[svn] / branches / dev-api-3 / xvidcore / src / motion / motion_est.c Repository:
ViewVC logotype

Annotation of /branches/dev-api-3/xvidcore/src/motion/motion_est.c

Parent Directory Parent Directory | Revision Log Revision Log

Revision 574 - (view) (download)

1 : Isibaar 3 /**************************************************************************
2 :     *
3 : chl 259 * XVID MPEG-4 VIDEO CODEC
4 :     * motion estimation
5 :     *
6 :     * This program is an implementation of a part of one or more MPEG-4
7 :     * Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :     * to use this software module in hardware or software products are
9 :     * advised that its use may infringe existing patents or copyrights, and
10 :     * any such use would be at such party's own risk. The original
11 :     * developer of this software module and his/her company, and subsequent
12 :     * editors and their companies, will have no liability for use of this
13 :     * software or modifications or derivatives thereof.
14 :     *
15 :     * This program is free software; you can redistribute it and/or modify
16 :     * it under the terms of the GNU General Public License as published by
17 :     * the Free Software Foundation; either version 2 of the License, or
18 :     * (at your option) any later version.
19 :     *
20 :     * This program is distributed in the hope that it will be useful,
21 :     * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :     * GNU General Public License for more details.
24 :     *
25 :     * You should have received a copy of the GNU General Public License
26 :     * along with this program; if not, write to the Free Software
27 :     * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :     *
29 :     *************************************************************************/
30 :    
31 : Isibaar 3 #include <assert.h>
32 :     #include <stdio.h>
33 : chl 96 #include <stdlib.h>
34 : Isibaar 3
35 :     #include "../encoder.h"
36 :     #include "../utils/mbfunctions.h"
37 :     #include "../prediction/mbprediction.h"
38 :     #include "../global.h"
39 :     #include "../utils/timer.h"
40 : chl 530 #include "motion_est.h"
41 : suxen_drol 118 #include "motion.h"
42 : Isibaar 3 #include "sad.h"
43 : Isibaar 539 #include "../utils/emms.h"
44 : Isibaar 3
45 : chl 530 #define INITIAL_SKIP_THRESH (10)
46 :     #define FINAL_SKIP_THRESH (50)
47 :     #define MAX_SAD00_FOR_SKIP (20)
48 : Isibaar 539 #define MAX_CHROMA_SAD_FOR_SKIP (22)
49 :     #define SKIP_THRESH_B (25)
50 : Isibaar 3
51 : chl 530 #define CHECK_CANDIDATE(X,Y,D) { \
52 :     (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
53 : Isibaar 3
54 : chl 530 #define iDiamondSize 2
55 : chl 141
56 : chl 530 static __inline int
57 :     d_mv_bits(int x, int y, const uint32_t iFcode)
58 : Isibaar 3 {
59 : chl 530 int xb, yb;
60 :    
61 :     if (x == 0) xb = 1;
62 :     else {
63 :     if (x < 0) x = -x;
64 :     x += (1 << (iFcode - 1)) - 1;
65 :     x >>= (iFcode - 1);
66 :     if (x > 32) x = 32;
67 :     xb = mvtab[x] + iFcode;
68 : edgomez 195 }
69 : Isibaar 3
70 : chl 530 if (y == 0) yb = 1;
71 :     else {
72 :     if (y < 0) y = -y;
73 :     y += (1 << (iFcode - 1)) - 1;
74 :     y >>= (iFcode - 1);
75 :     if (y > 32) y = 32;
76 :     yb = mvtab[y] + iFcode;
77 :     }
78 :     return xb + yb;
79 : Isibaar 3 }
80 :    
81 :    
82 : Isibaar 539 /* CHECK_CANDIATE FUNCTIONS START */
83 :    
84 : chl 530 static void
85 :     CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
86 : Isibaar 3 {
87 : chl 530 int32_t * const sad = data->temp;
88 :     int t;
89 :     const uint8_t * Reference;
90 : Isibaar 3
91 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
92 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
93 : suxen_drol 136
94 : chl 530 switch ( ((x&1)<<1) + (y&1) ) {
95 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
96 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
97 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
98 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
99 :     }
100 : chl 259
101 : chl 530 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1);
102 : Isibaar 3
103 : chl 530 t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
104 :     data->temp[0] += lambda_vec16[data->iQuant] * t;
105 :     data->temp[1] += lambda_vec8[data->iQuant] * t;
106 : chl 326
107 : chl 530 if (data->temp[0] < data->iMinSAD[0]) {
108 :     data->iMinSAD[0] = data->temp[0];
109 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
110 :     *dir = Direction; }
111 : chl 184
112 : chl 530 if (data->temp[1] < data->iMinSAD[1]) {
113 :     data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
114 :     if (data->temp[2] < data->iMinSAD[2]) {
115 :     data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
116 :     if (data->temp[3] < data->iMinSAD[3]) {
117 :     data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
118 :     if (data->temp[4] < data->iMinSAD[4]) {
119 :     data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
120 : chl 184
121 : Isibaar 3 }
122 :    
123 : chl 530 static void
124 :     CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
125 :     {
126 :     int32_t sad;
127 :     const uint8_t * Reference;
128 : chl 326
129 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
130 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
131 : Isibaar 3
132 : chl 530 switch ( ((x&1)<<1) + (y&1) )
133 :     {
134 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
135 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
136 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
137 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
138 :     }
139 : Isibaar 3
140 : chl 530 sad = lambda_vec16[data->iQuant] *
141 :     d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
142 :     sad += sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
143 : Isibaar 3
144 : chl 530 if (sad < *(data->iMinSAD)) {
145 :     *(data->iMinSAD) = sad;
146 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
147 :     *dir = Direction; }
148 : Isibaar 3 }
149 :    
150 : chl 530 static void
151 : Isibaar 539 CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
152 :     {
153 :     int32_t sad;
154 :    
155 :     if (( x > data->max_dx) || ( x < data->min_dx)
156 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
157 :    
158 :     sad = lambda_vec16[data->iQuant] *
159 :     d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
160 :    
161 :     sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
162 :     data->iEdgedWidth, 256*4096);
163 :    
164 :     if (sad < *(data->iMinSAD)) {
165 :     *(data->iMinSAD) = sad;
166 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
167 :     *dir = Direction; }
168 :     }
169 :    
170 :    
171 :     static void
172 : chl 530 CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
173 :     {
174 :     int32_t sad;
175 :     const int xb = data->currentMV[1].x;
176 :     const int yb = data->currentMV[1].y;
177 :     const uint8_t *ReferenceF, *ReferenceB;
178 : Isibaar 3
179 : chl 530 if (( xf > data->max_dx) || ( xf < data->min_dx)
180 :     || ( yf > data->max_dy) || (yf < data->min_dy)) return;
181 : Isibaar 3
182 : chl 530 switch ( ((xf&1)<<1) + (yf&1) ) {
183 :     case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break;
184 :     case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
185 :     case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break;
186 :     default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
187 :     }
188 : edgomez 195
189 : chl 530 switch ( ((xb&1)<<1) + (yb&1) ) {
190 :     case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break;
191 :     case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
192 :     case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break;
193 :     default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
194 :     }
195 : Isibaar 3
196 : chl 530 sad = lambda_vec16[data->iQuant] *
197 :     ( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) +
198 :     d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) );
199 : Isibaar 3
200 : chl 530 sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
201 : Isibaar 3
202 : chl 530 if (sad < *(data->iMinSAD)) {
203 :     *(data->iMinSAD) = sad;
204 :     data->currentMV->x = xf; data->currentMV->y = yf;
205 :     *dir = Direction; }
206 : Isibaar 3 }
207 :    
208 : chl 530 static void
209 :     CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
210 : Isibaar 3 {
211 : chl 530 int32_t sad;
212 :     int k;
213 :     const uint8_t *ReferenceF;
214 :     const uint8_t *ReferenceB;
215 :     VECTOR mvs, b_mvs;
216 : Isibaar 3
217 : chl 530 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
218 : Isibaar 3
219 : chl 530 sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
220 : Isibaar 3
221 : chl 530 for (k = 0; k < 4; k++) {
222 :     mvs.x = data->directmvF[k].x + x;
223 :     b_mvs.x = ((x == 0) ?
