[svn] / trunk / xvidcore / src / motion / estimation_pvop.c Repository:
ViewVC logotype

Annotation of /trunk/xvidcore/src/motion/estimation_pvop.c

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1701 - (view) (download)

1 : edgomez 1382 /*****************************************************************************
2 :     *
3 :     * XVID MPEG-4 VIDEO CODEC
4 :     * - Motion Estimation for P- and S- VOPs -
5 :     *
6 :     * Copyright(C) 2002 Christoph Lampert <gruel@web.de>
7 :     * 2002 Michael Militzer <michael@xvid.org>
8 :     * 2002-2003 Radoslaw Czyz <xvid@syskin.cjb.net>
9 :     *
10 :     * This program is free software ; you can redistribute it and/or modify
11 :     * it under the terms of the GNU General Public License as published by
12 :     * the Free Software Foundation ; either version 2 of the License, or
13 :     * (at your option) any later version.
14 :     *
15 :     * This program is distributed in the hope that it will be useful,
16 :     * but WITHOUT ANY WARRANTY ; without even the implied warranty of
17 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 :     * GNU General Public License for more details.
19 :     *
20 :     * You should have received a copy of the GNU General Public License
21 :     * along with this program ; if not, write to the Free Software
22 :     * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 :     *
24 : syskin 1701 * $Id: estimation_pvop.c,v 1.22 2006-04-19 15:42:19 syskin Exp $
25 : edgomez 1382 *
26 :     ****************************************************************************/
27 :    
28 :     #include <assert.h>
29 :     #include <stdio.h>
30 :     #include <stdlib.h>
31 :     #include <string.h> /* memcpy */
32 :    
33 :     #include "../encoder.h"
34 :     #include "../prediction/mbprediction.h"
35 :     #include "../global.h"
36 :     #include "../utils/timer.h"
37 :     #include "../image/interpolate8x8.h"
38 :     #include "estimation.h"
39 :     #include "motion.h"
40 :     #include "sad.h"
41 :     #include "motion_inlines.h"
42 : syskin 1682 #include "motion_smp.h"
43 : edgomez 1382
44 : syskin 1682
45 : edgomez 1382 static const int xvid_me_lambda_vec8[32] =
46 : Isibaar 1604 { 0 ,(int)(1.0 * NEIGH_TEND_8X8 + 0.5),
47 :     (int)(2.0*NEIGH_TEND_8X8 + 0.5), (int)(3.0*NEIGH_TEND_8X8 + 0.5),
48 :     (int)(4.0*NEIGH_TEND_8X8 + 0.5), (int)(5.0*NEIGH_TEND_8X8 + 0.5),
49 :     (int)(6.0*NEIGH_TEND_8X8 + 0.5), (int)(7.0*NEIGH_TEND_8X8 + 0.5),
50 :     (int)(8.0*NEIGH_TEND_8X8 + 0.5), (int)(9.0*NEIGH_TEND_8X8 + 0.5),
51 :     (int)(10.0*NEIGH_TEND_8X8 + 0.5), (int)(11.0*NEIGH_TEND_8X8 + 0.5),
52 :     (int)(12.0*NEIGH_TEND_8X8 + 0.5), (int)(13.0*NEIGH_TEND_8X8 + 0.5),
53 :     (int)(14.0*NEIGH_TEND_8X8 + 0.5), (int)(15.0*NEIGH_TEND_8X8 + 0.5),
54 :     (int)(16.0*NEIGH_TEND_8X8 + 0.5), (int)(17.0*NEIGH_TEND_8X8 + 0.5),
55 :     (int)(18.0*NEIGH_TEND_8X8 + 0.5), (int)(19.0*NEIGH_TEND_8X8 + 0.5),
56 :     (int)(20.0*NEIGH_TEND_8X8 + 0.5), (int)(21.0*NEIGH_TEND_8X8 + 0.5),
57 :     (int)(22.0*NEIGH_TEND_8X8 + 0.5), (int)(23.0*NEIGH_TEND_8X8 + 0.5),
58 :     (int)(24.0*NEIGH_TEND_8X8 + 0.5), (int)(25.0*NEIGH_TEND_8X8 + 0.5),
59 :     (int)(26.0*NEIGH_TEND_8X8 + 0.5), (int)(27.0*NEIGH_TEND_8X8 + 0.5),
60 :     (int)(28.0*NEIGH_TEND_8X8 + 0.5), (int)(29.0*NEIGH_TEND_8X8 + 0.5),
61 :     (int)(30.0*NEIGH_TEND_8X8 + 0.5), (int)(31.0*NEIGH_TEND_8X8 + 0.5)
62 : edgomez 1382 };
63 :    
64 :     static void
65 :     CheckCandidate16(const int x, const int y, SearchData * const data, const unsigned int Direction)
66 :     {
67 :     const uint8_t * Reference;
68 : syskin 1441 int32_t sad, xc, yc; uint32_t t;
69 :     VECTOR * current;
70 : edgomez 1382
71 :     if ( (x > data->max_dx) || (x < data->min_dx)
72 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
73 :    
74 : syskin 1441 if (data->qpel_precision) { /* x and y are in 1/4 precision */
75 :     Reference = xvid_me_interpolate16x16qpel(x, y, 0, data);
76 :     current = data->currentQMV;
77 :     xc = x/2; yc = y/2;
78 :     } else {
79 :     Reference = GetReference(x, y, data);
80 :     current = data->currentMV;
81 :     xc = x; yc = y;
82 :     }
83 : edgomez 1382
84 :     sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp);
85 : syskin 1564 t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision);
86 : edgomez 1382
87 : Isibaar 1604 sad += (data->lambda16 * t);
88 :     data->temp[0] += (data->lambda8 * t);
89 : edgomez 1382
90 :     if (data->chroma) {
91 :     if (sad >= data->iMinSAD[0]) goto no16;
92 : syskin 1441 sad += xvid_me_ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
93 :     (yc >> 1) + roundtab_79[yc & 0x3], data);
94 : edgomez 1382 }
95 :    
96 :     if (sad < data->iMinSAD[0]) {
97 :     data->iMinSAD[0] = sad;
98 : syskin 1441 current[0].x = x; current[0].y = y;
99 : edgomez 1382 data->dir = Direction;
100 :     }
101 :    
102 :     no16:
103 :     if (data->temp[0] < data->iMinSAD[1]) {
104 : syskin 1441 data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; }
105 : edgomez 1382 if (data->temp[1] < data->iMinSAD[2]) {
106 : syskin 1441 data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; }
107 : edgomez 1382 if (data->temp[2] < data->iMinSAD[3]) {
108 : syskin 1441 data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; }
109 : edgomez 1382 if (data->temp[3] < data->iMinSAD[4]) {
110 : syskin 1441 data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; }
111 : edgomez 1382 }
112 :    
113 :     static void
114 :     CheckCandidate8(const int x, const int y, SearchData * const data, const unsigned int Direction)
115 :     {
116 :     int32_t sad; uint32_t t;
117 :     const uint8_t * Reference;
118 :     VECTOR * current;
119 :    
120 :     if ( (x > data->max_dx) || (x < data->min_dx)
121 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
122 :    
123 :     if (!data->qpel_precision) {
124 :     Reference = GetReference(x, y, data);
125 :     current = data->currentMV;
126 :     } else { /* x and y are in 1/4 precision */
127 :     Reference = xvid_me_interpolate8x8qpel(x, y, 0, 0, data);
128 :     current = data->currentQMV;
