[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 631 - (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 : syskin 628 static int32_t
93 :     ChromaSAD(int dx, int dy, const SearchData * const data)
94 : Isibaar 3 {
95 : syskin 628 int sad;
96 :     dx = (dx >> 1) + roundtab_79[dx & 0x3];
97 :     dy = (dy >> 1) + roundtab_79[dy & 0x3];
98 : Isibaar 3
99 : syskin 628 switch (((dx & 1) << 1) + (dy & 1)) { // ((dx%2)?2:0)+((dy%2)?1:0)
100 :     case 0:
101 :     sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
102 :     sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2);
103 :     break;
104 :     case 1:
105 :     dx = dx / 2; dy = (dy - 1) / 2;
106 :     sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
107 :     sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + (dy+1) * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2);
108 :     break;
109 :     case 2:
110 :     dx = (dx - 1) / 2; dy = dy / 2;
111 :     sad = sad8bi(data->CurU, data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->RefCU + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
112 :     sad += sad8bi(data->CurV, data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->RefCV + dy * (data->iEdgedWidth/2) + dx+1, data->iEdgedWidth/2);
113 :     break;
114 :     default:
115 :     dx = (dx - 1) / 2; dy = (dy - 1) / 2;
116 :     interpolate8x8_halfpel_hv(data->RefQ,
117 :     data->RefCU + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
118 :     data->rounding);
119 :     sad = sad8(data->CurU, data->RefQ, data->iEdgedWidth/2);
120 :     interpolate8x8_halfpel_hv(data->RefQ,
121 :     data->RefCV + dy * (data->iEdgedWidth/2) + dx, data->iEdgedWidth/2,
122 :     data->rounding);
123 :     sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2);
124 :     break;
125 : chl 530 }
126 : syskin 628 return sad;
127 :     }
128 : Isibaar 3
129 : Isibaar 600
130 : syskin 628 /* CHECK_CANDIATE FUNCTIONS START */
131 : chl 184
132 :    
133 : chl 530 static void
134 : syskin 628 CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
135 : Isibaar 606 {
136 :     int t;
137 :     const uint8_t * Reference;
138 :    
139 :     if (( x > data->max_dx) || ( x < data->min_dx)
140 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
141 :    
142 :     switch ( ((x&1)<<1) + (y&1) ) {
143 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
144 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
145 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
146 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
147 :     }
148 :    
149 :     data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
150 :    
151 : syskin 628 if (data->qpel) t = d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
152 :     else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
153 : Isibaar 606
154 : syskin 628 data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
155 :     data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
156 :    
157 :     if (data->chroma) data->temp[0] += ChromaSAD(x, y, data);
158 :    
159 : Isibaar 606 if (data->temp[0] < data->iMinSAD[0]) {
160 :     data->iMinSAD[0] = data->temp[0];
161 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
162 :     *dir = Direction; }
163 :    
164 :     if (data->temp[1] < data->iMinSAD[1]) {
165 :     data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
166 :     if (data->temp[2] < data->iMinSAD[2]) {
167 :     data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
168 :     if (data->temp[3] < data->iMinSAD[3]) {
169 :     data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
170 :     if (data->temp[4] < data->iMinSAD[4]) {
171 :     data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
172 :    
173 :     }
174 :    
175 :     static void
176 : chl 530 CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
177 :     {
178 :     int32_t sad;
179 :     const uint8_t * Reference;
180 : chl 326
181 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
182 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
183 : Isibaar 3
184 : chl 530 switch ( ((x&1)<<1) + (y&1) )
185 :     {
186 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
187 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
188 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
189 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
190 :     }
191 : Isibaar 3
192 : syskin 628 sad = sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
193 :     sad += (data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode) * sad)/1000;
194 : Isibaar 3
195 : chl 530 if (sad < *(data->iMinSAD)) {
196 :     *(data->iMinSAD) = sad;
197 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
198 :     *dir = Direction; }
199 : Isibaar 3 }
200 :    
201 : chl 530 static void
202 : Isibaar 580 CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
203 :    
204 :     // CheckCandidate16 variant which expects x and y in quarter pixel resolution
205 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
206 :     // around currentMV!
207 :     {
208 :     int t;
209 : syskin 601 uint8_t * Reference = (uint8_t *)data->RefQ;
210 : Isibaar 580 const uint8_t *ref1, *ref2, *ref3, *ref4;
211 :     VECTOR halfpelMV = *(data->currentMV);
212 :    
213 :     int32_t iEdgedWidth = data->iEdgedWidth;
214 :     uint32_t rounding = data->rounding;
215 :    
216 :     if (( x > data->max_dx) || ( x < data->min_dx)
217 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
218 :    
219 : Isibaar 606 GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases
220 : Isibaar 580 switch( ((x&1)<<1) + (y&1) )
221 :     {
222 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
223 : syskin 601 GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
224 : Isibaar 580 break;
225 :    
226 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
227 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
228 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
229 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
230 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
231 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
232 :     break;
233 :    
234 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
235 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
236 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
237 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
238 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
239 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
240 :     break;
241 :    
242 :     default: // x and y in qpel resolution - the "corners" (top left/right and
243 :     // bottom left/right) during qpel refinement
244 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
245 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
246 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
247 :    
248 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
249 :     interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
250 :     interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
251 :     interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
252 :     break;
253 :     }
254 :    
255 : syskin 594 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1);
256 : Isibaar 580
257 :     t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
258 : syskin 628 data->temp[0] += (data->lambda16 * t * data->temp[0])/1000;
259 :     data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100;
260 : Isibaar 580
261 : syskin 628 if (data->chroma)
262 :     data->temp[0] += ChromaSAD(x/2, y/2, data);
263 :    
264 : Isibaar 580 if (data->temp[0] < data->iMinSAD[0]) {
265 :     data->iMinSAD[0] = data->temp[0];
266 :     data->currentQMV[0].x = x; data->currentQMV[0].y = y;
267 : syskin 594 /* *dir = Direction;*/ }
268 : Isibaar 580
269 :     if (data->temp[1] < data->iMinSAD[1]) {
270 :     data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; }
271 :     if (data->temp[2] < data->iMinSAD[2]) {
272 :     data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; }
273 :     if (data->temp[3] < data->iMinSAD[3]) {
274 :     data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; }
275 :     if (data->temp[4] < data->iMinSAD[4]) {
276 :     data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; }
277 :     }
278 :    
279 :     static void
280 : Isibaar 539 CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
281 :     {
282 :     int32_t sad;
283 :    
284 :     if (( x > data->max_dx) || ( x < data->min_dx)
285 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
286 :    
287 : syskin 601 sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
288 : Isibaar 539 data->iEdgedWidth, 256*4096);
289 :    
290 :     if (sad < *(data->iMinSAD)) {
291 :     *(data->iMinSAD) = sad;
292 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
293 :     *dir = Direction; }
294 :     }
295 :    
296 :    
297 :     static void
298 : chl 530 CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
299 :     {
300 :     int32_t sad;
301 :     const int xb = data->currentMV[1].x;
302 :     const int yb = data->currentMV[1].y;
303 :     const uint8_t *ReferenceF, *ReferenceB;
304 : Isibaar 3
305 : chl 530 if (( xf > data->max_dx) || ( xf < data->min_dx)
306 :     || ( yf > data->max_dy) || (yf < data->min_dy)) return;
307 : Isibaar 3
308 : chl 530 switch ( ((xf&1)<<1) + (yf&1) ) {
309 :     case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break;
310 :     case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
311 :     case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break;
312 :     default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
313 :     }
314 : edgomez 195
315 : chl 530 switch ( ((xb&1)<<1) + (yb&1) ) {
316 :     case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break;
317 :     case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
318 :     case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break;
319 :     default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
320 :     }
321 : Isibaar 3
322 : syskin 628 sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
323 :    
324 :     sad += (data->lambda16 *
325 : chl 530 ( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) +
326 : syskin 628 d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode)) * sad)/1000;
327 : Isibaar 3
328 : chl 530 if (sad < *(data->iMinSAD)) {
329 :     *(data->iMinSAD) = sad;
330 :     data->currentMV->x = xf; data->currentMV->y = yf;
331 :     *dir = Direction; }
332 : Isibaar 3 }
333 :    
334 : chl 530 static void
335 :     CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
336 : Isibaar 3 {
337 : syskin 628 int32_t sad = 0;
338 : chl 530 int k;
339 :     const uint8_t *ReferenceF;
340 :     const uint8_t *ReferenceB;
341 :     VECTOR mvs, b_mvs;
342 : Isibaar 3
343 : chl 530 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
344 : Isibaar 3
345 : chl 530 for (k = 0; k < 4; k++) {
346 :     mvs.x = data->directmvF[k].x + x;
347 :     b_mvs.x = ((x == 0) ?
348 :     data->directmvB[k].x
349 :     : mvs.x - data->referencemv[k].x);
350 : Isibaar 3
351 : chl 530 mvs.y = data->directmvF[k].y + y;
352 :     b_mvs.y = ((y == 0) ?
