[cvs] / xvidcore / src / motion / estimation_rd_based.c Repository:
ViewVC logotype

Annotation of /xvidcore/src/motion/estimation_rd_based.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1.10 - (view) (download)

1 : edgomez 1.2 /*****************************************************************************
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 1.10 * $Id: estimation_rd_based.c,v 1.9 2004/12/05 04:53:01 syskin Exp $
24 : edgomez 1.2 *
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 1.7 const unsigned int lambda,
60 : syskin 1.10 const uint16_t * mpeg_quant_matrices,
61 :     const unsigned int quant_sq)
62 : edgomez 1.2 {
63 :     int sum;
64 :     int bits;
65 :     int distortion = 0;
66 :    
67 :     fdct(data);
68 :    
69 :     if (quant_type) sum = quant_h263_inter(coeff, data, quant, mpeg_quant_matrices);
70 :     else sum = quant_mpeg_inter(coeff, data, quant, mpeg_quant_matrices);
71 :    
72 :     if (sum > 0) {
73 :     *cbp |= 1 << (5 - block);
74 :     bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table);
75 :    
76 :     if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
77 :     else dequant_mpeg_inter(dqcoeff, coeff, quant, mpeg_quant_matrices);
78 :    
79 :     distortion = sse8_16bit(data, dqcoeff, 8*sizeof(int16_t));
80 :     } else {
81 :     const static int16_t zero_block[64] =
82 :     {
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 :     0, 0, 0, 0, 0, 0, 0, 0,
91 :     };
92 :     bits = 0;
93 :     distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t));
94 :     }
95 :    
96 :    
97 : syskin 1.10 return bits + (lambda*distortion)/quant_sq;
98 : edgomez 1.2 }
99 :    
100 :     static __inline unsigned int
101 :     Block_CalcBitsIntra(MACROBLOCK * pMB,
102 :     const unsigned int x,
103 :     const unsigned int y,
104 :     const unsigned int mb_width,
105 :     const uint32_t block,
106 :     int16_t coeff[64],
107 :     int16_t qcoeff[64],
108 :     int16_t dqcoeff[64],
109 :     int16_t predictors[8],
110 :     const uint32_t quant,
111 :     const int quant_type,
112 :     unsigned int bits[2],
113 :     unsigned int cbp[2],
114 : syskin 1.7 unsigned int lambda,
115 : syskin 1.10 const uint16_t * mpeg_quant_matrices,
116 :     const unsigned int quant_sq)
117 : edgomez 1.2 {
118 :     int direction;
119 :     int16_t *pCurrent;
120 :     unsigned int i, coded;
121 :     unsigned int distortion = 0;
122 :     const uint32_t iDcScaler = get_dc_scaler(quant, block < 4);
123 :    
124 :     fdct(coeff);
125 :    
126 :     if (quant_type) {
127 :     quant_h263_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
128 :     dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
129 :     } else {
130 :     quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices);
131 :     dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices);
132 :     }
133 :    
134 :     predict_acdc(pMB-(x+mb_width*y), x, y, mb_width, block, qcoeff,
135 : edgomez 1.5 quant, iDcScaler, predictors, 0);
136 : edgomez 1.2
137 :     direction = pMB->acpred_directions[block];
138 :     pCurrent = pMB->pred_values[block];
139 :    
140 :     /* store current coeffs to pred_values[] for future prediction */
141 :     pCurrent[0] = qcoeff[0] * iDcScaler;
142 : edgomez 1.5 pCurrent[0] = CLIP(pCurrent[0], -2048, 2047);
143 : edgomez 1.2 for (i = 1; i < 8; i++) {
144 :     pCurrent[i] = qcoeff[i];
145 :     pCurrent[i + 7] = qcoeff[i * 8];
146 :     }
147 :    
148 :     /* dc prediction */
149 :     qcoeff[0] = qcoeff[0] - predictors[0];
150 :    
151 :     if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len;
152 :     else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len;
153 :    
154 :     /* calc cost before ac prediction */
155 :     bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]);
156 :     if (coded > 0) cbp[0] |= 1 << (5 - block);
157 :    
158 :     /* apply ac prediction & calc cost*/
159 :     if (direction == 1) {
160 :     for (i = 1; i < 8; i++) {
161 :     qcoeff[i] -= predictors[i];
162 :     predictors[i] = qcoeff[i];
163 :     }
164 :     } else { /* acpred_direction == 2 */
165 :     for (i = 1; i < 8; i++) {
166 :     qcoeff[i*8] -= predictors[i];
167 :     predictors[i] = qcoeff[i*8];
168 :     }
169 :     }
170 :    
171 :     bits[1] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[direction]);
172 :     if (coded > 0) cbp[1] |= 1 << (5 - block);
173 :    
174 :     distortion = sse8_16bit(coeff, dqcoeff, 8*sizeof(int16_t));
175 :    
176 : syskin 1.10 return (lambda*distortion)/quant_sq;
177 : edgomez 1.2 }
178 :    
179 :    
180 :    
181 :     static void
182 :     CheckCandidateRD16(const int x, const int y, SearchData * const data, const unsigned int Direction)
