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

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

Parent Directory Parent Directory | Revision Log Revision Log


Revision 619 - (view) (download)

1 : Isibaar 3 /**************************************************************************
2 :     *
3 : chl 259 * XVID MPEG-4 VIDEO CODEC
4 :     * motion estimation
5 :     *
6 :     * This program is an implementation of a part of one or more MPEG-4
7 :     * Video tools as specified in ISO/IEC 14496-2 standard. Those intending
8 :     * to use this software module in hardware or software products are
9 :     * advised that its use may infringe existing patents or copyrights, and
10 :     * any such use would be at such party's own risk. The original
11 :     * developer of this software module and his/her company, and subsequent
12 :     * editors and their companies, will have no liability for use of this
13 :     * software or modifications or derivatives thereof.
14 :     *
15 :     * This program is free software; you can redistribute it and/or modify
16 :     * it under the terms of the GNU General Public License as published by
17 :     * the Free Software Foundation; either version 2 of the License, or
18 :     * (at your option) any later version.
19 :     *
20 :     * This program is distributed in the hope that it will be useful,
21 :     * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 :     * GNU General Public License for more details.
24 :     *
25 :     * You should have received a copy of the GNU General Public License
26 :     * along with this program; if not, write to the Free Software
27 :     * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
28 :     *
29 :     *************************************************************************/
30 :    
31 : Isibaar 3 #include <assert.h>
32 :     #include <stdio.h>
33 : chl 96 #include <stdlib.h>
34 : Isibaar 3
35 :     #include "../encoder.h"
36 :     #include "../utils/mbfunctions.h"
37 :     #include "../prediction/mbprediction.h"
38 :     #include "../global.h"
39 :     #include "../utils/timer.h"
40 : Isibaar 580 #include "../image/interpolate8x8.h"
41 : chl 530 #include "motion_est.h"
42 : suxen_drol 118 #include "motion.h"
43 : Isibaar 3 #include "sad.h"
44 : Isibaar 539 #include "../utils/emms.h"
45 : Isibaar 3
46 : chl 530 #define INITIAL_SKIP_THRESH (10)
47 :     #define FINAL_SKIP_THRESH (50)
48 :     #define MAX_SAD00_FOR_SKIP (20)
49 : Isibaar 539 #define MAX_CHROMA_SAD_FOR_SKIP (22)
50 :     #define SKIP_THRESH_B (25)
51 : Isibaar 3
52 : chl 530 #define CHECK_CANDIDATE(X,Y,D) { \
53 :     (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
54 : Isibaar 3
55 : Isibaar 580 #define GET_REFERENCE(X, Y, REF) { \
56 : chl 596 switch ( (((X)&1)<<1) + ((Y)&1) ) \
57 : Isibaar 580 { \
58 : syskin 601 case 0 : REF = (uint8_t *)data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
59 :     case 1 : REF = (uint8_t *)data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
60 :     case 2 : REF = (uint8_t *)data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
61 :     default : REF = (uint8_t *)data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
62 : Isibaar 580 } \
63 :     }
64 :    
65 : chl 530 #define iDiamondSize 2
66 : chl 141
67 : chl 530 static __inline int
68 :     d_mv_bits(int x, int y, const uint32_t iFcode)
69 : Isibaar 3 {
70 : chl 530 int xb, yb;
71 :    
72 :     if (x == 0) xb = 1;
73 :     else {
74 :     if (x < 0) x = -x;
75 :     x += (1 << (iFcode - 1)) - 1;
76 :     x >>= (iFcode - 1);
77 :     if (x > 32) x = 32;
78 :     xb = mvtab[x] + iFcode;
79 : edgomez 195 }
80 : Isibaar 3
81 : chl 530 if (y == 0) yb = 1;
82 :     else {
83 :     if (y < 0) y = -y;
84 :     y += (1 << (iFcode - 1)) - 1;
85 :     y >>= (iFcode - 1);
86 :     if (y > 32) y = 32;
87 :     yb = mvtab[y] + iFcode;
88 :     }
89 :     return xb + yb;
90 : Isibaar 3 }
91 :    
92 :    
93 : Isibaar 539 /* CHECK_CANDIATE FUNCTIONS START */
94 :    
95 : chl 530 static void
96 :     CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
97 : Isibaar 3 {
98 : chl 530 int t;
99 :     const uint8_t * Reference;
100 : Isibaar 3
101 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
102 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
103 : suxen_drol 136
104 : chl 530 switch ( ((x&1)<<1) + (y&1) ) {
105 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
106 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
107 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
108 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
109 :     }
110 : chl 259
111 : syskin 594 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
112 : Isibaar 3
113 : syskin 601 t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
114 :     data->temp[0] += data->lambda16 * t;
115 :     data->temp[1] += data->lambda8 * t;
116 : Isibaar 600
117 : chl 530 if (data->temp[0] < data->iMinSAD[0]) {
118 :     data->iMinSAD[0] = data->temp[0];
119 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
120 :     *dir = Direction; }
121 : chl 184
122 : chl 530 if (data->temp[1] < data->iMinSAD[1]) {
123 :     data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
124 :     if (data->temp[2] < data->iMinSAD[2]) {
125 :     data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
126 :     if (data->temp[3] < data->iMinSAD[3]) {
127 :     data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
128 :     if (data->temp[4] < data->iMinSAD[4]) {
129 :     data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
130 : chl 184
131 : Isibaar 3 }
132 :    
133 : chl 530 static void
134 : Isibaar 606 CheckCandidate16Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
135 :     {
136 :     //this function expects X and Y in halfpel positions, but assumes that qpel positions are our goal
137 :     int t;
138 :     const uint8_t * Reference;
139 :    
140 :     if (( x > data->max_dx) || ( x < data->min_dx)
141 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
142 :    
143 :     switch ( ((x&1)<<1) + (y&1) ) {
144 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
145 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
146 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
147 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
148 :     }
149 :    
150 :     data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
151 :    
152 :     t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
153 :     data->temp[0] += data->lambda16 * t;
154 :     data->temp[1] += data->lambda8 * t;
155 :    
156 :     if (data->temp[0] < data->iMinSAD[0]) {
157 :     data->iMinSAD[0] = data->temp[0];
158 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
159 :     *dir = Direction; }
160 :    
161 :     if (data->temp[1] < data->iMinSAD[1]) {
162 :     data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
163 :     if (data->temp[2] < data->iMinSAD[2]) {
164 :     data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
165 :     if (data->temp[3] < data->iMinSAD[3]) {
166 :     data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
167 :     if (data->temp[4] < data->iMinSAD[4]) {
168 :     data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
169 :    
170 :     }
171 :    
172 :     static void
173 : chl 530 CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
174 :     {
175 :     int32_t sad;
176 :     const uint8_t * Reference;
177 : chl 326
178 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
179 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
180 : Isibaar 3
181 : chl 530 switch ( ((x&1)<<1) + (y&1) )
182 :     {
183 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
184 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
185 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
186 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
187 :     }
188 : Isibaar 3
189 : syskin 601 sad = data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
190 : Isibaar 580 sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
191 : Isibaar 3
192 : chl 530 if (sad < *(data->iMinSAD)) {
193 :     *(data->iMinSAD) = sad;
194 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
195 :     *dir = Direction; }
196 : Isibaar 3 }
197 :    
198 : chl 530 static void
199 : Isibaar 580 CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
200 :    
201 :     // CheckCandidate16 variant which expects x and y in quarter pixel resolution
202 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
203 :     // around currentMV!
204 :     {
205 :     int t;
206 : syskin 601 uint8_t * Reference = (uint8_t *)data->RefQ;
207 : Isibaar 580 const uint8_t *ref1, *ref2, *ref3, *ref4;
208 :     VECTOR halfpelMV = *(data->currentMV);
209 :    
210 :     int32_t iEdgedWidth = data->iEdgedWidth;
211 :     uint32_t rounding = data->rounding;
212 :    
213 :     if (( x > data->max_dx) || ( x < data->min_dx)
214 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
215 :    
216 : Isibaar 606 GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases
217 : Isibaar 580 switch( ((x&1)<<1) + (y&1) )
218 :     {
219 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
220 : syskin 601 GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
221 : Isibaar 580 break;
222 :    
223 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
224 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
225 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
226 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
227 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
228 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
229 :     break;
230 :    
231 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
232 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
233 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
234 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
235 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
236 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
237 :     break;
238 :    
239 :     default: // x and y in qpel resolution - the "corners" (top left/right and
240 :     // bottom left/right) during qpel refinement
241 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
242 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
243 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
244 :    
245 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
246 :     interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
247 :     interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
248 :     interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
249 :     break;
250 :     }
251 :    
252 : syskin 594 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1);
253 : Isibaar 580
254 :     t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
255 : syskin 601 data->temp[0] += data->lambda16 * t;
256 :     data->temp[1] += data->lambda8 * t;
257 : Isibaar 580
258 :     if (data->temp[0] < data->iMinSAD[0]) {
259 :     data->iMinSAD[0] = data->temp[0];
260 :     data->currentQMV[0].x = x; data->currentQMV[0].y = y;
261 : syskin 594 /* *dir = Direction;*/ }
262 : Isibaar 580
263 :     if (data->temp[1] < data->iMinSAD[1]) {
264 :     data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; }
265 :     if (data->temp[2] < data->iMinSAD[2]) {
266 :     data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; }
267 :     if (data->temp[3] < data->iMinSAD[3]) {
268 :     data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; }
269 :     if (data->temp[4] < data->iMinSAD[4]) {
270 :     data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; }
271 :     }
272 :    
273 :     static void
274 : Isibaar 539 CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
275 :     {
276 :     int32_t sad;
277 :    
278 :     if (( x > data->max_dx) || ( x < data->min_dx)
279 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
280 :    
281 : syskin 601 sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
282 : Isibaar 539 data->iEdgedWidth, 256*4096);
283 :    
284 :     if (sad < *(data->iMinSAD)) {
285 :     *(data->iMinSAD) = sad;
286 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
287 :     *dir = Direction; }
288 :     }
289 :    
290 :    
291 :     static void
292 : chl 530 CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
293 :     {
294 :     int32_t sad;
295 :     const int xb = data->currentMV[1].x;
296 :     const int yb = data->currentMV[1].y;
297 :     const uint8_t *ReferenceF, *ReferenceB;
298 : Isibaar 3
299 : chl 530 if (( xf > data->max_dx) || ( xf < data->min_dx)
300 :     || ( yf > data->max_dy) || (yf < data->min_dy)) return;
301 : Isibaar 3
302 : chl 530 switch ( ((xf&1)<<1) + (yf&1) ) {
303 :     case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break;
304 :     case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
305 :     case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break;
306 :     default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
307 :     }
308 : edgomez 195
309 : chl 530 switch ( ((xb&1)<<1) + (yb&1) ) {
310 :     case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break;
311 :     case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
312 :     case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break;
313 :     default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
314 :     }
315 : Isibaar 3
316 : syskin 601 sad = data->lambda16 *
317 : chl 530 ( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) +
318 :     d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) );
319 : Isibaar 3
320 : chl 530 sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
321 : Isibaar 3
322 : chl 530 if (sad < *(data->iMinSAD)) {
323 :     *(data->iMinSAD) = sad;
324 :     data->currentMV->x = xf; data->currentMV->y = yf;
325 :     *dir = Direction; }
326 : Isibaar 3 }
327 :    
328 : chl 530 static void
329 :     CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
330 : Isibaar 3 {
331 : chl 530 int32_t sad;
332 :     int k;
333 :     const uint8_t *ReferenceF;
334 :     const uint8_t *ReferenceB;
335 :     VECTOR mvs, b_mvs;
336 : Isibaar 3
337 : chl 530 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
338 : Isibaar 3
339 : syskin 601 sad = data->lambda16 * d_mv_bits(x, y, 1);
340 : Isibaar 3
341 : chl 530 for (k = 0; k < 4; k++) {
342 :     mvs.x = data->directmvF[k].x + x;
343 :     b_mvs.x = ((x == 0) ?