224 :     data->directmvB[k].x
225 :     : mvs.x - data->referencemv[k].x);
226 : Isibaar 3
227 : chl 530 mvs.y = data->directmvF[k].y + y;
228 :     b_mvs.y = ((y == 0) ?
229 :     data->directmvB[k].y
230 :     : mvs.y - data->referencemv[k].y);
231 :    
232 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
233 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
234 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
235 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
236 : Isibaar 3
237 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
238 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
239 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
240 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
241 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
242 :     }
243 : Isibaar 3
244 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
245 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
246 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
247 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
248 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
249 : chl 344 }
250 : chl 530
251 :     sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
252 :     ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
253 :     ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
254 :     data->iEdgedWidth);
255 :     if (sad > *(data->iMinSAD)) return;
256 : edgomez 78 }
257 : chl 530
258 :     if (sad < *(data->iMinSAD)) {
259 :     *(data->iMinSAD) = sad;
260 :     data->currentMV->x = x; data->currentMV->y = y;
261 :     *dir = Direction; }
262 : Isibaar 3 }
263 :    
264 : chl 530 static void
265 :     CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
266 : chl 96 {
267 : chl 530 int32_t sad;
268 :     const uint8_t *ReferenceF;
269 :     const uint8_t *ReferenceB;
270 :     VECTOR mvs, b_mvs;
271 : chl 96
272 : chl 530 if (( x > 31) || ( x < -31) || ( y > 31) || (y < -31)) return;
273 :    
274 :     sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
275 : edgomez 195
276 : chl 530 mvs.x = data->directmvF[0].x + x;
277 :     b_mvs.x = ((x == 0) ?
278 :     data->directmvB[0].x
279 :     : mvs.x - data->referencemv[0].x);
280 : edgomez 195
281 : chl 530 mvs.y = data->directmvF[0].y + y;
282 :     b_mvs.y = ((y == 0) ?
283 :     data->directmvB[0].y
284 :     : mvs.y - data->referencemv[0].y);
285 :    
286 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
287 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
288 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
289 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
290 : chl 96
291 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
292 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
293 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
294 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
295 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
296 :     }
297 : chl 96
298 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
299 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
300 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
301 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
302 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
303 :     }
304 :    
305 :     sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
306 : chl 96
307 : chl 530 if (sad < *(data->iMinSAD)) {
308 :     *(data->iMinSAD) = sad;
309 :     data->currentMV->x = x; data->currentMV->y = y;
310 :     *dir = Direction; }
311 :     }
312 : chl 96
313 : chl 530 static void
314 :     CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
315 :     {
316 :     int32_t sad;
317 :     const uint8_t * Reference;
318 : edgomez 195
319 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
320 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
321 : chl 96
322 : chl 530 switch ( ((x&1)<<1) + (y&1) )
323 :     {
324 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
325 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
326 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
327 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
328 :     }
329 : edgomez 195
330 : chl 530 sad = sad8(data->Cur, Reference, data->iEdgedWidth);
331 :     sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
332 : edgomez 195
333 : chl 530 if (sad < *(data->iMinSAD)) {
334 :     *(data->iMinSAD) = sad;
335 :     data->currentMV->x = x; data->currentMV->y = y;
336 :     *dir = Direction; }
337 : chl 96 }
338 :    
339 : Isibaar 539 /* CHECK_CANDIATE FUNCTIONS END */
340 : chl 96
341 : chl 530 /* MAINSEARCH FUNCTIONS START */
342 : edgomez 195
343 : chl 530 static void
344 :     AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
345 : chl 181 {
346 :    
347 :     /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
348 :    
349 : chl 530 int iDirection;
350 : edgomez 195
351 :     do {
352 : chl 181 iDirection = 0;
353 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
354 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
355 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
356 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
357 : chl 181
358 :     /* now we're doing diagonal checks near our candidate */
359 :    
360 : chl 530 if (iDirection) { //checking if anything found
361 : chl 181 bDirection = iDirection;
362 :     iDirection = 0;
363 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
364 :     if (bDirection & 3) { //our candidate is left or right
365 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
366 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
367 :     } else { // what remains here is up or down
368 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
369 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
370 : chl 181
371 : edgomez 195 if (iDirection) {
372 :     bDirection += iDirection;
373 : chl 530 x = data->currentMV->x; y = data->currentMV->y; }
374 :     } else { //about to quit, eh? not so fast....
375 : edgomez 195 switch (bDirection) {
376 : chl 181 case 2:
377 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
378 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
379 : chl 181 break;
380 :     case 1:
381 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
382 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
383 : chl 181 break;
384 : edgomez 195 case 2 + 4:
385 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
386 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
387 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
388 : chl 181 break;
389 :     case 4:
390 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
391 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
392 : chl 181 break;
393 :     case 8:
394 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
395 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
396 : chl 181 break;
397 : edgomez 195 case 1 + 4:
398 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
399 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
400 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
401 : chl 181 break;
402 : edgomez 195 case 2 + 8:
403 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
404 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
405 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
406 : chl 181 break;
407 : edgomez 195 case 1 + 8:
408 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
409 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
410 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
411 : chl 181 break;
412 : edgomez 195 default: //1+2+4+8 == we didn't find anything at all
413 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
414 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
415 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
416 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
417 : chl 181 break;
418 :     }
419 : chl 530 if (!iDirection) break; //ok, the end. really
420 :     bDirection = iDirection;
421 :     x = data->currentMV->x; y = data->currentMV->y;
422 : chl 181 }
423 :     }
424 : edgomez 195 while (1); //forever
425 : chl 181 }
426 :    
427 : chl 530 static void
428 :     SquareSearch(int x, int y, const SearchData * const data, int bDirection)
429 : chl 326 {
430 : chl 530 int iDirection;
431 : chl 326
432 : chl 530 do {
433 :     iDirection = 0;
434 :     if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
435 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
436 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
437 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
438 :     if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
439 :     if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
440 :     if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
441 :     if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
442 : chl 326
443 : chl 530 bDirection = iDirection;
444 :     x = data->currentMV->x; y = data->currentMV->y;
445 :     } while (iDirection);
446 : chl 346 }
447 :    
448 : chl 530 static void
449 :     DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
450 : chl 346 {
451 :    
452 : chl 181 /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
453 :    
454 : chl 530 int iDirection;
455 : edgomez 195
456 :     do {
457 : chl 181 iDirection = 0;
458 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
459 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
460 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
461 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
462 : chl 181
463 :     /* now we're doing diagonal checks near our candidate */
464 :    
465 : chl 530 if (iDirection) { //checking if anything found
466 : chl 181 bDirection = iDirection;
467 :     iDirection = 0;
468 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
