Parent Directory | Revision Log
Revision 1.1.2.8 - (view) (download)
1 : | edgomez | 1.1.2.1 | /***************************************************************************** |
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 | 1.1.2.8 | * $Id: estimation_rd_based.c,v 1.1.2.7 2003/10/07 13:02:35 edgomez Exp $ |
24 : | edgomez | 1.1.2.1 | * |
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 : | edgomez | 1.1.2.7 | #include "../quant/quant.h" |
44 : | edgomez | 1.1.2.1 | #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 : | edgomez | 1.1.2.6 | const int block, |
58 : | const uint16_t * scan_table) | ||
59 : | edgomez | 1.1.2.1 | { |
60 : | int sum; | ||
61 : | int bits; | ||
62 : | int distortion = 0; | ||
63 : | int i; | ||
64 : | |||
65 : | fdct(data); | ||
66 : | |||
67 : | edgomez | 1.1.2.7 | if (quant_type) sum = quant_h263_inter(coeff, data, quant); |
68 : | else sum = quant_mpeg_inter(coeff, data, quant); | ||
69 : | edgomez | 1.1.2.1 | |
70 : | if (sum > 0) { | ||
71 : | *cbp |= 1 << (5 - block); | ||
72 : | edgomez | 1.1.2.6 | bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table); |
73 : | edgomez | 1.1.2.1 | |
74 : | edgomez | 1.1.2.7 | if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant); |
75 : | else dequant_mpeg_inter(dqcoeff, coeff, quant); | ||
76 : | edgomez | 1.1.2.1 | |
77 : | for (i = 0; i < 64; i++) | ||
78 : | distortion += (data[i] - dqcoeff[i])*(data[i] - dqcoeff[i]); | ||
79 : | |||
80 : | } else { | ||
81 : | bits = 0; | ||
82 : | for (i = 0; i < 64; i++) | ||
83 : | distortion += data[i]*data[i]; | ||
84 : | } | ||
85 : | |||
86 : | return bits + (LAMBDA*distortion)/(quant*quant); | ||
87 : | } | ||
88 : | |||
89 : | static __inline unsigned int | ||
90 : | syskin | 1.1.2.8 | Block_CalcBitsIntra(MACROBLOCK * pMB, |
91 : | const unsigned int x, | ||
92 : | const unsigned int y, | ||
93 : | const unsigned int mb_width, | ||
94 : | const uint32_t block, | ||
95 : | int16_t coeff[64], | ||
96 : | int16_t qcoeff[64], | ||
97 : | int16_t dqcoeff[64], | ||
98 : | int16_t predictors[8], | ||
99 : | const uint32_t quant, | ||
100 : | const int quant_type, | ||
101 : | unsigned int bits[2], | ||
102 : | unsigned int cbp[2]) | ||
103 : | edgomez | 1.1.2.1 | { |
104 : | syskin | 1.1.2.8 | int direction; |
105 : | int16_t *pCurrent; | ||
106 : | unsigned int i, coded; | ||
107 : | unsigned int distortion = 0; | ||
108 : | const uint32_t iDcScaler = get_dc_scaler(quant, block < 4); | ||
109 : | |||
110 : | fdct(coeff); | ||
111 : | |||
112 : | if (quant_type) { | ||
113 : | quant_h263_intra(qcoeff, coeff, quant, iDcScaler); | ||
114 : | dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler); | ||
115 : | } else { | ||
116 : | quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler); | ||
117 : | dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler); | ||
118 : | } | ||
119 : | edgomez | 1.1.2.1 | |
120 : | syskin | 1.1.2.8 | predict_acdc(pMB-(x+mb_width*y), x, y, mb_width, block, qcoeff, |
121 : | quant, iDcScaler, predictors, 0); | ||
122 : | |||
123 : | direction = pMB->acpred_directions[block]; | ||
124 : | pCurrent = pMB->pred_values[block]; | ||
125 : | edgomez | 1.1.2.1 | |
126 : | syskin | 1.1.2.8 | /* store current coeffs to pred_values[] for future prediction */ |
127 : | pCurrent[0] = qcoeff[0] * iDcScaler; | ||
128 : | for (i = 1; i < 8; i++) { | ||
129 : | pCurrent[i] = qcoeff[i]; | ||
130 : | pCurrent[i + 7] = qcoeff[i * 8]; | ||
131 : | edgomez | 1.1.2.1 | } |
132 : | |||
133 : | syskin | 1.1.2.8 | /* dc prediction */ |
134 : | qcoeff[0] = qcoeff[0] - predictors[0]; | ||
135 : | edgomez | 1.1.2.1 | |
136 : | syskin | 1.1.2.8 | if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len; |
137 : | else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len; | ||
138 : | edgomez | 1.1.2.1 | |
139 : | syskin | 1.1.2.8 | /* calc cost before ac prediction */ |
140 : | bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]); | ||
141 : | if (coded > 0) cbp[0] |= 1 << (5 - block); | ||
142 : | |||
143 : | /* apply ac prediction & calc cost*/ | ||
144 : | if (direction == 1) { | ||
145 : | for (i = 1; i < 8; i++) { | ||
146 : | qcoeff[i] -= predictors[i]; | ||
147 : | predictors[i] = qcoeff[i]; | ||
148 : | } | ||
149 : | } else { /* acpred_direction == 2 */ | ||
150 : | for (i = 1; i < 8; i++) { | ||
151 : | qcoeff[i*8] -= predictors[i]; | ||
152 : | predictors[i] = qcoeff[i*8]; | ||
153 : | } | ||
154 : | } | ||
155 : | |||
156 : | bits[1] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[direction]); | ||
157 : | if (coded > 0) cbp[1] |= 1 << (5 - block); | ||
158 : | edgomez | 1.1.2.1 | |
159 : | for (i = 0; i < 64; i++) | ||