129 :     }
130 :    
131 :     sad = sad8(data->Cur, Reference, data->iEdgedWidth);
132 : syskin 1564 t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision);
133 : edgomez 1382
134 : Isibaar 1604 sad += (data->lambda8 * t);
135 : edgomez 1382
136 :     if (sad < *(data->iMinSAD)) {
137 :     *(data->iMinSAD) = sad;
138 :     current->x = x; current->y = y;
139 :     data->dir = Direction;
140 :     }
141 :     }
142 :    
143 :     int
144 :     xvid_me_SkipDecisionP(const IMAGE * current, const IMAGE * reference,
145 :     const int x, const int y,
146 : syskin 1564 const uint32_t stride, const uint32_t iQuant)
147 : edgomez 1382 {
148 :     int offset = (x + y*stride)*8;
149 : syskin 1564 uint32_t sadC = sad8(current->u + offset,
150 :     reference->u + offset, stride);
151 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
152 :     sadC += sad8(current->v + offset,
153 :     reference->v + offset, stride);
154 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
155 :     return 1;
156 : edgomez 1382 }
157 :    
158 :     /*
159 :     * pmv are filled with:
160 :     * [0]: Median (or whatever is correct in a special case)
161 :     * [1]: left neighbour
162 :     * [2]: top neighbour
163 :     * [3]: topright neighbour
164 :     * psad are filled with:
165 :     * [0]: minimum of [1] to [3]
166 :     * [1]: left neighbour's SAD (NB:[1] to [3] are actually not needed)
167 :     * [2]: top neighbour's SAD
168 :     * [3]: topright neighbour's SAD
169 :     */
170 :    
171 :     static __inline void
172 :     get_pmvdata2(const MACROBLOCK * const mbs,
173 :     const int mb_width,
174 :     const int bound,
175 :     const int x,
176 :     const int y,
177 :     VECTOR * const pmv,
178 :     int32_t * const psad)
179 :     {
180 :     int lx, ly, lz; /* left */
181 :     int tx, ty, tz; /* top */
182 :     int rx, ry, rz; /* top-right */
183 :     int lpos, tpos, rpos;
184 :     int num_cand = 0, last_cand = 1;
185 :    
186 :     lx = x - 1; ly = y; lz = 1;
187 :     tx = x; ty = y - 1; tz = 2;
188 :     rx = x + 1; ry = y - 1; rz = 2;
189 :    
190 :     lpos = lx + ly * mb_width;
191 :     rpos = rx + ry * mb_width;
192 :     tpos = tx + ty * mb_width;
193 :    
194 :     if (lpos >= bound && lx >= 0) {
195 :     num_cand++;
196 :     last_cand = 1;
197 :     pmv[1] = mbs[lpos].mvs[lz];
198 :     psad[1] = mbs[lpos].sad8[lz];
199 :     } else {
200 :     pmv[1] = zeroMV;
201 :     psad[1] = MV_MAX_ERROR;
202 :     }
203 :    
204 :     if (tpos >= bound) {
205 :     num_cand++;
206 :     last_cand = 2;
207 :     pmv[2]= mbs[tpos].mvs[tz];
208 :     psad[2] = mbs[tpos].sad8[tz];
209 :     } else {
210 :     pmv[2] = zeroMV;
211 :     psad[2] = MV_MAX_ERROR;
212 :     }
213 :    
214 :     if (rpos >= bound && rx < mb_width) {
215 :     num_cand++;
216 :     last_cand = 3;
217 :     pmv[3] = mbs[rpos].mvs[rz];
218 :     psad[3] = mbs[rpos].sad8[rz];
219 :     } else {
220 :     pmv[3] = zeroMV;
221 :     psad[3] = MV_MAX_ERROR;
222 :     }
223 :    
224 :     /* original pmvdata() compatibility hack */
225 :     if (x == 0 && y == 0) {
226 :     pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV;
227 :     psad[0] = 0;
228 :     psad[1] = psad[2] = psad[3] = MV_MAX_ERROR;
229 :     return;
230 :     }
231 :    
232 :     /* if only one valid candidate preictor, the invalid candiates are set to the canidate */
233 :     if (num_cand == 1) {
234 :     pmv[0] = pmv[last_cand];
235 :     psad[0] = psad[last_cand];
236 :     return;
237 :     }
238 :    
239 :     if ((MVequal(pmv[1], pmv[2])) && (MVequal(pmv[1], pmv[3]))) {
240 :     pmv[0] = pmv[1];
241 :     psad[0] = MIN(MIN(psad[1], psad[2]), psad[3]);
242 :     return;
243 :     }
244 :    
245 :     /* set median, minimum */
246 :    
247 :     pmv[0].x =
248 :     MIN(MAX(pmv[1].x, pmv[2].x),
249 :     MIN(MAX(pmv[2].x, pmv[3].x), MAX(pmv[1].x, pmv[3].x)));
250 :     pmv[0].y =
251 :     MIN(MAX(pmv[1].y, pmv[2].y),
252 :     MIN(MAX(pmv[2].y, pmv[3].y), MAX(pmv[1].y, pmv[3].y)));
253 :    
254 :     psad[0] = MIN(MIN(psad[1], psad[2]), psad[3]);
255 :    
256 :     }
257 :    
258 :    
259 :     static void
260 :     ModeDecision_SAD(SearchData * const Data,
261 :     MACROBLOCK * const pMB,
262 :     const MACROBLOCK * const pMBs,
263 :     const int x, const int y,
264 :     const MBParam * const pParam,
265 :     const uint32_t MotionFlags,
266 :     const uint32_t VopFlags,
267 :     const uint32_t VolFlags,
268 :     const IMAGE * const pCurrent,
269 :     const IMAGE * const pRef,
270 :     const IMAGE * const vGMC,
271 : syskin 1443 const int coding_type,
272 :     const int skip_sad)
273 : edgomez 1382 {
274 :     int mode = MODE_INTER;
275 :     int mcsel = 0;
276 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
277 :     const uint32_t iQuant = pMB->quant;
278 :    
279 :     const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0);
280 :    
281 :     int sad;
282 :     int InterBias = MV16_INTER_BIAS;
283 :    
284 :     pMB->mcsel = 0;
285 :    
286 :     if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
287 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
288 :     mode = MODE_INTER;
289 :     sad = Data->iMinSAD[0];
290 :     } else {
291 :     mode = MODE_INTER4V;
292 :     sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
293 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
294 :     Data->iMinSAD[0] = sad;
295 :     }
296 :    
297 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
298 : syskin 1443 if (skip_possible && (skip_sad < (int)iQuant * MAX_SAD00_FOR_SKIP))
299 : syskin 1701 if ( (100*skip_sad)/(pMB->sad16+1) < FINAL_SKIP_THRESH)
300 : syskin 1564 if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) {
301 : edgomez 1382 mode = MODE_NOT_CODED;
302 :     sad = 0;
303 :     }
304 :    
305 :     /* mcsel */
306 :     if (coding_type == S_VOP) {
307 :    
308 :     int32_t iSAD = sad16(Data->Cur,
309 :     vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
310 :    
311 :     if (Data->chroma) {
312 :     iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