353 :     data->directmvB[k].y
354 :     : mvs.y - data->referencemv[k].y);
355 :    
356 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
357 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
358 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
359 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
360 : Isibaar 3
361 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
362 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
363 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
364 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
365 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
366 :     }
367 : Isibaar 3
368 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
369 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
370 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
371 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
372 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
373 : chl 344 }
374 : chl 530
375 :     sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
376 :     ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
377 :     ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
378 :     data->iEdgedWidth);
379 :     if (sad > *(data->iMinSAD)) return;
380 : edgomez 78 }
381 : chl 530
382 : syskin 628 sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000;
383 :    
384 : chl 530 if (sad < *(data->iMinSAD)) {
385 :     *(data->iMinSAD) = sad;
386 :     data->currentMV->x = x; data->currentMV->y = y;
387 :     *dir = Direction; }
388 : Isibaar 3 }
389 :    
390 : chl 530 static void
391 :     CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
392 : chl 96 {
393 : chl 530 int32_t sad;
394 :     const uint8_t *ReferenceF;
395 :     const uint8_t *ReferenceB;
396 :     VECTOR mvs, b_mvs;
397 : chl 96
398 : syskin 576 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
399 : edgomez 195
400 : chl 530 mvs.x = data->directmvF[0].x + x;
401 :     b_mvs.x = ((x == 0) ?
402 :     data->directmvB[0].x
403 :     : mvs.x - data->referencemv[0].x);
404 : edgomez 195
405 : chl 530 mvs.y = data->directmvF[0].y + y;
406 :     b_mvs.y = ((y == 0) ?
407 :     data->directmvB[0].y
408 :     : mvs.y - data->referencemv[0].y);
409 :    
410 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
411 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
412 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
413 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
414 : chl 96
415 : chl 530 switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
416 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
417 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
418 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
419 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
420 :     }
421 : chl 96
422 : chl 530 switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
423 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
424 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
425 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
426 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
427 :     }
428 :    
429 : syskin 628 sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
430 :     sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000;
431 : chl 96
432 : chl 530 if (sad < *(data->iMinSAD)) {
433 :     *(data->iMinSAD) = sad;
434 :     data->currentMV->x = x; data->currentMV->y = y;
435 :     *dir = Direction; }
436 :     }
437 : chl 96
438 : chl 530 static void
439 :     CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
440 :     {
441 : syskin 628 int32_t sad; int t;
442 : chl 530 const uint8_t * Reference;
443 : edgomez 195
444 : chl 530 if (( x > data->max_dx) || ( x < data->min_dx)
445 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
446 : chl 96
447 : chl 530 switch ( ((x&1)<<1) + (y&1) )
448 :     {
449 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
450 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
451 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
452 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
453 :     }
454 : edgomez 195
455 : chl 530 sad = sad8(data->Cur, Reference, data->iEdgedWidth);
456 : syskin 628 if (data->qpel) t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
457 :     else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
458 : edgomez 195
459 : syskin 628 sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100;
460 : chl 96
461 : Isibaar 606 if (sad < *(data->iMinSAD)) {
462 :     *(data->iMinSAD) = sad;
463 :     data->currentMV->x = x; data->currentMV->y = y;
464 :     *dir = Direction; }
465 :     }
466 :    
467 :     static void
468 : Isibaar 580 CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
469 :     // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
470 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
471 :     // around currentMV!
472 :    
473 :     {
474 :     int32_t sad;
475 :     uint8_t *Reference = (uint8_t *) data->RefQ;
476 :     const uint8_t *ref1, *ref2, *ref3, *ref4;
477 :     VECTOR halfpelMV = *(data->currentMV);
478 :    
479 :     int32_t iEdgedWidth = data->iEdgedWidth;
480 :     uint32_t rounding = data->rounding;
481 :    
482 :     if (( x > data->max_dx) || ( x < data->min_dx)
483 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
484 :    
485 : Isibaar 606 GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
486 : Isibaar 580 switch( ((x&1)<<1) + (y&1) )
487 :     {
488 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
489 : syskin 601 GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference);
490 : Isibaar 580 break;
491 :    
492 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
493 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
494 :    
495 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
496 :     break;
497 :    
498 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
499 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
500 :    
501 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
502 :     break;
503 :    
504 :     default: // x and y in qpel resolution - the "corners" (top left/right and
505 :     // bottom left/right) during qpel refinement
506 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
507 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
508 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
509 :    
510 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
511 :     break;
512 :     }
513 :    
514 :     sad = sad8(data->Cur, Reference, data->iEdgedWidth);
515 : syskin 628 sad += (data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode) * (sad+NEIGH_8X8_BIAS))/100;
516 : Isibaar 580
517 :     if (sad < *(data->iMinSAD)) {
518 :     *(data->iMinSAD) = sad;
519 :     data->currentQMV->x = x; data->currentQMV->y = y;
520 :     *dir = Direction; }
521 :     }
522 :    
523 : Isibaar 539 /* CHECK_CANDIATE FUNCTIONS END */
524 : chl 96
525 : chl 530 /* MAINSEARCH FUNCTIONS START */
526 : edgomez 195
527 : chl 530 static void
528 :     AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
529 : chl 181 {
530 :    
531 :     /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
532 :    
533 : chl 530 int iDirection;
534 : edgomez 195
535 :     do {
536 : chl 181 iDirection = 0;
537 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
538 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
539 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
540 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
541 : chl 181
542 :     /* now we're doing diagonal checks near our candidate */
543 :    
544 : chl 530 if (iDirection) { //checking if anything found
545 : chl 181 bDirection = iDirection;
546 :     iDirection = 0;
547 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
548 :     if (bDirection & 3) { //our candidate is left or right
549 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
550 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
551 :     } else { // what remains here is up or down
552 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
553 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
554 : chl 181
555 : edgomez 195 if (iDirection) {
556 :     bDirection += iDirection;
557 : chl 530 x = data->currentMV->x; y = data->currentMV->y; }
558 :     } else { //about to quit, eh? not so fast....
559 : edgomez 195 switch (bDirection) {
560 : chl 181 case 2:
561 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
562 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
563 : chl 181 break;
564 :     case 1:
565 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
566 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
567 : chl 181 break;
568 : edgomez 195 case 2 + 4:
569 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
570 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
571 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
572 : chl 181 break;
573 :     case 4:
574 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
575 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
576 : chl 181 break;
577 :     case 8:
578 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
579 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
580 : chl 181 break;
581 : edgomez 195 case 1 + 4:
582 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
583 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
584 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
585 : chl 181 break;
586 : edgomez 195 case 2 + 8:
587 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
588 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
589 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
590 : chl 181 break;
591 : edgomez 195 case 1 + 8:
592 : chl 530 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
593 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
594 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
595 : chl 181 break;
596 : edgomez 195 default: //1+2+4+8 == we didn't find anything at all
597 : chl 530 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
598 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
599 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
600 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
601 : chl 181 break;
602 :     }
603 : chl 530 if (!iDirection) break; //ok, the end. really
604 :     bDirection = iDirection;
605 :     x = data->currentMV->x; y = data->currentMV->y;
606 : chl 181 }
607 :     }
608 : edgomez 195 while (1); //forever
609 : chl 181 }
610 :    
611 : chl 530 static void
612 :     SquareSearch(int x, int y, const SearchData * const data, int bDirection)
613 : chl 326 {
614 : chl 530 int iDirection;
615 : chl 326
616 : chl 530 do {
617 :     iDirection = 0;
618 :     if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
619 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
620 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
621 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
622 :     if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
623 :     if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
624 :     if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
625 :     if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
626 : chl 326
627 : chl 530 bDirection = iDirection;
628 :     x = data->currentMV->x; y = data->currentMV->y;
629 :     } while (iDirection);
630 : chl 346 }
631 :    
632 : chl 530 static void
633 :     DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
634 : chl 346 {
635 :    
636 : chl 181 /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
637 :    
638 : chl 530 int iDirection;
639 : edgomez 195
640 :     do {
641 : chl 181 iDirection = 0;
642 : chl 530 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
643 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
644 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
645 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
646 : chl 181
647 :     /* now we're doing diagonal checks near our candidate */
648 :    
649 : chl 530 if (iDirection) { //checking if anything found
650 : chl 181 bDirection = iDirection;
651 :     iDirection = 0;
652 : chl 530 x = data->currentMV->x; y = data->currentMV->y;
653 :     if (bDirection & 3) { //our candidate is left or right
654 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
655 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
656 :     } else { // what remains here is up or down
657 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
658 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
659 : chl 181
660 : chl 530 bDirection += iDirection;
661 :     x = data->currentMV->x; y = data->currentMV->y;
662 : chl 181 }
663 :     }
664 : chl 530 while (iDirection);
665 : chl 181 }
666 :    
667 : chl 530 /* MAINSEARCH FUNCTIONS END */
668 : chl 181
669 : chl 530 /* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
670 :    
671 :     static void
672 :     HalfpelRefine(const SearchData * const data)
673 : chl 96 {
674 : chl 530 /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */
675 : edgomez 195
676 : chl 530 VECTOR backupMV = *(data->currentMV);
677 :     int iDirection; //not needed
678 : chl 96
679 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
680 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
681 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
682 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
683 : edgomez 195
684 : chl 530 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
685 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
686 :    
687 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
688 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
689 : chl 96 }
690 :    
691 : Isibaar 580
692 :     static void
693 :     QuarterpelRefine(const SearchData * const data)
694 :     {
695 :     /* Perform quarter pixel refinement*/
696 :    
697 :     VECTOR backupMV = *(data->currentQMV);
698 :     int iDirection; //not needed
699 :    
700 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
701 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
702 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
703 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
704 :    
705 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
706 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
707 :    
708 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
709 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
710 :    
711 :     }
712 :    
713 : chl 530 static __inline int
714 :     SkipDecisionP(const IMAGE * current, const IMAGE * reference,
715 :     const int x, const int y,
716 :     const uint32_t iEdgedWidth, const uint32_t iQuant)
717 : chl 96
718 : Isibaar 3 {
719 : chl 530 /* keep repeating checks for all b-frames before this P frame,
720 :     to make sure that SKIP is possible (todo)
721 :     how: if skip is not possible set sad00 to a very high value */
722 : Isibaar 3
723 : chl 530 uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8,
724 :     reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
725 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
726 : h 545 sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8,
727 :     reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
728 : chl 530 if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
729 : edgomez 195
730 : chl 530 return 1;
731 :     }
732 : edgomez 195
733 : chl 530 static __inline void
734 :     SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
735 :     {
736 :     pMB->mode = MODE_NOT_CODED;
737 : Isibaar 539 pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
738 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
739 : syskin 628
740 : Isibaar 580 pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
741 :     pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
742 :    
743 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
744 : Isibaar 3 }
745 :    
746 : syskin 628 bool
747 : chl 530 MotionEstimation(MBParam * const pParam,
748 :     FRAMEINFO * const current,
749 :     FRAMEINFO * const reference,
750 :     const IMAGE * const pRefH,
751 :     const IMAGE * const pRefV,
752 :     const IMAGE * const pRefHV,
753 :     const uint32_t iLimit)
754 : Isibaar 3 {
755 : chl 530 MACROBLOCK *const pMBs = current->mbs;
756 :     const IMAGE *const pCurrent = &current->image;
757 :     const IMAGE *const pRef = &reference->image;
758 : Isibaar 3
759 : chl 530 const VECTOR zeroMV = { 0, 0 };
760 : Isibaar 3
761 : chl 530 uint32_t x, y;
762 :     uint32_t iIntra = 0;
763 : syskin 628 int32_t InterBias, quant = current->quant, sad00;
764 : Isibaar 580 uint8_t *qimage;
765 : edgomez 195
766 : Isibaar 539 // some pre-initialized thingies for SearchP
767 :     int32_t temp[5];
768 :     VECTOR currentMV[5];
769 : Isibaar 580 VECTOR currentQMV[5];
770 : Isibaar 539 int32_t iMinSAD[5];
771 :     SearchData Data;
772 :     Data.iEdgedWidth = pParam->edged_width;
773 :     Data.currentMV = currentMV;
774 : Isibaar 580 Data.currentQMV = currentQMV;
775 : Isibaar 539 Data.iMinSAD = iMinSAD;
776 :     Data.temp = temp;
777 :     Data.iFcode = current->fcode;
778 : Isibaar 580 Data.rounding = pParam->m_rounding_type;
779 : syskin 628 Data.qpel = pParam->m_quarterpel;
780 :     Data.chroma = current->global_flags & XVID_ME_COLOUR;
781 : Isibaar 539
782 : Isibaar 580 if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
783 : syskin 594 return 1; // allocate some mem for qpel interpolated blocks
784 : Isibaar 580 // somehow this is dirty since I think we shouldn't use malloc outside
785 :     // encoder_create() - so please fix me!