183 :     {
184 :    
185 :     int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
186 :     int32_t rd = 0;
187 :     VECTOR * current;
188 :     const uint8_t * ptr;
189 :     int i, t, xc, yc;
190 :     unsigned cbp = 0;
191 :    
192 :     if ( (x > data->max_dx) || (x < data->min_dx)
193 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
194 :    
195 :     if (!data->qpel_precision) {
196 :     ptr = GetReference(x, y, data);
197 :     current = data->currentMV;
198 :     xc = x; yc = y;
199 :     } else { /* x and y are in 1/4 precision */
200 :     ptr = xvid_me_interpolate16x16qpel(x, y, 0, data);
201 :     current = data->currentQMV;
202 :     xc = x/2; yc = y/2;
203 :     }
204 :    
205 :     for(i = 0; i < 4; i++) {
206 :     int s = 8*((i&1) + (i>>1)*data->iEdgedWidth);
207 :     transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth);
208 : syskin 1.10 rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant,
209 :     data->quant_type, &cbp, i, data->scan_table, data->lambda[i],
210 :     data->mpeg_quant_matrices, data->quant_sq);
211 : edgomez 1.2 }
212 :    
213 : syskin 1.9 rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision);
214 : edgomez 1.2
215 :     if (data->temp[0] + t < data->iMinSAD[1]) {
216 :     data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); }
217 :     if (data->temp[1] < data->iMinSAD[2]) {
218 :     data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); }
219 :     if (data->temp[2] < data->iMinSAD[3]) {
220 :     data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); }
221 :     if (data->temp[3] < data->iMinSAD[4]) {
222 :     data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); }
223 :    
224 :     rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
225 :    
226 :     if (rd >= data->iMinSAD[0]) return;
227 :    
228 :     /* chroma */
229 :     xc = (xc >> 1) + roundtab_79[xc & 0x3];
230 :     yc = (yc >> 1) + roundtab_79[yc & 0x3];
231 :    
232 :     /* chroma U */
233 :     ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
234 :     transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2);
235 : syskin 1.10 rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
236 :     &cbp, 4, data->scan_table, data->lambda[4],
237 :     data->mpeg_quant_matrices, data->quant_sq);
238 : edgomez 1.2 if (rd >= data->iMinSAD[0]) return;
239 :    
240 :     /* chroma V */
241 :     ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
242 :     transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2);
243 : syskin 1.10 rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
244 :     &cbp, 5, data->scan_table, data->lambda[5],
245 :     data->mpeg_quant_matrices, data->quant_sq);
246 : edgomez 1.2
247 :     rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
248 :    
249 :     if (rd < data->iMinSAD[0]) {
250 :     data->iMinSAD[0] = rd;
251 :     current[0].x = x; current[0].y = y;
252 :     data->dir = Direction;
253 :     *data->cbp = cbp;
254 :     }
255 :     }
256 :    
257 :     static void
258 :     CheckCandidateRD8(const int x, const int y, SearchData * const data, const unsigned int Direction)
259 :     {
260 :    
261 :     int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
262 :     int32_t rd;
263 :     VECTOR * current;
264 :     const uint8_t * ptr;
265 :     unsigned int cbp = 0;
266 :    
267 :     if ( (x > data->max_dx) || (x < data->min_dx)
268 :     || (y > data->max_dy) || (y < data->min_dy) ) return;
269 :    
270 :     if (!data->qpel_precision) {
271 :     ptr = GetReference(x, y, data);
272 :     current = data->currentMV;
273 :     } else { /* x and y are in 1/4 precision */
274 :     ptr = xvid_me_interpolate8x8qpel(x, y, 0, 0, data);
275 :     current = data->currentQMV;
276 :     }
277 :    
278 :     transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
279 : syskin 1.10 rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type,
280 :     &cbp, 5, data->scan_table, data->lambda[0],
281 :     data->mpeg_quant_matrices, data->quant_sq);
282 : syskin 1.9 rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision);
283 : edgomez 1.2
284 :     if (rd < data->iMinSAD[0]) {
285 :     *data->cbp = cbp;
286 :     data->iMinSAD[0] = rd;
287 :     current[0].x = x; current[0].y = y;
288 :     data->dir = Direction;
289 :     }
290 :     }
291 :    
292 :    
293 :     static int
294 :     findRD_inter(SearchData * const Data,
295 :     const int x, const int y,
296 :     const MBParam * const pParam,
297 :     const uint32_t MotionFlags)
298 :     {
299 :     int i;
300 :     int32_t bsad[5];
301 :    
302 :     if (Data->qpel) {
303 :     for(i = 0; i < 5; i++) {
304 :     Data->currentMV[i].x = Data->currentQMV[i].x/2;
