[svn] / branches / release-1_3-branch / xvidcore / src / motion / estimation_rd_based.c Repository:
ViewVC logotype

Annotation of /branches/release-1_3-branch/xvidcore/src/motion/estimation_rd_based.c

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1509 - (view) (download)
Original Path: trunk/xvidcore/src/motion/estimation_rd_based.c

1 : edgomez 1382 /*****************************************************************************
2 :     *
3 :     * XVID MPEG-4 VIDEO CODEC
4 :     * - Rate-Distortion Based Motion Estimation for P- and S- VOPs -
5 :     *
6 :     * Copyright(C) 2003 Radoslaw Czyz <xvid@syskin.cjb.net>
7 :     * 2003 Michael Militzer <michael@xvid.org>
8 :     *
9 :     * This program is free software ; you can redistribute it and/or modify
10 :     * it under the terms of the GNU General Public License as published by
11 :     * the Free Software Foundation ; either version 2 of the License, or
12 :     * (at your option) any later version.
13 :     *
14 :     * This program is distributed in the hope that it will be useful,
15 :     * but WITHOUT ANY WARRANTY ; without even the implied warranty of
16 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 :     * GNU General Public License for more details.
18 :     *
19 :     * You should have received a copy of the GNU General Public License
20 :     * along with this program ; if not, write to the Free Software
21 :     * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 :     *
23 : syskin 1509 * $Id: estimation_rd_based.c,v 1.8 2004-07-18 12:23:45 syskin Exp $
24 : edgomez 1382 *
25 :     ****************************************************************************/
26 :    
27 :     /* RD mode decision and search */
28 :    
29 :     #include <assert.h>
30 :     #include <stdio.h>
31 :     #include <stdlib.h>
32 :     #include <string.h> /* memcpy */
33 :    
34 :     #include "../encoder.h"
35 :     #include "../bitstream/mbcoding.h"
36 :     #include "../prediction/mbprediction.h"
37 :     #include "../global.h"
38 :     #include "../image/interpolate8x8.h"
39 :     #include "estimation.h"
40 :     #include "motion.h"
41 :     #include "sad.h"
42 :     #include "../bitstream/zigzag.h"
43 :     #include "../quant/quant.h"
44 :     #include "../bitstream/vlc_codes.h"
45 :     #include "../dct/fdct.h"
46 :     #include "motion_inlines.h"
47 :    
48 :     /* rd = BITS_MULT*bits + LAMBDA*distortion */
49 :     #define LAMBDA ( (int)(BITS_MULT*1.0) )
50 :    
51 :     static __inline unsigned int
52 :     Block_CalcBits( int16_t * const coeff,
53 :     int16_t * const data,
54 :     int16_t * const dqcoeff,
55 :     const uint32_t quant, const int quant_type,
56 :     uint32_t * cbp,
57 :     const int block,
58 :     const uint16_t * scan_table,
59 : syskin 1506 const unsigned int lambda,
60 : edgomez 1382 const uint16_t * mpeg_quant_matrices)
61 :     {
62 :     int sum;
63 :     int bits;
64 :     int distortion = 0;
65 :    
66 :     fdct(data);
67 :    
68 :     if (quant_type) sum = quant_h263_inter(coeff, data, quant, mpeg_quant_matrices);
69 :     else sum = quant_mpeg_inter(coeff, data, quant, mpeg_quant_matrices);
70 :    
71 :     if (sum > 0) {
72 :     *cbp |= 1 << (5 - block);
73 :     bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table);
74 :    
75 :     if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
76 :     else dequant_mpeg_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
77 :    
78 :     distortion = sse8_16bit(data, dqcoeff, 8*sizeof(int16_t));
79 :     } else {
80 :     const static int16_t zero_block[64] =
81 :     {
82 :     0, 0, 0, 0, 0, 0, 0, 0,
83 :     0, 0, 0, 0, 0, 0, 0, 0,
84 :     0, 0, 0, 0, 0, 0, 0, 0,
85 :     0, 0, 0, 0, 0, 0, 0, 0,
86 :     0, 0, 0, 0, 0, 0, 0, 0,
87 :     0, 0, 0, 0, 0, 0, 0, 0,
88 :     0, 0, 0, 0, 0, 0, 0, 0,
89 :     0, 0, 0, 0, 0, 0, 0, 0,
90 :     };
91 :     bits = 0;
92 :     distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t));
93 :     }
94 :    
95 :    
96 : syskin 1506 return bits + (lambda*distortion)/(quant*quant);
97 : edgomez 1382 }
98 :    
99 :     static __inline unsigned int
100 :     Block_CalcBitsIntra(MACROBLOCK * pMB,
101 :     const unsigned int x,
102 :     const unsigned int y,
103 :     const unsigned int mb_width,
104 :     const uint32_t block,
105 :     int16_t coeff[64],
106 :     int16_t qcoeff[64],
107 :     int16_t dqcoeff[64],
108 :     int16_t predictors[8],
109 :     const uint32_t quant,
110 :     const int quant_type,
111 :     unsigned int bits[2],
112 :     unsigned int cbp[2],
113 : syskin 1506 unsigned int lambda,
114 : edgomez 1382 const uint16_t * mpeg_quant_matrices)
115 :     {
116 :     int direction;
117 :     int16_t *pCurrent;
118 :     unsigned int i, coded;
119 :     unsigned int distortion = 0;