313 :     iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
314 :     }
315 :    
316 :     if (iSAD <= sad) { /* mode decision GMC */
317 :     mode = MODE_INTER;
318 :     mcsel = 1;
319 :     sad = iSAD;
320 :     }
321 :     }
322 :    
323 :     /* intra decision */
324 :    
325 :     if (iQuant > 10) InterBias += 60 * (iQuant - 10); /* to make high quants work */
326 :     if (y != 0)
327 :     if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
328 :     if (x != 0)
329 :     if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
330 :    
331 :     if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */
332 :    
333 :     if (InterBias < sad) {
334 : syskin 1564 int32_t deviation = dev16(Data->Cur, Data->iEdgedWidth);
335 : edgomez 1382 if (deviation < (sad - InterBias)) mode = MODE_INTRA;
336 :     }
337 :    
338 :     pMB->cbp = 63;
339 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
340 :    
341 :     if (mode == MODE_INTER && mcsel == 0) {
342 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
343 :    
344 :     if(Data->qpel) {
345 :     pMB->qmvs[0] = pMB->qmvs[1]
346 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
347 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
348 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
349 :     } else {
350 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
351 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
352 :     }
353 :    
354 :     } else if (mode == MODE_INTER ) { /* but mcsel == 1 */
355 :    
356 :     pMB->mcsel = 1;
357 :     if (Data->qpel) {
358 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
359 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
360 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
361 :     } else
362 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
363 :    
364 :     } else
365 :     if (mode == MODE_INTER4V) ; /* anything here? */
366 :     else /* INTRA, NOT_CODED */
367 :     ZeroMacroblockP(pMB, 0);
368 :    
369 :     pMB->mode = mode;
370 :     }
371 :    
372 :     static __inline void
373 :     PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount,
374 : syskin 1564 int iHcount, const MACROBLOCK * const prevMB)
375 : edgomez 1382 {
376 :    
377 :     if ( (y != 0) && (x < (iWcount-1)) ) { /* [5] top-right neighbour */
378 :     pmv[5].x = EVEN(pmv[3].x);
379 :     pmv[5].y = EVEN(pmv[3].y);
380 :     } else pmv[5].x = pmv[5].y = 0;
381 :    
382 :     if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }/* pmv[3] is left neighbour */
383 :     else pmv[3].x = pmv[3].y = 0;
384 :    
385 :     if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }/* [4] top neighbour */
386 :     else pmv[4].x = pmv[4].y = 0;
387 :    
388 :     /* [1] median prediction */
389 :     pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
390 :    
391 :     pmv[0].x = pmv[0].y = 0; /* [0] is zero; not used in the loop (checked before) but needed here for make_mask */
392 :    
393 :     pmv[2].x = EVEN(prevMB->mvs[0].x); /* [2] is last frame */
394 :     pmv[2].y = EVEN(prevMB->mvs[0].y);
395 :    
396 :     if ((x < iWcount-1) && (y < iHcount-1)) {
397 :     pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); /* [6] right-down neighbour in last frame */
398 :     pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
399 :     } else pmv[6].x = pmv[6].y = 0;
400 :     }
401 :    
402 :     static void
403 :     Search8(SearchData * const OldData,
404 :     const int x, const int y,
405 :     const uint32_t MotionFlags,
406 :     const MBParam * const pParam,
407 :     MACROBLOCK * const pMB,
408 :     const MACROBLOCK * const pMBs,
409 :     const int block,
410 :     SearchData * const Data)
411 :     {
412 :     int i = 0;
413 : syskin 1441 VECTOR vbest_q; int32_t sbest_q;
414 : edgomez 1382 *Data->iMinSAD = *(OldData->iMinSAD + 1 + block);
415 :     *Data->currentMV = *(OldData->currentMV + 1 + block);
416 :     *Data->currentQMV = *(OldData->currentQMV + 1 + block);
417 :    
418 :     if(Data->qpel) {
419 :     Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
420 :     if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y,
421 : syskin 1564 Data->predMV, Data->iFcode, 0);
422 : edgomez 1382 } else {
423 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block);
424 :     if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y,
425 : syskin 1564 Data->predMV, Data->iFcode, 0);
426 : edgomez 1382 }
427 :    
428 : Isibaar 1604 *(Data->iMinSAD) += (Data->lambda8 * i);
429 : edgomez 1382
430 :     if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) {
431 :    
432 : syskin 1441 vbest_q = Data->currentQMV[0];
433 :     sbest_q = Data->iMinSAD[0];
434 :    
435 : edgomez 1382
436 : syskin 1564 Data->RefP[0] = OldData->RefP[0] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
437 :     Data->RefP[1] = OldData->RefP[1] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
438 :     Data->RefP[2] = OldData->RefP[2] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
439 :     Data->RefP[3] = OldData->RefP[3] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
440 : edgomez 1382
441 : syskin 1564 Data->Cur = OldData->Cur + 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
442 : edgomez 1382 Data->qpel_precision = 0;
443 :    
444 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
445 : syskin 1564 pParam->width, pParam->height, Data->iFcode - Data->qpel, 1);
446 : edgomez 1382
447 :     if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) {
448 :    
449 :     MainSearchFunc *MainSearchPtr;
450 :     if (MotionFlags & XVID_ME_USESQUARES8) MainSearchPtr = xvid_me_SquareSearch;
451 :     else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = xvid_me_AdvDiamondSearch;
452 :     else MainSearchPtr = xvid_me_DiamondSearch;
453 :    
454 : syskin 1564 MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate8);
455 : edgomez 1382 }
456 :    
457 : syskin 1441 if(!Data->qpel) {
458 :     /* halfpel mode */
459 :     if (MotionFlags & XVID_ME_HALFPELREFINE8)
460 : syskin 1478 /* perform halfpel refine of current best vector */