305 :     Data->currentMV[i].y = Data->currentQMV[i].y/2;
306 :     }
307 :     Data->qpel_precision = 1;
308 :     CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255);
309 :    
310 :     if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */
311 :     for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i];
312 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
313 : syskin 1.9 pParam->width, pParam->height, Data->iFcode - Data->qpel, 1);
314 : edgomez 1.2 Data->qpel_precision = 0;
315 :     if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1)
316 :     CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
317 :     }
318 :    
319 :     } else { /* not qpel */
320 :    
321 :     CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
322 :     }
323 :    
324 :     if (MotionFlags&XVID_ME_EXTSEARCH_RD)
325 :     xvid_me_SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16);
326 :    
327 :     if (MotionFlags&XVID_ME_HALFPELREFINE16_RD)
328 : syskin 1.6 xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidateRD16, 0);
329 : edgomez 1.2
330 :     if (Data->qpel) {
331 :     if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */
332 :     for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) {
333 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */
334 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
335 :     }
336 :    
337 :     /* preparing for qpel-precision search */
338 :     Data->qpel_precision = 1;
339 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
340 : syskin 1.9 pParam->width, pParam->height, Data->iFcode, 2);
341 : edgomez 1.2 }
342 : syskin 1.3 if (MotionFlags & XVID_ME_QUARTERPELREFINE16_RD) {
343 :     if (MotionFlags & XVID_ME_FASTREFINE16)
344 :     FullRefine_Fast(Data, CheckCandidateRD16, 0);
345 :     else
346 : syskin 1.6 xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateRD16, 0);
347 : syskin 1.3 }
348 : edgomez 1.2 }
349 :    
350 :     if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */
351 :     VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV;
352 :     if (!MVequal(Data->predMV, *v))
353 :     CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255);
354 :     }
355 :     return Data->iMinSAD[0];
356 :     }
357 :    
358 :     static int
359 :     findRD_inter4v(SearchData * const Data,
360 :     MACROBLOCK * const pMB, const MACROBLOCK * const pMBs,
361 :     const int x, const int y,
362 :     const MBParam * const pParam, const uint32_t MotionFlags,
363 :     const VECTOR * const backup)
364 :     {
365 :    
366 :     unsigned int cbp = 0, bits = 0, t = 0, i;
367 :     SearchData Data2, *Data8 = &Data2;
368 :     int sumx = 0, sumy = 0;
369 :     int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64;
370 :     uint8_t * ptr;
371 :    
372 :     memcpy(Data8, Data, sizeof(SearchData));
373 :    
374 :     for (i = 0; i < 4; i++) { /* for all luma blocks */
375 :    
376 :     *Data8->iMinSAD = *(Data->iMinSAD + i + 1);
377 :     *Data8->currentMV = *(Data->currentMV + i + 1);
378 :     *Data8->currentQMV = *(Data->currentQMV + i + 1);
379 :     Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
380 :     Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
381 :     Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
382 :     Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
383 :     Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
384 :     *Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; /* copy corresponding cbp bit */
385 : syskin 1.7 Data8->lambda[0] = Data->lambda[i];
386 : edgomez 1.2
387 :     if(Data->qpel) {
388 :     Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i);
389 :     if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y,
390 : syskin 1.9 Data8->predMV, Data8->iFcode, 0);
391 : edgomez 1.2 } else {
392 :     Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i);
393 :     if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y,
394 : syskin 1.9 Data8->predMV, Data8->iFcode, 0);
395 : edgomez 1.2 }
396 :    
397 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
398 : syskin 1.9 pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1);
399 : edgomez 1.2
400 :     *Data8->iMinSAD += BITS_MULT*t;
401 :    
402 :     Data8->qpel_precision = Data8->qpel;
403 :     /* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */
404 :     {
405 :     VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV;
406 :     if (!MVequal (*v, backup[i+1]) )
407 :     CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255);
408 :     }
409 :    
410 :     if (Data8->qpel) {
411 : syskin 1.3 int bsad = Data8->iMinSAD[0];
412 :     int bx = Data8->currentQMV->x;
413 :     int by = Data8->currentQMV->y;
414 :    