469 :     if (bDirection & 3) { //our candidate is left or right
470 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
471 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
472 :     } else { // what remains here is up or down
473 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
474 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
475 : chl 181
476 : chl 530 bDirection += iDirection;
477 :     x = data->currentMV->x; y = data->currentMV->y;
478 : chl 181 }
479 :     }
480 : chl 530 while (iDirection);
481 : chl 181 }
482 :    
483 : chl 530 /* MAINSEARCH FUNCTIONS END */
484 : chl 181
485 : chl 530 /* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
486 :    
487 :     static void
488 :     HalfpelRefine(const SearchData * const data)
489 : chl 96 {
490 : chl 530 /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */
491 : edgomez 195
492 : chl 530 VECTOR backupMV = *(data->currentMV);
493 :     int iDirection; //not needed
494 : chl 96
495 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
496 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
497 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
498 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
499 : edgomez 195
500 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
501 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
502 :    
503 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
504 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
505 : chl 96 }
506 :    
507 : chl 530 static __inline int
508 :     SkipDecisionP(const IMAGE * current, const IMAGE * reference,
509 :     const int x, const int y,
510 :     const uint32_t iEdgedWidth, const uint32_t iQuant)
511 : chl 96
512 : Isibaar 3 {
513 : chl 530 /* keep repeating checks for all b-frames before this P frame,
514 :     to make sure that SKIP is possible (todo)
515 :     how: if skip is not possible set sad00 to a very high value */
516 : Isibaar 3
517 : chl 530 uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8,
518 :     reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
519 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
520 : h 545 sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8,
521 :     reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
522 : chl 530 if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
523 : edgomez 195
524 : chl 530 return 1;
525 :     }
526 : edgomez 195
527 : chl 530 static __inline void
528 :     SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
529 :     {
530 :     pMB->mode = MODE_NOT_CODED;
531 : Isibaar 539 pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
532 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
533 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
534 : Isibaar 3 }
535 :    
536 : chl 530 bool
537 :     MotionEstimation(MBParam * const pParam,
538 :     FRAMEINFO * const current,
539 :     FRAMEINFO * const reference,
540 :     const IMAGE * const pRefH,
541 :     const IMAGE * const pRefV,
542 :     const IMAGE * const pRefHV,
543 :     const uint32_t iLimit)
544 : Isibaar 3 {
545 : chl 530 MACROBLOCK *const pMBs = current->mbs;
546 :     const IMAGE *const pCurrent = &current->image;
547 :     const IMAGE *const pRef = &reference->image;
548 : Isibaar 3
549 : chl 530 const VECTOR zeroMV = { 0, 0 };
550 : Isibaar 3
551 : chl 530 uint32_t x, y;
552 :     uint32_t iIntra = 0;
553 : syskin 574 int32_t InterBias, quant = current->quant;
554 : edgomez 195
555 : Isibaar 539 // some pre-initialized thingies for SearchP
556 :     int32_t temp[5];
557 :     VECTOR currentMV[5];
558 :     int32_t iMinSAD[5];
559 :     SearchData Data;
560 :     Data.iEdgedWidth = pParam->edged_width;
561 :     Data.currentMV = currentMV;
562 :     Data.iMinSAD = iMinSAD;
563 :     Data.temp = temp;
564 :     Data.iFcode = current->fcode;
565 :    
566 : chl 530 if (sadInit) (*sadInit) ();
567 : edgomez 195
568 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
569 :     for (x = 0; x < pParam->mb_width; x++) {
570 : Isibaar 3
571 : chl 530 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
572 :     int32_t sad00 = pMB->sad16
573 :     = sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
574 :     pRef->y + (x + y * pParam->edged_width) * 16,
575 :     pParam->edged_width, pMB->sad8 );
576 : edgomez 195
577 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
578 :     pMB->dquant = NO_CHANGE;
579 :     pMB->quant = current->quant; }
580 : syskin 574 else
581 :     if (pMB->dquant != NO_CHANGE) {
582 :     quant += DQtab[pMB->dquant];
583 :     if (quant > 31) quant = 31;
584 :     else if (quant < 1) quant = 1;
585 :     pMB->quant = quant;
586 :     }
587 : chl 181
588 : chl 530 //initial skip decision
589 : Isibaar 3
590 : chl 530 if ((pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant)
591 :     && (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) ) {
592 :     if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) {
593 :     SkipMacroblockP(pMB, sad00);
594 :     continue;
595 :     }
596 :     } else sad00 = 256*4096; // skip not allowed - for final skip decision
597 : Isibaar 3
598 : chl 530 SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
599 :     y, current->motion_flags, pMB->quant,
600 : Isibaar 539 &Data, pParam, pMBs, reference->mbs,
601 : chl 530 current->global_flags & XVID_INTER4V, pMB);
602 : edgomez 195
603 : chl 530 /* final skip decision, a.k.a. "the vector you found, really that good?" */
604 :     if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
605 :     if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH)
606 :     { SkipMacroblockP(pMB, sad00); continue; }
607 :    
608 :     /* finally, intra decision */
609 : Isibaar 3
610 : chl 530 InterBias = MV16_INTER_BIAS;
611 :     if (pMB->quant > 8) InterBias += 50 * (pMB->quant - 8); // to make high quants work
612 :     if (y != 0)
613 :     if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 50;
614 :     if (x != 0)
615 :     if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 50;
616 : Isibaar 3
617 : chl 530 if (InterBias < pMB->sad16) {
618 :     const int32_t deviation =
619 :     dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
620 :     pParam->edged_width);
621 : Isibaar 3
622 : chl 530 if (deviation < (pMB->sad16 - InterBias)) {
623 :     if (++iIntra >= iLimit) return 1;
624 :     pMB->mode = MODE_INTRA;
625 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
626 :     pMB->mvs[3] = zeroMV;
627 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
628 :     pMB->sad8[3] = 0;
629 :     }
630 :     }
631 :     }
632 : Isibaar 3 }
633 : chl 530 return 0;
634 :     }
635 : Isibaar 3
636 : edgomez 195
637 : chl 530 #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16)
638 : Isibaar 3
639 : chl 530 static __inline int
640 :     make_mask(const VECTOR * const pmv, const int i)
641 :     {
642 : Isibaar 539 int mask = 255, j;
643 : chl 530 for (j = 0; j < i; j++) {
644 :     if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
645 :     if (pmv[i].x == pmv[j].x) {
646 :     if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
647 :     if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
648 :     } else
649 :     if (pmv[i].y == pmv[j].y) {
650 :     if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
651 :     if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
652 :     }
653 : Isibaar 3 }
654 : chl 530 return mask;
655 :     }
656 : edgomez 195
657 : chl 530 static __inline void
658 :     PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount,
659 :     const int iHcount, const MACROBLOCK * const prevMB)
660 :     {
661 : edgomez 195
662 : chl 530 //this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
663 : edgomez 195
664 : chl 530 if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour
665 :     pmv[5].x = EVEN(pmv[3].x);
666 : Isibaar 539 pmv[5].y = EVEN(pmv[3].y);
667 :     } else pmv[5].x = pmv[5].y = 0;
668 : chl 169
669 : chl 530 if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
670 :     else pmv[3].x = pmv[3].y = 0;
671 : Isibaar 3
672 : chl 530 if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
673 :     else pmv[4].x = pmv[4].y = 0;
674 : chl 169
675 : chl 530 // [1] median prediction
676 :     pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
677 : chl 169
678 : chl 530 pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
679 : chl 169
680 : chl 530 pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
681 :     pmv[2].y = EVEN(prevMB->mvs[0].y);
682 : chl 169
683 : chl 530 if ((x != iWcount-1) && (y != iHcount-1)) {
684 :     pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
685 : Isibaar 539 pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
686 :     } else pmv[6].x = pmv[6].y = 0;
687 : chl 530 }
688 : chl 169
689 : chl 530 static void
690 :     SearchP(const uint8_t * const pRef,
691 :     const uint8_t * const pRefH,
692 :     const uint8_t * const pRefV,
693 :     const uint8_t * const pRefHV,
694 :     const IMAGE * const pCur,
695 :     const int x,
696 :     const int y,
697 :     const uint32_t MotionFlags,
698 :     const uint32_t iQuant,
699 : Isibaar 539 SearchData * const Data,
700 : chl 530 const MBParam * const pParam,
701 :     const MACROBLOCK * const pMBs,
702 :     const MACROBLOCK * const prevMBs,
703 :     int inter4v,
704 :     MACROBLOCK * const pMB)
705 :     {
706 : chl 169
707 : chl 530 int i, iDirection = 255, mask, threshA;
708 : Isibaar 539 VECTOR pmv[7];
709 : Isibaar 3
710 : Isibaar 539 get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
711 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
712 :     pParam->width, pParam->height, Data->iFcode);
713 : Isibaar 3
714 : Isibaar 539 Data->predMV = pmv[0];