461 : syskin 1564 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate8, 0);
462 : syskin 1441 } else {
463 :     /* qpel mode */
464 :     Data->currentQMV->x = 2*Data->currentMV->x;
465 :     Data->currentQMV->y = 2*Data->currentMV->y;
466 :    
467 :     if(MotionFlags & XVID_ME_FASTREFINE8) {
468 :     /* fast */
469 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
470 : syskin 1564 pParam->width, pParam->height, Data->iFcode, 2);
471 : syskin 1441 FullRefine_Fast(Data, CheckCandidate8, 0);
472 :     } else if(MotionFlags & XVID_ME_QUARTERPELREFINE8) {
473 :     /* full */
474 :     if (MotionFlags & XVID_ME_HALFPELREFINE8) {
475 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate8, 0); /* hpel part */
476 : syskin 1441 Data->currentQMV->x = 2*Data->currentMV->x;
477 :     Data->currentQMV->y = 2*Data->currentMV->y;
478 :     }
479 : edgomez 1382
480 : syskin 1441 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
481 : syskin 1564 pParam->width, pParam->height, Data->iFcode, 2);
482 : syskin 1441 Data->qpel_precision = 1;
483 : edgomez 1382
484 : syskin 1478 xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate8, 0); /* qpel part */
485 : edgomez 1382 }
486 :     }
487 :    
488 : syskin 1441 if (sbest_q <= Data->iMinSAD[0]) /* we have not found a better match */
489 :     Data->currentQMV[0] = vbest_q;
490 : edgomez 1382
491 :     }
492 :    
493 : syskin 1564 if(Data->qpel) {
494 : edgomez 1382 pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x;
495 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y;
496 :     pMB->qmvs[block] = *Data->currentQMV;
497 :     } else {
498 :     pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
499 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
500 :     }
501 :    
502 :     *(OldData->iMinSAD + 1 + block) = *Data->iMinSAD;
503 :     *(OldData->currentMV + 1 + block) = *Data->currentMV;
504 :     *(OldData->currentQMV + 1 + block) = *Data->currentQMV;
505 :    
506 :     pMB->mvs[block] = *Data->currentMV;
507 :     pMB->sad8[block] = 4 * *Data->iMinSAD;
508 :     }
509 :    
510 :    
511 :    
512 :     static void
513 :     SearchP(const IMAGE * const pRef,
514 :     const uint8_t * const pRefH,
515 :     const uint8_t * const pRefV,
516 :     const uint8_t * const pRefHV,
517 :     const IMAGE * const pCur,
518 :     const int x,
519 :     const int y,
520 :     const uint32_t MotionFlags,
521 :     const uint32_t VopFlags,
522 :     SearchData * const Data,
523 :     const MBParam * const pParam,
524 :     const MACROBLOCK * const pMBs,
525 :     const MACROBLOCK * const prevMBs,
526 :     MACROBLOCK * const pMB)
527 :     {
528 :    
529 :     int i, threshA;
530 :     VECTOR pmv[7];
531 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
532 :     CheckFunc * CheckCandidate;
533 :    
534 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
535 : syskin 1564 pParam->width, pParam->height, Data->iFcode - Data->qpel, 1);
536 : edgomez 1382
537 :     get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, pmv, Data->temp);
538 :    
539 :     Data->chromaX = Data->chromaY = 0; /* chroma-sad cache */
540 : syskin 1564 Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
541 :     Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8;
542 :     Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8;
543 : edgomez 1382
544 : syskin 1564 Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16;
545 :     Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16;
546 :     Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16;
547 :     Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16;
548 :     Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
549 :     Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
550 : edgomez 1382
551 :     Data->lambda16 = xvid_me_lambda_vec16[pMB->quant];
552 :     Data->lambda8 = xvid_me_lambda_vec8[pMB->quant];
553 :     Data->qpel_precision = 0;
554 :     Data->dir = 0;
555 :    
556 :     memset(Data->currentMV, 0, 5*sizeof(VECTOR));
557 :    
558 :     if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
559 :     else Data->predMV = pmv[0];
560 :    
561 : syskin 1564 i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0);
562 : Isibaar 1604 Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i);
563 :     Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i);
564 : edgomez 1382 Data->iMinSAD[2] = pMB->sad8[1];
565 :     Data->iMinSAD[3] = pMB->sad8[2];
566 :     Data->iMinSAD[4] = pMB->sad8[3];
567 :    
568 :     if ((!(VopFlags & XVID_VOP_MODEDECISION_RD)) && (x | y)) {
569 :     threshA = Data->temp[0]; /* that's where we keep this SAD atm */
570 :     if (threshA < 512) threshA = 512;
571 :     else if (threshA > 1024) threshA = 1024;
572 :     } else
573 :     threshA = 512;
574 :    
575 :     PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
576 : syskin 1564 prevMBs + x + y * pParam->mb_width);
577 : edgomez 1382
578 : syskin 1564 if (inter4v) CheckCandidate = CheckCandidate16;
579 :     else CheckCandidate = CheckCandidate16no4v; /* for extra speed */
580 : edgomez 1382
581 :     /* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
582 :    
583 :     for (i = 1; i < 7; i++)
584 :     if (!vector_repeats(pmv, i)) {
585 :     CheckCandidate(pmv[i].x, pmv[i].y, Data, i);
586 :     if (Data->iMinSAD[0] <= threshA) { i++; break; }
587 :     }
588 :    
589 :     if ((Data->iMinSAD[0] <= threshA) ||
590 :     (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
591 :     (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
592 :     inter4v = 0;
593 :     else {
594 :    
595 :     MainSearchFunc * MainSearchPtr;
596 :     int mask = make_mask(pmv, i, Data->dir); /* all vectors pmv[0..i-1] have been checked */
597 :    
598 :     if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch;
599 :     else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch;
600 :     else MainSearchPtr = xvid_me_DiamondSearch;
601 :    
602 :     MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate);
603 :    
604 :     /* extended search, diamond starting in 0,0 and in prediction.