786 : syskin 628 Data.RefQ = qimage;
787 : chl 530 if (sadInit) (*sadInit) ();
788 : edgomez 195
789 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
790 :     for (x = 0; x < pParam->mb_width; x++) {
791 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
792 : syskin 628
793 :     pMB->sad16
794 : chl 530 = sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
795 :     pRef->y + (x + y * pParam->edged_width) * 16,
796 :     pParam->edged_width, pMB->sad8 );
797 : edgomez 195
798 : syskin 628 if (Data.chroma) {
799 :     pMB->sad16 += sad8(pCurrent->u + x*8 + y*(pParam->edged_width/2)*8,
800 :     pRef->u + x*8 + y*(pParam->edged_width/2)*8, pParam->edged_width/2);
801 :    
802 :     pMB->sad16 += sad8(pCurrent->v + (x + y*(pParam->edged_width/2))*8,
803 :     pRef->v + (x + y*(pParam->edged_width/2))*8, pParam->edged_width/2);
804 :     }
805 :    
806 :     sad00 = pMB->sad16; //if no gmc; else sad00 = (..)
807 :    
808 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
809 :     pMB->dquant = NO_CHANGE;
810 : syskin 601 pMB->quant = current->quant;
811 :     } else {
812 : syskin 574 if (pMB->dquant != NO_CHANGE) {
813 :     quant += DQtab[pMB->dquant];
814 :     if (quant > 31) quant = 31;
815 :     else if (quant < 1) quant = 1;
816 :     }
817 : syskin 601 pMB->quant = quant;
818 :     }
819 : chl 181
820 : syskin 628 //initial skip decision
821 : chl 619 /* no early skip for GMC (global vector = skip vector is unknown!) */
822 :     if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */
823 :     if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH)
824 : syskin 628 if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
825 : chl 619 SkipMacroblockP(pMB, sad00);
826 :     continue;
827 :     }
828 :     }
829 : syskin 628
830 :     SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
831 : chl 530 y, current->motion_flags, pMB->quant,
832 : Isibaar 539 &Data, pParam, pMBs, reference->mbs,
833 : chl 530 current->global_flags & XVID_INTER4V, pMB);
834 : edgomez 195
835 : chl 530 /* final skip decision, a.k.a. "the vector you found, really that good?" */
836 : chl 619 if (current->coding_type == P_VOP) {
837 :     if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
838 : syskin 601 && ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) )
839 : syskin 628 if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
840 : chl 619 SkipMacroblockP(pMB, sad00);
841 :     continue;
842 :     }
843 :     }
844 : syskin 628
845 : chl 530 /* finally, intra decision */
846 : Isibaar 3
847 : chl 530 InterBias = MV16_INTER_BIAS;
848 : syskin 628 if (pMB->quant > 8) InterBias += 80 * (pMB->quant - 8); // to make high quants work
849 : chl 530 if (y != 0)
850 : syskin 628 if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 80;
851 : chl 530 if (x != 0)
852 : syskin 628 if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 80;
853 :    
854 :     if (Data.chroma) InterBias += 50; // to compensate bigger SAD
855 : Isibaar 3
856 : chl 530 if (InterBias < pMB->sad16) {
857 :     const int32_t deviation =
858 :     dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
859 :     pParam->edged_width);
860 : Isibaar 3
861 : chl 530 if (deviation < (pMB->sad16 - InterBias)) {
862 : Isibaar 580 if (++iIntra >= iLimit) { free(qimage); return 1; }
863 : chl 530 pMB->mode = MODE_INTRA;
864 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
865 :     pMB->mvs[3] = zeroMV;
866 : Isibaar 580 pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
867 :     pMB->qmvs[3] = zeroMV;
868 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
869 :     pMB->sad8[3] = 0;
870 :     }
871 :     }
872 :     }
873 : Isibaar 3 }
874 : Isibaar 580 free(qimage);
875 : chl 619
876 :     if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */
877 : syskin 628 current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode );
878 : chl 619 else
879 :     current->GMC_MV = zeroMV;
880 :    
881 : syskin 628 return 0;
882 : chl 530 }
883 : Isibaar 3
884 : edgomez 195
885 : chl 530 #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16)
886 : Isibaar 3
887 : chl 530 static __inline int
888 :     make_mask(const VECTOR * const pmv, const int i)
889 :     {
890 : Isibaar 539 int mask = 255, j;
891 : chl 530 for (j = 0; j < i; j++) {
892 :     if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
893 :     if (pmv[i].x == pmv[j].x) {
894 :     if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
895 :     if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
896 :     } else
897 :     if (pmv[i].y == pmv[j].y) {
898 :     if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
899 :     if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
900 :     }
901 : Isibaar 3 }
902 : chl 530 return mask;
903 :     }
904 : edgomez 195
905 : chl 530 static __inline void
906 :     PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount,
907 :     const int iHcount, const MACROBLOCK * const prevMB)
908 :     {
909 : edgomez 195
910 : chl 530 //this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
911 : edgomez 195
912 : chl 530 if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour
913 :     pmv[5].x = EVEN(pmv[3].x);
914 : Isibaar 539 pmv[5].y = EVEN(pmv[3].y);
915 :     } else pmv[5].x = pmv[5].y = 0;
916 : chl 169
917 : chl 530 if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
918 :     else pmv[3].x = pmv[3].y = 0;
919 : Isibaar 3
920 : chl 530 if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
921 :     else pmv[4].x = pmv[4].y = 0;
922 : chl 169
923 : chl 530 // [1] median prediction
924 :     pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
925 : chl 169
926 : 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
927 : chl 169
928 : chl 530 pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
929 :     pmv[2].y = EVEN(prevMB->mvs[0].y);
930 : chl 169
931 : chl 530 if ((x != iWcount-1) && (y != iHcount-1)) {
932 :     pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
933 : Isibaar 539 pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
934 :     } else pmv[6].x = pmv[6].y = 0;
935 : chl 530 }
936 : chl 169
937 : chl 530 static void
938 : syskin 628 SearchP(const IMAGE * const pRef,
939 : chl 530 const uint8_t * const pRefH,
940 :     const uint8_t * const pRefV,
941 :     const uint8_t * const pRefHV,
942 :     const IMAGE * const pCur,
943 :     const int x,
944 :     const int y,
945 :     const uint32_t MotionFlags,
946 :     const uint32_t iQuant,
947 : Isibaar 539 SearchData * const Data,
948 : chl 530 const MBParam * const pParam,
949 :     const MACROBLOCK * const pMBs,
950 :     const MACROBLOCK * const prevMBs,
951 :     int inter4v,
952 :     MACROBLOCK * const pMB)
953 :     {
954 : chl 169
955 : chl 530 int i, iDirection = 255, mask, threshA;
956 : Isibaar 539 VECTOR pmv[7];
957 : Isibaar 3
958 : Isibaar 539 get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
959 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
960 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
961 : Isibaar 3
962 : Isibaar 539 Data->predMV = pmv[0];
963 : Isibaar 580
964 : Isibaar 539 Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
965 : syskin 628 Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8;
966 :     Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8;
967 :    
968 :     Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16;
969 : Isibaar 539 Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
970 :     Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
971 :     Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
972 : syskin 628 Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
973 :     Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
974 : Isibaar 3
975 : syskin 601 Data->lambda16 = lambda_vec16[iQuant];
976 : Isibaar 606 Data->lambda8 = lambda_vec8[iQuant];
977 : chl 169
978 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
979 : Isibaar 539 Data->min_dx = EVEN(Data->min_dx);
980 :     Data->max_dx = EVEN(Data->max_dx);
981 :     Data->min_dy = EVEN(Data->min_dy);
982 :     Data->max_dy = EVEN(Data->max_dy); }
983 : edgomez 568
984 :     if (pMB->dquant != NO_CHANGE) inter4v = 0;
985 : edgomez 195
986 : Isibaar 580 for(i = 0; i < 5; i++)
987 :     Data->currentMV[i].x = Data->currentMV[i].y = 0;
988 : Isibaar 539
989 : Isibaar 606 if (pParam->m_quarterpel) {
990 :     Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
991 :     i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode);
992 :     } else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
993 :    
994 : syskin 628 Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000;
995 :     Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100;
996 : Isibaar 539 Data->iMinSAD[2] = pMB->sad8[1];
997 :     Data->iMinSAD[3] = pMB->sad8[2];
998 :     Data->iMinSAD[4] = pMB->sad8[3];
999 :    
1000 : chl 530 if ((x == 0) && (y == 0)) threshA = 512;
1001 :     else {
1002 : edgomez 568 threshA = Data->temp[0]; // that's when we keep this SAD atm
1003 : chl 530 if (threshA < 512) threshA = 512;
1004 :     if (threshA > 1024) threshA = 1024; }
1005 : Isibaar 3
1006 : chl 530 PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1007 :     prevMBs + x + y * pParam->mb_width);
1008 : edgomez 195
1009 : syskin 628 if (inter4v || pParam->m_quarterpel || Data->chroma) CheckCandidate = CheckCandidate16;
1010 :     else CheckCandidate = CheckCandidate16no4v;
1011 : edgomez 568
1012 : chl 530 /* main loop. checking all predictions */
1013 :    
1014 :     for (i = 1; i < 7; i++) {
1015 :     if (!(mask = make_mask(pmv, i)) ) continue;
1016 : edgomez 568 (*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1017 : Isibaar 539 if (Data->iMinSAD[0] <= threshA) break;
1018 : chl 530 }
1019 :    
1020 : Isibaar 539 if ((Data->iMinSAD[0] <= threshA) ||
1021 :     (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1022 :     (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1023 : chl 530 inter4v = 0;
1024 : Isibaar 539 } else {
1025 : Isibaar 3
1026 : Isibaar 539 MainSearchFunc * MainSearchPtr;
1027 :     if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1028 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1029 :     else MainSearchPtr = DiamondSearch;
1030 : chl 181
1031 : Isibaar 539 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1032 : Isibaar 3
1033 : chl 530 /* extended search, diamond starting in 0,0 and in prediction.
1034 :     note that this search is/might be done in halfpel positions,
1035 :     which makes it more different than the diamond above */