715 :     Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
716 :     Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16;
717 :     Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
718 :     Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
719 :     Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
720 : Isibaar 3
721 : Isibaar 539 Data->iQuant = iQuant;
722 : chl 169
723 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
724 : Isibaar 539 Data->min_dx = EVEN(Data->min_dx);
725 :     Data->max_dx = EVEN(Data->max_dx);
726 :     Data->min_dy = EVEN(Data->min_dy);
727 :     Data->max_dy = EVEN(Data->max_dy); }
728 : edgomez 568
729 :     if (pMB->dquant != NO_CHANGE) inter4v = 0;
730 : edgomez 195
731 : Isibaar 539 if (inter4v) CheckCandidate = CheckCandidate16;
732 :     else CheckCandidate = CheckCandidate16no4v;
733 : chl 169
734 : Isibaar 539 for(i = 0; i < 5; i++) Data->currentMV[i].x = Data->currentMV[i].y = 0;
735 :    
736 :     i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
737 :     Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i;
738 :     Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i;
739 :     Data->iMinSAD[2] = pMB->sad8[1];
740 :     Data->iMinSAD[3] = pMB->sad8[2];
741 :     Data->iMinSAD[4] = pMB->sad8[3];
742 :    
743 : chl 530 if ((x == 0) && (y == 0)) threshA = 512;
744 :     else {
745 : edgomez 568 threshA = Data->temp[0]; // that's when we keep this SAD atm
746 : chl 530 if (threshA < 512) threshA = 512;
747 :     if (threshA > 1024) threshA = 1024; }
748 : Isibaar 3
749 : chl 530 PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
750 :     prevMBs + x + y * pParam->mb_width);
751 : edgomez 195
752 : edgomez 568 if (inter4v) CheckCandidate = CheckCandidate16;
753 :     else CheckCandidate = CheckCandidate16no4v;
754 :    
755 :    
756 : chl 530 /* main loop. checking all predictions */
757 :    
758 :     for (i = 1; i < 7; i++) {
759 :     if (!(mask = make_mask(pmv, i)) ) continue;
760 : edgomez 568 (*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
761 : Isibaar 539 if (Data->iMinSAD[0] <= threshA) break;
762 : chl 530 }
763 :    
764 : Isibaar 539 if ((Data->iMinSAD[0] <= threshA) ||
765 :     (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
766 :     (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
767 : chl 530 inter4v = 0;
768 : Isibaar 539 } else {
769 : Isibaar 3
770 : Isibaar 539 MainSearchFunc * MainSearchPtr;
771 :     if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
772 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
773 :     else MainSearchPtr = DiamondSearch;
774 : chl 181
775 : Isibaar 539 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
776 : Isibaar 3
777 : chl 530 /* extended search, diamond starting in 0,0 and in prediction.
778 :     note that this search is/might be done in halfpel positions,
779 :     which makes it more different than the diamond above */
780 : chl 259
781 : Isibaar 539 if (MotionFlags & PMV_EXTSEARCH16) {
782 :     int32_t bSAD;
783 :     VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
784 :     if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
785 :     startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
786 :     if (!(MVequal(startMV, backupMV))) {
787 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
788 : Isibaar 3
789 : Isibaar 539 CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
790 :     (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
791 :     if (bSAD < Data->iMinSAD[0]) {
792 :     Data->currentMV[0] = backupMV;
793 :     Data->iMinSAD[0] = bSAD; }
794 :     }
795 : Isibaar 3
796 : Isibaar 539 backupMV = Data->currentMV[0];
797 :     if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1;
798 :     else startMV.x = startMV.y = 0;
799 :     if (!(MVequal(startMV, backupMV))) {
800 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
801 : edgomez 195
802 : Isibaar 539 CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
803 :     (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
804 :     if (bSAD < Data->iMinSAD[0]) {
805 :     Data->currentMV[0] = backupMV;
806 :     Data->iMinSAD[0] = bSAD; }
807 :     }
808 : Isibaar 3 }
809 :     }
810 :    
811 : Isibaar 539 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
812 : Isibaar 3
813 : Isibaar 539 if (inter4v) {
814 :     SearchData Data8;
815 :     Data8.iFcode = Data->iFcode;
816 :     Data8.iQuant = Data->iQuant;
817 :     Data8.iEdgedWidth = Data->iEdgedWidth;
818 :     Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
819 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
820 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
821 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
822 :     }
823 : Isibaar 3
824 : chl 530 if (!(inter4v) ||
825 : Isibaar 539 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
826 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
827 : chl 530 // INTER MODE
828 :     pMB->mode = MODE_INTER;
829 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
830 :     = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
831 : Isibaar 3
832 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
833 : Isibaar 539 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
834 : Isibaar 3
835 : Isibaar 539 pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
836 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
837 : edgomez 195 } else {
838 : chl 530 // INTER4V MODE; all other things are already set in Search8
839 :     pMB->mode = MODE_INTER4V;
840 : Isibaar 539 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
841 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
842 : edgomez 78 }
843 : chl 530
844 : Isibaar 3 }
845 :    
846 : chl 530 static void
847 :     Search8(const SearchData * const OldData,
848 :     const int x, const int y,
849 :     const uint32_t MotionFlags,
850 :     const MBParam * const pParam,
851 :     MACROBLOCK * const pMB,
852 :     const MACROBLOCK * const pMBs,
853 : Isibaar 539 const int block,
854 :     SearchData * const Data)
855 : chl 345 {
856 : Isibaar 539 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
857 :     Data->iMinSAD = OldData->iMinSAD + 1 + block;
858 :     Data->currentMV = OldData->currentMV + 1 + block;
859 : chl 345
860 : chl 530 if (block != 0)
861 : Isibaar 539 *(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
862 :     d_mv_bits( Data->currentMV->x - Data->predMV.x,
863 :     Data->currentMV->y - Data->predMV.y,
864 :     Data->iFcode);
865 : chl 345
866 : chl 530 if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
867 : chl 345
868 : Isibaar 539 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
869 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
870 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
871 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
872 : chl 345
873 : Isibaar 539 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
874 : chl 530
875 : Isibaar 539 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
876 : chl 530 pParam->width, pParam->height, OldData->iFcode);
877 : chl 345
878 : chl 530 CheckCandidate = CheckCandidate8;
879 : chl 345
880 : chl 530 if (MotionFlags & PMV_EXTSEARCH8) {
881 : chl 345
882 : chl 530 MainSearchFunc *MainSearchPtr;
883 :     if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
884 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
885 :     else MainSearchPtr = DiamondSearch;
886 : chl 345
887 : Isibaar 539 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); }
888 : chl 345
889 : Isibaar 539 if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(Data);
890 : chl 345 }
891 : chl 530
892 : Isibaar 539 pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
893 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
894 :     pMB->mvs[block] = *(Data->currentMV);
895 :     pMB->sad8[block] = 4 * (*Data->iMinSAD);
896 :    
897 : chl 345 }
898 :    
899 : chl 530 /* B-frames code starts here */
900 : chl 345
901 : chl 530 static __inline VECTOR
902 :     ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
903 : Isibaar 3 {
904 : chl 530 /* the stupidiest function ever */
905 :     if (mode == MODE_FORWARD) return pMB->mvs[0];
906 :     else return pMB->b_mvs[0];
907 : Isibaar 3 }
908 :    
909 : chl 530 static void __inline
910 :     PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
911 :     const uint32_t iWcount,
912 :     const MACROBLOCK * const pMB,
913 :     const uint32_t mode_curr)
914 : Isibaar 3 {
915 :    
916 : chl 530 // [0] is prediction
917 :     pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
918 : Isibaar 3
919 : chl 530 pmv[1].x = pmv[1].y = 0; // [1] is zero
920 : Isibaar 3
921 : chl 530 pmv[2] = ChoosePred(pMB, mode_curr);
922 :     pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
923 : edgomez 195
924 : chl 530 if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
925 :     pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
926 : Isibaar 539 pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
927 :     } else pmv[3].x = pmv[3].y = 0;
928 : Isibaar 3
929 : chl 530 if (y != 0) {
930 :     pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
931 :     pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
932 :     } else pmv[4].x = pmv[4].y = 0;
933 : Isibaar 3
934 : chl 530 if (x != 0) {
935 :     pmv[5] = ChoosePred(pMB-1, mode_curr);
936 :     pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
937 :     } else pmv[5].x = pmv[5].y = 0;
938 : Isibaar 3
939 : chl 530 if ((x != 0)&&(y != 0)) {
940 :     pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
941 :     pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
942 :     } else pmv[6].x = pmv[6].y = 0;
943 : edgomez 195
944 : chl 530 // more?