160 : | syskin | 1.1.2.8 | distortion += (coeff[i] - dqcoeff[i])*(coeff[i] - dqcoeff[i]); |
161 : | |||
162 : | |||
163 : | return (LAMBDA*distortion)/(quant*quant); | ||
164 : | edgomez | 1.1.2.1 | |
165 : | } | ||
166 : | |||
167 : | syskin | 1.1.2.8 | |
168 : | |||
169 : | edgomez | 1.1.2.1 | static void |
170 : | CheckCandidateRD16(const int x, const int y, const SearchData * const data, const unsigned int Direction) | ||
171 : | { | ||
172 : | |||
173 : | int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; | ||
174 : | int32_t rd = 0; | ||
175 : | VECTOR * current; | ||
176 : | const uint8_t * ptr; | ||
177 : | int i, cbp = 0, t, xc, yc; | ||
178 : | |||
179 : | if ( (x > data->max_dx) || (x < data->min_dx) | ||
180 : | || (y > data->max_dy) || (y < data->min_dy) ) return; | ||
181 : | |||
182 : | if (!data->qpel_precision) { | ||
183 : | ptr = GetReference(x, y, data); | ||
184 : | current = data->currentMV; | ||
185 : | xc = x; yc = y; | ||
186 : | } else { /* x and y are in 1/4 precision */ | ||
187 : | ptr = xvid_me_interpolate16x16qpel(x, y, 0, data); | ||
188 : | current = data->currentQMV; | ||
189 : | xc = x/2; yc = y/2; | ||
190 : | } | ||
191 : | |||
192 : | for(i = 0; i < 4; i++) { | ||
193 : | int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); | ||
194 : | transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); | ||
195 : | edgomez | 1.1.2.6 | rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table); |
196 : | edgomez | 1.1.2.1 | } |
197 : | |||
198 : | rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); | ||
199 : | |||
200 : | if (data->temp[0] + t < data->iMinSAD[1]) { | ||
201 : | data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); } | ||
202 : | if (data->temp[1] < data->iMinSAD[2]) { | ||
203 : | data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); } | ||
204 : | if (data->temp[2] < data->iMinSAD[3]) { | ||
205 : | data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); } | ||
206 : | if (data->temp[3] < data->iMinSAD[4]) { | ||
207 : | data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); } | ||
208 : | |||
209 : | rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; | ||
210 : | |||
211 : | if (rd >= data->iMinSAD[0]) return; | ||
212 : | |||
213 : | /* chroma */ | ||
214 : | xc = (xc >> 1) + roundtab_79[xc & 0x3]; | ||
215 : | yc = (yc >> 1) + roundtab_79[yc & 0x3]; | ||
216 : | |||
217 : | /* chroma U */ | ||
218 : | ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); | ||
219 : | transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); | ||
220 : | edgomez | 1.1.2.6 | rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table); |
221 : | edgomez | 1.1.2.1 | if (rd >= data->iMinSAD[0]) return; |
222 : | |||
223 : | /* chroma V */ | ||
224 : | ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); | ||
225 : | transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); | ||
226 : | edgomez | 1.1.2.6 | rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table); |
227 : | edgomez | 1.1.2.1 | |
228 : | rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; | ||
229 : | |||
230 : | if (rd < data->iMinSAD[0]) { | ||
231 : | data->iMinSAD[0] = rd; | ||
232 : | current[0].x = x; current[0].y = y; | ||
233 : | *data->dir = Direction; | ||
234 : | *data->cbp = cbp; | ||
235 : | } | ||
236 : | } | ||
237 : | |||
238 : | static void | ||
239 : | CheckCandidateRD8(const int x, const int y, const SearchData * const data, const unsigned int Direction) | ||
240 : | { | ||
241 : | |||
242 : | int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; | ||
243 : | int32_t rd; | ||
244 : | VECTOR * current; | ||
245 : | const uint8_t * ptr; | ||
246 : | int cbp = 0; | ||
247 : | |||
248 : | if ( (x > data->max_dx) || (x < data->min_dx) | ||
249 : | || (y > data->max_dy) || (y < data->min_dy) ) return; | ||
250 : | |||
251 : | if (!data->qpel_precision) { | ||
252 : | ptr = GetReference(x, y, data); | ||
253 : | current = data->currentMV; | ||
254 : | } else { /* x and y are in 1/4 precision */ | ||
255 : | ptr = xvid_me_interpolate8x8qpel(x, y, 0, 0, data); | ||
256 : | current = data->currentQMV; | ||
257 : | } | ||
258 : | |||
259 : | transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); | ||
260 : | edgomez | 1.1.2.6 | rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table); |
261 : | edgomez | 1.1.2.1 | rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
262 : | |||
263 : | if (rd < data->iMinSAD[0]) { | ||
264 : | *data->cbp = cbp; | ||
265 : | data->iMinSAD[0] = rd; | ||
266 : | current[0].x = x; current[0].y = y; | ||
267 : | *data->dir = Direction; | ||