415 :     Data8->currentMV->x = Data8->currentQMV->x/2;
416 :     Data8->currentMV->y = Data8->currentQMV->y/2;
417 :    
418 : edgomez 1.2 if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */
419 :     Data8->qpel_precision = 0;
420 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
421 : syskin 1.9 pParam->width, pParam->height, Data8->iFcode - 1, 1);
422 : edgomez 1.2
423 :     if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1)
424 :     CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255);
425 :    
426 :     if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD)
427 :     xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
428 :    
429 :     if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
430 : syskin 1.6 xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0);
431 : edgomez 1.2
432 : syskin 1.3 if(bsad > *Data8->iMinSAD) { /* we have found a better match */
433 :     bx = Data8->currentQMV->x = 2*Data8->currentMV->x;
434 :     by = Data8->currentQMV->y = 2*Data8->currentMV->y;
435 :     bsad = Data8->iMinSAD[0];
436 : edgomez 1.2 }
437 :    
438 :     Data8->qpel_precision = 1;
439 :     get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
440 : syskin 1.9 pParam->width, pParam->height, Data8->iFcode, 2);
441 : edgomez 1.2
442 :     }
443 : syskin 1.3
444 :     if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) {
445 :     if (MotionFlags & XVID_ME_FASTREFINE8)
446 :     FullRefine_Fast(Data8, CheckCandidateRD8, 0);
447 : syskin 1.6 else xvid_me_SubpelRefine(Data->currentQMV[0], Data8, CheckCandidateRD8, 0);
448 : syskin 1.3 }
449 :    
450 :     if (bsad <= Data->iMinSAD[0]) {
451 :     /* we have not found a better match */
452 :     Data8->iMinSAD[0] = bsad;
453 :     Data8->currentQMV->x = bx;
454 :     Data8->currentQMV->y = by;
455 :     }
456 : edgomez 1.2
457 :     } else { /* not qpel */
458 :    
459 :     if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */
460 :     xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
461 :    
462 :     if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
463 : syskin 1.6 xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0); /* halfpel refinement */
464 : edgomez 1.2 }
465 :    
466 :     /* checking vector equal to predicion */
467 :     if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) {
468 :     const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV;
469 :     if (!MVequal(*v, Data8->predMV))
470 :     CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255);
471 :     }
472 :    
473 :     bits += *Data8->iMinSAD;
474 :     if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */
475 :    
476 :     /* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */
477 :     if(Data->qpel) {
478 :     pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x;
479 :     pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y;
480 :     pMB->qmvs[i] = *Data8->currentQMV;
481 :     sumx += Data8->currentQMV->x/2;
482 :     sumy += Data8->currentQMV->y/2;
483 :     } else {
484 :     pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x;
485 :     pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y;
486 :     sumx += Data8->currentMV->x;
487 :     sumy += Data8->currentMV->y;
488 :     }
489 :     pMB->mvs[i] = *Data8->currentMV;
490 :     pMB->sad8[i] = 4 * *Data8->iMinSAD;
491 :     if (Data8->cbp[0]) cbp |= 1 << (5 - i);
492 :    
493 :     } /* end - for all luma blocks */
494 :    
495 :     bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
496 :    
497 :     /* let's check chroma */
498 :     sumx = (sumx >> 3) + roundtab_76[sumx & 0xf];
499 :     sumy = (sumy >> 3) + roundtab_76[sumy & 0xf];
500 :    
501 :     /* chroma U */
502 :     ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
503 :     transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2);
504 : syskin 1.10 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4,
505 :     Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq);
506 : edgomez 1.2
507 :     if (bits >= *Data->iMinSAD) return bits;
508 :    
509 :     /* chroma V */
510 :     ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
511 :     transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2);
512 : syskin 1.10 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5,
513 :     Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq);
514 : edgomez 1.2
515 :     bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len;
516 :    
517 :     *Data->cbp = cbp;
518 :     return bits;
519 :     }
520 :    
521 :     static int
522 :     findRD_intra(SearchData * const Data, MACROBLOCK * pMB,
523 :     const int x, const int y, const int mb_width)
524 :     {
525 :     unsigned int cbp[2] = {0, 0}, bits[2], i;