945 :     }
946 : chl 181
947 : edgomez 170
948 : chl 530 /* search backward or forward, for b-frames */
949 :     static void
950 :     SearchBF( const uint8_t * const pRef,
951 :     const uint8_t * const pRefH,
952 :     const uint8_t * const pRefV,
953 :     const uint8_t * const pRefHV,
954 :     const IMAGE * const pCur,
955 :     const int x, const int y,
956 :     const uint32_t MotionFlags,
957 :     const uint32_t iFcode,
958 :     const MBParam * const pParam,
959 :     MACROBLOCK * const pMB,
960 :     const VECTOR * const predMV,
961 :     int32_t * const best_sad,
962 : h 545 const int32_t mode_current,
963 :     SearchData * const Data)
964 : chl 530 {
965 : Isibaar 3
966 : chl 530 const int32_t iEdgedWidth = pParam->edged_width;
967 :    
968 :     int i, iDirection, mask;
969 : h 545 VECTOR pmv[7];
970 : chl 530 MainSearchFunc *MainSearchPtr;
971 : h 545 *Data->iMinSAD = MV_MAX_ERROR;
972 :     Data->iFcode = iFcode;
973 : Isibaar 3
974 : h 545 Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
975 :     Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
976 :     Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
977 :     Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
978 : Isibaar 3
979 : h 545 Data->predMV = *predMV;
980 : edgomez 195
981 : h 545 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
982 : chl 530 pParam->width, pParam->height, iFcode);
983 : Isibaar 3
984 : h 545 pmv[0] = Data->predMV;
985 :     PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
986 : edgomez 195
987 : h 545 Data->currentMV->x = Data->currentMV->y = 0;
988 : Isibaar 3
989 : chl 530 CheckCandidate = CheckCandidate16no4v;
990 : chl 181
991 : chl 530 // main loop. checking all predictions
992 :     for (i = 0; i < 8; i++) {
993 :     if (!(mask = make_mask(pmv, i)) ) continue;
994 : h 545 CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
995 : edgomez 78 }
996 : Isibaar 3
997 : chl 530 if (MotionFlags & PMV_USESQUARES16)
998 :     MainSearchPtr = SquareSearch;
999 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1000 :     MainSearchPtr = AdvDiamondSearch;
1001 :     else MainSearchPtr = DiamondSearch;
1002 : chl 96
1003 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1004 : chl 169
1005 : h 545 HalfpelRefine(Data);
1006 : chl 169
1007 : chl 530 // three bits are needed to code backward mode. four for forward
1008 :     // we treat the bits just like they were vector's
1009 : h 545 if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant];
1010 :     else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant];
1011 : chl 169
1012 :    
1013 : h 545 if (*Data->iMinSAD < *best_sad) {
1014 :     *best_sad = *Data->iMinSAD;
1015 : chl 530 pMB->mode = mode_current;
1016 : h 545 pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1017 :     pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1018 :     if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1019 :     else pMB->b_mvs[0] = *Data->currentMV;
1020 : chl 169 }
1021 : chl 530
1022 :     }
1023 : chl 169
1024 : chl 530 static int32_t
1025 : Isibaar 539 SearchDirect(const IMAGE * const f_Ref,
1026 : chl 530 const uint8_t * const f_RefH,
1027 :     const uint8_t * const f_RefV,
1028 :     const uint8_t * const f_RefHV,
1029 : Isibaar 539 const IMAGE * const b_Ref,
1030 : chl 530 const uint8_t * const b_RefH,
1031 :     const uint8_t * const b_RefV,
1032 :     const uint8_t * const b_RefHV,
1033 :     const IMAGE * const pCur,
1034 :     const int x, const int y,
1035 :     const uint32_t MotionFlags,
1036 :     const int32_t TRB, const int32_t TRD,
1037 :     const MBParam * const pParam,
1038 :     MACROBLOCK * const pMB,
1039 :     const MACROBLOCK * const b_mb,
1040 : h 545 int32_t * const best_sad,
1041 :     SearchData * const Data)
1042 : edgomez 195
1043 : chl 530 {
1044 : h 545 int32_t skip_sad;
1045 : chl 530 int k;
1046 : h 545
1047 : chl 530 MainSearchFunc *MainSearchPtr;
1048 : Isibaar 3
1049 : h 545 *Data->iMinSAD = 256*4096;
1050 :     Data->referencemv = b_mb->mvs;
1051 : chl 140
1052 : h 545 Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1053 :     Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16;
1054 :     Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16;
1055 :     Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16;
1056 :     Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1057 :     Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16;
1058 :     Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16;
1059 :     Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16;
1060 : chl 140
1061 : h 545 Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1062 :     Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1063 :     Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1064 :     Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1065 : Isibaar 3
1066 : chl 530 for (k = 0; k < 4; k++) {
1067 : h 545 pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1068 :     pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1069 :     pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1070 :     pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1071 : Isibaar 3
1072 : h 545 if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx )
1073 :     || ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) {
1074 : Isibaar 3
1075 : Isibaar 539 *best_sad = 256*4096; // in that case, we won't use direct mode
1076 :     pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1077 :     pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1078 :     return 0;
1079 :     }
1080 :     if (b_mb->mode != MODE_INTER4V) {
1081 :     pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1082 :     pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1083 : h 545 Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1084 :     Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1085 : Isibaar 539 break;
1086 :     }
1087 :     }
1088 : Isibaar 3
1089 : Isibaar 539 if (b_mb->mode == MODE_INTER4V)
1090 :     CheckCandidate = CheckCandidateDirect;
1091 :     else CheckCandidate = CheckCandidateDirectno4v;
1092 : edgomez 195
1093 : h 545 (*CheckCandidate)(0, 0, 255, &k, Data);
1094 : Isibaar 3
1095 : chl 530 // skip decision
1096 : h 545 if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) {
1097 : Isibaar 539 //checking chroma. everything copied from MC
1098 :     //this is not full chroma compensation, only it's fullpel approximation. should work though
1099 :     int sum, dx, dy, b_dx, b_dy;
1100 : Isibaar 3
1101 : Isibaar 539 sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1102 :     dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1103 :    
1104 :     sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1105 :     dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1106 :    
1107 :     sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
1108 :     b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1109 :    
1110 :     sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
1111 :     b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1112 :    
1113 : h 545 sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2),
1114 :     f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1115 :     b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1116 :     Data->iEdgedWidth/2);
1117 :     sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2),
1118 :     f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1119 :     b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1120 :     Data->iEdgedWidth/2);
1121 : Isibaar 539
1122 : h 545 if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) {
1123 : Isibaar 539 pMB->mode = MODE_DIRECT_NONE_MV;
1124 : h 545 return *Data->iMinSAD;
1125 : Isibaar 539 }
1126 :     }
1127 :    
1128 : h 545 skip_sad = *Data->iMinSAD;
1129 : edgomez 195
1130 : chl 530 // DIRECT MODE DELTA VECTOR SEARCH.