605 :     note that this search is/might be done in halfpel positions,
606 :     which makes it more different than the diamond above */
607 :    
608 :     if (MotionFlags & XVID_ME_EXTSEARCH16) {
609 :     int32_t bSAD;
610 :     VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
611 :     if (Data->qpel) {
612 :     startMV.x /= 2;
613 :     startMV.y /= 2;
614 :     }
615 :     if (!(MVequal(startMV, backupMV))) {
616 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
617 :    
618 :     CheckCandidate(startMV.x, startMV.y, Data, 255);
619 :     xvid_me_DiamondSearch(startMV.x, startMV.y, Data, 255, CheckCandidate);
620 :     if (bSAD < Data->iMinSAD[0]) {
621 :     Data->currentMV[0] = backupMV;
622 : syskin 1564 Data->iMinSAD[0] = bSAD;
623 :     }
624 : edgomez 1382 }
625 :    
626 :     backupMV = Data->currentMV[0];
627 :     startMV.x = startMV.y = 1;
628 :     if (!(MVequal(startMV, backupMV))) {
629 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
630 :    
631 :     CheckCandidate(startMV.x, startMV.y, Data, 255);
632 :     xvid_me_DiamondSearch(startMV.x, startMV.y, Data, 255, CheckCandidate);
633 :     if (bSAD < Data->iMinSAD[0]) {
634 :     Data->currentMV[0] = backupMV;
635 :     Data->iMinSAD[0] = bSAD;
636 :     }
637 :     }
638 :     }
639 :     }
640 :    
641 :    
642 : syskin 1441 if(!Data->qpel) {
643 :     /* halfpel mode */
644 :     if (MotionFlags & XVID_ME_HALFPELREFINE16)
645 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate, 0);
646 : syskin 1441 } else {
647 :     /* qpel mode */
648 :    
649 :     for(i = 0; i < 5; i++) {
650 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */
651 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
652 :     }
653 :     if(MotionFlags & XVID_ME_FASTREFINE16 && MotionFlags & XVID_ME_QUARTERPELREFINE16) {
654 :     /* fast */
655 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
656 : syskin 1564 pParam->width, pParam->height, Data->iFcode, 2);
657 : syskin 1441 FullRefine_Fast(Data, CheckCandidate, 0);
658 :     } else {
659 :     if(MotionFlags & (XVID_ME_QUARTERPELREFINE16 | XVID_ME_QUARTERPELREFINE16_RD)) {
660 :     /* full */
661 :     if (MotionFlags & XVID_ME_HALFPELREFINE16) {
662 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate, 0); /* hpel part */
663 : syskin 1441 for(i = 0; i < 5; i++) {
664 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x;
665 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
666 :     }
667 :     }
668 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
669 : syskin 1564 pParam->width, pParam->height, Data->iFcode, 2);
670 : syskin 1441 Data->qpel_precision = 1;
671 :     if(MotionFlags & XVID_ME_QUARTERPELREFINE16)
672 : syskin 1478 xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate, 0); /* qpel part */
673 : syskin 1441 }
674 : edgomez 1382 }
675 :     }
676 :    
677 : syskin 1564 if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30 * ((MotionFlags & XVID_ME_FASTREFINE16) ? 8 : 1))
678 : edgomez 1382 inter4v = 0;
679 :    
680 :     if (inter4v) {
681 :     SearchData Data8;
682 :     memcpy(&Data8, Data, sizeof(SearchData)); /* quick copy of common data */
683 :    
684 :     Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
685 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
686 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
687 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
688 :    
689 :     if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_RD))) {
690 : syskin 1441 /* chroma is only used for comparison to INTER. if the comparison will be done in RD domain, it will not be used */
691 : edgomez 1382 int sumx = 0, sumy = 0;
692 :    
693 :     if (Data->qpel)
694 :     for (i = 1; i < 5; i++) {
695 :     sumx += Data->currentQMV[i].x/2;
696 :     sumy += Data->currentQMV[i].y/2;
697 :     }
698 :     else
699 :     for (i = 1; i < 5; i++) {
700 :     sumx += Data->currentMV[i].x;
701 :     sumy += Data->currentMV[i].y;
702 :     }
703 :    
704 :     Data->iMinSAD[1] += xvid_me_ChromaSAD((sumx >> 3) + roundtab_76[sumx & 0xf],
705 :     (sumy >> 3) + roundtab_76[sumy & 0xf], Data);
706 :     }
707 :     } else Data->iMinSAD[1] = 4096*256;
708 :     }
709 :    
710 : syskin 1668 static int
711 :     InitialSkipDecisionP(int sad00,
712 :     const MBParam * pParam,
713 :     const FRAMEINFO * current,
714 :     MACROBLOCK * pMB,
715 :     const MACROBLOCK * prevMB,
716 :     int x, int y,
717 :     const SearchData * Data,
718 :     const IMAGE * const pGMC,
719 :     const IMAGE * const pCurrent,
720 :     const IMAGE * const pRef,
721 :     const uint32_t MotionFlags)
722 :     {
723 :     const unsigned int iEdgedWidth = pParam->edged_width;
724 :    
725 :     int skip_thresh = INITIAL_SKIP_THRESH * \
726 :     (current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1);
727 :     int stat_thresh = 0;
728 :    
729 :     /* initial skip decision */
730 :     if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */
731 :     if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh)
732 :     if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant)) {
733 :     ZeroMacroblockP(pMB, sad00);
734 :     pMB->mode = MODE_NOT_CODED;
735 :     return 1;
736 :     }
737 :     }
738 :    
739 :     if(MotionFlags & XVID_ME_DETECT_STATIC_MOTION) {
740 :     VECTOR *cmpMV;
741 :     VECTOR staticMV = { 0, 0 };
742 :     const MACROBLOCK * pMBs = current->mbs;