120 :     const uint32_t iDcScaler = get_dc_scaler(quant, block < 4);
121 :    
122 :     fdct(coeff);
123 :    
124 :     if (quant_type) {
125 :     quant_h263_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
126 :     dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
127 :     } else {
128 :     quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
129 :     dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
130 :     }
131 :    
132 :     predict_acdc(pMB-(x+mb_width*y), x, y, mb_width, block, qcoeff,
133 : edgomez 1472 quant, iDcScaler, predictors, 0);
134 : edgomez 1382
135 :     direction = pMB->acpred_directions[block];
136 :     pCurrent = pMB->pred_values[block];
137 :    
138 :     /* store current coeffs to pred_values[] for future prediction */
139 :     pCurrent[0] = qcoeff[0] * iDcScaler;
140 : edgomez 1472 pCurrent[0] = CLIP(pCurrent[0], -2048, 2047);
141 : edgomez 1382 for (i = 1; i < 8; i++) {
142 :     pCurrent[i] = qcoeff[i];
143 :     pCurrent[i + 7] = qcoeff[i * 8];
144 :     }
145 :    
146 :     /* dc prediction */
147 :     qcoeff[0] = qcoeff[0] - predictors[0];
148 :    
149 :     if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len;
150 :     else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len;
151 :    
152 :     /* calc cost before ac prediction */
153 :     bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]);
154 :     if (coded > 0) cbp[0] |= 1 << (5 - block);
155 :    
156 :     /* apply ac prediction & calc cost*/
157 :     if (direction == 1) {
158 :     for (i = 1; i < 8; i++) {
159 :     qcoeff[i] -= predictors[i];
160 :     predictors[i] = qcoeff[i];
161 :     }
162 :     } else { /* acpred_direction == 2 */
163 :     for (i = 1; i < 8; i++) {
164 :     qcoeff[i*8] -= predictors[i];
165 :     predictors[i] = qcoeff[i*8];
166 :     }
167 :     }
168 :    
169 :     bits[1] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[direction]);
170 :     if (coded > 0) cbp[1] |= 1 << (5 - block);
171 :    
172 :     distortion = sse8_16bit(coeff, dqcoeff, 8*sizeof(int16_t));
173 :    
174 : syskin 1506 return (lambda*distortion)/(quant*quant);
175 : edgomez 1382 }
176 :    
177 :    
178 :    
179 :     static void
180 :     CheckCandidateRD16(const int x, const int y, SearchData * const data, const unsigned int Direction)
181 :     {
182 :    
183 :     int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
184 :     int32_t rd = 0;
185 :     VECTOR * current;
186 :     const uint8_t * ptr;
187 :     int i, t, xc, yc;
188 :     unsigned cbp = 0;
189 :    
190 :     if ( (x > data->max_dx) || (x < data->min_dx)
191 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
192 :    
193 :     if (!data->qpel_precision) {
194 :     ptr = GetReference(x, y, data);
195 :     current = data->currentMV;
196 :     xc = x; yc = y;
197 :     } else { /* x and y are in 1/4 precision */
198 :     ptr = xvid_me_interpolate16x16qpel(x, y, 0, data);
199 :     current = data->currentQMV;
200 :     xc = x/2; yc = y/2;
201 :     }
202 :    
203 :     for(i = 0; i < 4; i++) {
204 :     int s = 8*((i&1) + (i>>1)*data->iEdgedWidth);
205 :     transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth);
206 : syskin 1506 rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices);
207 : edgomez 1382 }
208 :    
209 :     rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
210 :    
211 :     if (data->temp[0] + t < data->iMinSAD[1]) {
212 :     data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); }
213 :     if (data->temp[1] < data->iMinSAD[2]) {
214 :     data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); }
215 :     if (data->temp[2] < data->iMinSAD[3]) {
216 :     data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); }
217 :     if (data->temp[3] < data->iMinSAD[4]) {
218 :     data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); }
219 :    
220 :     rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
221 :    
222 :     if (rd >= data->iMinSAD[0]) return;
223 :    
224 :     /* chroma */
225 :     xc = (xc >> 1) + roundtab_79[xc & 0x3];
226 :     yc = (yc >> 1) + roundtab_79[yc & 0x3];
227 :    
228 :     /* chroma U */
229 :     ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
230 :     transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2);
231 : syskin 1506 rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices);
232 : edgomez 1382 if (rd >= data->iMinSAD[0]) return;
233 :    
234 :     /* chroma V */
235 :     ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
236 :     transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2);
237 : syskin 1506 rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices);
238 : edgomez 1382
239 :     rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
240 :    
241 :     if (rd < data->iMinSAD[0]) {
242 :     data->iMinSAD[0] = rd;