268 : | } | ||
269 : | } | ||
270 : | |||
271 : | |||
272 : | static int | ||
273 : | findRD_inter(SearchData * const Data, | ||
274 : | const int x, const int y, | ||
275 : | const MBParam * const pParam, | ||
276 : | const uint32_t MotionFlags) | ||
277 : | { | ||
278 : | int i; | ||
279 : | int32_t bsad[5]; | ||
280 : | |||
281 : | if (Data->qpel) { | ||
282 : | for(i = 0; i < 5; i++) { | ||
283 : | Data->currentMV[i].x = Data->currentQMV[i].x/2; | ||
284 : | Data->currentMV[i].y = Data->currentQMV[i].y/2; | ||
285 : | } | ||
286 : | Data->qpel_precision = 1; | ||
287 : | CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255); | ||
288 : | |||
289 : | if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ | ||
290 : | for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; | ||
291 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
292 : | pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); | ||
293 : | Data->qpel_precision = 0; | ||
294 : | if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) | ||
295 : | CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); | ||
296 : | } | ||
297 : | |||
298 : | } else { /* not qpel */ | ||
299 : | |||
300 : | CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); | ||
301 : | } | ||
302 : | |||
303 : | if (MotionFlags&XVID_ME_EXTSEARCH_RD) | ||
304 : | xvid_me_SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16); | ||
305 : | |||
306 : | if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) | ||
307 : | xvid_me_SubpelRefine(Data, CheckCandidateRD16); | ||
308 : | |||
309 : | if (Data->qpel) { | ||
310 : | if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ | ||
311 : | for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { | ||
312 : | Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ | ||
313 : | Data->currentQMV[i].y = 2 * Data->currentMV[i].y; | ||
314 : | } | ||
315 : | |||
316 : | /* preparing for qpel-precision search */ | ||
317 : | Data->qpel_precision = 1; | ||
318 : | get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, | ||
319 : | pParam->width, pParam->height, Data->iFcode, 2, 0); | ||
320 : | } | ||
321 : | if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) | ||
322 : | xvid_me_SubpelRefine(Data, CheckCandidateRD16); | ||
323 : | } | ||
324 : | |||
325 : | if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ | ||
326 : | VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; | ||
327 : | if (!MVequal(Data->predMV, *v)) | ||
328 : | CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255); | ||
329 : | } | ||
330 : | return Data->iMinSAD[0]; | ||
331 : | } | ||
332 : | |||
333 : | static int | ||
334 : | findRD_inter4v(const SearchData * const Data, | ||
335 : | MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, | ||
336 : | const int x, const int y, | ||
337 : | const MBParam * const pParam, const uint32_t MotionFlags, | ||
338 : | const VECTOR * const backup) | ||
339 : | { | ||
340 : | |||
341 : | int cbp = 0, bits = 0, t = 0, i; | ||
342 : | SearchData Data2, *Data8 = &Data2; | ||
343 : | int sumx = 0, sumy = 0; | ||
344 : | int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; | ||
345 : | uint8_t * ptr; | ||
346 : | |||
347 : | memcpy(Data8, Data, sizeof(SearchData)); | ||
348 : | |||
349 : | for (i = 0; i < 4; i++) { /* for all luma blocks */ | ||
350 : | |||
351 : | Data8->iMinSAD = Data->iMinSAD + i + 1; | ||
352 : | Data8->currentMV = Data->currentMV + i + 1; | ||
353 : | Data8->currentQMV = Data->currentQMV + i + 1; | ||
354 : | Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
355 : | Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
356 : | Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
357 : | Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
358 : | Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
359 : | edgomez | 1.1.2.3 | *Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; /* copy corresponding cbp bit */ |
360 : | edgomez | 1.1.2.1 | |
361 : | if(Data->qpel) { | ||
362 : | Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); | ||
363 : | if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, | ||
364 : | Data8->predMV, Data8->iFcode, 0, 0); | ||
365 : | } else { | ||
366 : | Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); | ||
367 : | if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, | ||
368 : | Data8->predMV, Data8->iFcode, 0, 0); | ||
369 : | } | ||
370 : | |||
371 : | get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, | ||
372 : | pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0); | ||
373 : | |||
374 : | *Data8->iMinSAD += BITS_MULT*t; | ||
375 : | |||
376 : | Data8->qpel_precision = Data8->qpel; | ||