743 :    
744 :     if (current->coding_type == S_VOP)
745 :     cmpMV = &pMB->amv;
746 :     else
747 :     cmpMV = &staticMV;
748 :    
749 :     if(x > 0 && y > 0 && x < pParam->mb_width) {
750 :     if(MVequal((&pMBs[(x-1) + y * pParam->mb_width])->mvs[0], *cmpMV) &&
751 :     MVequal((&pMBs[x + (y-1) * pParam->mb_width])->mvs[0], *cmpMV) &&
752 :     MVequal((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mvs[0], *cmpMV) &&
753 :     MVequal(prevMB->mvs[0], *cmpMV)) {
754 :     stat_thresh = MAX((&pMBs[(x-1) + y * pParam->mb_width])->sad16,
755 :     MAX((&pMBs[x + (y-1) * pParam->mb_width])->sad16,
756 :     MAX((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16,
757 :     prevMB->sad16)));
758 :     } else {
759 :     stat_thresh = MIN((&pMBs[(x-1) + y * pParam->mb_width])->sad16,
760 :     MIN((&pMBs[x + (y-1) * pParam->mb_width])->sad16,
761 :     MIN((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16,
762 :     prevMB->sad16)));
763 :     }
764 :     }
765 :     }
766 :    
767 :     /* favorize (0,0) or global vector for cartoons */
768 :     if (current->vop_flags & XVID_VOP_CARTOON) {
769 :     if (current->coding_type == S_VOP) {
770 :     int32_t iSAD = sad16(pCurrent->y + (x + y * iEdgedWidth) * 16,
771 :     pGMC->y + 16*y*iEdgedWidth + 16*x, iEdgedWidth, 65536);
772 :    
773 :     if (Data->chroma) {
774 :     iSAD += sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, pGMC->u + 8*y*(iEdgedWidth/2) + 8*x, iEdgedWidth/2);
775 :     iSAD += sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, pGMC->v + 8*y*(iEdgedWidth/2) + 8*x, iEdgedWidth/2);
776 :     }
777 :    
778 :     if (iSAD <= stat_thresh) { /* mode decision GMC */
779 :     pMB->mode = MODE_INTER;
780 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = iSAD;
781 :     pMB->mcsel = 1;
782 :     if (Data->qpel) {
783 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
784 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
785 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
786 :     } else
787 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
788 :    
789 :     return 1;
790 :     }
791 :     }
792 :     else if (sad00 < stat_thresh) {
793 :     VECTOR predMV;
794 :     if (Data->qpel)
795 :     predMV = get_qpmv2(current->mbs, pParam->mb_width, 0, x, y, 0);
796 :     else
797 :     predMV = get_pmv2(current->mbs, pParam->mb_width, 0, x, y, 0);
798 :    
799 :     ZeroMacroblockP(pMB, sad00);
800 :     pMB->cbp = 0x3f;
801 :     pMB->pmvs[0].x = - predMV.x;
802 :     pMB->pmvs[0].y = - predMV.y;
803 :     return 1;
804 :     }
805 :     }
806 :    
807 :     return 0;
808 :     }
809 :    
810 : edgomez 1382 static __inline uint32_t
811 :     MakeGoodMotionFlags(const uint32_t MotionFlags, const uint32_t VopFlags, const uint32_t VolFlags)
812 :     {
813 :     uint32_t Flags = MotionFlags;
814 :    
815 :     if (!(VopFlags & XVID_VOP_MODEDECISION_RD))
816 :     Flags &= ~(XVID_ME_QUARTERPELREFINE16_RD+XVID_ME_QUARTERPELREFINE8_RD+XVID_ME_HALFPELREFINE16_RD+XVID_ME_HALFPELREFINE8_RD+XVID_ME_EXTSEARCH_RD);
817 :    
818 :     if (Flags & XVID_ME_EXTSEARCH_RD)
819 :     Flags |= XVID_ME_HALFPELREFINE16_RD;
820 :    
821 :     if (Flags & XVID_ME_EXTSEARCH_RD && MotionFlags & XVID_ME_EXTSEARCH8)
822 :     Flags |= XVID_ME_HALFPELREFINE8_RD;
823 :    
824 :     if (Flags & XVID_ME_HALFPELREFINE16_RD)
825 :     Flags |= XVID_ME_QUARTERPELREFINE16_RD;
826 :    
827 :     if (Flags & XVID_ME_HALFPELREFINE8_RD) {
828 :     Flags |= XVID_ME_QUARTERPELREFINE8_RD;
829 :     Flags &= ~XVID_ME_HALFPELREFINE8;
830 :     }
831 :    
832 :     if (Flags & XVID_ME_QUARTERPELREFINE8_RD)
833 :     Flags &= ~XVID_ME_QUARTERPELREFINE8;
834 :    
835 : edgomez 1423 if (Flags & XVID_ME_QUARTERPELREFINE16_RD)
836 :     Flags &= ~XVID_ME_QUARTERPELREFINE16;
837 :    
838 : edgomez 1382 if (!(VolFlags & XVID_VOL_QUARTERPEL))
839 :     Flags &= ~(XVID_ME_QUARTERPELREFINE16+XVID_ME_QUARTERPELREFINE8+XVID_ME_QUARTERPELREFINE16_RD+XVID_ME_QUARTERPELREFINE8_RD);
840 :    
841 :     if (!(VopFlags & XVID_VOP_HALFPEL))
842 :     Flags &= ~(XVID_ME_EXTSEARCH16+XVID_ME_HALFPELREFINE16+XVID_ME_HALFPELREFINE8+XVID_ME_HALFPELREFINE16_RD+XVID_ME_HALFPELREFINE8_RD);
843 :    
844 : syskin 1564 if (VopFlags & XVID_VOP_GREYSCALE)
845 : edgomez 1382 Flags &= ~(XVID_ME_CHROMA_PVOP + XVID_ME_CHROMA_BVOP);
846 :    
847 : syskin 1441 if (Flags & XVID_ME_FASTREFINE8)
848 :     Flags &= ~XVID_ME_HALFPELREFINE8_RD;
849 :    
850 :     if (Flags & XVID_ME_FASTREFINE16)
851 :     Flags &= ~XVID_ME_HALFPELREFINE16_RD;
852 :    
853 : edgomez 1382 return Flags;
854 :     }
855 :    
856 : syskin 1567 static __inline void
857 :     motionStatsPVOP(int * const MVmax, int * const mvCount, int * const mvSum,
858 :     const MACROBLOCK * const pMB, const int qpel)
859 :     {
860 :     const VECTOR * const mv = qpel ? pMB->qmvs : pMB->mvs;
861 :     int i;
862 :     int max = *MVmax;
863 :    
864 :     switch (pMB->mode) {
865 :     case MODE_INTER4V:
866 :     *mvCount += 3;
867 :     for(i = 3; i; i--) {
868 :     if (mv[i].x > max) max = mv[i].x;
869 :     else if (-mv[i].x - 1 > max) max = -mv[i].x - 1;
870 :     *mvSum += mv[i].x * mv[i].x;
871 :     if (mv[i].y > max) max = mv[i].y;
872 :     else if (-mv[i].y - 1 > max) max = -mv[i].y - 1;
873 :     *mvSum += mv[i].y * mv[i].y;
874 :     }
875 :     case MODE_INTER:
876 : syskin 1600 (*mvCount)++;