344 :     data->directmvB[k].x
345 :     : mvs.x - data->referencemv[k].x);
346 : Isibaar 3
347 : chl 530 mvs.y = data->directmvF[k].y + y;
348 :     b_mvs.y = ((y == 0) ?
349 :     data->directmvB[k].y
350 :     : mvs.y - data->referencemv[k].y);
351 :    
352 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
353 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
354 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
355 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
356 : Isibaar 3
357 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
358 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
359 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
360 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
361 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
362 :     }
363 : Isibaar 3
364 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
365 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
366 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
367 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
368 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
369 : chl 344 }
370 : chl 530
371 :     sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
372 :     ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
373 :     ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
374 :     data->iEdgedWidth);
375 :     if (sad > *(data->iMinSAD)) return;
376 : edgomez 78 }
377 : chl 530
378 :     if (sad < *(data->iMinSAD)) {
379 :     *(data->iMinSAD) = sad;
380 :     data->currentMV->x = x; data->currentMV->y = y;
381 :     *dir = Direction; }
382 : Isibaar 3 }
383 :    
384 : chl 530 static void
385 :     CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
386 : chl 96 {
387 : chl 530 int32_t sad;
388 :     const uint8_t *ReferenceF;
389 :     const uint8_t *ReferenceB;
390 :     VECTOR mvs, b_mvs;
391 : chl 96
392 : syskin 576 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
393 : chl 530
394 : syskin 601 sad = data->lambda16 * d_mv_bits(x, y, 1);
395 : edgomez 195
396 : chl 530 mvs.x = data->directmvF[0].x + x;
397 :     b_mvs.x = ((x == 0) ?
398 :     data->directmvB[0].x
399 :     : mvs.x - data->referencemv[0].x);
400 : edgomez 195
401 : chl 530 mvs.y = data->directmvF[0].y + y;
402 :     b_mvs.y = ((y == 0) ?
403 :     data->directmvB[0].y
404 :     : mvs.y - data->referencemv[0].y);
405 :    
406 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
407 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
408 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
409 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
410 : chl 96
411 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
412 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
413 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
414 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
415 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
416 :     }
417 : chl 96
418 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
419 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
420 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
421 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
422 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
423 :     }
424 :    
425 :     sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
426 : chl 96
427 : chl 530 if (sad < *(data->iMinSAD)) {
428 :     *(data->iMinSAD) = sad;
429 :     data->currentMV->x = x; data->currentMV->y = y;
430 :     *dir = Direction; }
431 :     }
432 : chl 96
433 : chl 530 static void
434 :     CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
435 :     {
436 :     int32_t sad;
437 :     const uint8_t * Reference;
438 : edgomez 195
439 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
440 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
441 : chl 96
442 : chl 530 switch ( ((x&1)<<1) + (y&1) )
443 :     {
444 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
445 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
446 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
447 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
448 :     }
449 : edgomez 195
450 : chl 530 sad = sad8(data->Cur, Reference, data->iEdgedWidth);
451 : syskin 601 sad += data->lambda8 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
452 : edgomez 195
453 : chl 530 if (sad < *(data->iMinSAD)) {
454 :     *(data->iMinSAD) = sad;
455 :     data->currentMV->x = x; data->currentMV->y = y;
456 :     *dir = Direction; }
457 : chl 96 }
458 :    
459 : Isibaar 580 static void
460 : Isibaar 606 CheckCandidate8Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
461 :     {
462 :     int32_t sad;
463 :     const uint8_t * Reference;
464 :    
465 :     if (( x > data->max_dx) || ( x < data->min_dx)
466 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
467 :    
468 :     switch ( ((x&1)<<1) + (y&1) )
469 :     {
470 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
471 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
472 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
473 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
474 :     }
475 :    
476 :     sad = sad8(data->Cur, Reference, data->iEdgedWidth);
477 :     sad += data->lambda8 * d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
478 :    
479 :     if (sad < *(data->iMinSAD)) {
480 :     *(data->iMinSAD) = sad;
481 :     data->currentMV->x = x; data->currentMV->y = y;
482 :     *dir = Direction; }
483 :     }
484 :    
485 :     static void
486 : Isibaar 580 CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
487 :     // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
488 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
489 :     // around currentMV!
490 :    
491 :     {
492 :     int32_t sad;
493 :     uint8_t *Reference = (uint8_t *) data->RefQ;
494 :     const uint8_t *ref1, *ref2, *ref3, *ref4;
495 :     VECTOR halfpelMV = *(data->currentMV);
496 :    
497 :     int32_t iEdgedWidth = data->iEdgedWidth;
498 :     uint32_t rounding = data->rounding;
499 :    
500 :     if (( x > data->max_dx) || ( x < data->min_dx)
501 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
502 :    
503 : Isibaar 606 GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
504 : Isibaar 580 switch( ((x&1)<<1) + (y&1) )
505 :     {
506 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
507 : syskin 601 GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
508 : Isibaar 580 break;
509 :    
510 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
511 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
512 :    
513 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
514 :     break;
515 :    
516 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
517 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
518 :    
519 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
520 :     break;
521 :    
522 :     default: // x and y in qpel resolution - the "corners" (top left/right and
523 :     // bottom left/right) during qpel refinement
524 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
525 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
526 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
527 :    
528 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
529 :     break;
530 :     }
531 :    
532 :     sad = sad8(data->Cur, Reference, data->iEdgedWidth);
533 : syskin 601 sad += data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
534 : Isibaar 580
535 :     if (sad < *(data->iMinSAD)) {
536 :     *(data->iMinSAD) = sad;
537 :     data->currentQMV->x = x; data->currentQMV->y = y;
538 :     *dir = Direction; }
539 :     }
540 :    
541 : Isibaar 539 /* CHECK_CANDIATE FUNCTIONS END */
542 : chl 96
543 : chl 530 /* MAINSEARCH FUNCTIONS START */
544 : edgomez 195
545 : chl 530 static void
546 :     AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
547 : chl 181 {
548 :    
549 :     /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
550 :    
551 : chl 530 int iDirection;
552 : edgomez 195
553 :     do {
554 : chl 181 iDirection = 0;
555 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
556 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
557 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
558 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
559 : chl 181
560 :     /* now we're doing diagonal checks near our candidate */
561 :    
562 : chl 530 if (iDirection) { //checking if anything found
563 : chl 181 bDirection = iDirection;
564 :     iDirection = 0;
565 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
566 :     if (bDirection & 3) { //our candidate is left or right
567 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
568 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
569 :     } else { // what remains here is up or down
570 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
571 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
572 : chl 181
573 : edgomez 195 if (iDirection) {
574 :     bDirection += iDirection;
575 : chl 530 x = data->currentMV->x; y = data->currentMV->y; }
576 :     } else { //about to quit, eh? not so fast....