377 : | /* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */ | ||
378 : | { | ||
379 : | VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; | ||
380 : | if (!MVequal (*v, backup[i+1]) ) | ||
381 : | CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255); | ||
382 : | } | ||
383 : | |||
384 : | if (Data8->qpel) { | ||
385 : | if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ | ||
386 : | int32_t s = *Data8->iMinSAD; | ||
387 : | Data8->currentMV->x = Data8->currentQMV->x/2; | ||
388 : | Data8->currentMV->y = Data8->currentQMV->y/2; | ||
389 : | Data8->qpel_precision = 0; | ||
390 : | get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, | ||
391 : | pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); | ||
392 : | |||
393 : | if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) | ||
394 : | CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255); | ||
395 : | |||
396 : | if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) | ||
397 : | xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); | ||
398 : | |||
399 : | if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) | ||
400 : | xvid_me_SubpelRefine(Data8, CheckCandidateRD8); | ||
401 : | |||
402 : | if(s > *Data8->iMinSAD) { /* we have found a better match */ | ||
403 : | Data8->currentQMV->x = 2*Data8->currentMV->x; | ||
404 : | Data8->currentQMV->y = 2*Data8->currentMV->y; | ||
405 : | } | ||
406 : | |||
407 : | Data8->qpel_precision = 1; | ||
408 : | get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, | ||
409 : | pParam->width, pParam->height, Data8->iFcode, 2, 0); | ||
410 : | |||
411 : | } | ||
412 : | if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) | ||
413 : | xvid_me_SubpelRefine(Data8, CheckCandidateRD8); | ||
414 : | |||
415 : | } else { /* not qpel */ | ||
416 : | |||
417 : | if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */ | ||
418 : | xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); | ||
419 : | |||
420 : | if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) | ||
421 : | xvid_me_SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */ | ||
422 : | } | ||
423 : | |||
424 : | /* checking vector equal to predicion */ | ||
425 : | if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) { | ||
426 : | const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; | ||
427 : | if (!MVequal(*v, Data8->predMV)) | ||
428 : | CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255); | ||
429 : | } | ||
430 : | |||
431 : | bits += *Data8->iMinSAD; | ||
432 : | if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */ | ||
433 : | |||
434 : | /* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */ | ||
435 : | if(Data->qpel) { | ||
436 : | pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; | ||
437 : | pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; | ||
438 : | pMB->qmvs[i] = *Data8->currentQMV; | ||
439 : | sumx += Data8->currentQMV->x/2; | ||
440 : | sumy += Data8->currentQMV->y/2; | ||
441 : | } else { | ||
442 : | pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x; | ||
443 : | pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y; | ||
444 : | sumx += Data8->currentMV->x; | ||
445 : | sumy += Data8->currentMV->y; | ||
446 : | } | ||
447 : | pMB->mvs[i] = *Data8->currentMV; | ||
448 : | pMB->sad8[i] = 4 * *Data8->iMinSAD; | ||
449 : | if (Data8->cbp[0]) cbp |= 1 << (5 - i); | ||
450 : | |||
451 : | } /* end - for all luma blocks */ | ||
452 : | |||
453 : | bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; | ||
454 : | |||
455 : | /* let's check chroma */ | ||
456 : | sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; | ||
457 : | sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; | ||
458 : | |||
459 : | /* chroma U */ | ||
460 : | ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); | ||
461 : | transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); | ||
462 : | edgomez | 1.1.2.6 | bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table); |
463 : | edgomez | 1.1.2.1 | |
464 : | if (bits >= *Data->iMinSAD) return bits; | ||
465 : | |||
466 : | /* chroma V */ | ||
467 : | ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); | ||
468 : | transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); | ||
469 : | edgomez | 1.1.2.6 | bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table); |
470 : | edgomez | 1.1.2.1 | |
471 : | bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; | ||
472 : | |||
473 : | *Data->cbp = cbp; | ||
474 : | return bits; | ||
475 : | } | ||
476 : | |||
477 : | static int | ||
478 : | syskin | 1.1.2.8 | findRD_intra(const SearchData * const Data, MACROBLOCK * pMB, |