1036 : chl 259
1037 : Isibaar 539 if (MotionFlags & PMV_EXTSEARCH16) {
1038 :     int32_t bSAD;
1039 :     VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1040 :     if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1041 :     startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1042 :     if (!(MVequal(startMV, backupMV))) {
1043 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1044 : Isibaar 3
1045 : Isibaar 606 (*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1046 : Isibaar 539 (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1047 :     if (bSAD < Data->iMinSAD[0]) {
1048 :     Data->currentMV[0] = backupMV;
1049 :     Data->iMinSAD[0] = bSAD; }
1050 :     }
1051 : Isibaar 3
1052 : Isibaar 539 backupMV = Data->currentMV[0];
1053 :     if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1;
1054 :     else startMV.x = startMV.y = 0;
1055 :     if (!(MVequal(startMV, backupMV))) {
1056 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1057 : edgomez 195
1058 : Isibaar 606 (*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
1059 : Isibaar 539 (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1060 :     if (bSAD < Data->iMinSAD[0]) {
1061 :     Data->currentMV[0] = backupMV;
1062 :     Data->iMinSAD[0] = bSAD; }
1063 :     }
1064 : Isibaar 3 }
1065 :     }
1066 :    
1067 : Isibaar 539 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1068 : Isibaar 3
1069 : Isibaar 580 for(i = 0; i < 5; i++) {
1070 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1071 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1072 :     }
1073 :    
1074 :     if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1075 :    
1076 : Isibaar 606 CheckCandidate = CheckCandidate16_qpel;
1077 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1078 : syskin 601 pParam->width, pParam->height, Data->iFcode, 0);
1079 :    
1080 : Isibaar 580 QuarterpelRefine(Data);
1081 :     }
1082 :    
1083 : syskin 628 if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0;
1084 : Isibaar 539 if (inter4v) {
1085 :     SearchData Data8;
1086 :     Data8.iFcode = Data->iFcode;
1087 : Isibaar 606 Data8.lambda8 = Data->lambda8;
1088 : Isibaar 539 Data8.iEdgedWidth = Data->iEdgedWidth;
1089 : Isibaar 606 Data8.RefQ = Data->RefQ;
1090 : syskin 628 Data8.qpel = Data->qpel;
1091 : Isibaar 539 Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1092 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1093 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1094 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1095 : syskin 628
1096 :     if (Data->chroma) {
1097 :     int sum, dx, dy;
1098 :    
1099 :     if(pParam->m_quarterpel) {
1100 :     sum = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2;
1101 :     } else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1102 :     dy = (sum >> 3) + roundtab_76[sum & 0xf];
1103 :    
1104 :     if(pParam->m_quarterpel) {
1105 :     sum = pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2;
1106 :     } else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1107 :     dx = (sum >> 3) + roundtab_76[sum & 0xf];
1108 :    
1109 :     Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1110 :     }
1111 : Isibaar 539 }
1112 : Isibaar 3
1113 : chl 530 if (!(inter4v) ||
1114 : Isibaar 539 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1115 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1116 : chl 530 // INTER MODE
1117 :     pMB->mode = MODE_INTER;
1118 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
1119 :     = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1120 : Isibaar 3
1121 : Isibaar 580 pMB->qmvs[0] = pMB->qmvs[1]
1122 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1123 :    
1124 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1125 : Isibaar 539 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1126 : Isibaar 3
1127 : Isibaar 580 if(pParam->m_quarterpel) {
1128 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1129 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1130 : syskin 628 } else {
1131 : Isibaar 580 pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1132 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1133 :     }
1134 : edgomez 195 } else {
1135 : chl 530 // INTER4V MODE; all other things are already set in Search8
1136 :     pMB->mode = MODE_INTER4V;
1137 : Isibaar 539 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1138 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1139 : edgomez 78 }
1140 : Isibaar 3 }
1141 :    
1142 : chl 530 static void
1143 :     Search8(const SearchData * const OldData,
1144 :     const int x, const int y,
1145 :     const uint32_t MotionFlags,
1146 :     const MBParam * const pParam,
1147 :     MACROBLOCK * const pMB,
1148 :     const MACROBLOCK * const pMBs,
1149 : Isibaar 539 const int block,
1150 :     SearchData * const Data)
1151 : chl 345 {
1152 : Isibaar 539 Data->iMinSAD = OldData->iMinSAD + 1 + block;
1153 :     Data->currentMV = OldData->currentMV + 1 + block;
1154 : Isibaar 580 Data->currentQMV = OldData->currentQMV + 1 + block;
1155 : chl 345
1156 : syskin 601 if(pParam->m_quarterpel) {
1157 : Isibaar 606 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1158 : syskin 628 if (block != 0) *(Data->iMinSAD) += (Data->lambda8 *
1159 : syskin 601 d_mv_bits( Data->currentQMV->x - Data->predQMV.x,
1160 :     Data->currentQMV->y - Data->predQMV.y,
1161 : syskin 628 Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1162 : Isibaar 606 } else {
1163 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1164 : syskin 628 if (block != 0) *(Data->iMinSAD) += (Data->lambda8 *
1165 : syskin 601 d_mv_bits( Data->currentMV->x - Data->predMV.x,
1166 :     Data->currentMV->y - Data->predMV.y,
1167 : syskin 628 Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100;
1168 : Isibaar 606 }
1169 : chl 345
1170 : chl 530 if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
1171 : chl 345
1172 : Isibaar 539 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1173 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1174 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1175 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1176 : chl 345
1177 : Isibaar 539 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1178 : chl 530
1179 : Isibaar 539 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1180 : Isibaar 580 pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1181 : syskin 628 CheckCandidate = CheckCandidate8;
1182 : chl 345
1183 : chl 530 if (MotionFlags & PMV_EXTSEARCH8) {
1184 : Isibaar 580 int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1185 :    
1186 : chl 530 MainSearchFunc *MainSearchPtr;
1187 :     if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1188 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1189 :     else MainSearchPtr = DiamondSearch;
1190 : chl 345
1191 : Isibaar 580 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1192 : chl 345
1193 : syskin 601 if(*(Data->iMinSAD) < temp_sad) {
1194 : Isibaar 580 Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1195 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1196 :     }
1197 :     }
1198 :    
1199 :     if (MotionFlags & PMV_HALFPELREFINE8) {
1200 :     int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1201 :    
1202 :     HalfpelRefine(Data); // perform halfpel refine of current best vector
1203 :    
1204 :     if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1205 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1206 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1207 :     }
1208 :     }
1209 :    
1210 : syskin 601 if(pParam->m_quarterpel) {
1211 :     if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1212 :     (MotionFlags & PMV_QUARTERPELREFINE8)) {
1213 : Isibaar 600 CheckCandidate = CheckCandidate8_qpel;
1214 : Isibaar 606 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1215 :     pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1216 :     QuarterpelRefine(Data);
1217 : syskin 601 }
1218 : Isibaar 580 }
1219 : chl 345 }
1220 : chl 530
1221 : Isibaar 580 if(pParam->m_quarterpel) {
1222 :     pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1223 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1224 :     }
1225 :     else {
1226 :     pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1227 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1228 :     }
1229 :    
1230 : Isibaar 539 pMB->mvs[block] = *(Data->currentMV);
1231 : Isibaar 580 pMB->qmvs[block] = *(Data->currentQMV);
1232 :    
1233 : syskin 601 pMB->sad8[block] = 4 * (*Data->iMinSAD);
1234 : chl 345 }
1235 :    
1236 : chl 530 /* B-frames code starts here */
1237 : chl 345
1238 : chl 530 static __inline VECTOR
1239 :     ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1240 : Isibaar 3 {
1241 : chl 530 /* the stupidiest function ever */
1242 :     if (mode == MODE_FORWARD) return pMB->mvs[0];
1243 :     else return pMB->b_mvs[0];
1244 : Isibaar 3 }
1245 :    
1246 : chl 530 static void __inline
1247 :     PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1248 :     const uint32_t iWcount,
1249 :     const MACROBLOCK * const pMB,
1250 :     const uint32_t mode_curr)
1251 : Isibaar 3 {
1252 :    
1253 : chl 530 // [0] is prediction
1254 :     pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1255 : Isibaar 3
1256 : chl 530 pmv[1].x = pmv[1].y = 0; // [1] is zero
1257 : Isibaar 3
1258 : chl 530 pmv[2] = ChoosePred(pMB, mode_curr);
1259 :     pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1260 : edgomez 195
1261 : chl 530 if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1262 :     pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1263 : Isibaar 539 pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1264 :     } else pmv[3].x = pmv[3].y = 0;
1265 : Isibaar 3
1266 : chl 530 if (y != 0) {
1267 :     pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1268 :     pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1269 :     } else pmv[4].x = pmv[4].y = 0;
1270 : Isibaar 3
1271 : chl 530 if (x != 0) {
1272 :     pmv[5] = ChoosePred(pMB-1, mode_curr);
1273 :     pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1274 :     } else pmv[5].x = pmv[5].y = 0;
1275 : Isibaar 3
1276 : chl 530 if ((x != 0)&&(y != 0)) {
1277 :     pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1278 :     pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1279 :     } else pmv[6].x = pmv[6].y = 0;
1280 : edgomez 195
1281 : chl 530 // more?