577 : edgomez 195 switch (bDirection) {
578 : chl 181 case 2:
579 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
580 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
581 : chl 181 break;
582 :     case 1:
583 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
584 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
585 : chl 181 break;
586 : edgomez 195 case 2 + 4:
587 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
588 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
589 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
590 : chl 181 break;
591 :     case 4:
592 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
593 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
594 : chl 181 break;
595 :     case 8:
596 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
597 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
598 : chl 181 break;
599 : edgomez 195 case 1 + 4:
600 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
601 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
602 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
603 : chl 181 break;
604 : edgomez 195 case 2 + 8:
605 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
606 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
607 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
608 : chl 181 break;
609 : edgomez 195 case 1 + 8:
610 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
611 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
612 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
613 : chl 181 break;
614 : edgomez 195 default: //1+2+4+8 == we didn't find anything at all
615 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
616 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
617 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
618 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
619 : chl 181 break;
620 :     }
621 : chl 530 if (!iDirection) break; //ok, the end. really
622 :     bDirection = iDirection;
623 :     x = data->currentMV->x; y = data->currentMV->y;
624 : chl 181 }
625 :     }
626 : edgomez 195 while (1); //forever
627 : chl 181 }
628 :    
629 : chl 530 static void
630 :     SquareSearch(int x, int y, const SearchData * const data, int bDirection)
631 : chl 326 {
632 : chl 530 int iDirection;
633 : chl 326
634 : chl 530 do {
635 :     iDirection = 0;
636 :     if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
637 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
638 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
639 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
640 :     if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
641 :     if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
642 :     if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
643 :     if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
644 : chl 326
645 : chl 530 bDirection = iDirection;
646 :     x = data->currentMV->x; y = data->currentMV->y;
647 :     } while (iDirection);
648 : chl 346 }
649 :    
650 : chl 530 static void
651 :     DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
652 : chl 346 {
653 :    
654 : chl 181 /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
655 :    
656 : chl 530 int iDirection;
657 : edgomez 195
658 :     do {
659 : chl 181 iDirection = 0;
660 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
661 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
662 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
663 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
664 : chl 181
665 :     /* now we're doing diagonal checks near our candidate */
666 :    
667 : chl 530 if (iDirection) { //checking if anything found
668 : chl 181 bDirection = iDirection;
669 :     iDirection = 0;
670 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
671 :     if (bDirection & 3) { //our candidate is left or right
672 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
673 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
674 :     } else { // what remains here is up or down
675 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
676 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
677 : chl 181
678 : chl 530 bDirection += iDirection;
679 :     x = data->currentMV->x; y = data->currentMV->y;
680 : chl 181 }
681 :     }
682 : chl 530 while (iDirection);
683 : chl 181 }
684 :    
685 : chl 530 /* MAINSEARCH FUNCTIONS END */
686 : chl 181
687 : chl 530 /* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
688 :    
689 :     static void
690 :     HalfpelRefine(const SearchData * const data)
691 : chl 96 {
692 : chl 530 /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */
693 : edgomez 195
694 : chl 530 VECTOR backupMV = *(data->currentMV);
695 :     int iDirection; //not needed
696 : chl 96
697 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
698 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
699 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
700 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
701 : edgomez 195
702 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
703 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
704 :    
705 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
706 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
707 : chl 96 }
708 :    
709 : Isibaar 580
710 :     static void
711 :     QuarterpelRefine(const SearchData * const data)
712 :     {
713 :     /* Perform quarter pixel refinement*/
714 :    
715 :     VECTOR backupMV = *(data->currentQMV);
716 :     int iDirection; //not needed
717 :    
718 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
719 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
720 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
721 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
722 :    
723 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
724 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
725 :    
726 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
727 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
728 :    
729 :     }
730 :    
731 : chl 530 static __inline int
732 :     SkipDecisionP(const IMAGE * current, const IMAGE * reference,
733 :     const int x, const int y,
734 :     const uint32_t iEdgedWidth, const uint32_t iQuant)
735 : chl 96
736 : Isibaar 3 {
737 : chl 530 /* keep repeating checks for all b-frames before this P frame,
738 :     to make sure that SKIP is possible (todo)
739 :     how: if skip is not possible set sad00 to a very high value */
740 : Isibaar 3
741 : chl 530 uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8,
742 :     reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
743 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
744 : h 545 sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8,
745 :     reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
746 : chl 530 if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
747 : edgomez 195
748 : chl 530 return 1;
749 :     }
750 : edgomez 195
751 : chl 530 static __inline void
752 :     SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
753 :     {
754 :     pMB->mode = MODE_NOT_CODED;
755 : Isibaar 539 pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
756 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
757 : Isibaar 580 pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
758 :     pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
759 :    
760 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
761 : Isibaar 3 }
762 :    
763 : chl 619 bool
764 : chl 530 MotionEstimation(MBParam * const pParam,
765 :     FRAMEINFO * const current,
766 :     FRAMEINFO * const reference,
767 :     const IMAGE * const pRefH,
768 :     const IMAGE * const pRefV,
769 :     const IMAGE * const pRefHV,
770 :     const uint32_t iLimit)
771 : Isibaar 3 {
772 : chl 530 MACROBLOCK *const pMBs = current->mbs;
773 :     const IMAGE *const pCurrent = &current->image;
774 :     const IMAGE *const pRef = &reference->image;
775 : Isibaar 3
776 : chl 530 const VECTOR zeroMV = { 0, 0 };
777 : Isibaar 3
778 : chl 530 uint32_t x, y;
779 :     uint32_t iIntra = 0;
780 : syskin 574 int32_t InterBias, quant = current->quant;
781 : Isibaar 580 uint8_t *qimage;
782 : edgomez 195
783 : Isibaar 539 // some pre-initialized thingies for SearchP
784 :     int32_t temp[5];
785 :     VECTOR currentMV[5];
786 : Isibaar 580 VECTOR currentQMV[5];
787 : Isibaar 539 int32_t iMinSAD[5];
788 :     SearchData Data;
789 :     Data.iEdgedWidth = pParam->edged_width;
790 :     Data.currentMV = currentMV;
791 : Isibaar 580 Data.currentQMV = currentQMV;
792 : Isibaar 539 Data.iMinSAD = iMinSAD;
793 :     Data.temp = temp;
794 :     Data.iFcode = current->fcode;
795 : Isibaar 580 Data.rounding = pParam->m_rounding_type;
796 : Isibaar 539
797 : Isibaar 580 if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
798 : syskin 594 return 1; // allocate some mem for qpel interpolated blocks
799 : Isibaar 580 // somehow this is dirty since I think we shouldn't use malloc outside
800 :     // encoder_create() - so please fix me!
801 :    
802 : chl 530 if (sadInit) (*sadInit) ();
803 : edgomez 195
804 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
805 :     for (x = 0; x < pParam->mb_width; x++) {
806 : Isibaar 3
807 : chl 530 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
808 :     int32_t sad00 = pMB->sad16
809 :     = sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
810 :     pRef->y + (x + y * pParam->edged_width) * 16,
811 :     pParam->edged_width, pMB->sad8 );
812 : edgomez 195
813 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
814 :     pMB->dquant = NO_CHANGE;
815 : syskin 601 pMB->quant = current->quant;
816 :     } else {
817 : syskin 574 if (pMB->dquant != NO_CHANGE) {
818 :     quant += DQtab[pMB->dquant];
819 :     if (quant > 31) quant = 31;
820 :     else if (quant < 1) quant = 1;
821 :     }
822 : syskin 601 pMB->quant = quant;
823 :     }
824 : chl 181
825 : chl 619 /* initial skip decision */
826 :     /* no early skip for GMC (global vector = skip vector is unknown!) */
827 : Isibaar 3
828 : chl 619 if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
829 :     if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH)
830 :     if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
831 :     SkipMacroblockP(pMB, sad00);
832 :     continue;
833 :     }
834 :     }
835 : Isibaar 3
836 : Isibaar 580 SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x,
837 : chl 530 y, current->motion_flags, pMB->quant,
838 : Isibaar 539 &Data, pParam, pMBs, reference->mbs,
839 : chl 530 current->global_flags & XVID_INTER4V, pMB);
840 : edgomez 195
841 : chl 530 /* final skip decision, a.k.a. "the vector you found, really that good?" */
842 : chl 619 if (current->coding_type == P_VOP) {
843 :     if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
844 : syskin 601 && ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) )
845 : chl 619 if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
846 :     SkipMacroblockP(pMB, sad00);
847 :     continue;
848 :     }
849 :     }
850 :    
851 : chl 530 /* finally, intra decision */
852 : Isibaar 3
853 : chl 530 InterBias = MV16_INTER_BIAS;
854 :     if (pMB->quant > 8) InterBias += 50 * (pMB->quant - 8); // to make high quants work
855 :     if (y != 0)
856 :     if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 50;
857 :     if (x != 0)
858 :     if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 50;
859 : Isibaar 3
860 : chl 530 if (InterBias < pMB->sad16) {
861 :     const int32_t deviation =
862 :     dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
863 :     pParam->edged_width);
864 : Isibaar 3
865 : chl 530 if (deviation < (pMB->sad16 - InterBias)) {
866 : Isibaar 580 if (++iIntra >= iLimit) { free(qimage); return 1; }
867 : chl 530 pMB->mode = MODE_INTRA;
868 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
869 :     pMB->mvs[3] = zeroMV;
870 : Isibaar 580 pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
871 :     pMB->qmvs[3] = zeroMV;
872 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
873 :     pMB->sad8[3] = 0;
874 :     }
875 :     }
876 :     }
877 : Isibaar 3 }
878 : Isibaar 580 free(qimage);