479 : | const int x, const int y, const int mb_width) | ||
480 : | edgomez | 1.1.2.1 | { |
481 : | syskin | 1.1.2.8 | int cbp[2] = {0, 0}, bits[2], i; |
482 : | int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ | ||
483 : | int distortion = 0; | ||
484 : | |||
485 : | int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128; | ||
486 : | const uint32_t iQuant = Data->iQuant; | ||
487 : | int16_t predictors[6][8]; | ||
488 : | edgomez | 1.1.2.1 | |
489 : | for(i = 0; i < 4; i++) { | ||
490 : | int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
491 : | transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); | ||
492 : | syskin | 1.1.2.8 | |
493 : | edgomez | 1.1.2.1 | |
494 : | syskin | 1.1.2.8 | distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff, |
495 : | predictors[i], iQuant, Data->quant_type, bits, cbp); | ||
496 : | bits1 += distortion + BITS_MULT * bits[0]; | ||
497 : | bits2 += distortion + BITS_MULT * bits[1]; | ||
498 : | |||
499 : | if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0]) | ||
500 : | return bits1; | ||
501 : | edgomez | 1.1.2.1 | } |
502 : | |||
503 : | syskin | 1.1.2.8 | bits1 += BITS_MULT*xvid_cbpy_tab[cbp[0]>>2].len; |
504 : | bits2 += BITS_MULT*xvid_cbpy_tab[cbp[1]>>2].len; | ||
505 : | edgomez | 1.1.2.1 | |
506 : | /*chroma U */ | ||
507 : | transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); | ||
508 : | syskin | 1.1.2.8 | distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff, |
509 : | predictors[4], iQuant, Data->quant_type, bits, cbp); | ||
510 : | bits1 += distortion + BITS_MULT * bits[0]; | ||
511 : | bits2 += distortion + BITS_MULT * bits[1]; | ||
512 : | edgomez | 1.1.2.1 | |
513 : | syskin | 1.1.2.8 | if (bits1 >= Data->iMinSAD[0] && bits2 >= Data->iMinSAD[0]) |
514 : | return bits1; | ||
515 : | edgomez | 1.1.2.1 | |
516 : | /* chroma V */ | ||
517 : | transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); | ||
518 : | syskin | 1.1.2.8 | distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff, |
519 : | predictors[5], iQuant, Data->quant_type, bits, cbp); | ||
520 : | edgomez | 1.1.2.1 | |
521 : | syskin | 1.1.2.8 | bits1 += distortion + BITS_MULT * bits[0]; |
522 : | bits2 += distortion + BITS_MULT * bits[1]; | ||
523 : | edgomez | 1.1.2.1 | |
524 : | syskin | 1.1.2.8 | bits1 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[0] & 3) << 3)].len; |
525 : | bits2 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[1] & 3) << 3)].len; | ||
526 : | |||
527 : | *Data->cbp = bits1 <= bits2 ? cbp[0] : cbp[1]; | ||
528 : | |||
529 : | return MIN(bits1, bits2); | ||
530 : | edgomez | 1.1.2.1 | } |
531 : | |||
532 : | syskin | 1.1.2.8 | |
533 : | edgomez | 1.1.2.1 | static int |
534 : | findRD_gmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) | ||
535 : | { | ||
536 : | int bits = BITS_MULT*1; /* this one is mcsel */ | ||
537 : | int cbp = 0, i; | ||
538 : | int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; | ||
539 : | |||
540 : | for(i = 0; i < 4; i++) { | ||
541 : | int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); | ||
542 : | transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); | ||
543 : | edgomez | 1.1.2.6 | bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, Data->scan_table); |
544 : | edgomez | 1.1.2.1 | if (bits >= Data->iMinSAD[0]) return bits; |
545 : | } | ||
546 : | |||
547 : | bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; | ||
548 : | |||
549 : | /*chroma U */ | ||
550 : | transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); | ||
551 : | edgomez | 1.1.2.6 | bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table); |
552 : | edgomez | 1.1.2.1 | |
553 : | if (bits >= Data->iMinSAD[0]) return bits; | ||
554 : | |||
555 : | /* chroma V */ | ||
556 : | transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); | ||
557 : | edgomez | 1.1.2.6 | bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table); |
558 : | edgomez | 1.1.2.1 | |
559 : | bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; | ||
560 : | |||
561 : | *Data->cbp = cbp; | ||
562 : | |||
563 : | return bits; | ||
564 : | } | ||
565 : | |||
566 : | void | ||
567 : | xvid_me_ModeDecision_RD(SearchData * const Data, | ||
568 : | MACROBLOCK * const pMB, | ||
569 : | const MACROBLOCK * const pMBs, | ||
570 : | const int x, const int y, | ||
571 : | const MBParam * const pParam, | ||
572 : | const uint32_t MotionFlags, | ||
573 : | const uint32_t VopFlags, | ||
574 : | const uint32_t VolFlags, | ||
575 : | const IMAGE * const pCurrent, | ||
576 : | const IMAGE * const pRef, | ||
577 : | const IMAGE * const vGMC, | ||