1282 :     }
1283 : chl 181
1284 : edgomez 170
1285 : chl 530 /* search backward or forward, for b-frames */
1286 :     static void
1287 :     SearchBF( const uint8_t * const pRef,
1288 :     const uint8_t * const pRefH,
1289 :     const uint8_t * const pRefV,
1290 :     const uint8_t * const pRefHV,
1291 :     const IMAGE * const pCur,
1292 :     const int x, const int y,
1293 :     const uint32_t MotionFlags,
1294 :     const uint32_t iFcode,
1295 :     const MBParam * const pParam,
1296 :     MACROBLOCK * const pMB,
1297 :     const VECTOR * const predMV,
1298 :     int32_t * const best_sad,
1299 : h 545 const int32_t mode_current,
1300 :     SearchData * const Data)
1301 : chl 530 {
1302 : Isibaar 3
1303 : chl 530 const int32_t iEdgedWidth = pParam->edged_width;
1304 :    
1305 :     int i, iDirection, mask;
1306 : h 545 VECTOR pmv[7];
1307 : chl 530 MainSearchFunc *MainSearchPtr;
1308 : h 545 *Data->iMinSAD = MV_MAX_ERROR;
1309 :     Data->iFcode = iFcode;
1310 : Isibaar 3
1311 : h 545 Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1312 :     Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1313 :     Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1314 :     Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1315 : Isibaar 3
1316 : h 545 Data->predMV = *predMV;
1317 : edgomez 195
1318 : h 545 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1319 : Isibaar 580 pParam->width, pParam->height, iFcode, pParam->m_quarterpel);
1320 : Isibaar 3
1321 : h 545 pmv[0] = Data->predMV;
1322 :     PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1323 : edgomez 195
1324 : h 545 Data->currentMV->x = Data->currentMV->y = 0;
1325 : Isibaar 3
1326 : chl 530 CheckCandidate = CheckCandidate16no4v;
1327 : chl 181
1328 : chl 530 // main loop. checking all predictions
1329 :     for (i = 0; i < 8; i++) {
1330 :     if (!(mask = make_mask(pmv, i)) ) continue;
1331 : h 545 CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1332 : edgomez 78 }
1333 : Isibaar 3
1334 : chl 530 if (MotionFlags & PMV_USESQUARES16)
1335 :     MainSearchPtr = SquareSearch;
1336 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1337 :     MainSearchPtr = AdvDiamondSearch;
1338 :     else MainSearchPtr = DiamondSearch;
1339 : chl 96
1340 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1341 : chl 169
1342 : h 545 HalfpelRefine(Data);
1343 : chl 169
1344 : chl 530 // three bits are needed to code backward mode. four for forward
1345 :     // we treat the bits just like they were vector's
1346 : syskin 628 if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16;
1347 :     else *Data->iMinSAD += 3 * Data->lambda16;
1348 : chl 169
1349 : h 545 if (*Data->iMinSAD < *best_sad) {
1350 :     *best_sad = *Data->iMinSAD;
1351 : chl 530 pMB->mode = mode_current;
1352 : h 545 pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1353 :     pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1354 : syskin 601 if (mode_current == MODE_FORWARD) pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV;
1355 :     else pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search
1356 : chl 169 }
1357 : chl 530
1358 :     }
1359 : chl 169
1360 : chl 530 static int32_t
1361 : Isibaar 539 SearchDirect(const IMAGE * const f_Ref,
1362 : chl 530 const uint8_t * const f_RefH,
1363 :     const uint8_t * const f_RefV,
1364 :     const uint8_t * const f_RefHV,
1365 : Isibaar 539 const IMAGE * const b_Ref,
1366 : chl 530 const uint8_t * const b_RefH,
1367 :     const uint8_t * const b_RefV,
1368 :     const uint8_t * const b_RefHV,
1369 :     const IMAGE * const pCur,
1370 :     const int x, const int y,
1371 :     const uint32_t MotionFlags,
1372 :     const int32_t TRB, const int32_t TRD,
1373 :     const MBParam * const pParam,
1374 :     MACROBLOCK * const pMB,
1375 :     const MACROBLOCK * const b_mb,
1376 : h 545 int32_t * const best_sad,
1377 :     SearchData * const Data)
1378 : edgomez 195
1379 : chl 530 {
1380 : h 545 int32_t skip_sad;
1381 : chl 530 int k;
1382 : h 545
1383 : chl 530 MainSearchFunc *MainSearchPtr;
1384 : Isibaar 3
1385 : h 545 *Data->iMinSAD = 256*4096;
1386 :     Data->referencemv = b_mb->mvs;
1387 : chl 140
1388 : h 545 Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1389 :     Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16;
1390 :     Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16;
1391 :     Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16;
1392 :     Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1393 :     Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16;
1394 :     Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16;
1395 :     Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16;
1396 : chl 140
1397 : h 545 Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1398 :     Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1399 :     Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1400 :     Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1401 : Isibaar 3
1402 : chl 530 for (k = 0; k < 4; k++) {
1403 : h 545 pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1404 :     pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1405 :     pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1406 :     pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1407 : Isibaar 3
1408 : h 545 if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx )
1409 :     || ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) {
1410 : Isibaar 3
1411 : Isibaar 539 *best_sad = 256*4096; // in that case, we won't use direct mode
1412 :     pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1413 :     pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1414 :     return 0;
1415 :     }
1416 :     if (b_mb->mode != MODE_INTER4V) {
1417 :     pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1418 :     pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1419 : h 545 Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1420 :     Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1421 : Isibaar 539 break;
1422 :     }
1423 :     }
1424 : Isibaar 3
1425 : Isibaar 539 if (b_mb->mode == MODE_INTER4V)
1426 :     CheckCandidate = CheckCandidateDirect;
1427 :     else CheckCandidate = CheckCandidateDirectno4v;
1428 : edgomez 195
1429 : h 545 (*CheckCandidate)(0, 0, 255, &k, Data);
1430 : Isibaar 3
1431 : chl 530 // skip decision
1432 : syskin 628 if (*Data->iMinSAD < pMB->quant * SKIP_THRESH_B) {
1433 : syskin 594 //possible skip - checking chroma. everything copied from MC
1434 : Isibaar 539 //this is not full chroma compensation, only it's fullpel approximation. should work though
1435 :     int sum, dx, dy, b_dx, b_dy;
1436 : Isibaar 3
1437 : Isibaar 539 sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1438 :     dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1439 :    
1440 :     sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1441 :     dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1442 :    
1443 :     sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
1444 :     b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1445 :    
1446 :     sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
1447 :     b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1448 :    
1449 : h 545 sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2),
1450 :     f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1451 :     b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1452 :     Data->iEdgedWidth/2);
1453 :     sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2),
1454 :     f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1455 :     b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1456 :     Data->iEdgedWidth/2);
1457 : Isibaar 539
1458 : syskin 601 if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) {
1459 : Isibaar 539 pMB->mode = MODE_DIRECT_NONE_MV;
1460 : h 545 return *Data->iMinSAD;
1461 : Isibaar 539 }
1462 :     }
1463 :    
1464 : h 545 skip_sad = *Data->iMinSAD;
1465 : edgomez 195
1466 : chl 530 // DIRECT MODE DELTA VECTOR SEARCH.