1131 :     // This has to be made more effective, but at the moment I'm happy it's running at all
1132 : Isibaar 3
1133 : chl 530 if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1134 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1135 :     else MainSearchPtr = DiamondSearch;
1136 : edgomez 195
1137 : h 545 (*MainSearchPtr)(0, 0, Data, 255);
1138 : Isibaar 3
1139 : h 545 HalfpelRefine(Data);
1140 : Isibaar 3
1141 : h 545 *Data->iMinSAD += 1 * lambda_vec16[Data->iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's
1142 :     *best_sad = *Data->iMinSAD;
1143 : Isibaar 3
1144 : chl 530 if (b_mb->mode == MODE_INTER4V)
1145 :     pMB->mode = MODE_DIRECT;
1146 :     else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1147 : edgomez 195
1148 : h 545 pMB->pmvs[3] = *Data->currentMV;
1149 : edgomez 195
1150 : chl 530 for (k = 0; k < 4; k++) {
1151 : h 545 pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1152 :     pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
1153 :     ? Data->directmvB[k].x
1154 :     : pMB->mvs[k].x - Data->referencemv[k].x);
1155 :     pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1156 :     pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1157 :     ? Data->directmvB[k].y
1158 :     : pMB->mvs[k].y - Data->referencemv[k].y);
1159 : chl 530 if (b_mb->mode != MODE_INTER4V) {
1160 :     pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1161 :     pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1162 :     break;
1163 :     }
1164 :     }
1165 : h 545 return skip_sad;
1166 : Isibaar 3 }
1167 : chl 96
1168 : h 545
1169 : chl 530 static __inline void
1170 :     SearchInterpolate(const uint8_t * const f_Ref,
1171 :     const uint8_t * const f_RefH,
1172 :     const uint8_t * const f_RefV,
1173 :     const uint8_t * const f_RefHV,
1174 :     const uint8_t * const b_Ref,
1175 :     const uint8_t * const b_RefH,
1176 :     const uint8_t * const b_RefV,
1177 :     const uint8_t * const b_RefHV,
1178 :     const IMAGE * const pCur,
1179 :     const int x, const int y,
1180 :     const uint32_t fcode,
1181 :     const uint32_t bcode,
1182 :     const uint32_t MotionFlags,
1183 :     const MBParam * const pParam,
1184 :     const VECTOR * const f_predMV,
1185 :     const VECTOR * const b_predMV,
1186 :     MACROBLOCK * const pMB,
1187 : h 545 int32_t * const best_sad,
1188 :     SearchData * const fData)
1189 : chl 530
1190 : chl 96 {
1191 :    
1192 : edgomez 195 const int32_t iEdgedWidth = pParam->edged_width;
1193 : chl 96
1194 : chl 530 int iDirection, i, j;
1195 : h 545 SearchData bData;
1196 : chl 96
1197 : h 545 bData.iMinSAD = fData->iMinSAD;
1198 :     *bData.iMinSAD = 4096*256;
1199 :     bData.Cur = fData->Cur;
1200 :     fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
1201 :     bData.currentMV = fData->currentMV + 1;
1202 :     bData.iQuant = fData->iQuant;
1203 :     fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1204 : edgomez 195
1205 : h 545 bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
1206 :     bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
1207 :     bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
1208 :     bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
1209 :     bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
1210 :     bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
1211 :     bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
1212 :     bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
1213 : chl 96
1214 : h 545 bData.bpredMV = fData->predMV = *f_predMV;
1215 :     fData->bpredMV = bData.predMV = *b_predMV;
1216 : edgomez 195
1217 : h 545 fData->currentMV[0] = pMB->mvs[0];
1218 :     fData->currentMV[1] = pMB->b_mvs[0];
1219 :     get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode);
1220 : chl 530 get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode);
1221 : chl 96
1222 : h 545 if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1223 :     if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
1224 :     if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
1225 :     if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1226 : Isibaar 539
1227 : h 545 if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1228 :     if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
1229 :     if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
1230 :     if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1231 : Isibaar 539
1232 : h 545 CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1233 : chl 96
1234 : chl 530 //diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1235 : chl 96
1236 : chl 530 do {
1237 :     iDirection = 255;
1238 :     // forward MV moves
1239 : h 545 i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1240 : chl 96
1241 : h 545 CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1242 :     CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1243 :     CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1244 :     CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1245 : chl 96
1246 : chl 530 // backward MV moves
1247 : h 545 i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1248 :     fData->currentMV[2] = fData->currentMV[0];
1249 : chl 326
1250 : Isibaar 539 CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1251 :     CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1252 :     CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1253 :     CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1254 : chl 96
1255 : chl 530 } while (!(iDirection));
1256 : edgomez 195
1257 : chl 530 // two bits are needed to code interpolate mode. we treat the bits just like they were vector's
1258 : h 545 *fData->iMinSAD += 2 * lambda_vec16[fData->iQuant];
1259 :     if (*fData->iMinSAD < *best_sad) {
1260 :     *best_sad = *fData->iMinSAD;
1261 :     pMB->mvs[0] = fData->currentMV[0];
1262 :     pMB->b_mvs[0] = fData->currentMV[1];
1263 : chl 530 pMB->mode = MODE_INTERPOLATE;
1264 : chl 96
1265 : chl 530 pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1266 :     pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1267 :     pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1268 :     pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1269 :     }
1270 :     }
1271 : chl 96
1272 : h 545
1273 : chl 530 void
1274 :     MotionEstimationBVOP(MBParam * const pParam,
1275 :     FRAMEINFO * const frame,
1276 :     const int32_t time_bp,
1277 :     const int32_t time_pp,
1278 :     // forward (past) reference
1279 :     const MACROBLOCK * const f_mbs,
1280 :     const IMAGE * const f_ref,
1281 :     const IMAGE * const f_refH,
1282 :     const IMAGE * const f_refV,
1283 :     const IMAGE * const f_refHV,
1284 :     // backward (future) reference
1285 :     const MACROBLOCK * const b_mbs,
1286 :     const IMAGE * const b_ref,
1287 :     const IMAGE * const b_refH,
1288 :     const IMAGE * const b_refV,
1289 :     const IMAGE * const b_refHV)
1290 :     {
1291 :     uint32_t i, j;
1292 :     int32_t best_sad, skip_sad;
1293 :     int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1294 :     static const VECTOR zeroMV={0,0};
1295 : chl 96
1296 : chl 530 VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1297 : chl 96
1298 : chl 530 const int32_t TRB = time_pp - time_bp;
1299 :     const int32_t TRD = time_pp;
1300 : chl 96
1301 : h 545 // some pre-inintialized data for the rest of the search
1302 :    
1303 :     SearchData Data;
1304 :     int32_t iMinSAD;
1305 :     VECTOR currentMV[3];
1306 :     Data.iEdgedWidth = pParam->edged_width;
1307 :     Data.currentMV = currentMV;
1308 :     Data.iMinSAD = &iMinSAD;
1309 :     Data.iQuant = frame->quant;
1310 :    
1311 : chl 530 // note: i==horizontal, j==vertical
1312 : chl 96
1313 : chl 530 for (j = 0; j < pParam->mb_height; j++) {
1314 : chl 96
1315 : chl 530 f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1316 : edgomez 195
1317 : chl 530 for (i = 0; i < pParam->mb_width; i++) {
1318 :     MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1319 :     const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1320 : edgomez 195
1321 : chl 530 /* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */
1322 :     if (b_mb->mode == MODE_NOT_CODED) {
1323 :     pMB->mode = MODE_NOT_CODED;
1324 :     continue;
1325 : chl 96 }
1326 :    
1327 : h 545 Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1328 : chl 530 /* direct search comes first, because it (1) checks for SKIP-mode
1329 :     and (2) sets very good predictions for forward and backward search */
1330 : chl 96
1331 : Isibaar 539 skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1332 :     b_ref, b_refH->y, b_refV->y, b_refHV->y,
1333 : chl 530 &frame->image,
1334 :     i, j,
1335 :     frame->motion_flags,
1336 :     TRB, TRD,
1337 :     pParam,
1338 :     pMB, b_mb,
1339 : h 545 &best_sad,
1340 :     &Data);
1341 : chl 96
1342 : Isibaar 539 if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1343 : chl 96
1344 : chl 530 // best_sad = 256*4096; //uncomment to disable Directsearch.