877 : syskin 1567 *mvSum += mv[0].x * mv[0].x;
878 :     *mvSum += mv[0].y * mv[0].y;
879 : syskin 1600 if (mv[0].x > max) max = mv[0].x;
880 :     else if (-mv[0].x - 1 > max) max = -mv[0].x - 1;
881 :     if (mv[0].y > max) max = mv[0].y;
882 :     else if (-mv[0].y - 1 > max) max = -mv[0].y - 1;
883 :     *MVmax = max;
884 : syskin 1567 default:
885 :     break;
886 :     }
887 :     }
888 :    
889 : syskin 1668 void
890 : edgomez 1382 MotionEstimation(MBParam * const pParam,
891 :     FRAMEINFO * const current,
892 :     FRAMEINFO * const reference,
893 :     const IMAGE * const pRefH,
894 :     const IMAGE * const pRefV,
895 :     const IMAGE * const pRefHV,
896 : syskin 1668 const IMAGE * const pGMC,
897 : edgomez 1382 const uint32_t iLimit)
898 :     {
899 :     MACROBLOCK *const pMBs = current->mbs;
900 :     const IMAGE *const pCurrent = &current->image;
901 :     const IMAGE *const pRef = &reference->image;
902 :    
903 : syskin 1564 const uint32_t mb_width = pParam->mb_width;
904 :     const uint32_t mb_height = pParam->mb_height;
905 : edgomez 1382 const uint32_t iEdgedWidth = pParam->edged_width;
906 :     const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags);
907 :     int stat_thresh = 0;
908 : syskin 1567 int MVmax = 0, mvSum = 0, mvCount = 0;
909 : edgomez 1382
910 :     uint32_t x, y;
911 : syskin 1668 int sad00;
912 : edgomez 1382 int skip_thresh = INITIAL_SKIP_THRESH * \
913 :     (current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1);
914 : syskin 1668 int block = 0;
915 : edgomez 1382
916 :     /* some pre-initialized thingies for SearchP */
917 :     DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE);
918 :     SearchData Data;
919 :     memset(&Data, 0, sizeof(SearchData));
920 :     Data.iEdgedWidth = iEdgedWidth;
921 :     Data.iFcode = current->fcode;
922 :     Data.rounding = pParam->m_rounding_type;
923 :     Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0);
924 :     Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP;
925 :     Data.dctSpace = dct_space;
926 :     Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT);
927 :     Data.mpeg_quant_matrices = pParam->mpeg_quant_matrices;
928 :    
929 :     Data.RefQ = pRefV->u; /* a good place, also used in MC (for similar purpose) */
930 :     if (sadInit) (*sadInit) ();
931 :    
932 :     for (y = 0; y < mb_height; y++) {
933 :     for (x = 0; x < mb_width; x++) {
934 : syskin 1668 MACROBLOCK *pMB = &pMBs[block];
935 :     MACROBLOCK *prevMB = &reference->mbs[block];
936 :     int skip;
937 :     block++;
938 : edgomez 1382
939 : syskin 1564 pMB->sad16 =
940 : edgomez 1382 sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
941 :     pRef->y + (x + y * iEdgedWidth) * 16,
942 : syskin 1564 pParam->edged_width, pMB->sad8);
943 : syskin 1668
944 : syskin 1443 sad00 = 4*MAX(MAX(pMB->sad8[0], pMB->sad8[1]), MAX(pMB->sad8[2], pMB->sad8[3]));
945 : edgomez 1382
946 :     if (Data.chroma) {
947 :     Data.chromaSAD = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8,
948 :     pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2)
949 :     + sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8,
950 :     pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
951 :     pMB->sad16 += Data.chromaSAD;
952 : syskin 1443 sad00 += Data.chromaSAD;
953 : edgomez 1382 }
954 :    
955 : syskin 1668 skip = InitialSkipDecisionP(sad00, pParam, current, pMB, prevMB, x, y, &Data, pGMC,
956 :     pCurrent, pRef, MotionFlags);
957 :     if (skip) continue;
958 : edgomez 1382
959 :     SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
960 :     y, MotionFlags, current->vop_flags,
961 :     &Data, pParam, pMBs, reference->mbs, pMB);
962 :    
963 :     if (current->vop_flags & XVID_VOP_MODEDECISION_RD)
964 :     xvid_me_ModeDecision_RD(&Data, pMB, pMBs, x, y, pParam,
965 :     MotionFlags, current->vop_flags, current->vol_flags,
966 :     pCurrent, pRef, pGMC, current->coding_type);
967 :    
968 :     else if (current->vop_flags & XVID_VOP_FAST_MODEDECISION_RD)
969 :     xvid_me_ModeDecision_Fast(&Data, pMB, pMBs, x, y, pParam,
970 :     MotionFlags, current->vop_flags, current->vol_flags,
971 :     pCurrent, pRef, pGMC, current->coding_type);
972 :     else
973 :     ModeDecision_SAD(&Data, pMB, pMBs, x, y, pParam,
974 :     MotionFlags, current->vop_flags, current->vol_flags,
975 : syskin 1443 pCurrent, pRef, pGMC, current->coding_type, sad00);
976 : edgomez 1382
977 :    
978 : syskin 1567 motionStatsPVOP(&MVmax, &mvCount, &mvSum, pMB, Data.qpel);
979 : edgomez 1382 }
980 :     }
981 :    
982 : syskin 1567 current->fcode = getMinFcode(MVmax);
983 :     current->sStat.iMvSum = mvSum;
984 :     current->sStat.iMvCount = mvCount;
985 : edgomez 1570 }
986 : syskin 1682
987 :     void
988 :     MotionEstimateSMP(SMPmotionData * h)
989 :     {
990 :     const MBParam * const pParam = h->pParam;
991 :     const FRAMEINFO * const current = h->current;
992 :     const FRAMEINFO * const reference = h->reference;
993 :     const IMAGE * const pRefH = h->pRefH;
994 :     const IMAGE * const pRefV = h->pRefV;
995 :     const IMAGE * const pRefHV = h->pRefHV;
996 :     const IMAGE * const pGMC = h->pGMC;
997 :     uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags,
998 :     current->vop_flags,
999 :     current->vol_flags);
1000 :    
1001 :     MACROBLOCK *const pMBs = current->mbs;