1467 :     // This has to be made more effective, but at the moment I'm happy it's running at all
1468 : Isibaar 3
1469 : chl 530 if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1470 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1471 :     else MainSearchPtr = DiamondSearch;
1472 : edgomez 195
1473 : h 545 (*MainSearchPtr)(0, 0, Data, 255);
1474 : Isibaar 3
1475 : h 545 HalfpelRefine(Data);
1476 : Isibaar 3
1477 : syskin 628 *Data->iMinSAD += 1 * Data->lambda16; // one bit is needed to code direct mode
1478 : h 545 *best_sad = *Data->iMinSAD;
1479 : Isibaar 3
1480 : chl 530 if (b_mb->mode == MODE_INTER4V)
1481 :     pMB->mode = MODE_DIRECT;
1482 :     else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1483 : edgomez 195
1484 : h 545 pMB->pmvs[3] = *Data->currentMV;
1485 : edgomez 195
1486 : chl 530 for (k = 0; k < 4; k++) {
1487 : h 545 pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1488 :     pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
1489 :     ? Data->directmvB[k].x
1490 :     : pMB->mvs[k].x - Data->referencemv[k].x);
1491 :     pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1492 :     pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1493 :     ? Data->directmvB[k].y
1494 :     : pMB->mvs[k].y - Data->referencemv[k].y);
1495 : chl 530 if (b_mb->mode != MODE_INTER4V) {
1496 :     pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1497 :     pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1498 :     break;
1499 :     }
1500 :     }
1501 : h 545 return skip_sad;
1502 : Isibaar 3 }
1503 : chl 96
1504 : h 545
1505 : chl 530 static __inline void
1506 :     SearchInterpolate(const uint8_t * const f_Ref,
1507 :     const uint8_t * const f_RefH,
1508 :     const uint8_t * const f_RefV,
1509 :     const uint8_t * const f_RefHV,
1510 :     const uint8_t * const b_Ref,
1511 :     const uint8_t * const b_RefH,
1512 :     const uint8_t * const b_RefV,
1513 :     const uint8_t * const b_RefHV,
1514 :     const IMAGE * const pCur,
1515 :     const int x, const int y,
1516 :     const uint32_t fcode,
1517 :     const uint32_t bcode,
1518 :     const uint32_t MotionFlags,
1519 :     const MBParam * const pParam,
1520 :     const VECTOR * const f_predMV,
1521 :     const VECTOR * const b_predMV,
1522 :     MACROBLOCK * const pMB,
1523 : h 545 int32_t * const best_sad,
1524 :     SearchData * const fData)
1525 : chl 530
1526 : chl 96 {
1527 :    
1528 : edgomez 195 const int32_t iEdgedWidth = pParam->edged_width;
1529 : chl 96
1530 : chl 530 int iDirection, i, j;
1531 : h 545 SearchData bData;
1532 : chl 96
1533 : syskin 601 *(bData.iMinSAD = fData->iMinSAD) = 4096*256;
1534 : h 545 bData.Cur = fData->Cur;
1535 :     fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
1536 :     bData.currentMV = fData->currentMV + 1;
1537 : syskin 601 bData.lambda16 = fData->lambda16;
1538 : h 545 fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1539 : edgomez 195
1540 : h 545 bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
1541 :     bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
1542 :     bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
1543 :     bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
1544 :     bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
1545 :     bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
1546 :     bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
1547 :     bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
1548 : chl 96
1549 : h 545 bData.bpredMV = fData->predMV = *f_predMV;
1550 :     fData->bpredMV = bData.predMV = *b_predMV;
1551 : edgomez 195
1552 : syskin 601 fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed
1553 : 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);
1554 :     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);
1555 : chl 96
1556 : h 545 if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1557 :     if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
1558 :     if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
1559 :     if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1560 : Isibaar 539
1561 : h 545 if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1562 :     if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
1563 :     if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
1564 :     if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1565 : Isibaar 539
1566 : h 545 CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1567 : chl 96
1568 : chl 530 //diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1569 : chl 96
1570 : chl 530 do {
1571 :     iDirection = 255;
1572 :     // forward MV moves
1573 : h 545 i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1574 : chl 96
1575 : h 545 CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1576 :     CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1577 :     CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1578 :     CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1579 : chl 96
1580 : chl 530 // backward MV moves
1581 : h 545 i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1582 :     fData->currentMV[2] = fData->currentMV[0];
1583 : chl 326
1584 : Isibaar 539 CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1585 :     CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1586 :     CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1587 :     CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1588 : chl 96
1589 : chl 530 } while (!(iDirection));
1590 : edgomez 195
1591 : syskin 628 *fData->iMinSAD += 2 * fData->lambda16; // two bits are needed to code interpolate mode.
1592 : Isibaar 606
1593 : h 545 if (*fData->iMinSAD < *best_sad) {
1594 :     *best_sad = *fData->iMinSAD;
1595 :     pMB->mvs[0] = fData->currentMV[0];
1596 :     pMB->b_mvs[0] = fData->currentMV[1];
1597 : chl 530 pMB->mode = MODE_INTERPOLATE;
1598 : chl 96
1599 : chl 530 pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1600 :     pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1601 :     pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1602 :     pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1603 :     }
1604 :     }
1605 : chl 96
1606 : chl 530 void
1607 :     MotionEstimationBVOP(MBParam * const pParam,
1608 :     FRAMEINFO * const frame,
1609 :     const int32_t time_bp,
1610 :     const int32_t time_pp,
1611 :     // forward (past) reference
1612 :     const MACROBLOCK * const f_mbs,
1613 :     const IMAGE * const f_ref,
1614 :     const IMAGE * const f_refH,
1615 :     const IMAGE * const f_refV,
1616 :     const IMAGE * const f_refHV,
1617 :     // backward (future) reference
1618 :     const MACROBLOCK * const b_mbs,
1619 :     const IMAGE * const b_ref,
1620 :     const IMAGE * const b_refH,
1621 :     const IMAGE * const b_refV,
1622 :     const IMAGE * const b_refHV)
1623 :     {
1624 :     uint32_t i, j;
1625 :     int32_t best_sad, skip_sad;
1626 :     int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1627 :     static const VECTOR zeroMV={0,0};
1628 : chl 96
1629 : chl 530 VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1630 : chl 96
1631 : chl 530 const int32_t TRB = time_pp - time_bp;
1632 :     const int32_t TRD = time_pp;
1633 : chl 96
1634 : h 545 // some pre-inintialized data for the rest of the search
1635 :    
1636 :     SearchData Data;
1637 :     int32_t iMinSAD;
1638 :     VECTOR currentMV[3];
1639 :     Data.iEdgedWidth = pParam->edged_width;
1640 :     Data.currentMV = currentMV;
1641 :     Data.iMinSAD = &iMinSAD;
1642 : syskin 601 Data.lambda16 = lambda_vec16[frame->quant];
1643 : h 545
1644 : chl 530 // note: i==horizontal, j==vertical
1645 : chl 96
1646 : chl 530 for (j = 0; j < pParam->mb_height; j++) {
1647 : chl 96
1648 : chl 530 f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1649 : edgomez 195
1650 : chl 530 for (i = 0; i < pParam->mb_width; i++) {
1651 :     MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1652 :     const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1653 : edgomez 195
1654 : chl 619 /* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */
1655 : syskin 628 if (b_mb->mode == MODE_NOT_CODED) {
1656 : chl 530 pMB->mode = MODE_NOT_CODED;
1657 :     continue;
1658 : chl 96 }
1659 :    
1660 : h 545 Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1661 : syskin 601 pMB->quant = frame->quant;
1662 : syskin 628
1663 : chl 530 /* direct search comes first, because it (1) checks for SKIP-mode
1664 :     and (2) sets very good predictions for forward and backward search */
1665 : Isibaar 539 skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1666 :     b_ref, b_refH->y, b_refV->y, b_refHV->y,
1667 : chl 530 &frame->image,
1668 :     i, j,
1669 :     frame->motion_flags,
1670 :     TRB, TRD,
1671 :     pParam,
1672 :     pMB, b_mb,
1673 : h 545 &best_sad,
1674 :     &Data);
1675 : chl 96
1676 : Isibaar 539 if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1677 : chl 96
1678 : chl 530 // forward search
1679 :     SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1680 :     &frame->image, i, j,
1681 :     frame->motion_flags,
1682 : h 545 frame->fcode, pParam,
1683 : chl 530 pMB, &f_predMV, &best_sad,
1684 : h 545 MODE_FORWARD, &Data);
1685 : chl 96
1686 : chl 530 // backward search
1687 :     SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1688 :     &frame->image, i, j,