578 : | const int coding_type) | ||
579 : | { | ||
580 : | int mode = MODE_INTER; | ||
581 : | int mcsel = 0; | ||
582 : | int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); | ||
583 : | const uint32_t iQuant = pMB->quant; | ||
584 : | |||
585 : | int min_rd, intra_rd, i, cbp, c[2] = {0, 0}; | ||
586 : | VECTOR backup[5], *v; | ||
587 : | Data->iQuant = iQuant; | ||
588 : | Data->cbp = c; | ||
589 : | syskin | 1.1.2.5 | Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ? |
590 : | scan_tables[2] : scan_tables[0]; | ||
591 : | syskin | 1.1.2.2 | |
592 : | pMB->mcsel = 0; | ||
593 : | edgomez | 1.1.2.1 | |
594 : | v = Data->qpel ? Data->currentQMV : Data->currentMV; | ||
595 : | for (i = 0; i < 5; i++) { | ||
596 : | Data->iMinSAD[i] = 256*4096; | ||
597 : | backup[i] = v[i]; | ||
598 : | } | ||
599 : | |||
600 : | min_rd = findRD_inter(Data, x, y, pParam, MotionFlags); | ||
601 : | cbp = *Data->cbp; | ||
602 : | |||
603 : | if (coding_type == S_VOP) { | ||
604 : | int gmc_rd; | ||
605 : | *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ | ||
606 : | gmc_rd = findRD_gmc(Data, vGMC, x, y); | ||
607 : | if (gmc_rd < min_rd) { | ||
608 : | mcsel = 1; | ||
609 : | *Data->iMinSAD = min_rd = gmc_rd; | ||
610 : | mode = MODE_INTER; | ||
611 : | cbp = *Data->cbp; | ||
612 : | } | ||
613 : | } | ||
614 : | |||
615 : | if (inter4v) { | ||
616 : | int v4_rd; | ||
617 : | v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); | ||
618 : | if (v4_rd < min_rd) { | ||
619 : | Data->iMinSAD[0] = min_rd = v4_rd; | ||
620 : | mode = MODE_INTER4V; | ||
621 : | cbp = *Data->cbp; | ||
622 : | } | ||
623 : | } | ||
624 : | |||
625 : | syskin | 1.1.2.8 | intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width); |
626 : | edgomez | 1.1.2.1 | if (intra_rd < min_rd) { |
627 : | *Data->iMinSAD = min_rd = intra_rd; | ||
628 : | mode = MODE_INTRA; | ||
629 : | syskin | 1.1.2.8 | cbp = *Data->cbp; |
630 : | edgomez | 1.1.2.1 | } |
631 : | |||
632 : | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; | ||
633 : | pMB->cbp = cbp; | ||
634 : | |||
635 : | |||
636 : | if (Data->rrv) { | ||
637 : | Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); | ||
638 : | Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); | ||
639 : | } | ||
640 : | |||
641 : | if (mode == MODE_INTER && mcsel == 0) { | ||
642 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; | ||
643 : | |||
644 : | if(Data->qpel) { | ||
645 : | pMB->qmvs[0] = pMB->qmvs[1] | ||
646 : | = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; | ||
647 : | pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; | ||
648 : | pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; | ||
649 : | } else { | ||
650 : | pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; | ||
651 : | pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; | ||
652 : | } | ||
653 : | |||
654 : | edgomez | 1.1.2.3 | } else if (mode == MODE_INTER ) { /* but mcsel == 1 */ |
655 : | edgomez | 1.1.2.1 | |
656 : | pMB->mcsel = 1; | ||
657 : | if (Data->qpel) { | ||
658 : | pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; | ||
659 : | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; | ||
660 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; | ||
661 : | } else | ||
662 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; | ||
663 : | |||
664 : | } else | ||
665 : | if (mode == MODE_INTER4V) ; /* anything here? */ | ||
666 : | else /* INTRA, NOT_CODED */ | ||
667 : | ZeroMacroblockP(pMB, 0); | ||
668 : | |||
669 : | pMB->mode = mode; | ||
670 : | } | ||
671 : | |||
672 : | void | ||
673 : | xvid_me_ModeDecision_Fast(SearchData * const Data, | ||
674 : | MACROBLOCK * const pMB, | ||
675 : | const MACROBLOCK * const pMBs, | ||
676 : | const int x, const int y, | ||
677 : | const MBParam * const pParam, | ||
678 : | const uint32_t MotionFlags, | ||
679 : | const uint32_t VopFlags, | ||
680 : | const uint32_t VolFlags, | ||
681 : | const IMAGE * const pCurrent, | ||
682 : | const IMAGE * const pRef, | ||
683 : | const IMAGE * const vGMC, | ||
684 : | const int coding_type) | ||
685 : | { | ||
686 : | int mode = MODE_INTER; | ||
687 : | int mcsel = 0; | ||
688 : | int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); | ||
689 : | const uint32_t iQuant = pMB->quant; | ||
690 : | const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0); | ||
691 : | edgomez | 1.1.2.4 | int sad; |
692 : | edgomez | 1.1.2.1 | int min_rd = -1, intra_rd, i, cbp = 63, c[2] = {0, 0}; |
693 : | VECTOR backup[5], *v; | ||
694 : | int sad_backup[5]; | ||
695 : | int InterBias = MV16_INTER_BIAS; | ||
696 : | int thresh = 0; | ||