1345 :     // To disable any other mode, just comment the function call
1346 : chl 96
1347 : chl 530 // forward search
1348 :     SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1349 :     &frame->image, i, j,
1350 :     frame->motion_flags,
1351 : h 545 frame->fcode, pParam,
1352 : chl 530 pMB, &f_predMV, &best_sad,
1353 : h 545 MODE_FORWARD, &Data);
1354 : chl 96
1355 : chl 530 // backward search
1356 :     SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1357 :     &frame->image, i, j,
1358 :     frame->motion_flags,
1359 : h 545 frame->bcode, pParam,
1360 : chl 530 pMB, &b_predMV, &best_sad,
1361 : h 545 MODE_BACKWARD, &Data);
1362 : chl 96
1363 : chl 530 // interpolate search comes last, because it uses data from forward and backward as prediction
1364 : chl 96
1365 : chl 530 SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1366 :     b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1367 :     &frame->image,
1368 :     i, j,
1369 :     frame->fcode, frame->bcode,
1370 :     frame->motion_flags,
1371 : h 545 pParam,
1372 : chl 530 &f_predMV, &b_predMV,
1373 : h 545 pMB, &best_sad,
1374 :     &Data);
1375 : chl 96
1376 : chl 530 switch (pMB->mode) {
1377 :     case MODE_FORWARD:
1378 :     f_count++;
1379 :     f_predMV = pMB->mvs[0];
1380 :     break;
1381 :     case MODE_BACKWARD:
1382 :     b_count++;
1383 :     b_predMV = pMB->b_mvs[0];
1384 :     break;
1385 :     case MODE_INTERPOLATE:
1386 :     i_count++;
1387 :     f_predMV = pMB->mvs[0];
1388 :     b_predMV = pMB->b_mvs[0];
1389 :     break;
1390 :     case MODE_DIRECT:
1391 :     case MODE_DIRECT_NO4V:
1392 :     d_count++;
1393 :     break;
1394 :     default:
1395 :     break;
1396 : chl 96 }
1397 :     }
1398 :     }
1399 :    
1400 : chl 530 // fprintf(debug,"B-Stat: F: %04d B: %04d I: %04d D: %04d, N: %04d\n",
1401 :     // f_count,b_count,i_count,d_count,n_count);
1402 : chl 96
1403 :     }
1404 :    
1405 : chl 530 /* Hinted ME starts here */
1406 : chl 96
1407 : chl 530 static __inline void
1408 :     Search8hinted( const SearchData * const OldData,
1409 :     const int x, const int y,
1410 :     const uint32_t MotionFlags,
1411 :     const MBParam * const pParam,
1412 :     MACROBLOCK * const pMB,
1413 :     const MACROBLOCK * const pMBs,
1414 :     const int block)
1415 : chl 96 {
1416 : chl 530 SearchData Data;
1417 :     MainSearchFunc *MainSearchPtr;
1418 : chl 141
1419 : chl 530 Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1420 :     Data.iMinSAD = OldData->iMinSAD + 1 + block;
1421 :     Data.currentMV = OldData->currentMV+1+block;
1422 :     Data.iFcode = OldData->iFcode;
1423 :     Data.iQuant = OldData->iQuant;
1424 : chl 96
1425 : chl 530 Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1426 :     Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1427 :     Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1428 :     Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1429 :     Data.iEdgedWidth = pParam->edged_width;
1430 :     Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1431 : chl 96
1432 : chl 530 CheckCandidate = CheckCandidate8;
1433 : edgomez 195
1434 : chl 530 if (block != 0)
1435 :     *(Data.iMinSAD) += lambda_vec8[Data.iQuant] *
1436 :     d_mv_bits( Data.currentMV->x - Data.predMV.x,
1437 :     Data.currentMV->y - Data.predMV.y,
1438 :     Data.iFcode);
1439 : edgomez 195
1440 :    
1441 : chl 530 get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8,
1442 :     pParam->width, pParam->height, OldData->iFcode);
1443 : chl 96
1444 : chl 530 if (pMB->mode == MODE_INTER4V) {
1445 :     int dummy;
1446 :     CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); }
1447 : chl 96
1448 : chl 530 if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1449 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1450 :     else MainSearchPtr = DiamondSearch;
1451 : chl 96
1452 : chl 530 (*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255);
1453 : edgomez 195
1454 : chl 530 if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data);
1455 : chl 96
1456 : chl 530 pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x;
1457 :     pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y;
1458 :     pMB->mvs[block] = *(Data.currentMV);
1459 :     pMB->sad8[block] = 4 * (*(Data.iMinSAD));
1460 : chl 96 }
1461 :    
1462 : suxen_drol 118
1463 : chl 530 static void
1464 :     SearchPhinted ( const uint8_t * const pRef,
1465 : chl 289 const uint8_t * const pRefH,
1466 :     const uint8_t * const pRefV,
1467 :     const uint8_t * const pRefHV,
1468 :     const IMAGE * const pCur,
1469 :     const int x,
1470 :     const int y,
1471 :     const uint32_t MotionFlags,
1472 :     const uint32_t iQuant,
1473 :     const MBParam * const pParam,
1474 :     const MACROBLOCK * const pMBs,
1475 : chl 530 int inter4v,
1476 : h 545 MACROBLOCK * const pMB,
1477 : edgomez 568 SearchData * const Data)
1478 : chl 289 {
1479 : chl 530
1480 : chl 289 const int32_t iEdgedWidth = pParam->edged_width;
1481 : chl 530
1482 : edgomez 568 int i, t;
1483 : chl 530 MainSearchFunc * MainSearchPtr;
1484 : suxen_drol 118
1485 : h 545 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1486 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1487 :     pParam->width, pParam->height, Data->iFcode);
1488 : suxen_drol 118
1489 : h 545 Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16;
1490 :     Data->Ref = pRef + (x + iEdgedWidth*y)*16;
1491 :     Data->RefH = pRefH + (x + iEdgedWidth*y) * 16;
1492 :     Data->RefV = pRefV + (x + iEdgedWidth*y) * 16;
1493 :     Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
1494 :     Data->iQuant = iQuant;
1495 : edgomez 568
1496 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
1497 : h 545 Data->min_dx = EVEN(Data->min_dx);
1498 :     Data->max_dx = EVEN(Data->max_dx);
1499 :     Data->min_dy = EVEN(Data->min_dy);
1500 :     Data->max_dy = EVEN(Data->max_dy);
1501 : chl 530 }
1502 : chl 289
1503 : h 545 for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1504 : chl 289
1505 : chl 530 if (pMB->dquant != NO_CHANGE) inter4v = 0;
1506 : chl 289
1507 : chl 530 if (inter4v)
1508 :     CheckCandidate = CheckCandidate16;
1509 :     else CheckCandidate = CheckCandidate16no4v;
1510 : chl 289
1511 :    
1512 : chl 530 pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
1513 :     pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
1514 : h 545 if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
1515 :     if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
1516 :     if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
1517 :     if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
1518 : chl 370
1519 : edgomez 568 (*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
1520 : chl 289
1521 : chl 530 if (pMB->mode == MODE_INTER4V)
1522 :     for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
1523 :     pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
1524 :     pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
1525 :     if (!(make_mask(pMB->mvs, i)))
1526 : edgomez 568 (*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