243 :     current[0].x = x; current[0].y = y;
244 :     data->dir = Direction;
245 :     *data->cbp = cbp;
246 :     }
247 :     }
248 :    
249 :     static void
250 :     CheckCandidateRD8(const int x, const int y, SearchData * const data, const unsigned int Direction)
251 :     {
252 :    
253 :     int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
254 :     int32_t rd;
255 :     VECTOR * current;
256 :     const uint8_t * ptr;
257 :     unsigned int cbp = 0;
258 :    
259 :     if ( (x > data->max_dx) || (x < data->min_dx)
260 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
261 :    
262 :     if (!data->qpel_precision) {
263 :     ptr = GetReference(x, y, data);
264 :     current = data->currentMV;
265 :     } else { /* x and y are in 1/4 precision */
266 :     ptr = xvid_me_interpolate8x8qpel(x, y, 0, 0, data);
267 :     current = data->currentQMV;
268 :     }
269 :    
270 :     transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
271 : syskin 1506 rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->lambda[0], data->mpeg_quant_matrices);
272 : edgomez 1382 rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
273 :    
274 :     if (rd < data->iMinSAD[0]) {
275 :     *data->cbp = cbp;
276 :     data->iMinSAD[0] = rd;
277 :     current[0].x = x; current[0].y = y;
278 :     data->dir = Direction;
279 :     }
280 :     }
281 :    
282 :    
283 :     static int
284 :     findRD_inter(SearchData * const Data,
285 :     const int x, const int y,
286 :     const MBParam * const pParam,
287 :     const uint32_t MotionFlags)
288 :     {
289 :     int i;
290 :     int32_t bsad[5];
291 :    
292 :     if (Data->qpel) {
293 :     for(i = 0; i < 5; i++) {
294 :     Data->currentMV[i].x = Data->currentQMV[i].x/2;
295 :     Data->currentMV[i].y = Data->currentQMV[i].y/2;
296 :     }
297 :     Data->qpel_precision = 1;
298 :     CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255);
299 :    
300 :     if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */
301 :     for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i];
302 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
303 :     pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
304 :     Data->qpel_precision = 0;
305 :     if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1)
306 :     CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
307 :     }
308 :    
309 :     } else { /* not qpel */
310 :    
311 :     CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
312 :     }
313 :    
314 :     if (MotionFlags&XVID_ME_EXTSEARCH_RD)
315 :     xvid_me_SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16);
316 :    
317 :     if (MotionFlags&XVID_ME_HALFPELREFINE16_RD)
318 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidateRD16, 0);
319 : edgomez 1382
320 :     if (Data->qpel) {
321 :     if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */
322 :     for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) {
323 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */
324 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
325 :     }
326 :    
327 :     /* preparing for qpel-precision search */
328 :     Data->qpel_precision = 1;
329 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
330 :     pParam->width, pParam->height, Data->iFcode, 2, 0);
331 :     }
332 : syskin 1441 if (MotionFlags & XVID_ME_QUARTERPELREFINE16_RD) {
333 :     if (MotionFlags & XVID_ME_FASTREFINE16)
334 :     FullRefine_Fast(Data, CheckCandidateRD16, 0);
335 :     else
336 : syskin 1478 xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateRD16, 0);
337 : syskin 1441 }
338 : edgomez 1382 }
339 :    
340 :     if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */
341 :     VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV;
342 :     if (!MVequal(Data->predMV, *v))
343 :     CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255);
344 :     }
345 :     return Data->iMinSAD[0];
346 :     }
347 :    
348 :     static int
349 :     findRD_inter4v(SearchData * const Data,
350 :     MACROBLOCK * const pMB, const MACROBLOCK * const pMBs,
351 :     const int x, const int y,
352 :     const MBParam * const pParam, const uint32_t MotionFlags,
353 :     const VECTOR * const backup)
354 :     {
355 :    
356 :     unsigned int cbp = 0, bits = 0, t = 0, i;
357 :     SearchData Data2, *Data8 = &Data2;
358 :     int sumx = 0, sumy = 0;
359 :     int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64;
360 :     uint8_t * ptr;
361 :    
362 :     memcpy(Data8, Data, sizeof(SearchData));
363 :    
364 :     for (i = 0; i < 4; i++) { /* for all luma blocks */
365 :    
366 :     *Data8->iMinSAD = *(Data->iMinSAD + i + 1);
367 :     *Data8->currentMV = *(Data->currentMV + i + 1);
368 :     *Data8->currentQMV = *(Data->currentQMV + i + 1);