697 : | int top = 0, top_right = 0, left = 0; | ||
698 : | syskin | 1.1.2.5 | Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ? |
699 : | scan_tables[2] : scan_tables[0]; | ||
700 : | edgomez | 1.1.2.1 | |
701 : | pMB->mcsel = 0; | ||
702 : | |||
703 : | /* INTER <-> INTER4V decision */ | ||
704 : | if ((Data->iMinSAD[0] + 75 < Data->iMinSAD[1] + | ||
705 : | Data->iMinSAD[2] + Data->iMinSAD[3] + Data->iMinSAD[4])) { /* normal, fast, SAD-based mode decision */ | ||
706 : | if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + | ||
707 : | Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { | ||
708 : | mode = MODE_INTER; | ||
709 : | sad = Data->iMinSAD[0]; | ||
710 : | } else { | ||
711 : | mode = MODE_INTER4V; | ||
712 : | sad = Data->iMinSAD[1] + Data->iMinSAD[2] + | ||
713 : | Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant; | ||
714 : | Data->iMinSAD[0] = sad; | ||
715 : | } | ||
716 : | |||
717 : | /* final skip decision, a.k.a. "the vector you found, really that good?" */ | ||
718 : | if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) | ||
719 : | if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) | ||
720 : | if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { | ||
721 : | mode = MODE_NOT_CODED; | ||
722 : | sad = 0; /* Compiler warning */ | ||
723 : | goto early_out; | ||
724 : | } | ||
725 : | |||
726 : | /* mcsel */ | ||
727 : | if (coding_type == S_VOP) { | ||
728 : | |||
729 : | int32_t iSAD = sad16(Data->Cur, | ||
730 : | vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); | ||
731 : | |||
732 : | if (Data->chroma) { | ||
733 : | iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); | ||
734 : | iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); | ||
735 : | } | ||
736 : | |||
737 : | if (iSAD <= sad) { /* mode decision GMC */ | ||
738 : | mode = MODE_INTER; | ||
739 : | mcsel = 1; | ||
740 : | sad = iSAD; | ||
741 : | } | ||
742 : | |||
743 : | } | ||
744 : | } else { /* Rate-Distortion INTER<->INTER4V */ | ||
745 : | Data->iQuant = iQuant; | ||
746 : | Data->cbp = c; | ||
747 : | v = Data->qpel ? Data->currentQMV : Data->currentMV; | ||
748 : | |||
749 : | /* final skip decision, a.k.a. "the vector you found, really that good?" */ | ||
750 : | if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) | ||
751 : | if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH) | ||
752 : | if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { | ||
753 : | mode = MODE_NOT_CODED; | ||
754 : | sad = 0; /* Compiler warning */ | ||
755 : | goto early_out; | ||
756 : | } | ||
757 : | |||
758 : | for (i = 0; i < 5; i++) { | ||
759 : | sad_backup[i] = Data->iMinSAD[i]; | ||
760 : | Data->iMinSAD[i] = 256*4096; | ||
761 : | backup[i] = v[i]; | ||
762 : | } | ||
763 : | |||
764 : | min_rd = findRD_inter(Data, x, y, pParam, MotionFlags); | ||
765 : | cbp = *Data->cbp; | ||
766 : | sad = sad_backup[0]; | ||
767 : | |||
768 : | if (coding_type == S_VOP) { | ||
769 : | int gmc_rd; | ||
770 : | |||
771 : | *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ | ||
772 : | gmc_rd = findRD_gmc(Data, vGMC, x, y); | ||
773 : | if (gmc_rd < min_rd) { | ||
774 : | mcsel = 1; | ||
775 : | *Data->iMinSAD = min_rd = gmc_rd; | ||
776 : | mode = MODE_INTER; | ||
777 : | cbp = *Data->cbp; | ||
778 : | sad = sad16(Data->Cur, | ||
779 : | vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); | ||
780 : | if (Data->chroma) { | ||
781 : | sad += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); | ||
782 : | sad += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); | ||
783 : | } | ||
784 : | } | ||
785 : | } | ||
786 : | |||
787 : | if (inter4v) { | ||
788 : | int v4_rd; | ||
789 : | v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); | ||
790 : | if (v4_rd < min_rd) { | ||
791 : | Data->iMinSAD[0] = min_rd = v4_rd; | ||
792 : | mode = MODE_INTER4V; | ||
793 : | cbp = *Data->cbp; | ||
794 : | sad = sad_backup[1] + sad_backup[2] + | ||
795 : | sad_backup[3] + sad_backup[4] + IMV16X16 * (int32_t)iQuant; | ||
796 : | } | ||
797 : | } | ||
798 : | } | ||
799 : | |||
800 : | left = top = top_right = -1; | ||
801 : | thresh = 0; | ||
802 : | |||
803 : | if((x > 0) && (y > 0) && (x < (int32_t) pParam->mb_width)) { | ||
804 : | edgomez | 1.1.2.3 | left = (&pMBs[(x-1) + y * pParam->mb_width])->sad16; /* left */ |
805 : | top = (&pMBs[x + (y-1) * pParam->mb_width])->sad16; /* top */ | ||
806 : | top_right = (&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16; /* top right */ | ||
807 : | edgomez | 1.1.2.1 | |
808 : | if(((&pMBs[(x-1) + y * pParam->mb_width])->mode != MODE_INTRA) && | ||
809 : | ((&pMBs[x + (y-1) * pParam->mb_width])->mode != MODE_INTRA) && | ||