1689 :     frame->motion_flags,
1690 : h 545 frame->bcode, pParam,
1691 : chl 530 pMB, &b_predMV, &best_sad,
1692 : h 545 MODE_BACKWARD, &Data);
1693 : chl 96
1694 : chl 530 // interpolate search comes last, because it uses data from forward and backward as prediction
1695 : chl 96
1696 : chl 530 SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1697 :     b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1698 :     &frame->image,
1699 :     i, j,
1700 :     frame->fcode, frame->bcode,
1701 :     frame->motion_flags,
1702 : h 545 pParam,
1703 : chl 530 &f_predMV, &b_predMV,
1704 : h 545 pMB, &best_sad,
1705 :     &Data);
1706 : chl 96
1707 : chl 530 switch (pMB->mode) {
1708 :     case MODE_FORWARD:
1709 :     f_count++;
1710 :     f_predMV = pMB->mvs[0];
1711 :     break;
1712 :     case MODE_BACKWARD:
1713 :     b_count++;
1714 :     b_predMV = pMB->b_mvs[0];
1715 :     break;
1716 :     case MODE_INTERPOLATE:
1717 :     i_count++;
1718 :     f_predMV = pMB->mvs[0];
1719 :     b_predMV = pMB->b_mvs[0];
1720 :     break;
1721 :     case MODE_DIRECT:
1722 :     case MODE_DIRECT_NO4V:
1723 :     d_count++;
1724 :     break;
1725 :     default:
1726 :     break;
1727 : chl 96 }
1728 :     }
1729 :     }
1730 :     }
1731 :    
1732 : chl 530 /* Hinted ME starts here */
1733 : chl 96
1734 : syskin 594 static void
1735 : syskin 628 SearchPhinted ( const IMAGE * const pRef,
1736 : chl 289 const uint8_t * const pRefH,
1737 :     const uint8_t * const pRefV,
1738 :     const uint8_t * const pRefHV,
1739 :     const IMAGE * const pCur,
1740 :     const int x,
1741 :     const int y,
1742 :     const uint32_t MotionFlags,
1743 :     const uint32_t iQuant,
1744 :     const MBParam * const pParam,
1745 :     const MACROBLOCK * const pMBs,
1746 : chl 530 int inter4v,
1747 : h 545 MACROBLOCK * const pMB,
1748 : edgomez 568 SearchData * const Data)
1749 : chl 289 {
1750 : chl 530
1751 : edgomez 568 int i, t;
1752 : chl 530 MainSearchFunc * MainSearchPtr;
1753 : suxen_drol 118
1754 : h 545 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1755 : syskin 628 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1756 : h 545 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1757 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1758 : suxen_drol 118
1759 : syskin 628 Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
1760 :     Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8;
1761 :     Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8;
1762 : Isibaar 600
1763 : syskin 628 Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16;
1764 :     Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
1765 :     Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
1766 :     Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
1767 :     Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
1768 :     Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
1769 :    
1770 : chl 530 if (!(MotionFlags & PMV_HALFPEL16)) {
1771 : h 545 Data->min_dx = EVEN(Data->min_dx);
1772 :     Data->max_dx = EVEN(Data->max_dx);
1773 :     Data->min_dy = EVEN(Data->min_dy);
1774 :     Data->max_dy = EVEN(Data->max_dy);
1775 : chl 530 }
1776 : chl 289
1777 : h 545 for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1778 : chl 289
1779 : chl 530 if (pMB->dquant != NO_CHANGE) inter4v = 0;
1780 : chl 289
1781 : syskin 628 if (inter4v || pParam->m_quarterpel || Data->chroma) CheckCandidate = CheckCandidate16;
1782 :     else CheckCandidate = CheckCandidate16no4v;
1783 : chl 289
1784 : chl 530 pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
1785 :     pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
1786 : h 545 if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
1787 :     if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
1788 :     if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
1789 :     if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
1790 : chl 370
1791 : edgomez 568 (*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
1792 : chl 289
1793 : chl 530 if (pMB->mode == MODE_INTER4V)
1794 :     for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
1795 :     pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
1796 :     pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
1797 :     if (!(make_mask(pMB->mvs, i)))
1798 : edgomez 568 (*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
1799 : chl 289 }
1800 :    
1801 :     if (MotionFlags & PMV_USESQUARES16)
1802 : chl 530 MainSearchPtr = SquareSearch;
1803 : chl 289 else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1804 : chl 530 MainSearchPtr = AdvDiamondSearch;
1805 :     else MainSearchPtr = DiamondSearch;
1806 : chl 289
1807 : h 545 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1808 : chl 289
1809 : h 545 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1810 : chl 289
1811 : syskin 594 for(i = 0; i < 5; i++) {
1812 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1813 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1814 :     }
1815 : chl 289
1816 : syskin 594 if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1817 : Isibaar 606 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1818 :     pParam->width, pParam->height, Data->iFcode, 0);
1819 :     CheckCandidate = CheckCandidate16_qpel;
1820 : syskin 594 QuarterpelRefine(Data);
1821 :     }
1822 :    
1823 :     if (inter4v) {
1824 :     SearchData Data8;
1825 :     Data8.iFcode = Data->iFcode;
1826 : Isibaar 606 Data8.lambda8 = Data->lambda8;
1827 : syskin 594 Data8.iEdgedWidth = Data->iEdgedWidth;
1828 : Isibaar 606 Data8.RefQ = Data->RefQ;
1829 : syskin 628 Data8.qpel = Data->qpel;
1830 :     Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1831 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1832 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1833 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1834 :    
1835 :     if (Data->chroma) {
1836 :     int sum, dx, dy;
1837 :    
1838 :     if(pParam->m_quarterpel)
1839 :     sum = (pMB->qmvs[0].y + pMB->qmvs[1].y + pMB->qmvs[2].y + pMB->qmvs[3].y)/2;
1840 :     else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1841 :     dy = (sum ? SIGN(sum) *
1842 :     (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
1843 :    
1844 :     if(pParam->m_quarterpel)
1845 :     sum = (pMB->qmvs[0].x + pMB->qmvs[1].x + pMB->qmvs[2].x + pMB->qmvs[3].x)/2;
1846 :     else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1847 :     dx = (sum ? SIGN(sum) *
1848 :     (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
1849 :     Data->iMinSAD[1] += ChromaSAD(dx, dy, Data);
1850 :     }
1851 : syskin 594 }
1852 :    
1853 : chl 530 if (!(inter4v) ||
1854 : h 545 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] +
1855 :     Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1856 : chl 530 // INTER MODE
1857 :     pMB->mode = MODE_INTER;
1858 : Isibaar 539 pMB->mvs[0] = pMB->mvs[1]
1859 : h 545 = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1860 : chl 289
1861 : Isibaar 606 pMB->qmvs[0] = pMB->qmvs[1]
1862 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1863 :    
1864 : chl 530 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1865 : h 545 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1866 : chl 289
1867 : Isibaar 606 if(pParam->m_quarterpel) {
1868 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1869 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1870 : syskin 628 } else {
1871 : Isibaar 606 pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1872 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1873 :     }
1874 : chl 530 } else {
1875 : syskin 628 // INTER4V MODE; all other things are already set in Search8
1876 : chl 530 pMB->mode = MODE_INTER4V;
1877 : h 545 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3]
1878 :     + Data->iMinSAD[4] + IMV16X16 * iQuant;
1879 : chl 289 }
1880 :    
1881 :     }
1882 :    
1883 : edgomez 195 void
1884 : chl 530 MotionEstimationHinted( MBParam * const pParam,
1885 :     FRAMEINFO * const current,
1886 :     FRAMEINFO * const reference,
1887 :     const IMAGE * const pRefH,
1888 :     const IMAGE * const pRefV,
1889 :     const IMAGE * const pRefHV)
1890 : suxen_drol 118 {
1891 : chl 530 MACROBLOCK *const pMBs = current->mbs;
1892 :     const IMAGE *const pCurrent = &current->image;
1893 :     const IMAGE *const pRef = &reference->image;
1894 : suxen_drol 118
1895 : chl 530 uint32_t x, y;
1896 : syskin 601 uint8_t * qimage;
1897 : syskin 574 int32_t temp[5], quant = current->quant;
1898 : h 545 int32_t iMinSAD[5];
1899 : Isibaar 606 VECTOR currentMV[5], currentQMV[5];
1900 : h 545 SearchData Data;
1901 :     Data.iEdgedWidth = pParam->edged_width;
1902 :     Data.currentMV = currentMV;
1903 : Isibaar 606 Data.currentQMV = currentQMV;
1904 : h 545 Data.iMinSAD = iMinSAD;
1905 :     Data.temp = temp;
1906 :     Data.iFcode = current->fcode;
1907 : syskin 594 Data.rounding = pParam->m_rounding_type;
1908 : syskin 628 Data.qpel = pParam->m_quarterpel;
1909 :     Data.chroma = current->global_flags & XVID_ME_COLOUR;
1910 : h 545
1911 : syskin 594 if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
1912 :     return; // allocate some mem for qpel interpolated blocks
1913 :     // somehow this is dirty since I think we shouldn't use malloc outside
1914 :     // encoder_create() - so please fix me!