369 :     Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
370 :     Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
371 :     Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
372 :     Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
373 :     Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
374 :     *Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; /* copy corresponding cbp bit */
375 : syskin 1506 Data8->lambda[0] = Data->lambda[i];
376 : edgomez 1382
377 :     if(Data->qpel) {
378 :     Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i);
379 :     if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y,
380 :     Data8->predMV, Data8->iFcode, 0, 0);
381 :     } else {
382 :     Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i);
383 :     if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y,
384 :     Data8->predMV, Data8->iFcode, 0, 0);
385 :     }
386 :    
387 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
388 :     pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0);
389 :    
390 :     *Data8->iMinSAD += BITS_MULT*t;
391 :    
392 :     Data8->qpel_precision = Data8->qpel;
393 :     /* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */
394 :     {
395 :     VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV;
396 :     if (!MVequal (*v, backup[i+1]) )
397 :     CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255);
398 :     }
399 :    
400 :     if (Data8->qpel) {
401 : syskin 1441 int bsad = Data8->iMinSAD[0];
402 :     int bx = Data8->currentQMV->x;
403 :     int by = Data8->currentQMV->y;
404 :    
405 :     Data8->currentMV->x = Data8->currentQMV->x/2;
406 :     Data8->currentMV->y = Data8->currentQMV->y/2;
407 :    
408 : edgomez 1382 if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */
409 :     Data8->qpel_precision = 0;
410 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
411 :     pParam->width, pParam->height, Data8->iFcode - 1, 1, 0);
412 :    
413 :     if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1)
414 :     CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255);
415 :    
416 :     if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD)
417 :     xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
418 :    
419 :     if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
420 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0);
421 : edgomez 1382
422 : syskin 1441 if(bsad > *Data8->iMinSAD) { /* we have found a better match */
423 :     bx = Data8->currentQMV->x = 2*Data8->currentMV->x;
424 :     by = Data8->currentQMV->y = 2*Data8->currentMV->y;
425 :     bsad = Data8->iMinSAD[0];
426 : edgomez 1382 }
427 :    
428 :     Data8->qpel_precision = 1;
429 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
430 :     pParam->width, pParam->height, Data8->iFcode, 2, 0);
431 :    
432 :     }
433 :    
434 : syskin 1441 if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) {
435 :     if (MotionFlags & XVID_ME_FASTREFINE8)
436 :     FullRefine_Fast(Data8, CheckCandidateRD8, 0);
437 : syskin 1478 else xvid_me_SubpelRefine(Data->currentQMV[0], Data8, CheckCandidateRD8, 0);
438 : syskin 1441 }
439 :    
440 :     if (bsad <= Data->iMinSAD[0]) {
441 :     /* we have not found a better match */
442 :     Data8->iMinSAD[0] = bsad;
443 :     Data8->currentQMV->x = bx;
444 :     Data8->currentQMV->y = by;
445 :     }
446 :    
447 : edgomez 1382 } else { /* not qpel */
448 :    
449 :     if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */
450 :     xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
451 :    
452 :     if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
453 : syskin 1478 xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0); /* halfpel refinement */
454 : edgomez 1382 }
455 :    
456 :     /* checking vector equal to predicion */
457 :     if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) {
458 :     const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV;
459 :     if (!MVequal(*v, Data8->predMV))
460 :     CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255);
461 :     }
462 :    
463 :     bits += *Data8->iMinSAD;
464 :     if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */
465 :    
466 :     /* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */
467 :     if(Data->qpel) {
468 :     pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x;
469 :     pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y;
470 :     pMB->qmvs[i] = *Data8->currentQMV;
471 :     sumx += Data8->currentQMV->x/2;
472 :     sumy += Data8->currentQMV->y/2;
473 :     } else {
474 :     pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x;
475 :     pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y;
476 :     sumx += Data8->currentMV->x;
477 :     sumy += Data8->currentMV->y;
478 :     }