1002 :     const IMAGE *const pCurrent = &current->image;
1003 :     const IMAGE *const pRef = &reference->image;
1004 :    
1005 :     const uint32_t mb_width = pParam->mb_width;
1006 :     const uint32_t mb_height = pParam->mb_height;
1007 :     const uint32_t iEdgedWidth = pParam->edged_width;
1008 :     int stat_thresh = 0;
1009 :     int MVmax = 0, mvSum = 0, mvCount = 0;
1010 :     int y_step = h->y_step;
1011 :     int start_y = h->start_y;
1012 :    
1013 :     uint32_t x, y;
1014 :     int sad00;
1015 :     int skip_thresh = INITIAL_SKIP_THRESH * \
1016 :     (current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1);
1017 :     int block = start_y*mb_width;
1018 :     int * complete_count_self = h->complete_count_self;
1019 : syskin 1691 const volatile int * complete_count_above = h->complete_count_above;
1020 : syskin 1682 int max_mbs;
1021 :     int current_mb = 0;
1022 :    
1023 :     /* some pre-initialized thingies for SearchP */
1024 :     DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE);
1025 :     SearchData Data;
1026 :     memset(&Data, 0, sizeof(SearchData));
1027 :     Data.iEdgedWidth = iEdgedWidth;
1028 :     Data.iFcode = current->fcode;
1029 :     Data.rounding = pParam->m_rounding_type;
1030 :     Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0);
1031 :     Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP;
1032 :     Data.dctSpace = dct_space;
1033 :     Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT);
1034 :     Data.mpeg_quant_matrices = pParam->mpeg_quant_matrices;
1035 :    
1036 :     /* todo: sort out temp memory space */
1037 :     Data.RefQ = h->RefQ;
1038 :     if (sadInit) (*sadInit) ();
1039 :    
1040 :     max_mbs = 0;
1041 :    
1042 :     for (y = start_y; y < mb_height; y += y_step) {
1043 :     if (y == 0) max_mbs = mb_width; /* we can process all blocks of the first row */
1044 :    
1045 :     for (x = 0; x < mb_width; x++) {
1046 :    
1047 :     MACROBLOCK *pMB, *prevMB;
1048 :     int skip;
1049 :    
1050 :     if (current_mb >= max_mbs) {
1051 :     /* we ME-ed all macroblocks we safely could. grab next portion */
1052 :     int above_count = *complete_count_above; /* sync point */
1053 :     if (above_count == mb_width) {
1054 :     /* full line above is ready */
1055 :     above_count = mb_width+1;
1056 :     if (y < mb_height-y_step) {
1057 :     /* this is not last line, grab a portion of MBs from the next line too */
1058 :     above_count += MAX(0, complete_count_above[1] - 1);
1059 :     }
1060 :     }
1061 :    
1062 :     max_mbs = current_mb + above_count - x - 1;
1063 :    
1064 :     if (current_mb >= max_mbs) {
1065 :     /* current workload is zero */
1066 :     x--;
1067 :     sched_yield();
1068 :     continue;
1069 :     }
1070 :     }
1071 :    
1072 :    
1073 : syskin 1691 pMB = &pMBs[block];
1074 :     prevMB = &reference->mbs[block];
1075 :    
1076 :     pMB->sad16 =
1077 :     sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16,
1078 :     pRef->y + (x + y * iEdgedWidth) * 16,
1079 :     pParam->edged_width, pMB->sad8);
1080 :    
1081 :     sad00 = 4*MAX(MAX(pMB->sad8[0], pMB->sad8[1]), MAX(pMB->sad8[2], pMB->sad8[3]));
1082 :    
1083 :     if (Data.chroma) {
1084 :     Data.chromaSAD = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8,
1085 :     pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2)
1086 :     + sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8,
1087 :     pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
1088 :     pMB->sad16 += Data.chromaSAD;
1089 :     sad00 += Data.chromaSAD;
1090 : syskin 1682 }
1091 :    
1092 : syskin 1691 skip = InitialSkipDecisionP(sad00, pParam, current, pMB, prevMB, x, y, &Data, pGMC,
1093 :     pCurrent, pRef, MotionFlags);
1094 :    
1095 : syskin 1682 if (skip) {
1096 :     current_mb++;
1097 :     block++;
1098 :     *complete_count_self = x+1;
1099 :     continue;
1100 :     }
1101 :    
1102 :     SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
1103 :     y, MotionFlags, current->vop_flags,
1104 :     &Data, pParam, pMBs, reference->mbs, pMB);
1105 :    
1106 :     if (current->vop_flags & XVID_VOP_MODEDECISION_RD)
1107 :     xvid_me_ModeDecision_RD(&Data, pMB, pMBs, x, y, pParam,
1108 :     MotionFlags, current->vop_flags, current->vol_flags,
1109 :     pCurrent, pRef, pGMC, current->coding_type);
1110 :    
1111 :     else if (current->vop_flags & XVID_VOP_FAST_MODEDECISION_RD)
1112 :     xvid_me_ModeDecision_Fast(&Data, pMB, pMBs, x, y, pParam,
1113 :     MotionFlags, current->vop_flags, current->vol_flags,
1114 :     pCurrent, pRef, pGMC, current->coding_type);
1115 :     else
1116 :     ModeDecision_SAD(&Data, pMB, pMBs, x, y, pParam,
1117 :     MotionFlags, current->vop_flags, current->vol_flags,
1118 :     pCurrent, pRef, pGMC, current->coding_type, sad00);
1119 :    
1120 :     *complete_count_self = x+1;
1121 :    
1122 :     current_mb++;
1123 :     block++;
1124 :    
1125 :     motionStatsPVOP(&MVmax, &mvCount, &mvSum, pMB, Data.qpel);
1126 :    
1127 :     }
1128 :     block += (y_step-1)*pParam->mb_width;
1129 :     complete_count_self++;
1130 :     complete_count_above++;
1131 :     }
1132 :    
1133 : syskin 1687 h->minfcode = getMinFcode(MVmax);
1134 :    
1135 : syskin 1682 h->MVmax = MVmax;
1136 :     h->mvSum = mvSum;
1137 :     h->mvCount = mvCount;
1138 :     }
1139 :    
1140 :    
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4