526 :     unsigned int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */
527 :     unsigned int distortion = 0;
528 :    
529 :     int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128;
530 :     const uint32_t iQuant = Data->iQuant;
531 :     int16_t predictors[6][8];
532 :    
533 :     for(i = 0; i < 4; i++) {
534 :     int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth);
535 :     transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth);
536 :    
537 :    
538 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff,
539 : syskin 1.10 predictors[i], iQuant, Data->quant_type, bits, cbp,
540 :     Data->lambda[i], Data->mpeg_quant_matrices, Data->quant_sq);
541 : edgomez 1.2 bits1 += distortion + BITS_MULT * bits[0];
542 :     bits2 += distortion + BITS_MULT * bits[1];
543 :    
544 :     if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0])
545 :     return bits1;
546 :     }
547 :    
548 :     bits1 += BITS_MULT*xvid_cbpy_tab[cbp[0]>>2].len;
549 :     bits2 += BITS_MULT*xvid_cbpy_tab[cbp[1]>>2].len;
550 :    
551 :     /*chroma U */
552 :     transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
553 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff,
554 : syskin 1.10 predictors[4], iQuant, Data->quant_type, bits, cbp,
555 :     Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq);
556 : edgomez 1.2 bits1 += distortion + BITS_MULT * bits[0];
557 :     bits2 += distortion + BITS_MULT * bits[1];
558 :    
559 :     if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0])
560 :     return bits1;
561 :    
562 :     /* chroma V */
563 :     transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
564 :     distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff,
565 : syskin 1.10 predictors[5], iQuant, Data->quant_type, bits, cbp,
566 :     Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq);
567 : edgomez 1.2
568 :     bits1 += distortion + BITS_MULT * bits[0];
569 :     bits2 += distortion + BITS_MULT * bits[1];
570 :    
571 :     bits1 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[0] & 3) << 3)].len;
572 :     bits2 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[1] & 3) << 3)].len;
573 :    
574 :     *Data->cbp = bits1 <= bits2 ? cbp[0] : cbp[1];
575 :    
576 :     return MIN(bits1, bits2);
577 :     }
578 :    
579 :    
580 :     static int
581 :     findRD_gmc(SearchData * const Data, const IMAGE * const vGMC, const int x, const int y)
582 :     {
583 :     int bits = BITS_MULT*1; /* this one is mcsel */
584 :     unsigned int cbp = 0, i;
585 :     int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64;
586 :    
587 :     for(i = 0; i < 4; i++) {
588 :     int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth);
589 :     transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth);
590 : syskin 1.10 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i,
591 :     Data->scan_table, Data->lambda[i], Data->mpeg_quant_matrices, Data->quant_sq);
592 : edgomez 1.2 if (bits >= Data->iMinSAD[0]) return bits;
593 :     }
594 :    
595 :     bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
596 :    
597 :     /*chroma U */
598 :     transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
599 : syskin 1.10 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4,
600 :     Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq);
601 : edgomez 1.2
602 :     if (bits >= Data->iMinSAD[0]) return bits;
603 :    
604 :     /* chroma V */
605 :     transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2);
606 : syskin 1.10 bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5,
607 :     Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq);
608 : edgomez 1.2
609 :     bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
610 :    
611 :     *Data->cbp = cbp;
612 :    
613 :     return bits;
614 :     }
615 :    
616 :     void
617 :     xvid_me_ModeDecision_RD(SearchData * const Data,
618 :     MACROBLOCK * const pMB,
619 :     const MACROBLOCK * const pMBs,
620 :     const int x, const int y,
621 :     const MBParam * const pParam,
622 :     const uint32_t MotionFlags,
623 :     const uint32_t VopFlags,
624 :     const uint32_t VolFlags,
625 :     const IMAGE * const pCurrent,
626 :     const IMAGE * const pRef,
627 :     const IMAGE * const vGMC,
628 :     const int coding_type)
629 :     {
630 :     int mode = MODE_INTER;
631 :     int mcsel = 0;
632 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
633 :     const uint32_t iQuant = pMB->quant;
634 :    
635 :     int min_rd, intra_rd, i, cbp;
636 :     VECTOR backup[5], *v;
637 :     Data->iQuant = iQuant;
638 : syskin 1.10 Data->quant_sq = iQuant*iQuant;
639 : edgomez 1.2 Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ?