479 :     pMB->mvs[i] = *Data8->currentMV;
480 :     pMB->sad8[i] = 4 * *Data8->iMinSAD;
481 :     if (Data8->cbp[0]) cbp |= 1 << (5 - i);
482 :    
483 :     } /* end - for all luma blocks */
484 :    
485 :     bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
486 :    
487 :     /* let's check chroma */
488 :     sumx = (sumx >> 3) + roundtab_76[sumx & 0xf];
489 :     sumy = (sumy >> 3) + roundtab_76[sumy & 0xf];
490 :    
491 :     /* chroma U */
492 :     ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
493 :     transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2);
494 : syskin 1506 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices);
495 : edgomez 1382
496 :     if (bits >= *Data->iMinSAD) return bits;
497 :    
498 :     /* chroma V */
499 :     ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
500 :     transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2);
501 : syskin 1506 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices);
502 : edgomez 1382
503 :     bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len;
504 :    
505 :     *Data->cbp = cbp;
506 :     return bits;
507 :     }
508 :    
509 :     static int
510 :     findRD_intra(SearchData * const Data, MACROBLOCK * pMB,
511 :     const int x, const int y, const int mb_width)
512 :     {
513 :     unsigned int cbp[2] = {0, 0}, bits[2], i;
514 :     unsigned int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */
515 :     unsigned int distortion = 0;
516 :    
517 :     int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128;
518 :     const uint32_t iQuant = Data->iQuant;
519 :     int16_t predictors[6][8];
520 :    
521 :     for(i = 0; i < 4; i++) {
522 :     int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth);
523 :     transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth);
524 :    
525 :    
526 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff,
527 : syskin 1506 predictors[i], iQuant, Data->quant_type, bits, cbp, Data->lambda[i], Data->mpeg_quant_matrices);
528 : edgomez 1382 bits1 += distortion + BITS_MULT * bits[0];
529 :     bits2 += distortion + BITS_MULT * bits[1];
530 :    
531 :     if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0])
532 :     return bits1;
533 :     }
534 :    
535 :     bits1 += BITS_MULT*xvid_cbpy_tab[cbp[0]>>2].len;
536 :     bits2 += BITS_MULT*xvid_cbpy_tab[cbp[1]>>2].len;
537 :    
538 :     /*chroma U */
539 :     transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
540 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff,
541 : syskin 1506 predictors[4], iQuant, Data->quant_type, bits, cbp, Data->lambda[4], Data->mpeg_quant_matrices);
542 : edgomez 1382 bits1 += distortion + BITS_MULT * bits[0];
543 :     bits2 += distortion + BITS_MULT * bits[1];
544 :    
545 :     if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0])
546 :     return bits1;
547 :    
548 :     /* chroma V */
549 :     transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
550 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff,
551 : syskin 1506 predictors[5], iQuant, Data->quant_type, bits, cbp, Data->lambda[5], Data->mpeg_quant_matrices);
552 : edgomez 1382
553 :     bits1 += distortion + BITS_MULT * bits[0];
554 :     bits2 += distortion + BITS_MULT * bits[1];
555 :    
556 :     bits1 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[0] & 3) << 3)].len;
557 :     bits2 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[1] & 3) << 3)].len;
558 :    
559 :     *Data->cbp = bits1 <= bits2 ? cbp[0] : cbp[1];
560 :    
561 :     return MIN(bits1, bits2);
562 :     }
563 :    
564 :    
565 :     static int
566 :     findRD_gmc(SearchData * const Data, const IMAGE * const vGMC, const int x, const int y)
567 :     {
568 :     int bits = BITS_MULT*1; /* this one is mcsel */
569 :     unsigned int cbp = 0, i;
570 :     int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64;
571 :    
572 :     for(i = 0; i < 4; i++) {
573 :     int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth);
574 :     transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth);
575 : syskin 1506 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, Data->scan_table, Data->lambda[i], Data->mpeg_quant_matrices);
576 : edgomez 1382 if (bits >= Data->iMinSAD[0]) return bits;
577 :     }
578 :    
579 :     bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
580 :    
581 :     /*chroma U */
582 :     transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
583 : syskin 1506 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices);
584 : edgomez 1382
585 :     if (bits >= Data->iMinSAD[0]) return bits;
586 :    
587 :     /* chroma V */
588 :     transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
589 : syskin 1506 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices);