879 : chl 619
880 :     if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
881 :     current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
882 :     else
883 :     current->GMC_MV = zeroMV;
884 :    
885 :     return 0;
886 : chl 530 }
887 : Isibaar 3
888 : edgomez 195
889 : chl 530 #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16)
890 : Isibaar 3
891 : chl 530 static __inline int
892 :     make_mask(const VECTOR * const pmv, const int i)
893 :     {
894 : Isibaar 539 int mask = 255, j;
895 : chl 530 for (j = 0; j < i; j++) {
896 :     if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
897 :     if (pmv[i].x == pmv[j].x) {
898 :     if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
899 :     if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
900 :     } else
901 :     if (pmv[i].y == pmv[j].y) {
902 :     if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
903 :     if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
904 :     }
905 : Isibaar 3 }
906 : chl 530 return mask;
907 :     }
908 : edgomez 195
909 : chl 530 static __inline void
910 :     PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount,
911 :     const int iHcount, const MACROBLOCK * const prevMB)
912 :     {
913 : edgomez 195
914 : chl 530 //this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
915 : edgomez 195
916 : chl 530 if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour
917 :     pmv[5].x = EVEN(pmv[3].x);
918 : Isibaar 539 pmv[5].y = EVEN(pmv[3].y);
919 :     } else pmv[5].x = pmv[5].y = 0;
920 : chl 169
921 : chl 530 if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
922 :     else pmv[3].x = pmv[3].y = 0;
923 : Isibaar 3
924 : chl 530 if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
925 :     else pmv[4].x = pmv[4].y = 0;
926 : chl 169
927 : chl 530 // [1] median prediction
928 :     pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
929 : chl 169
930 : chl 530 pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
931 : chl 169
932 : chl 530 pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
933 :     pmv[2].y = EVEN(prevMB->mvs[0].y);
934 : chl 169
935 : chl 530 if ((x != iWcount-1) && (y != iHcount-1)) {
936 :     pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
937 : Isibaar 539 pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
938 :     } else pmv[6].x = pmv[6].y = 0;
939 : chl 530 }
940 : chl 169
941 : chl 530 static void
942 :     SearchP(const uint8_t * const pRef,
943 :     const uint8_t * const pRefH,
944 :     const uint8_t * const pRefV,
945 :     const uint8_t * const pRefHV,
946 : Isibaar 580 const uint8_t * const pRefQ,
947 : chl 530 const IMAGE * const pCur,
948 :     const int x,
949 :     const int y,
950 :     const uint32_t MotionFlags,
951 :     const uint32_t iQuant,
952 : Isibaar 539 SearchData * const Data,
953 : chl 530 const MBParam * const pParam,
954 :     const MACROBLOCK * const pMBs,
955 :     const MACROBLOCK * const prevMBs,
956 :     int inter4v,
957 :     MACROBLOCK * const pMB)
958 :     {
959 : chl 169
960 : chl 530 int i, iDirection = 255, mask, threshA;
961 : Isibaar 539 VECTOR pmv[7];
962 : Isibaar 3
963 : Isibaar 539 get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
964 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
965 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
966 : Isibaar 3
967 : Isibaar 539 Data->predMV = pmv[0];
968 : Isibaar 580
969 : Isibaar 539 Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
970 :     Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16;
971 :     Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
972 :     Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
973 :     Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
974 : Isibaar 580 Data->RefQ = pRefQ;
975 : Isibaar 3
976 : syskin 601 Data->lambda16 = lambda_vec16[iQuant];
977 : Isibaar 606 Data->lambda8 = lambda_vec8[iQuant];
978 : chl 169
979 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
980 : Isibaar 539 Data->min_dx = EVEN(Data->min_dx);
981 :     Data->max_dx = EVEN(Data->max_dx);
982 :     Data->min_dy = EVEN(Data->min_dy);
983 :     Data->max_dy = EVEN(Data->max_dy); }
984 : edgomez 568
985 :     if (pMB->dquant != NO_CHANGE) inter4v = 0;
986 : edgomez 195
987 : Isibaar 580 for(i = 0; i < 5; i++)
988 :     Data->currentMV[i].x = Data->currentMV[i].y = 0;
989 : Isibaar 539
990 : Isibaar 606 if (pParam->m_quarterpel) {
991 :     Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
992 :     i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode);
993 :     } else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
994 :    
995 : Isibaar 539 Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i;
996 :     Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i;
997 :     Data->iMinSAD[2] = pMB->sad8[1];
998 :     Data->iMinSAD[3] = pMB->sad8[2];
999 :     Data->iMinSAD[4] = pMB->sad8[3];
1000 :    
1001 : chl 530 if ((x == 0) && (y == 0)) threshA = 512;
1002 :     else {
1003 : edgomez 568 threshA = Data->temp[0]; // that's when we keep this SAD atm
1004 : chl 530 if (threshA < 512) threshA = 512;
1005 :     if (threshA > 1024) threshA = 1024; }
1006 : Isibaar 3
1007 : chl 530 PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1008 :     prevMBs + x + y * pParam->mb_width);
1009 : edgomez 195
1010 : Isibaar 606 if (pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q;
1011 :     else
1012 :     if (inter4v) CheckCandidate = CheckCandidate16;
1013 :     else CheckCandidate = CheckCandidate16no4v;
1014 : edgomez 568
1015 : chl 530 /* main loop. checking all predictions */
1016 :    
1017 :     for (i = 1; i < 7; i++) {
1018 :     if (!(mask = make_mask(pmv, i)) ) continue;
1019 : edgomez 568 (*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1020 : Isibaar 539 if (Data->iMinSAD[0] <= threshA) break;
1021 : chl 530 }
1022 :    
1023 : Isibaar 539 if ((Data->iMinSAD[0] <= threshA) ||
1024 :     (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1025 :     (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1026 : chl 530 inter4v = 0;
1027 : Isibaar 539 } else {
1028 : Isibaar 3
1029 : Isibaar 539 MainSearchFunc * MainSearchPtr;
1030 :     if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1031 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1032 :     else MainSearchPtr = DiamondSearch;
1033 : chl 181
1034 : Isibaar 539 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1035 : Isibaar 3
1036 : chl 530 /* extended search, diamond starting in 0,0 and in prediction.
1037 :     note that this search is/might be done in halfpel positions,
1038 :     which makes it more different than the diamond above */
1039 : chl 259
1040 : Isibaar 539 if (MotionFlags & PMV_EXTSEARCH16) {
1041 :     int32_t bSAD;
1042 :     VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1043 :     if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1044 :     startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1045 :     if (!(MVequal(startMV, backupMV))) {
1046 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1047 : Isibaar 3
1048 : Isibaar 606 (*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1049 : Isibaar 539 (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1050 :     if (bSAD < Data->iMinSAD[0]) {
1051 :     Data->currentMV[0] = backupMV;
1052 :     Data->iMinSAD[0] = bSAD; }
1053 :     }
1054 : Isibaar 3
1055 : Isibaar 539 backupMV = Data->currentMV[0];
1056 :     if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1;
1057 :     else startMV.x = startMV.y = 0;
1058 :     if (!(MVequal(startMV, backupMV))) {
1059 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1060 : edgomez 195
1061 : Isibaar 606 (*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1062 : Isibaar 539 (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1063 :     if (bSAD < Data->iMinSAD[0]) {
1064 :     Data->currentMV[0] = backupMV;
1065 :     Data->iMinSAD[0] = bSAD; }
1066 :     }
1067 : Isibaar 3 }
1068 :     }
1069 :    
1070 : Isibaar 539 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1071 : Isibaar 3
1072 : Isibaar 580 for(i = 0; i < 5; i++) {
1073 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1074 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1075 :     }
1076 :    
1077 :     if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1078 :    
1079 : Isibaar 606 CheckCandidate = CheckCandidate16_qpel;
1080 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1081 : syskin 601 pParam->width, pParam->height, Data->iFcode, 0);
1082 :    
1083 : Isibaar 580 QuarterpelRefine(Data);
1084 :     }
1085 :    
1086 : Isibaar 539 if (inter4v) {
1087 :     SearchData Data8;
1088 :     Data8.iFcode = Data->iFcode;
1089 : Isibaar 606 Data8.lambda8 = Data->lambda8;
1090 : Isibaar 539 Data8.iEdgedWidth = Data->iEdgedWidth;
1091 : Isibaar 606 Data8.RefQ = Data->RefQ;
1092 : Isibaar 539 Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1093 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1094 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1095 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1096 :     }
1097 : Isibaar 3
1098 : chl 530 if (!(inter4v) ||
1099 : Isibaar 539 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1100 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1101 : chl 530 // INTER MODE
1102 :     pMB->mode = MODE_INTER;
1103 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
1104 :     = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1105 : Isibaar 3
1106 : Isibaar 580 pMB->qmvs[0] = pMB->qmvs[1]
1107 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1108 :    
1109 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1110 : Isibaar 539 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1111 : Isibaar 3
1112 : Isibaar 580 if(pParam->m_quarterpel) {
1113 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1114 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1115 :     }
1116 :     else {
1117 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1118 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1119 :     }
1120 : edgomez 195 } else {
1121 : chl 530 // INTER4V MODE; all other things are already set in Search8
1122 :     pMB->mode = MODE_INTER4V;
1123 : Isibaar 539 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1124 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1125 : edgomez 78 }
1126 : Isibaar 3 }
1127 :    
1128 : chl 530 static void
1129 :     Search8(const SearchData * const OldData,
1130 :     const int x, const int y,
1131 :     const uint32_t MotionFlags,
1132 :     const MBParam * const pParam,
1133 :     MACROBLOCK * const pMB,
1134 :     const MACROBLOCK * const pMBs,
1135 : Isibaar 539 const int block,
1136 :     SearchData * const Data)
1137 : chl 345 {
1138 : Isibaar 539 Data->iMinSAD = OldData->iMinSAD + 1 + block;
1139 :     Data->currentMV = OldData->currentMV + 1 + block;
1140 : Isibaar 580 Data->currentQMV = OldData->currentQMV + 1 + block;
1141 : chl 345
1142 : syskin 601 if(pParam->m_quarterpel) {
1143 : Isibaar 606 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1144 :     if (block != 0) *(Data->iMinSAD) += Data->lambda8 *
1145 : syskin 601 d_mv_bits( Data->currentQMV->x - Data->predQMV.x,
1146 :     Data->currentQMV->y - Data->predQMV.y,
1147 :     Data->iFcode);
1148 : Isibaar 606 CheckCandidate = CheckCandidate8Q;
1149 :     } else {
1150 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1151 : syskin 601 if (block != 0) *(Data->iMinSAD) += Data->lambda8 *
1152 :     d_mv_bits( Data->currentMV->x - Data->predMV.x,
1153 :     Data->currentMV->y - Data->predMV.y,
1154 :     Data->iFcode);
1155 : Isibaar 606 CheckCandidate = CheckCandidate8;
1156 :     }
1157 : chl 345
1158 : chl 530 if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
1159 : chl 345
1160 : Isibaar 539 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1161 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1162 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1163 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1164 : chl 345
1165 : Isibaar 539 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1166 : chl 530