640 :     scan_tables[2] : scan_tables[0];
641 :    
642 :     pMB->mcsel = 0;
643 :    
644 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
645 :     for (i = 0; i < 5; i++) {
646 :     Data->iMinSAD[i] = 256*4096;
647 :     backup[i] = v[i];
648 :     }
649 :    
650 : syskin 1.7 for (i = 0; i < 6; i++) {
651 :     /* HVS models, anyone ? */
652 :     Data->lambda[i] = LAMBDA;
653 :     }
654 :    
655 : edgomez 1.2 min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
656 :     cbp = *Data->cbp;
657 :    
658 :     if (coding_type == S_VOP) {
659 :     int gmc_rd;
660 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
661 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
662 :     if (gmc_rd < min_rd) {
663 :     mcsel = 1;
664 :     *Data->iMinSAD = min_rd = gmc_rd;
665 :     mode = MODE_INTER;
666 :     cbp = *Data->cbp;
667 :     }
668 :     }
669 :    
670 :     if (inter4v) {
671 :     int v4_rd;
672 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
673 :     if (v4_rd < min_rd) {
674 :     Data->iMinSAD[0] = min_rd = v4_rd;
675 :     mode = MODE_INTER4V;
676 :     cbp = *Data->cbp;
677 :     }
678 :     }
679 :    
680 :     intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width);
681 :     if (intra_rd < min_rd) {
682 :     *Data->iMinSAD = min_rd = intra_rd;
683 :     mode = MODE_INTRA;
684 :     cbp = *Data->cbp;
685 :     }
686 :    
687 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0;
688 :     pMB->cbp = cbp;
689 :    
690 :     if (mode == MODE_INTER && mcsel == 0) {
691 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
692 :    
693 :     if(Data->qpel) {
694 :     pMB->qmvs[0] = pMB->qmvs[1]
695 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
696 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
697 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
698 :     } else {
699 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
700 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
701 :     }
702 :    
703 :     } else if (mode == MODE_INTER ) { /* but mcsel == 1 */
704 :    
705 :     pMB->mcsel = 1;
706 :     if (Data->qpel) {
707 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
708 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
709 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
710 :     } else
711 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
712 :    
713 :     } else
714 :     if (mode == MODE_INTER4V) ; /* anything here? */
715 :     else /* INTRA, NOT_CODED */
716 :     ZeroMacroblockP(pMB, 0);
717 :    
718 :     pMB->mode = mode;
719 :     }
720 :    
721 :     void
722 :     xvid_me_ModeDecision_Fast(SearchData * const Data,
723 :     MACROBLOCK * const pMB,
724 :     const MACROBLOCK * const pMBs,
725 :     const int x, const int y,
726 :     const MBParam * const pParam,
727 :     const uint32_t MotionFlags,
728 :     const uint32_t VopFlags,
729 :     const uint32_t VolFlags,
730 :     const IMAGE * const pCurrent,
731 :     const IMAGE * const pRef,
732 :     const IMAGE * const vGMC,
733 :     const int coding_type)
734 :     {
735 :     int mode = MODE_INTER;
736 :     int mcsel = 0;
737 :     int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
738 :     const uint32_t iQuant = pMB->quant;
739 :     const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0);
740 :     int sad;
741 :     int min_rd = -1, intra_rd, i, cbp = 63;
742 :     VECTOR backup[5], *v;
743 :     int sad_backup[5];
744 :     int InterBias = MV16_INTER_BIAS;
745 :     int thresh = 0;
746 :     int top = 0, top_right = 0, left = 0;
747 :     Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ?