590 : edgomez 1382
591 :     bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
592 :    
593 :     *Data->cbp = cbp;
594 :    
595 :     return bits;
596 :     }
597 :    
598 :     void
599 :     xvid_me_ModeDecision_RD(SearchData * const Data,
600 :     MACROBLOCK * const pMB,
601 :     const MACROBLOCK * const pMBs,
602 :     const int x, const int y,
603 :     const MBParam * const pParam,
604 :     const uint32_t MotionFlags,
605 :     const uint32_t VopFlags,
606 :     const uint32_t VolFlags,
607 :     const IMAGE * const pCurrent,
608 :     const IMAGE * const pRef,
609 :     const IMAGE * const vGMC,
610 :     const int coding_type)
611 :     {
612 :     int mode = MODE_INTER;
613 :     int mcsel = 0;
614 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
615 :     const uint32_t iQuant = pMB->quant;
616 :    
617 :     int min_rd, intra_rd, i, cbp;
618 :     VECTOR backup[5], *v;
619 :     Data->iQuant = iQuant;
620 :     Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ?
621 :     scan_tables[2] : scan_tables[0];
622 :    
623 :     pMB->mcsel = 0;
624 :    
625 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
626 :     for (i = 0; i < 5; i++) {
627 :     Data->iMinSAD[i] = 256*4096;
628 :     backup[i] = v[i];
629 :     }
630 :    
631 : syskin 1506 for (i = 0; i < 6; i++) {
632 :     /* HVS models, anyone ? */
633 :     Data->lambda[i] = LAMBDA;
634 :     }
635 :    
636 : edgomez 1382 min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
637 :     cbp = *Data->cbp;
638 :    
639 :     if (coding_type == S_VOP) {
640 :     int gmc_rd;
641 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
642 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
643 :     if (gmc_rd < min_rd) {
644 :     mcsel = 1;
645 :     *Data->iMinSAD = min_rd = gmc_rd;
646 :     mode = MODE_INTER;
647 :     cbp = *Data->cbp;
648 :     }
649 :     }
650 :    
651 :     if (inter4v) {
652 :     int v4_rd;
653 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
654 :     if (v4_rd < min_rd) {
655 :     Data->iMinSAD[0] = min_rd = v4_rd;
656 :     mode = MODE_INTER4V;
657 :     cbp = *Data->cbp;
658 :     }
659 :     }
660 :    
661 :     intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width);
662 :     if (intra_rd < min_rd) {
663 :     *Data->iMinSAD = min_rd = intra_rd;
664 :     mode = MODE_INTRA;
665 :     cbp = *Data->cbp;
666 :     }
667 :    
668 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0;
669 :     pMB->cbp = cbp;
670 :    
671 :    
672 :     if (Data->rrv) {
673 :     Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
674 :     Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
675 :     }
676 :    
677 :     if (mode == MODE_INTER && mcsel == 0) {
678 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
679 :    
680 :     if(Data->qpel) {
681 :     pMB->qmvs[0] = pMB->qmvs[1]
682 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
683 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
684 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
685 :     } else {
686 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
687 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
688 :     }
689 :    
690 :     } else if (mode == MODE_INTER ) { /* but mcsel == 1 */
691 :    
692 :     pMB->mcsel = 1;
693 :     if (Data->qpel) {
694 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
695 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
696 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
697 :     } else
698 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
699 :    
700 :     } else
701 :     if (mode == MODE_INTER4V) ; /* anything here? */
702 :     else /* INTRA, NOT_CODED */
703 :     ZeroMacroblockP(pMB, 0);
704 :    
705 :     pMB->mode = mode;
706 :     }
707 :    
708 :     void
709 :     xvid_me_ModeDecision_Fast(SearchData * const Data,
710 :     MACROBLOCK * const pMB,
711 :     const MACROBLOCK * const pMBs,
712 :     const int x, const int y,
713 :     const MBParam * const pParam,
714 :     const uint32_t MotionFlags,
715 :     const uint32_t VopFlags,
716 :     const uint32_t VolFlags,
717 :     const IMAGE * const pCurrent,
718 :     const IMAGE * const pRef,
719 :     const IMAGE * const vGMC,
720 :     const int coding_type)
721 :     {
722 :     int mode = MODE_INTER;
723 :     int mcsel = 0;
724 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
725 :     const uint32_t iQuant = pMB->quant;
726 :     const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0);
727 :     int sad;
728 :     int min_rd = -1, intra_rd, i, cbp = 63;
729 :     VECTOR backup[5], *v;
730 :     int sad_backup[5];
731 :     int InterBias = MV16_INTER_BIAS;
732 :     int thresh = 0;
733 :     int top = 0, top_right = 0, left = 0;
734 :     Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ?
