063
|
|
1064 // printf("%d %d %d \n", best[0], best[1], dmin);
|
|
1065 return dmin;
|
|
1066 }
|
|
1067 #endif /* !CMP_DIRECT */
|