748 :     scan_tables[2] : scan_tables[0];
749 :    
750 :     pMB->mcsel = 0;
751 : syskin 1.10 Data->iQuant = iQuant;
752 :     Data->quant_sq = iQuant*iQuant;
753 : edgomez 1.2
754 : syskin 1.8 for (i = 0; i < 6; i++) {
755 :     /* HVS models, anyone ? */
756 :     Data->lambda[i] = LAMBDA;
757 :     }
758 :    
759 : edgomez 1.2 /* INTER <-> INTER4V decision */
760 :     if ((Data->iMinSAD[0] + 75 < Data->iMinSAD[1] +
761 :     Data->iMinSAD[2] + Data->iMinSAD[3] + Data->iMinSAD[4])) { /* normal, fast, SAD-based mode decision */
762 :     if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
763 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
764 :     mode = MODE_INTER;
765 :     sad = Data->iMinSAD[0];
766 :     } else {
767 :     mode = MODE_INTER4V;
768 :     sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
769 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
770 :     Data->iMinSAD[0] = sad;
771 :     }
772 :    
773 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
774 :     if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
775 :     if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH)
776 : syskin 1.9 if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) {
777 : edgomez 1.2 mode = MODE_NOT_CODED;
778 :     sad = 0; /* Compiler warning */
779 :     goto early_out;
780 :     }
781 :    
782 :     /* mcsel */
783 :     if (coding_type == S_VOP) {
784 :    
785 :     int32_t iSAD = sad16(Data->Cur,
786 :     vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
787 :    
788 :     if (Data->chroma) {
789 :     iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
790 :     iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
791 :     }
792 :    
793 :     if (iSAD <= sad) { /* mode decision GMC */
794 :     mode = MODE_INTER;
795 :     mcsel = 1;
796 :     sad = iSAD;
797 :     }
798 :    
799 :     }
800 :     } else { /* Rate-Distortion INTER<->INTER4V */
801 :     Data->iQuant = iQuant;
802 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
803 :    
804 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
805 :     if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
806 :     if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH)
807 : syskin 1.9 if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) {
808 : edgomez 1.2 mode = MODE_NOT_CODED;
809 :     sad = 0; /* Compiler warning */
810 :     goto early_out;
811 :     }
812 :    
813 :     for (i = 0; i < 5; i++) {
814 :     sad_backup[i] = Data->iMinSAD[i];
815 :     Data->iMinSAD[i] = 256*4096;
816 :     backup[i] = v[i];
817 :     }
818 :    
819 :     min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
820 :     cbp = *Data->cbp;
821 :     sad = sad_backup[0];
822 :    
823 :     if (coding_type == S_VOP) {
824 :     int gmc_rd;
825 :    
826 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
827 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
828 :     if (gmc_rd < min_rd) {
829 :     mcsel = 1;
830 :     *Data->iMinSAD = min_rd = gmc_rd;
831 :     mode = MODE_INTER;
832 :     cbp = *Data->cbp;
833 :     sad = sad16(Data->Cur,
834 :     vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536);
835 :     if (Data->chroma) {
836 :     sad += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
837 :     sad += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2);
838 :     }
839 :     }
840 :     }
841 :    
842 :     if (inter4v) {
843 :     int v4_rd;
844 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
845 :     if (v4_rd < min_rd) {
846 :     Data->iMinSAD[0] = min_rd = v4_rd;
847 :     mode = MODE_INTER4V;
848 :     cbp = *Data->cbp;
849 :     sad = sad_backup[1] + sad_backup[2] +
850 :     sad_backup[3] + sad_backup[4] + IMV16X16 * (int32_t)iQuant;
851 :     }
852 :     }
853 :     }
854 :    
855 :     left = top = top_right = -1;
856 :     thresh = 0;
857 :    
858 :     if((x > 0) && (y > 0) && (x < (int32_t) pParam->mb_width)) {
859 :     left = (&pMBs[(x-1) + y * pParam->mb_width])->sad16; /* left */