735 :     scan_tables[2] : scan_tables[0];
736 :    
737 :     pMB->mcsel = 0;
738 :    
739 : syskin 1509 for (i = 0; i < 6; i++) {
740 :     /* HVS models, anyone ? */
741 :     Data->lambda[i] = LAMBDA;
742 :     }
743 :    
744 : edgomez 1382 /* INTER <-> INTER4V decision */
745 :     if ((Data->iMinSAD[0] + 75 < Data->iMinSAD[1] +
746 :     Data->iMinSAD[2] + Data->iMinSAD[3] + Data->iMinSAD[4])) { /* normal, fast, SAD-based mode decision */
747 :     if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
748 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
749 :     mode = MODE_INTER;
750 :     sad = Data->iMinSAD[0];
751 :     } else {
752 :     mode = MODE_INTER4V;
753 :     sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
754 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
755 :     Data->iMinSAD[0] = sad;
756 :     }
757 :    
758 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
759 :     if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
760 :     if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH)
761 :     if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
762 :     mode = MODE_NOT_CODED;
763 :     sad = 0; /* Compiler warning */
764 :     goto early_out;
765 :     }
766 :    
767 :     /* mcsel */
768 :     if (coding_type == S_VOP) {
769 :    
770 :     int32_t iSAD = sad16(Data->Cur,
771 :     vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
772 :    
773 :     if (Data->chroma) {
774 :     iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
775 :     iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
776 :     }
777 :    
778 :     if (iSAD <= sad) { /* mode decision GMC */
779 :     mode = MODE_INTER;
780 :     mcsel = 1;
781 :     sad = iSAD;
782 :     }
783 :    
784 :     }
785 :     } else { /* Rate-Distortion INTER<->INTER4V */
786 :     Data->iQuant = iQuant;
787 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
788 :    
789 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
790 :     if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
791 :     if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH)
792 :     if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
793 :     mode = MODE_NOT_CODED;
794 :     sad = 0; /* Compiler warning */
795 :     goto early_out;
796 :     }
797 :    
798 :     for (i = 0; i < 5; i++) {
799 :     sad_backup[i] = Data->iMinSAD[i];
800 :     Data->iMinSAD[i] = 256*4096;
801 :     backup[i] = v[i];
802 :     }
803 :    
804 :     min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
805 :     cbp = *Data->cbp;
806 :     sad = sad_backup[0];
807 :    
808 :     if (coding_type == S_VOP) {
809 :     int gmc_rd;
810 :    
811 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
812 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
813 :     if (gmc_rd < min_rd) {
814 :     mcsel = 1;
815 :     *Data->iMinSAD = min_rd = gmc_rd;
816 :     mode = MODE_INTER;
817 :     cbp = *Data->cbp;
818 :     sad = sad16(Data->Cur,
819 :     vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
820 :     if (Data->chroma) {
821 :     sad += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
822 :     sad += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
823 :     }
824 :     }
825 :     }
826 :    
827 :     if (inter4v) {
828 :     int v4_rd;
829 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
830 :     if (v4_rd < min_rd) {
831 :     Data->iMinSAD[0] = min_rd = v4_rd;
832 :     mode = MODE_INTER4V;
833 :     cbp = *Data->cbp;
834 :     sad = sad_backup[1] + sad_backup[2] +
835 :     sad_backup[3] + sad_backup[4] + IMV16X16 * (int32_t)iQuant;
836 :     }
837 :     }
838 :     }
839 :    
840 :     left = top = top_right = -1;
841 :     thresh = 0;
842 :    
843 :     if((x > 0) && (y > 0) && (x < (int32_t) pParam->mb_width)) {
844 :     left = (&pMBs[(x-1) + y * pParam->mb_width])->sad16; /* left */
845 :     top = (&pMBs[x + (y-1) * pParam->mb_width])->sad16; /* top */
846 :     top_right = (&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16; /* top right */
847 :    
848 :     if(((&pMBs[(x-1) + y * pParam->mb_width])->mode != MODE_INTRA) &&