1527 : chl 289 }
1528 :    
1529 :     if (MotionFlags & PMV_USESQUARES16)
1530 : chl 530 MainSearchPtr = SquareSearch;
1531 : chl 289 else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1532 : chl 530 MainSearchPtr = AdvDiamondSearch;
1533 :     else MainSearchPtr = DiamondSearch;
1534 : chl 289
1535 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1536 : chl 289
1537 : h 545 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1538 : chl 289
1539 : chl 530 if (inter4v)
1540 :     for(i = 0; i < 4; i++)
1541 : h 545 Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i);
1542 : chl 289
1543 : chl 530 if (!(inter4v) ||
1544 : h 545 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] +
1545 :     Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1546 : chl 530 // INTER MODE
1547 : chl 289
1548 : chl 530 pMB->mode = MODE_INTER;
1549 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
1550 : h 545 = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1551 : chl 289
1552 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1553 : h 545 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1554 : chl 289
1555 : h 545 pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1556 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1557 : chl 530 } else {
1558 :     // INTER4V MODE; all other things are already set in Search8hinted
1559 :     pMB->mode = MODE_INTER4V;
1560 : h 545 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3]
1561 :     + Data->iMinSAD[4] + IMV16X16 * iQuant;
1562 : chl 289 }
1563 :    
1564 :     }
1565 :    
1566 : edgomez 195 void
1567 : chl 530 MotionEstimationHinted( MBParam * const pParam,
1568 :     FRAMEINFO * const current,
1569 :     FRAMEINFO * const reference,
1570 :     const IMAGE * const pRefH,
1571 :     const IMAGE * const pRefV,
1572 :     const IMAGE * const pRefHV)
1573 : suxen_drol 118 {
1574 : chl 530 MACROBLOCK *const pMBs = current->mbs;
1575 :     const IMAGE *const pCurrent = &current->image;
1576 :     const IMAGE *const pRef = &reference->image;
1577 : suxen_drol 118
1578 : chl 530 uint32_t x, y;
1579 : syskin 574 int32_t temp[5], quant = current->quant;
1580 : h 545 int32_t iMinSAD[5];
1581 :     VECTOR currentMV[5];
1582 :     SearchData Data;
1583 :     Data.iEdgedWidth = pParam->edged_width;
1584 :     Data.currentMV = currentMV;
1585 :     Data.iMinSAD = iMinSAD;
1586 :     Data.temp = temp;
1587 :     Data.iFcode = current->fcode;
1588 :    
1589 : chl 530 if (sadInit) (*sadInit) ();
1590 : suxen_drol 118
1591 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
1592 :     for (x = 0; x < pParam->mb_width; x++) {
1593 : chl 312
1594 : chl 530 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
1595 : suxen_drol 118
1596 : chl 530 //intra mode is copied from the first pass. At least for the time being
1597 :     if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue;
1598 : chl 346
1599 : syskin 574
1600 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
1601 :     pMB->dquant = NO_CHANGE;
1602 :     pMB->quant = current->quant; }
1603 : syskin 574 else
1604 :     if (pMB->dquant != NO_CHANGE) {
1605 :     quant += DQtab[pMB->dquant];
1606 :     if (quant > 31) quant = 31;
1607 :     else if (quant < 1) quant = 1;
1608 :     pMB->quant = quant;
1609 :     }
1610 : chl 341
1611 : Isibaar 539 SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
1612 :     y, current->motion_flags, pMB->quant,
1613 : h 545 pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
1614 :     &Data);
1615 : chl 317
1616 : Isibaar 539 }
1617 :     }
1618 :     }
1619 : suxen_drol 118
1620 : Isibaar 539 static __inline int
1621 :     MEanalyzeMB ( const uint8_t * const pRef,
1622 :     const uint8_t * const pCur,
1623 :     const int x,
1624 :     const int y,
1625 :     const MBParam * const pParam,
1626 :     const MACROBLOCK * const pMBs,
1627 : h 545 MACROBLOCK * const pMB,
1628 :     SearchData * const Data)
1629 : Isibaar 539 {
1630 : chl 317
1631 : Isibaar 539 int i, mask;
1632 : h 545 VECTOR pmv[3];
1633 : suxen_drol 118
1634 : h 545 *(Data->iMinSAD) = MV_MAX_ERROR;
1635 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1636 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1637 :     pParam->width, pParam->height, Data->iFcode);
1638 : chl 326
1639 : h 545 Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1640 :     Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1641 : Isibaar 539
1642 :     CheckCandidate = CheckCandidate16no4vI;
1643 : chl 326
1644 : Isibaar 539 pmv[1].x = EVEN(pMB->mvs[0].x);
1645 :     pmv[1].y = EVEN(pMB->mvs[0].y);
1646 : h 545 pmv[0].x = EVEN(Data->predMV.x);
1647 :     pmv[0].y = EVEN(Data->predMV.y);
1648 : Isibaar 539 pmv[2].x = pmv[2].y = 0;
1649 : chl 326
1650 : h 545 CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
1651 : Isibaar 539 if (!(mask = make_mask(pmv, 1)))
1652 : h 545 CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
1653 : Isibaar 539 if (!(mask = make_mask(pmv, 2)))
1654 : h 545 CheckCandidate16no4vI(0, 0, mask, &i, Data);
1655 : Isibaar 539
1656 : h 545 DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1657 : Isibaar 539
1658 :     pMB->mvs[0] = pMB->mvs[1]
1659 : h 545 = pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv()
1660 : Isibaar 539
1661 : h 545 return *(Data->iMinSAD);
1662 : Isibaar 539 }
1663 :    
1664 :     #define INTRA_THRESH 1350
1665 :     #define INTER_THRESH 900
1666 :    
1667 :     int
1668 :     MEanalysis( const IMAGE * const pRef,
1669 :     const IMAGE * const pCurrent,
1670 :     MBParam * const pParam,
1671 :     MACROBLOCK * const pMBs,
1672 :     const uint32_t iFcode)
1673 :     {
1674 :     uint32_t x, y, intra = 0;
1675 :     int sSAD = 0;
1676 : h 545
1677 :     VECTOR currentMV;
1678 :     int32_t iMinSAD;
1679 :     SearchData Data;
1680 :     Data.iEdgedWidth = pParam->edged_width;
1681 :     Data.currentMV = &currentMV;
1682 :     Data.iMinSAD = &iMinSAD;
1683 :     Data.iFcode = iFcode;
1684 :     Data.iQuant = 2;
1685 :    
1686 : Isibaar 539 if (sadInit) (*sadInit) ();
1687 :    
1688 :     for (y = 0; y < pParam->mb_height-1; y++) {
1689 :     for (x = 0; x < pParam->mb_width; x++) {
1690 :     int sad, dev;
1691 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
1692 :    
1693 :     sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
1694 : h 545 pParam, pMBs, pMB, &Data);
1695 : Isibaar 539
1696 :     if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work
1697 :     if (sad > INTRA_THRESH) {
1698 :     dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
1699 :     pParam->edged_width);
1700 :     if (dev + INTRA_THRESH < sad) intra++;
1701 :     if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame
1702 :     }
1703 :     sSAD += sad;
1704 :     }
1705 :    
1706 : suxen_drol 118 }
1707 :     }
1708 : Isibaar 539 sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
1709 :     if (sSAD > INTER_THRESH ) return 1; //P frame
1710 :     emms();
1711 :     return 0; // B frame
1712 :    
1713 : suxen_drol 118 }
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4