860 :     top = (&pMBs[x + (y-1) * pParam->mb_width])->sad16; /* top */
861 :     top_right = (&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16; /* top right */
862 :    
863 :     if(((&pMBs[(x-1) + y * pParam->mb_width])->mode != MODE_INTRA) &&
864 :     ((&pMBs[x + (y-1) * pParam->mb_width])->mode != MODE_INTRA) &&
865 :     ((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mode != MODE_INTRA)) {
866 :     thresh = MAX(MAX(top, left), top_right);
867 :     }
868 :     else
869 :     thresh = MIN(MIN(top, left), top_right);
870 :     }
871 :    
872 :     /* INTRA <-> INTER decision */
873 :     if (sad < thresh) { /* normal, fast, SAD-based mode decision */
874 :     /* intra decision */
875 :    
876 :     if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */
877 :     if (y != 0)
878 :     if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
879 :     if (x != 0)
880 :     if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
881 :    
882 :     if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */
883 :    
884 :     if (InterBias < sad) {
885 : syskin 1.9 int32_t deviation = dev16(Data->Cur, Data->iEdgedWidth);
886 : edgomez 1.2 if (deviation < (sad - InterBias)) mode = MODE_INTRA;
887 :     }
888 :    
889 :     pMB->cbp = 63;
890 :     } else { /* Rate-Distortion INTRA<->INTER */
891 :     if(min_rd < 0) {
892 :     Data->iQuant = iQuant;
893 :     v = Data->qpel ? Data->currentQMV : Data->currentMV;
894 :    
895 :     for (i = 0; i < 5; i++) {
896 :     Data->iMinSAD[i] = 256*4096;
897 :     backup[i] = v[i];
898 :     }
899 :    
900 :     if(mode == MODE_INTER) {
901 :     min_rd = findRD_inter(Data, x, y, pParam, MotionFlags);
902 :     cbp = *Data->cbp;
903 :    
904 :     if (coding_type == S_VOP) {
905 :     int gmc_rd;
906 :    
907 :     *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
908 :     gmc_rd = findRD_gmc(Data, vGMC, x, y);
909 :     if (gmc_rd < min_rd) {
910 :     mcsel = 1;
911 :     *Data->iMinSAD = min_rd = gmc_rd;
912 :     mode = MODE_INTER;
913 :     cbp = *Data->cbp;
914 :     }
915 :     }
916 :     }
917 :    
918 :     if(mode == MODE_INTER4V) {
919 :     int v4_rd;
920 :     v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
921 :     if (v4_rd < min_rd) {
922 :     Data->iMinSAD[0] = min_rd = v4_rd;
923 :     mode = MODE_INTER4V;
924 :     cbp = *Data->cbp;
925 :     }
926 :     }
927 :     }
928 :    
929 :     intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width);
930 :     if (intra_rd < min_rd) {
931 :     *Data->iMinSAD = min_rd = intra_rd;
932 :     mode = MODE_INTRA;
933 :     }
934 :    
935 :     pMB->cbp = cbp;
936 :     }
937 :    
938 :     early_out:
939 :     pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
940 :    
941 :     if (mode == MODE_INTER && mcsel == 0) {
942 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
943 :    
944 :     if(Data->qpel) {
945 :     pMB->qmvs[0] = pMB->qmvs[1]
946 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
947 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
948 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
949 :     } else {
950 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
951 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
952 :     }
953 :    
954 :     } else if (mode == MODE_INTER ) { /* but mcsel == 1 */
955 :    
956 :     pMB->mcsel = 1;
957 :     if (Data->qpel) {
958 :     pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv;
959 :     pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2;
960 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2;
961 :     } else
962 :     pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv;
963 :    
964 :     } else
965 :     if (mode == MODE_INTER4V) ; /* anything here? */
966 :     else /* INTRA, NOT_CODED */
967 :     ZeroMacroblockP(pMB, 0);
968 :    
969 :     pMB->mode = mode;
970 :     }

No admin address has been configured
ViewVC Help
Powered by ViewVC 1.0.4