1915 :    
1916 : syskin 601 Data.RefQ = qimage;
1917 :    
1918 : chl 530 if (sadInit) (*sadInit) ();
1919 : suxen_drol 118
1920 : chl 530 for (y = 0; y < pParam->mb_height; y++) {
1921 :     for (x = 0; x < pParam->mb_width; x++) {
1922 : chl 312
1923 : chl 530 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
1924 : suxen_drol 118
1925 : chl 530 //intra mode is copied from the first pass. At least for the time being
1926 :     if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue;
1927 : chl 346
1928 : chl 530 if (!(current->global_flags & XVID_LUMIMASKING)) {
1929 :     pMB->dquant = NO_CHANGE;
1930 :     pMB->quant = current->quant; }
1931 : Isibaar 606 else {
1932 : syskin 574 if (pMB->dquant != NO_CHANGE) {
1933 :     quant += DQtab[pMB->dquant];
1934 :     if (quant > 31) quant = 31;
1935 :     else if (quant < 1) quant = 1;
1936 :     }
1937 : Isibaar 606 pMB->quant = quant;
1938 :     }
1939 : chl 341
1940 : syskin 628 SearchPhinted(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
1941 : Isibaar 539 y, current->motion_flags, pMB->quant,
1942 : h 545 pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
1943 :     &Data);
1944 : chl 317
1945 : Isibaar 539 }
1946 :     }
1947 : syskin 594 free(qimage);
1948 : Isibaar 539 }
1949 : suxen_drol 118
1950 : Isibaar 539 static __inline int
1951 :     MEanalyzeMB ( const uint8_t * const pRef,
1952 :     const uint8_t * const pCur,
1953 :     const int x,
1954 :     const int y,
1955 :     const MBParam * const pParam,
1956 :     const MACROBLOCK * const pMBs,
1957 : h 545 MACROBLOCK * const pMB,
1958 :     SearchData * const Data)
1959 : Isibaar 539 {
1960 : chl 317
1961 : Isibaar 606 int i = 255, mask;
1962 : h 545 VECTOR pmv[3];
1963 : suxen_drol 118
1964 : h 545 *(Data->iMinSAD) = MV_MAX_ERROR;
1965 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1966 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1967 : Isibaar 580 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1968 : chl 326
1969 : h 545 Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
1970 :     Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
1971 : Isibaar 539
1972 :     CheckCandidate = CheckCandidate16no4vI;
1973 : chl 326
1974 : Isibaar 539 pmv[1].x = EVEN(pMB->mvs[0].x);
1975 :     pmv[1].y = EVEN(pMB->mvs[0].y);
1976 : h 545 pmv[0].x = EVEN(Data->predMV.x);
1977 :     pmv[0].y = EVEN(Data->predMV.y);
1978 : Isibaar 539 pmv[2].x = pmv[2].y = 0;
1979 : chl 326
1980 : h 545 CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
1981 : Isibaar 539 if (!(mask = make_mask(pmv, 1)))
1982 : h 545 CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
1983 : Isibaar 539 if (!(mask = make_mask(pmv, 2)))
1984 : h 545 CheckCandidate16no4vI(0, 0, mask, &i, Data);
1985 : Isibaar 539
1986 : Isibaar 606 DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
1987 : Isibaar 539
1988 : Isibaar 606 pMB->mvs[0] = *Data->currentMV;
1989 : syskin 628 pMB->mode = MODE_INTER;
1990 : Isibaar 539
1991 : h 545 return *(Data->iMinSAD);
1992 : Isibaar 539 }
1993 :    
1994 :     #define INTRA_THRESH 1350
1995 :     #define INTER_THRESH 900
1996 :    
1997 :     int
1998 :     MEanalysis( const IMAGE * const pRef,
1999 :     const IMAGE * const pCurrent,
2000 :     MBParam * const pParam,
2001 :     MACROBLOCK * const pMBs,
2002 :     const uint32_t iFcode)
2003 :     {
2004 :     uint32_t x, y, intra = 0;
2005 :     int sSAD = 0;
2006 : h 545
2007 :     VECTOR currentMV;
2008 :     int32_t iMinSAD;
2009 :     SearchData Data;
2010 :     Data.iEdgedWidth = pParam->edged_width;
2011 :     Data.currentMV = &currentMV;
2012 :     Data.iMinSAD = &iMinSAD;
2013 :     Data.iFcode = iFcode;
2014 :    
2015 : Isibaar 539 if (sadInit) (*sadInit) ();
2016 :    
2017 : syskin 601 for (y = 1; y < pParam->mb_height-1; y++) {
2018 :     for (x = 1; x < pParam->mb_width-1; x++) {
2019 : Isibaar 539 int sad, dev;
2020 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
2021 :    
2022 :     sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
2023 : h 545 pParam, pMBs, pMB, &Data);
2024 : Isibaar 539
2025 : syskin 601 if (sad > INTRA_THRESH) {
2026 :     dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
2027 :     pParam->edged_width);
2028 : syskin 628 if (dev + INTRA_THRESH < sad) { intra++; pMB->mode = MODE_INTRA; }
2029 : syskin 601 if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame
2030 : Isibaar 539 }
2031 : syskin 601 sSAD += sad;
2032 : suxen_drol 118 }
2033 :     }
2034 : Isibaar 539 sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2035 :     if (sSAD > INTER_THRESH ) return 1; //P frame
2036 :     emms();
2037 :     return 0; // B frame
2038 :    
2039 : suxen_drol 118 }
2040 : syskin 576
2041 :     int
2042 :     FindFcode( const MBParam * const pParam,
2043 :     const FRAMEINFO * const current)
2044 :     {
2045 :     uint32_t x, y;
2046 :     int max = 0, min = 0, i;
2047 :    
2048 :     for (y = 0; y < pParam->mb_height; y++) {
2049 :     for (x = 0; x < pParam->mb_width; x++) {
2050 :    
2051 :     MACROBLOCK *pMB = &current->mbs[x + y * pParam->mb_width];
2052 :     for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) {
2053 :     if (pMB->mvs[i].x > max) max = pMB->mvs[i].x;
2054 :     if (pMB->mvs[i].y > max) max = pMB->mvs[i].y;
2055 :    
2056 :     if (pMB->mvs[i].x < min) min = pMB->mvs[i].x;
2057 :     if (pMB->mvs[i].y < min) min = pMB->mvs[i].y;
2058 :     }
2059 :     }
2060 :     }
2061 :    
2062 :     min = -min;
2063 :     max += 1;
2064 :     if (min > max) max = min;
2065 : Isibaar 606 if (pParam->m_quarterpel) max *= 2;
2066 : syskin 576
2067 :     for (i = 1; (max > 32 << (i - 1)); i++);
2068 :     return i;
2069 :     }
2070 : chl 619
2071 : syskin 628 static void
2072 :     CheckGMC(int x, int y, const int dir, int * iDirection,
2073 :     const MACROBLOCK * const pMBs, uint32_t * bestcount, VECTOR * GMC,
2074 :     const MBParam * const pParam)
2075 :     {
2076 :     uint32_t mx, my, a, count = 0;
2077 : chl 619
2078 : syskin 628 for (my = 1; my < pParam->mb_height-1; my++)
2079 :     for (mx = 1; mx < pParam->mb_width-1; mx++) {
2080 :     VECTOR mv;
2081 :     const MACROBLOCK *pMB = &pMBs[mx + my * pParam->mb_width];
2082 :     if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) continue;
2083 :     mv = pMB->mvs[0];
2084 :     a = ABS(mv.x - x) + ABS(mv.y - y);
2085 :     if (a < 6) count += 6 - a;
2086 :     }
2087 : chl 619
2088 : syskin 628 if (count > *bestcount) {
2089 :     *bestcount = count;
2090 :     *iDirection = dir;
2091 :     GMC->x = x; GMC->y = y;
2092 :     }
2093 :     }
2094 :    
2095 :    
2096 : chl 619 static VECTOR
2097 :     GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode)
2098 :     {
2099 :    
2100 : syskin 628 uint32_t count, bestcount = 0;
2101 : chl 619 int x, y;
2102 : syskin 628 VECTOR gmc = {0,0};
2103 : chl 619 int step, min_x, max_x, min_y, max_y;
2104 :     uint32_t mx, my;
2105 : syskin 628 int iDirection, bDirection;
2106 : chl 619
2107 :     min_x = min_y = -32<<iFcode;
2108 :     max_x = max_y = 32<<iFcode;
2109 :    
2110 : syskin 628 //step1: let's find a rough camera panning
2111 : chl 619 for (step = 32; step >= 2; step /= 2) {
2112 :     bestcount = 0;
2113 :     for (y = min_y; y <= max_y; y += step)
2114 :     for (x = min_x ; x <= max_x; x += step) {
2115 :     count = 0;
2116 :     //for all macroblocks
2117 :     for (my = 1; my < pParam->mb_height-1; my++)
2118 :     for (mx = 1; mx < pParam->mb_width-1; mx++) {
2119 :     const MACROBLOCK *pMB = &pMBs[mx + my * pParam->mb_width];
2120 :     VECTOR mv;
2121 :    
2122 :     if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED)
2123 :     continue;
2124 :    
2125 :     mv = pMB->mvs[0];
2126 :     if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */
2127 :     count++;
2128 :     }
2129 :     if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; }
2130 :     }
2131 :     min_x = gmc.x - step;
2132 :     max_x = gmc.x + step;
2133 :     min_y = gmc.y - step;
2134 :     max_y = gmc.y + step;
2135 :    
2136 :     }
2137 : syskin 628
2138 :     if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10)
2139 :     gmc.x = gmc.y = 0; //no camara pan, no GMC
2140 :    
2141 :     // step2: let's refine camera panning using gradiend-descent approach.
2142 :     // TODO: more warping points may be evaluated here (like in interpolate mode search - two vectors in one diamond)
2143 :     bestcount = 0;
2144 :     CheckGMC(gmc.x, gmc.y, 255, &iDirection, pMBs, &bestcount, &gmc, pParam);
2145 :     do {
2146 :     x = gmc.x; y = gmc.y;
2147 :     bDirection = iDirection; iDirection = 0;
2148 :     if (bDirection & 1) CheckGMC(x - 1, y, 1+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2149 :     if (bDirection & 2) CheckGMC(x + 1, y, 2+4+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2150 :     if (bDirection & 4) CheckGMC(x, y - 1, 1+2+4, &iDirection, pMBs, &bestcount, &gmc, pParam);
2151 :     if (bDirection & 8) CheckGMC(x, y + 1, 1+2+8, &iDirection, pMBs, &bestcount, &gmc, pParam);
2152 :    
2153 :     } while (iDirection);
2154 :    
2155 : chl 619 if (pParam->m_quarterpel) {
2156 :     gmc.x *= 2;
2157 :     gmc.y *= 2; /* we store the halfpel value as pseudo-qpel to make comparison easier */
2158 :     }
2159 :    
2160 :     return gmc;
2161 :     }

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