849 :     ((&pMBs[x + (y-1) * pParam->mb_width])->mode != MODE_INTRA) &&
850 :     ((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mode != MODE_INTRA)) {
851 :     thresh = MAX(MAX(top, left), top_right);
852 :     }
853 :     else
854 :     thresh = MIN(MIN(top, left), top_right);
855 :     }
856 :    
857 :     /* INTRA <-> INTER decision */
858 :     if (sad < thresh) { /* normal, fast, SAD-based mode decision */
859 :     /* intra decision */
860 :    
861 :     if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */
862 :     if (y != 0)
863 :     if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
864 :     if (x != 0)
865 :     if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
866 :    
867 :     if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */
868 :     if (Data->rrv) InterBias *= 4;
869 :    
870 :     if (InterBias < sad) {
871 :     int32_t deviation;
872 :     if (!Data->rrv)
873 :     deviation = dev16(Data->Cur, Data->iEdgedWidth);
874 :     else
875 :     deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */
876 :     dev16(Data->Cur+16, Data->iEdgedWidth) +
877 :     dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) +
878 :     dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth);
879 :    
880 :     if (deviation < (sad - InterBias)) mode = MODE_INTRA;
881 :     }
882 :    
883 :     pMB->cbp = 63;
884 :     } else { /* Rate-Distortion INTRA<->INTER */
885 :     if(min_rd < 0) {
886 :     Data->iQuant = iQuant;
887 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
888 :    
889 :     for (i = 0; i < 5; i++) {
890 :     Data->iMinSAD[i] = 256*4096;
891 :     backup[i] = v[i];
892 :     }
893 :    
894 :     if(mode == MODE_INTER) {
895 :     min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
896 :     cbp = *Data->cbp;
897 :    
898 :     if (coding_type == S_VOP) {
899 :     int gmc_rd;
900 :    
901 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
902 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
903 :     if (gmc_rd < min_rd) {
904 :     mcsel = 1;
905 :     *Data->iMinSAD = min_rd = gmc_rd;
906 :     mode = MODE_INTER;
907 :     cbp = *Data->cbp;
908 :     }
909 :     }
910 :     }
911 :    
912 :     if(mode == MODE_INTER4V) {
913 :     int v4_rd;
914 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
915 :     if (v4_rd < min_rd) {
916 :     Data->iMinSAD[0] = min_rd = v4_rd;
917 :     mode = MODE_INTER4V;
918 :     cbp = *Data->cbp;
919 :     }
920 :     }
921 :     }
922 :    
923 :     intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width);
924 :     if (intra_rd < min_rd) {
925 :     *Data->iMinSAD = min_rd = intra_rd;
926 :     mode = MODE_INTRA;
927 :     }
928 :    
929 :     pMB->cbp = cbp;
930 :     }
931 :    
932 :     early_out:
933 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
934 :    
935 :     if (Data->rrv) {
936 :     Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
937 :     Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
938 :     }
939 :    
940 :     if (mode == MODE_INTER && mcsel == 0) {
941 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
942 :    
943 :     if(Data->qpel) {
944 :     pMB->qmvs[0] = pMB->qmvs[1]
945 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
946 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
947 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
948 :     } else {
949 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
950 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
951 :     }
952 :    
953 :     } else if (mode == MODE_INTER ) { /* but mcsel == 1 */
954 :    
955 :     pMB->mcsel = 1;
956 :     if (Data->qpel) {
957 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
958 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
959 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
960 :     } else
961 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
962 :    
963 :     } else
964 :     if (mode == MODE_INTER4V) ; /* anything here? */
965 :     else /* INTRA, NOT_CODED */
966 :     ZeroMacroblockP(pMB, 0);
967 :    
968 :     pMB->mode = mode;
969 :     }
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4