1167 : Isibaar 539 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1168 : Isibaar 580 pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1169 : chl 345
1170 : chl 530 if (MotionFlags & PMV_EXTSEARCH8) {
1171 : Isibaar 580 int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1172 :    
1173 : chl 530 MainSearchFunc *MainSearchPtr;
1174 :     if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1175 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1176 :     else MainSearchPtr = DiamondSearch;
1177 : chl 345
1178 : Isibaar 580 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1179 : chl 345
1180 : syskin 601 if(*(Data->iMinSAD) < temp_sad) {
1181 : Isibaar 580 Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1182 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1183 :     }
1184 :     }
1185 :    
1186 :     if (MotionFlags & PMV_HALFPELREFINE8) {
1187 :     int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1188 :    
1189 :     HalfpelRefine(Data); // perform halfpel refine of current best vector
1190 :    
1191 :     if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1192 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1193 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1194 :     }
1195 :     }
1196 :    
1197 : syskin 601 if(pParam->m_quarterpel) {
1198 :     if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1199 :     (MotionFlags & PMV_QUARTERPELREFINE8)) {
1200 : Isibaar 600 CheckCandidate = CheckCandidate8_qpel;
1201 : Isibaar 606 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1202 :     pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1203 :     QuarterpelRefine(Data);
1204 : syskin 601 }
1205 : Isibaar 580 }
1206 : chl 345 }
1207 : chl 530
1208 : Isibaar 580 if(pParam->m_quarterpel) {
1209 :     pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1210 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1211 :     }
1212 :     else {
1213 :     pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1214 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1215 :     }
1216 :    
1217 : Isibaar 539 pMB->mvs[block] = *(Data->currentMV);
1218 : Isibaar 580 pMB->qmvs[block] = *(Data->currentQMV);
1219 :    
1220 : syskin 601 pMB->sad8[block] = 4 * (*Data->iMinSAD);
1221 : chl 345 }
1222 :    
1223 : chl 530 /* B-frames code starts here */
1224 : chl 345
1225 : chl 530 static __inline VECTOR
1226 :     ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1227 : Isibaar 3 {
1228 : chl 530 /* the stupidiest function ever */
1229 :     if (mode == MODE_FORWARD) return pMB->mvs[0];
1230 :     else return pMB->b_mvs[0];
1231 : Isibaar 3 }
1232 :    
1233 : chl 530 static void __inline
1234 :     PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1235 :     const uint32_t iWcount,
1236 :     const MACROBLOCK * const pMB,
1237 :     const uint32_t mode_curr)
1238 : Isibaar 3 {
1239 :    
1240 : chl 530 // [0] is prediction
1241 :     pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1242 : Isibaar 3
1243 : chl 530 pmv[1].x = pmv[1].y = 0; // [1] is zero
1244 : Isibaar 3
1245 : chl 530 pmv[2] = ChoosePred(pMB, mode_curr);
1246 :     pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1247 : edgomez 195
1248 : chl 530 if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1249 :     pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1250 : Isibaar 539 pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1251 :     } else pmv[3].x = pmv[3].y = 0;
1252 : Isibaar 3
1253 : chl 530 if (y != 0) {
1254 :     pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1255 :     pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1256 :     } else pmv[4].x = pmv[4].y = 0;
1257 : Isibaar 3
1258 : chl 530 if (x != 0) {
1259 :     pmv[5] = ChoosePred(pMB-1, mode_curr);
1260 :     pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1261 :     } else pmv[5].x = pmv[5].y = 0;
1262 : Isibaar 3
1263 : chl 530 if ((x != 0)&&(y != 0)) {
1264 :     pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1265 :     pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1266 :     } else pmv[6].x = pmv[6].y = 0;
1267 : edgomez 195
1268 : chl 530 // more?
1269 :     }
1270 : chl 181
1271 : edgomez 170
1272 : chl 530 /* search backward or forward, for b-frames */
1273 :     static void
1274 :     SearchBF( const uint8_t * const pRef,
1275 :     const uint8_t * const pRefH,
1276 :     const uint8_t * const pRefV,
1277 :     const uint8_t * const pRefHV,
1278 :     const IMAGE * const pCur,
1279 :     const int x, const int y,
1280 :     const uint32_t MotionFlags,
1281 :     const uint32_t iFcode,
1282 :     const MBParam * const pParam,
1283 :     MACROBLOCK * const pMB,
1284 :     const VECTOR * const predMV,
1285 :     int32_t * const best_sad,
1286 : h 545 const int32_t mode_current,
1287 :     SearchData * const Data)
1288 : chl 530 {
1289 : Isibaar 3
1290 : chl 530 const int32_t iEdgedWidth = pParam->edged_width;
1291 :    
1292 :     int i, iDirection, mask;
1293 : h 545 VECTOR pmv[7];
1294 : chl 530 MainSearchFunc *MainSearchPtr;
1295 : h 545 *Data->iMinSAD = MV_MAX_ERROR;
1296 :     Data->iFcode = iFcode;
1297 : Isibaar 3
1298 : h 545 Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1299 :     Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1300 :     Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1301 :     Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1302 : Isibaar 3
1303 : h 545 Data->predMV = *predMV;
1304 : edgomez 195
1305 : h 545 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1306 : Isibaar 580 pParam->width, pParam->height, iFcode, pParam->m_quarterpel);
1307 : Isibaar 3
1308 : h 545 pmv[0] = Data->predMV;
1309 :     PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1310 : edgomez 195
1311 : h 545 Data->currentMV->x = Data->currentMV->y = 0;
1312 : Isibaar 3
1313 : chl 530 CheckCandidate = CheckCandidate16no4v;
1314 : chl 181
1315 : chl 530 // main loop. checking all predictions
1316 :     for (i = 0; i < 8; i++) {
1317 :     if (!(mask = make_mask(pmv, i)) ) continue;
1318 : h 545 CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1319 : edgomez 78 }
1320 : Isibaar 3
1321 : chl 530 if (MotionFlags & PMV_USESQUARES16)
1322 :     MainSearchPtr = SquareSearch;
1323 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1324 :     MainSearchPtr = AdvDiamondSearch;
1325 :     else MainSearchPtr = DiamondSearch;
1326 : chl 96
1327 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1328 : chl 169
1329 : h 545 HalfpelRefine(Data);
1330 : chl 169
1331 : chl 530 // three bits are needed to code backward mode. four for forward
1332 :     // we treat the bits just like they were vector's
1333 : syskin 601 if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16 * 2;
1334 :     else *Data->iMinSAD += 3 * Data->lambda16 * 2;
1335 : chl 169
1336 : h 545 if (*Data->iMinSAD < *best_sad) {
1337 :     *best_sad = *Data->iMinSAD;
1338 : chl 530 pMB->mode = mode_current;
1339 : h 545 pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1340 :     pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1341 : syskin 601 if (mode_current == MODE_FORWARD) pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV;
1342 :     else pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search
1343 : chl 169 }
1344 : chl 530
1345 :     }
1346 : chl 169
1347 : chl 530 static int32_t
1348 : Isibaar 539 SearchDirect(const IMAGE * const f_Ref,
1349 : chl 530 const uint8_t * const f_RefH,
1350 :     const uint8_t * const f_RefV,
1351 :     const uint8_t * const f_RefHV,
1352 : Isibaar 539 const IMAGE * const b_Ref,
1353 : chl 530 const uint8_t * const b_RefH,
1354 :     const uint8_t * const b_RefV,
1355 :     const uint8_t * const b_RefHV,
1356 :     const IMAGE * const pCur,
1357 :     const int x, const int y,
1358 :     const uint32_t MotionFlags,
1359 :     const int32_t TRB, const int32_t TRD,
1360 :     const MBParam * const pParam,
1361 :     MACROBLOCK * const pMB,
1362 :     const MACROBLOCK * const b_mb,
1363 : h 545 int32_t * const best_sad,
1364 :     SearchData * const Data)
1365 : edgomez 195
1366 : chl 530 {
1367 : h 545 int32_t skip_sad;
1368 : chl 530 int k;
1369 : h 545
1370 : chl 530 MainSearchFunc *MainSearchPtr;
1371 : Isibaar 3
1372 : h 545 *Data->iMinSAD = 256*4096;
1373 :     Data->referencemv = b_mb->mvs;
1374 : chl 140
1375 : h 545 Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1376 :     Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16;
1377 :     Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16;
1378 :     Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16;
1379 :     Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1380 :     Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16;
1381 :     Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16;
1382 :     Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16;
1383 : chl 140
1384 : h 545 Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1385 :     Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1386 :     Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1387 :     Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1388 : Isibaar 3
1389 : chl 530 for (k = 0; k < 4; k++) {
1390 : h 545 pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1391 :     pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1392 :     pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1393 :     pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1394 : Isibaar 3
1395 : h 545 if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx )
1396 :     || ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) {
1397 : Isibaar 3
1398 : Isibaar 539 *best_sad = 256*4096; // in that case, we won't use direct mode
1399 :     pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1400 :     pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1401 :     return 0;
1402 :     }
1403 :     if (b_mb->mode != MODE_INTER4V) {
1404 :     pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1405 :     pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1406 : h 545 Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1407 :     Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1408 : Isibaar 539 break;
1409 :     }
1410 :     }
1411 : Isibaar 3
1412 : Isibaar 539 if (b_mb->mode == MODE_INTER4V)
1413 :     CheckCandidate = CheckCandidateDirect;
1414 :     else CheckCandidate = CheckCandidateDirectno4v;
1415 : edgomez 195
1416 : h 545 (*CheckCandidate)(0, 0, 255, &k, Data);
1417 : Isibaar 3
1418 : chl 530 // skip decision
1419 : syskin 601 if (*Data->iMinSAD - 2 * Data->lambda16 < (uint32_t)pMB->quant * SKIP_THRESH_B) {
1420 : syskin 594 //possible skip - checking chroma. everything copied from MC
1421 : Isibaar 539 //this is not full chroma compensation, only it's fullpel approximation. should work though
1422 :     int sum, dx, dy, b_dx, b_dy;
1423 : Isibaar 3
1424 : Isibaar 539 sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1425 :     dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1426 :    
1427 :     sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1428 :     dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1429 :    
1430 :     sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
1431 :     b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1432 :    
1433 :     sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
1434 :     b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1435 :    
1436 : h 545 sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2),
1437 :     f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1438 :     b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1439 :     Data->iEdgedWidth/2);
1440 :     sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2),
1441 :     f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1442 :     b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1443 :     Data->iEdgedWidth/2);
1444 : Isibaar 539
1445 : syskin 601 if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) {
1446 : Isibaar 539 pMB->mode = MODE_DIRECT_NONE_MV;
1447 : h 545 return *Data->iMinSAD;
1448 : Isibaar 539 }
1449 :     }
1450 :    
1451 : h 545 skip_sad = *Data->iMinSAD;
1452 : edgomez 195
1453 : chl 530 // DIRECT MODE DELTA VECTOR SEARCH.
1454 :     // This has to be made more effective, but at the moment I'm happy it's running at all
1455 : Isibaar 3
1456 : chl 530 if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1457 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1458 :     else MainSearchPtr = DiamondSearch;
1459 : edgomez 195
1460 : h 545 (*MainSearchPtr)(0, 0, Data, 255);
1461 : Isibaar 3
1462 : h 545 HalfpelRefine(Data);
1463 : Isibaar 3
1464 : syskin 601 *Data->iMinSAD += 1 * Data->lambda16 * 2; // one bit is needed to code direct mode
1465 : h 545 *best_sad = *Data->iMinSAD;
1466 : Isibaar 3
1467 : chl 530 if (b_mb->mode == MODE_INTER4V)
1468 :     pMB->mode = MODE_DIRECT;
1469 :     else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1470 : edgomez 195
1471 : h 545 pMB->pmvs[3] = *Data->currentMV;
1472 : edgomez 195
1473 : chl 530 for (k = 0; k < 4; k++) {
1474 : h 545 pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1475 :     pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
1476 :     ? Data->directmvB[k].x
1477 :     : pMB->mvs[k].x - Data->referencemv[k].x);
1478 :     pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1479 :     pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1480 :     ? Data->directmvB[k].y
1481 :     : pMB->mvs[k].y - Data->referencemv[k].y);
1482 : chl 530 if (b_mb->mode != MODE_INTER4V) {
1483 :     pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1484 :     pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1485 :     break;
1486 :     }
1487 :     }
1488 : h 545 return skip_sad;
1489 : Isibaar 3 }
1490 : chl 96
1491 : h 545
1492 : chl 530 static __inline void
1493 :     SearchInterpolate(const uint8_t * const f_Ref,
1494 :     const uint8_t * const f_RefH,
1495 :     const uint8_t * const f_RefV,
1496 :     const uint8_t * const f_RefHV,
1497 :     const uint8_t * const b_Ref,
1498 :     const uint8_t * const b_RefH,
1499 :     const uint8_t * const b_RefV,
1500 :     const uint8_t * const b_RefHV,
1501 :     const IMAGE * const pCur,
1502 :     const int x, const int y,
1503 :     const uint32_t fcode,
1504 :     const uint32_t bcode,
1505 :     const uint32_t MotionFlags,
1506 :     const MBParam * const pParam,
1507 :     const VECTOR * const f_predMV,
1508 :     const VECTOR * const b_predMV,
1509 :     MACROBLOCK * const pMB,
1510 : h 545 int32_t * const best_sad,
1511 :     SearchData * const fData)
1512 : chl 530
1513 : chl 96 {
1514 :    
1515 : edgomez 195 const int32_t iEdgedWidth = pParam->edged_width;
1516 : chl 96
1517 : chl 530 int iDirection, i, j;
1518 : h 545 SearchData bData;
1519 : chl 96
1520 : syskin 601 *(bData.iMinSAD = fData->iMinSAD) = 4096*256;
1521 : h 545 bData.Cur = fData->Cur;
1522 :     fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
1523 :     bData.currentMV = fData->currentMV + 1;
1524 : syskin 601 bData.lambda16 = fData->lambda16;
1525 : h 545 fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1526 : edgomez 195
1527 : h 545 bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
1528 :     bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
1529 :     bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
1530 :     bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
1531 :     bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
1532 :     bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
1533 :     bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
1534 :     bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
1535 : chl 96
1536 : h 545 bData.bpredMV = fData->predMV = *f_predMV;
1537 :     fData->bpredMV = bData.predMV = *b_predMV;
1538 : edgomez 195
1539 : syskin 601 fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed
1540 : Isibaar 580 get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel);
1541 :     get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel);
1542 : chl 96
1543 : h 545 if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1544 :     if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
1545 :     if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
1546 :     if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1547 : Isibaar 539
1548 : h 545 if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1549 :     if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
1550 :     if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
1551 :     if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1552 : Isibaar 539
1553 : h 545 CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1554 : chl 96
1555 : chl 530 //diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1556 : chl 96
1557 : chl 530 do {
1558 :     iDirection = 255;
1559 :     // forward MV moves
1560 : h 545 i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1561 : chl 96
1562 : h 545 CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1563 :     CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1564 :     CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1565 :     CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1566 : chl 96
1567 : chl 530 // backward MV moves
1568 : h 545 i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1569 :     fData->currentMV[2] = fData->currentMV[0];
1570 : chl 326
1571 : Isibaar 539 CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1572 :     CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1573 :     CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1574 :     CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1575 : chl 96
1576 : chl 530 } while (!(iDirection));
1577 : edgomez 195
1578 : chl 530 // two bits are needed to code interpolate mode. we treat the bits just like they were vector's
1579 : Isibaar 606 *fData->iMinSAD += 2 * fData->lambda16 * 2;
1580 :    
1581 : h 545 if (*fData->iMinSAD < *best_sad) {
1582 :     *best_sad = *fData->iMinSAD;
1583 :     pMB->mvs[0] = fData->currentMV[0];
1584 :     pMB->b_mvs[0] = fData->currentMV[1];
1585 : chl 530 pMB->mode = MODE_INTERPOLATE;
1586 : chl 96
1587 : chl 530 pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1588 :     pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1589 :     pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1590 :     pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1591 :     }
1592 :     }
1593 : chl 96
1594 : h 545
1595 : chl 530 void
1596 :     MotionEstimationBVOP(MBParam * const pParam,
1597 :     FRAMEINFO * const frame,
1598 :     const int32_t time_bp,
1599 :     const int32_t time_pp,
1600 :     // forward (past) reference
1601 :     const MACROBLOCK * const f_mbs,
1602 :     const IMAGE * const f_ref,
1603 :     const IMAGE * const f_refH,
1604 :     const IMAGE * const f_refV,
1605 :     const IMAGE * const f_refHV,
1606 :     // backward (future) reference
1607 :     const MACROBLOCK * const b_mbs,
1608 :     const IMAGE * const b_ref,
1609 :     const IMAGE * const b_refH,
1610 :     const IMAGE * const b_refV,
1611 :     const IMAGE * const b_refHV)
1612 :     {
1613 :     uint32_t i, j;
1614 :     int32_t best_sad, skip_sad;
1615 :     int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1616 :     static const VECTOR zeroMV={0,0};
1617 : chl 96
1618 : chl 530 VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1619 : chl 96
1620 : chl 530 const int32_t TRB = time_pp - time_bp;
1621 :     const int32_t TRD = time_pp;
1622 : chl 96
1623 : h 545 // some pre-inintialized data for the rest of the search
1624 :    
1625 :     SearchData Data;
1626 :     int32_t iMinSAD;
1627 :     VECTOR currentMV[3];
1628 :     Data.iEdgedWidth = pParam->edged_width;
1629 :     Data.currentMV = currentMV;
1630 :     Data.iMinSAD = &iMinSAD;
1631 : syskin 601 Data.lambda16 = lambda_vec16[frame->quant];
1632 : h 545
1633 : chl 530 // note: i==horizontal, j==vertical
1634 : chl 96
1635 : chl 530 for (j = 0; j < pParam->mb_height; j++) {
1636 : chl 96
1637 : chl 530 f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1638 : edgomez 195
1639 : chl 530 for (i = 0; i < pParam->mb_width; i++) {
1640 :     MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1641 :     const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1642 : edgomez 195
1643 : chl 619 /* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1644 :     if ( (b_mb->mode == MODE_NOT_CODED) ) {
1645 : chl 530 pMB->mode = MODE_NOT_CODED;
1646 :     continue;
1647 : chl 96 }
1648 :    
1649 : h 545 Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1650 : syskin 601 pMB->quant = frame->quant;
1651 : chl 530 /* direct search comes first, because it (1) checks for SKIP-mode
1652 :     and (2) sets very good predictions for forward and backward search */
1653 : chl 96
1654 : Isibaar 539 skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1655 :     b_ref, b_refH->y, b_refV->y, b_refHV->y,
1656 : chl 530 &frame->image,
1657 :     i, j,
1658 :     frame->motion_flags,
1659 :     TRB, TRD,
1660 :     pParam,
1661 :     pMB, b_mb,
1662 : h 545 &best_sad,
1663 :     &Data);
1664 : chl 96
1665 : Isibaar 539 if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1666 : chl 96
1667 : chl 530 // best_sad = 256*4096; //uncomment to disable Directsearch.
1668 :     // To disable any other mode, just comment the function call
1669 : chl 96
1670 : chl 530 // forward search
1671 :     SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1672 :     &frame->image, i, j,
1673 :     frame->motion_flags,
1674 : h 545 frame->fcode, pParam,
1675 : chl 530 pMB, &f_predMV, &best_sad,
1676 : h 545 MODE_FORWARD, &Data);
1677 : chl 96
1678 : chl 530 // backward search
1679 :     SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1680 :     &frame->image, i, j,
1681 :     frame->motion_flags,
1682 : h 545 frame->bcode, pParam,
1683 : chl 530 pMB, &b_predMV, &best_sad,
1684 : h 545 MODE_BACKWARD, &Data);
1685 : chl 96
1686 : chl 530 // interpolate search comes last, because it uses data from forward and backward as prediction
1687 : chl 96
1688 : chl 530 SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1689 :     b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1690 :     &frame->image,
1691 :     i, j,
1692 :     frame->fcode, frame->bcode,
1693 :     frame->motion_flags,
1694 : h 545 pParam,
1695 : chl 530 &f_predMV, &b_predMV,
1696 : h 545 pMB, &best_sad,
1697 :     &Data);
1698 : chl 96
1699 : chl 530 switch (pMB->mode) {
1700 :     case MODE_FORWARD:
1701 :     f_count++;
1702 :     f_predMV = pMB->mvs[0];
1703 :     break;
1704 :     case MODE_BACKWARD:
1705 :     b_count++;
1706 :     b_predMV = pMB->b_mvs[0];
1707 :     break;
1708 :     case MODE_INTERPOLATE:
1709 :     i_count++;
1710 :     f_predMV = pMB->mvs[0];
1711 :     b_predMV = pMB->b_mvs[0];
1712 :     break;
1713 :     case MODE_DIRECT:
1714 :     case MODE_DIRECT_NO4V:
1715 :     d_count++;
1716 :     break;
1717 :     default:
1718 :     break;
1719 : chl 96 }
1720 :     }
1721 :     }
1722 :    
1723 : chl 530 // fprintf(debug,"B-Stat: F: %04d B: %04d I: %04d D: %04d, N: %04d\n",
1724 :     // f_count,b_count,i_count,d_count,n_count);
1725 : chl 96
1726 :     }
1727 :    
1728 : chl 530 /* Hinted ME starts here */
1729 : chl 96
1730 : syskin 594 static void
1731 :     Search8hinted(const SearchData * const OldData,
1732 : syskin 601 const int x, const int y,
1733 :     const uint32_t MotionFlags,
1734 :     const MBParam * const pParam,
1735 :     MACROBLOCK * const pMB,
1736 :     const MACROBLOCK * const pMBs,
1737 :     const int block,
1738 :     SearchData * const Data)
1739 : chl 96 {
1740 : syskin 594 int32_t temp_sad;
1741 : chl 530 MainSearchFunc *MainSearchPtr;
1742 : syskin 594 Data->iMinSAD = OldData->iMinSAD + 1 + block;
1743 :     Data->currentMV = OldData->currentMV + 1 + block;
1744 :     Data->currentQMV = OldData->currentQMV + 1 + block;
1745 : Isibaar 606 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1746 : chl 141
1747 : syskin 601 if(pParam->m_quarterpel) {
1748 : Isibaar 606 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1749 :     if (block != 0) *(Data->iMinSAD) += Data->lambda8 *
1750 : syskin 594 d_mv_bits( Data->currentQMV->x - Data->predQMV.x,
1751 :     Data->currentQMV->y - Data->predQMV.y,
1752 :     Data->iFcode);
1753 : Isibaar 606 CheckCandidate = CheckCandidate8Q;
1754 :     } else {
1755 : syskin 601 if (block != 0) *(Data->iMinSAD) += Data->lambda8 *
1756 :     d_mv_bits( Data->currentMV->x - Data->predMV.x,
1757 :     Data->currentMV->y - Data->predMV.y,
1758 :     Data->iFcode);
1759 : Isibaar 606 CheckCandidate = CheckCandidate8;
1760 :     }
1761 : chl 96
1762 : syskin 594 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1763 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1764 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1765 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1766 : edgomez 195
1767 : syskin 594 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1768 :    
1769 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1770 : Isibaar 580 pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1771 : chl 96
1772 : syskin 601 temp_sad = *(Data->iMinSAD); // store current MinSAD
1773 : syskin 594
1774 : chl 530 if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1775 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1776 :     else MainSearchPtr = DiamondSearch;
1777 : chl 96
1778 : syskin 594 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1779 : edgomez 195
1780 : syskin 594 if(*(Data->iMinSAD) < temp_sad) {
1781 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1782 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1783 :     }
1784 : chl 96
1785 : syskin 594 if (MotionFlags & PMV_HALFPELREFINE8) {
1786 :     temp_sad = *(Data->iMinSAD); // store current MinSAD
1787 :    
1788 :     HalfpelRefine(Data); // perform halfpel refine of current best vector
1789 :    
1790 :     if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1791 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1792 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1793 :     }
1794 :     }
1795 :    
1796 : syskin 601 if(pParam->m_quarterpel) {
1797 :     if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1798 :     (MotionFlags & PMV_QUARTERPELREFINE8)) {
1799 : Isibaar 606 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1800 :     pParam->width, pParam->height, Data->iFcode, 0);
1801 : syskin 601 CheckCandidate = CheckCandidate8_qpel;
1802 : Isibaar 606 QuarterpelRefine(Data);
1803 : syskin 601 }
1804 :     pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1805 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1806 :     } else {
1807 : syskin 594 pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1808 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1809 :     }
1810 :    
1811 :     pMB->mvs[block] = *(Data->currentMV);
1812 :     pMB->qmvs[block] = *(Data->currentQMV);
1813 :    
1814 :     pMB->sad8[block] = 4 * (*Data->iMinSAD);
1815 : chl 96 }
1816 :    
1817 : chl 530 static void
1818 :     SearchPhinted ( const uint8_t * const pRef,
1819 : chl 289 const uint8_t * const pRefH,
1820 :     const uint8_t * const pRefV,
1821 :     const uint8_t * const pRefHV,
1822 :     const IMAGE * const pCur,
1823 :     const int x,
1824 :     const int y,
1825 :     const uint32_t MotionFlags,
1826 :     const uint32_t iQuant,
1827 :     const MBParam * const pParam,
1828 :     const MACROBLOCK * const pMBs,
1829 : chl 530 int inter4v,
1830 : h 545 MACROBLOCK * const pMB,
1831 : edgomez 568 SearchData * const Data)
1832 : chl 289 {
1833 : chl 530
1834 : chl 289 const int32_t iEdgedWidth = pParam->edged_width;
1835 : chl 530
1836 : edgomez 568 int i, t;
1837 : chl 530 MainSearchFunc * MainSearchPtr;
1838 : suxen_drol 118
1839 : h 545 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1840 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1841 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1842 : suxen_drol 118
1843 : h 545 Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16;
1844 :     Data->Ref = pRef + (x + iEdgedWidth*y)*16;
1845 :     Data->RefH = pRefH + (x + iEdgedWidth*y) * 16;
1846 :     Data->RefV = pRefV + (x + iEdgedWidth*y) * 16;
1847 :     Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
1848 : syskin 601 Data->lambda16 = lambda_vec16[iQuant];
1849 : Isibaar 606 Data->lambda8 = lambda_vec8[iQuant];
1850 : Isibaar 600
1851 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
1852 : h 545 Data->min_dx = EVEN(Data->min_dx);
1853 :     Data->max_dx = EVEN(Data->max_dx);
1854 :     Data->min_dy = EVEN(Data->min_dy);
1855 :     Data->max_dy = EVEN(Data->max_dy);
1856 : chl 530 }
1857 : chl 289
1858 : h 545 for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1859 : chl 289
1860 : chl 530 if (pMB->dquant != NO_CHANGE) inter4v = 0;
1861 : chl 289
1862 : Isibaar 606 if(pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q;
1863 :     else
1864 :     if (inter4v) CheckCandidate = CheckCandidate16;
1865 :     else CheckCandidate = CheckCandidate16no4v;
1866 : chl 289
1867 : chl 530 pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
1868 :     pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
1869 : h 545 if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
1870 :     if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
1871 :     if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
1872 :     if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
1873 : chl 370
1874 : edgomez 568 (*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
1875 : chl 289
1876 : chl 530 if (pMB->mode == MODE_INTER4V)
1877 :     for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
1878 :     pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
1879 :     pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
1880 :     if (!(make_mask(pMB->mvs, i)))
1881 : edgomez 568 (*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
1882 : chl 289 }
1883 :    
1884 :     if (MotionFlags & PMV_USESQUARES16)
1885 : chl 530 MainSearchPtr = SquareSearch;
1886 : chl 289 else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1887 : chl 530 MainSearchPtr = AdvDiamondSearch;
1888 :     else MainSearchPtr = DiamondSearch;
1889 : chl 289
1890 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1891 : chl 289
1892 : h 545 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1893 : chl 289
1894 : syskin 594 for(i = 0; i < 5; i++) {
1895 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1896 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1897 :     }
1898 : chl 289
1899 : syskin 594 if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1900 : Isibaar 606 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1901 :     pParam->width, pParam->height, Data->iFcode, 0);
1902 :     CheckCandidate = CheckCandidate16_qpel;
1903 : syskin 594 QuarterpelRefine(Data);
1904 :     }
1905 :    
1906 :     if (inter4v) {
1907 :     SearchData Data8;
1908 :     Data8.iFcode = Data->iFcode;
1909 : Isibaar 606 Data8.lambda8 = Data->lambda8;
1910 : syskin 594 Data8.iEdgedWidth = Data->iEdgedWidth;
1911 : Isibaar 606 Data8.RefQ = Data->RefQ;
1912 : syskin 594 Search8hinted(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1913 :     Search8hinted(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1914 :     Search8hinted(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1915 :     Search8hinted(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1916 :     }
1917 :    
1918 : chl 530 if (!(inter4v) ||
1919 : h 545 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] +
1920 :     Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1921 : chl 530 // INTER MODE
1922 :     pMB->mode = MODE_INTER;
1923 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
1924 : h 545 = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1925 : chl 289
1926 : Isibaar 606 pMB->qmvs[0] = pMB->qmvs[1]
1927 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1928 :    
1929 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1930 : h 545 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1931 : chl 289
1932 : Isibaar 606 if(pParam->m_quarterpel) {
1933 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1934 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1935 :     }
1936 :     else {
1937 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1938 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1939 :     }
1940 : chl 530 } else {
1941 :     // INTER4V MODE; all other things are already set in Search8hinted
1942 :     pMB->mode = MODE_INTER4V;
1943 : h 545 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3]
1944 :     + Data->iMinSAD[4] + IMV16X16 * iQuant;
1945 : chl 289 }
1946 :    
1947 :     }
1948 :    
1949 : edgomez 195 void
1950 : chl 530 MotionEstimationHinted( MBParam * const pParam,
1951 :     FRAMEINFO * const current,
1952 :     FRAMEINFO * const reference,
1953 :     const IMAGE * const pRefH,
1954 :     const IMAGE * const pRefV,
1955 :     const IMAGE * const pRefHV)
1956 : suxen_drol 118 {
1957 : chl 530 MACROBLOCK *const pMBs = current->mbs;
1958 :     const IMAGE *const pCurrent = &current->image;
1959 :     const IMAGE *const pRef = &reference->image;
1960 : suxen_drol 118
1961 : chl 530 uint32_t x, y;
1962 : syskin 601 uint8_t * qimage;
1963 : syskin 574 int32_t temp[5], quant = current->quant;
1964 : h 545 int32_t iMinSAD[5];
1965 : Isibaar 606 VECTOR currentMV[5], currentQMV[5];
1966 : h 545 SearchData Data;
1967 :     Data.iEdgedWidth = pParam->edged_width;
1968 :     Data.currentMV = currentMV;
1969 : Isibaar 606 Data.currentQMV = currentQMV;
1970 : h 545 Data.iMinSAD = iMinSAD;
1971 :     Data.temp = temp;
1972 :     Data.iFcode = current->fcode;
1973 : syskin 594 Data.rounding = pParam->m_rounding_type;
1974 : h 545
1975 : syskin 594 if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
1976 :     return; // allocate some mem for qpel interpolated blocks
1977 :     // somehow this is dirty since I think we shouldn't use malloc outside
1978 :     // encoder_create() - so please fix me!
1979 :    
1980 : syskin 601 Data.RefQ = qimage;
1981 :    
1982 : chl 530 if (sadInit) (*sadInit) ();
1983 : suxen_drol 118
1984 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
1985 :     for (x = 0; x < pParam->mb_width; x++) {
1986 : chl 312
1987 : chl 530 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
1988 : suxen_drol 118
1989 : chl 530 //intra mode is copied from the first pass. At least for the time being
1990 :     if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue;
1991 : chl 346
1992 : syskin 574
1993 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
1994 :     pMB->dquant = NO_CHANGE;
1995 :     pMB->quant = current->quant; }
1996 : Isibaar 606 else {
1997 : syskin 574 if (pMB->dquant != NO_CHANGE) {
1998 :     quant += DQtab[pMB->dquant];
1999 :     if (quant > 31) quant = 31;
2000 :     else if (quant < 1) quant = 1;
2001 :     }
2002 : Isibaar 606 pMB->quant = quant;
2003 :     }
2004 : chl 341
2005 : syskin 601 SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
2006 : Isibaar 539 y, current->motion_flags, pMB->quant,
2007 : h 545 pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
2008 :     &Data);
2009 : chl 317
2010 : Isibaar 539 }
2011 :     }
2012 : syskin 594 free(qimage);
2013 : Isibaar 539 }
2014 : suxen_drol 118
2015 : Isibaar 539 static __inline int
2016 :     MEanalyzeMB ( const uint8_t * const pRef,
2017 :     const uint8_t * const pCur,
2018 :     const int x,
2019 :     const int y,
2020 :     const MBParam * const pParam,
2021 :     const MACROBLOCK * const pMBs,
2022 : h 545 MACROBLOCK * const pMB,
2023 :     SearchData * const Data)
2024 : Isibaar 539 {
2025 : chl 317
2026 : Isibaar 606 int i = 255, mask;
2027 : h 545 VECTOR pmv[3];
2028 : suxen_drol 118
2029 : h 545 *(Data->iMinSAD) = MV_MAX_ERROR;
2030 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
2031 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
2032 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
2033 : chl 326
2034 : h 545 Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
2035 :     Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
2036 : Isibaar 539
2037 :     CheckCandidate = CheckCandidate16no4vI;
2038 : chl 326
2039 : Isibaar 539 pmv[1].x = EVEN(pMB->mvs[0].x);
2040 :     pmv[1].y = EVEN(pMB->mvs[0].y);
2041 : h 545 pmv[0].x = EVEN(Data->predMV.x);
2042 :     pmv[0].y = EVEN(Data->predMV.y);
2043 : Isibaar 539 pmv[2].x = pmv[2].y = 0;
2044 : chl 326
2045 : h 545 CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
2046 : Isibaar 539 if (!(mask = make_mask(pmv, 1)))
2047 : h 545 CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
2048 : Isibaar 539 if (!(mask = make_mask(pmv, 2)))
2049 : h 545 CheckCandidate16no4vI(0, 0, mask, &i, Data);
2050 : Isibaar 539
2051 : Isibaar 606 DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
2052 : Isibaar 539
2053 : Isibaar 606 pMB->mvs[0] = *Data->currentMV;
2054 : Isibaar 539
2055 : h 545 return *(Data->iMinSAD);
2056 : Isibaar 539 }
2057 :    
2058 :     #define INTRA_THRESH 1350
2059 :     #define INTER_THRESH 900
2060 :    
2061 :     int
2062 :     MEanalysis( const IMAGE * const pRef,
2063 :     const IMAGE * const pCurrent,
2064 :     MBParam * const pParam,
2065 :     MACROBLOCK * const pMBs,
2066 :     const uint32_t iFcode)
2067 :     {
2068 :     uint32_t x, y, intra = 0;
2069 :     int sSAD = 0;
2070 : h 545
2071 :     VECTOR currentMV;
2072 :     int32_t iMinSAD;
2073 :     SearchData Data;
2074 :     Data.iEdgedWidth = pParam->edged_width;
2075 :     Data.currentMV = &currentMV;
2076 :     Data.iMinSAD = &iMinSAD;
2077 :     Data.iFcode = iFcode;
2078 :    
2079 : Isibaar 539 if (sadInit) (*sadInit) ();
2080 :    
2081 : syskin 601 for (y = 1; y < pParam->mb_height-1; y++) {
2082 :     for (x = 1; x < pParam->mb_width-1; x++) {
2083 : Isibaar 539 int sad, dev;
2084 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
2085 :    
2086 :     sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
2087 : h 545 pParam, pMBs, pMB, &Data);
2088 : Isibaar 539
2089 : syskin 601 if (sad > INTRA_THRESH) {
2090 :     dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
2091 :     pParam->edged_width);
2092 :     if (dev + INTRA_THRESH < sad) intra++;
2093 :     if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame
2094 : Isibaar 539 }
2095 : syskin 601 sSAD += sad;
2096 : suxen_drol 118 }
2097 :     }
2098 : Isibaar 539 sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2099 :     if (sSAD > INTER_THRESH ) return 1; //P frame
2100 :     emms();
2101 :     return 0; // B frame
2102 :    
2103 : suxen_drol 118 }
2104 : syskin 576
2105 :     int
2106 :     FindFcode( const MBParam * const pParam,
2107 :     const FRAMEINFO * const current)
2108 :     {
2109 :     uint32_t x, y;
2110 :     int max = 0, min = 0, i;
2111 :    
2112 :     for (y = 0; y < pParam->mb_height; y++) {
2113 :     for (x = 0; x < pParam->mb_width; x++) {
2114 :    
2115 :     MACROBLOCK *pMB = &current->mbs[x + y * pParam->mb_width];
2116 :     for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) {
2117 :     if (pMB->mvs[i].x > max) max = pMB->mvs[i].x;
2118 :     if (pMB->mvs[i].y > max) max = pMB->mvs[i].y;
2119 :    
2120 :     if (pMB->mvs[i].x < min) min = pMB->mvs[i].x;
2121 :     if (pMB->mvs[i].y < min) min = pMB->mvs[i].y;
2122 :     }
2123 :     }
2124 :     }
2125 :    
2126 :     min = -min;
2127 :     max += 1;
2128 :     if (min > max) max = min;
2129 : Isibaar 606 if (pParam->m_quarterpel) max *= 2;
2130 : syskin 576
2131 :     for (i = 1; (max > 32 << (i - 1)); i++);
2132 :     return i;
2133 :     }
2134 : chl 619
2135 :    
2136 :     /* Global Motion Estimation, (C) 2002 by sysKin */
2137 :     /* calculate global translation from local (halfpel) motion field */
2138 :    
2139 :     static VECTOR
2140 :     GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2141 :     {
2142 :    
2143 :     int count, bestcount = 0;
2144 :     int x, y;
2145 :     VECTOR gmc;
2146 :     int step, min_x, max_x, min_y, max_y;
2147 :     uint32_t mx, my;
2148 :    
2149 :     min_x = min_y = -32<<iFcode;
2150 :     max_x = max_y = 32<<iFcode;
2151 :    
2152 :     for (step = 32; step >= 2; step /= 2) {
2153 :     bestcount = 0;
2154 :     for (y = min_y; y <= max_y; y += step)
2155 :     for (x = min_x ; x <= max_x; x += step) {
2156 :     count = 0;
2157 :     //for all macroblocks
2158 :     for (my = 1; my < pParam->mb_height-1; my++)
2159 :     for (mx = 1; mx < pParam->mb_width-1; mx++) {
2160 :     const MACROBLOCK *pMB = &pMBs[mx + my * pParam->mb_width];
2161 :     VECTOR mv;
2162 :    
2163 :     if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED)
2164 :     continue;
2165 :    
2166 :     mv = pMB->mvs[0];
2167 :     if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2168 :     count++;
2169 :     }
2170 :     if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2171 :     }
2172 :     min_x = gmc.x - step;
2173 :     max_x = gmc.x + step;
2174 :     min_y = gmc.y - step;
2175 :     max_y = gmc.y + step;
2176 :    
2177 :     }
2178 :     if (pParam->m_quarterpel) {
2179 :     gmc.x *= 2;
2180 :     gmc.y *= 2; /* we store the halfpel value as pseudo-qpel to make comparison easier */
2181 :     }
2182 :    
2183 :     if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2184 :     gmc.x = gmc.y = 0; //no camara pan, no GMC
2185 :    
2186 :     return gmc;
2187 :     }

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