810 : | ((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mode != MODE_INTRA)) { | ||
811 : | thresh = MAX(MAX(top, left), top_right); | ||
812 : | } | ||
813 : | else | ||
814 : | thresh = MIN(MIN(top, left), top_right); | ||
815 : | } | ||
816 : | |||
817 : | /* INTRA <-> INTER decision */ | ||
818 : | if (sad < thresh) { /* normal, fast, SAD-based mode decision */ | ||
819 : | /* intra decision */ | ||
820 : | |||
821 : | if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ | ||
822 : | if (y != 0) | ||
823 : | if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; | ||
824 : | if (x != 0) | ||
825 : | if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; | ||
826 : | |||
827 : | if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ | ||
828 : | if (Data->rrv) InterBias *= 4; | ||
829 : | |||
830 : | if (InterBias < sad) { | ||
831 : | int32_t deviation; | ||
832 : | if (!Data->rrv) | ||
833 : | deviation = dev16(Data->Cur, Data->iEdgedWidth); | ||
834 : | else | ||
835 : | deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ | ||
836 : | dev16(Data->Cur+16, Data->iEdgedWidth) + | ||
837 : | dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + | ||
838 : | dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); | ||
839 : | |||
840 : | if (deviation < (sad - InterBias)) mode = MODE_INTRA; | ||
841 : | } | ||
842 : | |||
843 : | pMB->cbp = 63; | ||
844 : | } else { /* Rate-Distortion INTRA<->INTER */ | ||
845 : | if(min_rd < 0) { | ||
846 : | Data->iQuant = iQuant; | ||
847 : | Data->cbp = c; | ||
848 : | v = Data->qpel ? Data->currentQMV : Data->currentMV; | ||
849 : | |||
850 : | for (i = 0; i < 5; i++) { | ||
851 : | Data->iMinSAD[i] = 256*4096; | ||
852 : | backup[i] = v[i]; | ||
853 : | } | ||
854 : | |||
855 : | if(mode == MODE_INTER) { | ||
856 : | min_rd = findRD_inter(Data, x, y, pParam, MotionFlags); | ||
857 : | cbp = *Data->cbp; | ||
858 : | |||
859 : | if (coding_type == S_VOP) { | ||
860 : | int gmc_rd; | ||
861 : | |||
862 : | *Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ | ||
863 : | gmc_rd = findRD_gmc(Data, vGMC, x, y); | ||
864 : | if (gmc_rd < min_rd) { | ||
865 : | mcsel = 1; | ||
866 : | *Data->iMinSAD = min_rd = gmc_rd; | ||
867 : | mode = MODE_INTER; | ||
868 : | cbp = *Data->cbp; | ||
869 : | } | ||
870 : | } | ||
871 : | } | ||
872 : | |||
873 : | if(mode == MODE_INTER4V) { | ||
874 : | int v4_rd; | ||
875 : | v4_rd = findRD_inter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); | ||
876 : | if (v4_rd < min_rd) { | ||
877 : | Data->iMinSAD[0] = min_rd = v4_rd; | ||
878 : | mode = MODE_INTER4V; | ||
879 : | cbp = *Data->cbp; | ||
880 : | } | ||
881 : | } | ||
882 : | } | ||
883 : | |||
884 : | syskin | 1.1.2.8 | intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width); |
885 : | edgomez | 1.1.2.1 | if (intra_rd < min_rd) { |
886 : | *Data->iMinSAD = min_rd = intra_rd; | ||
887 : | mode = MODE_INTRA; | ||
888 : | } | ||
889 : | |||
890 : | pMB->cbp = cbp; | ||
891 : | } | ||
892 : | |||
893 : | early_out: | ||
894 : | pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; | ||
895 : | |||
896 : | if (Data->rrv) { | ||
897 : | Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); | ||
898 : | Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); | ||
899 : | } | ||
900 : | |||
901 : | if (mode == MODE_INTER && mcsel == 0) { | ||
902 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; | ||
903 : | |||
904 : | if(Data->qpel) { | ||
905 : | pMB->qmvs[0] = pMB->qmvs[1] | ||
906 : | = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; | ||
907 : | pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; | ||
908 : | pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; | ||
909 : | } else { | ||
910 : | pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; | ||
911 : | pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; | ||
912 : | } | ||
913 : | |||
914 : | edgomez | 1.1.2.3 | } else if (mode == MODE_INTER ) { /* but mcsel == 1 */ |
915 : | edgomez | 1.1.2.1 | |
916 : | pMB->mcsel = 1; | ||
917 : | if (Data->qpel) { | ||
918 : | pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; | ||
919 : | pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; | ||
920 : | pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; | ||
921 : | } else | ||
922 : | pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; | ||
923 : | |||
924 : | } else | ||
925 : | if (mode == MODE_INTER4V) ; /* anything here? */ | ||
926 : | else /* INTRA, NOT_CODED */ | ||
927 : | ZeroMacroblockP(pMB, 0); | ||
928 : | |||
929 : | pMB->mode = mode; | ||
930 : | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |