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

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

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1.44.2.10 - (view) (download)

1 : Isibaar 1.1 /**************************************************************************
2 :     *
3 : chl 1.26 * 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 1.1 #include <assert.h>
32 :     #include <stdio.h>
33 : chl 1.3 #include <stdlib.h>
34 : Isibaar 1.1
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 1.44.2.7 #include "../image/interpolate8x8.h"
41 : chl 1.44.2.1 #include "motion_est.h"
42 : suxen_drol 1.6 #include "motion.h"
43 : Isibaar 1.1 #include "sad.h"
44 : Isibaar 1.44.2.2 #include "../utils/emms.h"
45 : Isibaar 1.1
46 : chl 1.44.2.1 #define INITIAL_SKIP_THRESH (10)
47 :     #define FINAL_SKIP_THRESH (50)
48 :     #define MAX_SAD00_FOR_SKIP (20)
49 : Isibaar 1.44.2.2 #define MAX_CHROMA_SAD_FOR_SKIP (22)
50 :     #define SKIP_THRESH_B (25)
51 : chl 1.44.2.1
52 :     #define CHECK_CANDIDATE(X,Y,D) { \
53 :     (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
54 : edgomez 1.20
55 : Isibaar 1.44.2.7 #define GET_REFERENCE(X, Y, REF) { \
56 : chl 1.44.2.9 switch ( (((X)&1)<<1) + ((Y)&1) ) \
57 : Isibaar 1.44.2.7 { \
58 :     case 0 : REF = data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
59 :     case 1 : REF = data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
60 :     case 2 : REF = data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
61 :     default : REF = data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
62 :     } \
63 :     }
64 :    
65 : chl 1.44.2.1 #define iDiamondSize 2
66 : edgomez 1.20
67 : chl 1.44.2.1 static __inline int
68 :     d_mv_bits(int x, int y, const uint32_t iFcode)
69 :     {
70 :     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 :     }
80 :    
81 :     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 :     }
91 :    
92 : Isibaar 1.44.2.2
93 :     /* CHECK_CANDIATE FUNCTIONS START */
94 : chl 1.44.2.1
95 :     static void
96 :     CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
97 :     {
98 :     int t;
99 :     const uint8_t * Reference;
100 :    
101 :     if (( x > data->max_dx) || ( x < data->min_dx)
102 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
103 :    
104 :     switch ( ((x&1)<<1) + (y&1) ) {
105 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
106 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
107 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
108 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
109 :     }
110 :    
111 : syskin 1.44.2.8 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
112 : chl 1.10
113 : Isibaar 1.44.2.10 if(data->quarterpel)
114 :     t = d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
115 :     else
116 :     t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
117 :    
118 : chl 1.44.2.1 data->temp[0] += lambda_vec16[data->iQuant] * t;
119 :     data->temp[1] += lambda_vec8[data->iQuant] * t;
120 : chl 1.10
121 : chl 1.44.2.1 if (data->temp[0] < data->iMinSAD[0]) {
122 :     data->iMinSAD[0] = data->temp[0];
123 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
124 :     *dir = Direction; }
125 : chl 1.10
126 : chl 1.44.2.1 if (data->temp[1] < data->iMinSAD[1]) {
127 :     data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
128 :     if (data->temp[2] < data->iMinSAD[2]) {
129 :     data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
130 :     if (data->temp[3] < data->iMinSAD[3]) {
131 :     data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
132 :     if (data->temp[4] < data->iMinSAD[4]) {
133 :     data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
134 : Isibaar 1.1
135 : chl 1.44.2.1 }
136 : Isibaar 1.1
137 : chl 1.44.2.1 static void
138 :     CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
139 : Isibaar 1.1 {
140 : chl 1.44.2.1 int32_t sad;
141 :     const uint8_t * Reference;
142 : Isibaar 1.1
143 : chl 1.44.2.1 if (( x > data->max_dx) || ( x < data->min_dx)
144 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
145 : Isibaar 1.1
146 : chl 1.44.2.1 switch ( ((x&1)<<1) + (y&1) )
147 :     {
148 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
149 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
150 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
151 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
152 : edgomez 1.20 }
153 : Isibaar 1.1
154 : Isibaar 1.44.2.10 if(data->quarterpel)
155 :     sad = lambda_vec16[data->iQuant] *
156 :     d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
157 :     else
158 :     sad = lambda_vec16[data->iQuant] *
159 :     d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
160 :    
161 : Isibaar 1.44.2.7 sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
162 : Isibaar 1.1
163 : chl 1.44.2.1 if (sad < *(data->iMinSAD)) {
164 :     *(data->iMinSAD) = sad;
165 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
166 :     *dir = Direction; }
167 : Isibaar 1.1 }
168 :    
169 : chl 1.44.2.1 static void
170 : Isibaar 1.44.2.7 CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
171 :    
172 :     // CheckCandidate16 variant which expects x and y in quarter pixel resolution
173 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
174 :     // around currentMV!
175 :     {
176 :     int t;
177 :     uint8_t * Reference = (uint8_t *) data->RefQ;
178 :     const uint8_t *ref1, *ref2, *ref3, *ref4;
179 :     VECTOR halfpelMV = *(data->currentMV);
180 :    
181 :     int32_t iEdgedWidth = data->iEdgedWidth;
182 :     uint32_t rounding = data->rounding;
183 :    
184 :     if (( x > data->max_dx) || ( x < data->min_dx)
185 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
186 :    
187 :     switch( ((x&1)<<1) + (y&1) )
188 :     {
189 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
190 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference);
191 :     break;
192 :    
193 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
194 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
195 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
196 :    
197 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
198 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
199 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
200 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
201 :     break;
202 :    
203 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
204 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
205 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
206 :    
207 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
208 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
209 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
210 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
211 :     break;
212 :    
213 :     default: // x and y in qpel resolution - the "corners" (top left/right and
214 :     // bottom left/right) during qpel refinement
215 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
216 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
217 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
218 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
219 :    
220 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
221 :     interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
222 :     interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
223 :     interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
224 :     break;
225 :     }
226 :    
227 : syskin 1.44.2.8 data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1);
228 : Isibaar 1.44.2.7
229 :     t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
230 :     data->temp[0] += lambda_vec16[data->iQuant] * t;
231 :     data->temp[1] += lambda_vec8[data->iQuant] * t;
232 :    
233 :     if (data->temp[0] < data->iMinSAD[0]) {
234 :     data->iMinSAD[0] = data->temp[0];
235 :     data->currentQMV[0].x = x; data->currentQMV[0].y = y;
236 : syskin 1.44.2.8 /* *dir = Direction;*/ }
237 : Isibaar 1.44.2.7
238 :     if (data->temp[1] < data->iMinSAD[1]) {
239 :     data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; }
240 :     if (data->temp[2] < data->iMinSAD[2]) {
241 :     data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; }
242 :     if (data->temp[3] < data->iMinSAD[3]) {
243 :     data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; }
244 :     if (data->temp[4] < data->iMinSAD[4]) {
245 :     data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; }
246 :     }
247 :    
248 :     static void
249 :     CheckCandidate16no4v_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
250 :    
251 :     // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
252 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
253 :     // around currentMV!
254 :     {
255 :     int32_t sad;
256 :     uint8_t * Reference = (uint8_t *) data->RefQ;
257 :     const uint8_t *ref1, *ref2, *ref3, *ref4;
258 :     VECTOR halfpelMV = *(data->currentMV);
259 :    
260 :     int32_t iEdgedWidth = data->iEdgedWidth;
261 :     uint32_t rounding = data->rounding;
262 :    
263 :     if (( x > data->max_dx) || ( x < data->min_dx)
264 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
265 :    
266 :     switch( ((x&1)<<1) + (y&1) )
267 :     {
268 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
269 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference);
270 :     break;
271 :    
272 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
273 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
274 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
275 :    
276 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
277 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
278 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
279 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
280 :     break;
281 :    
282 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
283 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
284 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
285 :    
286 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
287 :     interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
288 :     interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
289 :     interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
290 :     break;
291 :    
292 :     default: // x and y in qpel resolution - the "corners" (top left/right and
293 :     // bottom left/right) during qpel refinement
294 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
295 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
296 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
297 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
298 :    
299 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
300 :     interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
301 :     interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
302 :     interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
303 :     break;
304 :     }
305 :    
306 :     sad = lambda_vec16[data->iQuant] *
307 :     d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
308 :     sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
309 :    
310 :     if (sad < *(data->iMinSAD)) {
311 :     *(data->iMinSAD) = sad;
312 :     data->currentQMV[0].x = x; data->currentQMV[0].y = y;
313 : syskin 1.44.2.8 // *dir = Direction;
314 : Isibaar 1.44.2.7 }
315 :     }
316 :    
317 :     static void
318 : Isibaar 1.44.2.2 CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
319 :     {
320 :     int32_t sad;
321 :    
322 :     if (( x > data->max_dx) || ( x < data->min_dx)
323 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
324 :    
325 :     sad = lambda_vec16[data->iQuant] *
326 :     d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
327 :    
328 :     sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
329 :     data->iEdgedWidth, 256*4096);
330 :    
331 :     if (sad < *(data->iMinSAD)) {
332 :     *(data->iMinSAD) = sad;
333 :     data->currentMV[0].x = x; data->currentMV[0].y = y;
334 :     *dir = Direction; }
335 :     }
336 :    
337 :    
338 :     static void
339 : chl 1.44.2.1 CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
340 : Isibaar 1.1 {
341 : chl 1.44.2.1 int32_t sad;
342 :     const int xb = data->currentMV[1].x;
343 :     const int yb = data->currentMV[1].y;
344 :     const uint8_t *ReferenceF, *ReferenceB;
345 : suxen_drol 1.8
346 : chl 1.44.2.1 if (( xf > data->max_dx) || ( xf < data->min_dx)
347 :     || ( yf > data->max_dy) || (yf < data->min_dy)) return;
348 : suxen_drol 1.8
349 : chl 1.44.2.1 switch ( ((xf&1)<<1) + (yf&1) ) {
350 :     case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break;
351 :     case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
352 :     case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break;
353 :     default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break;
354 :     }
355 : chl 1.19
356 : chl 1.44.2.1 switch ( ((xb&1)<<1) + (yb&1) ) {
357 :     case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break;
358 :     case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
359 :     case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break;
360 :     default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
361 :     }
362 : Isibaar 1.1
363 : chl 1.44.2.1 sad = lambda_vec16[data->iQuant] *
364 :     ( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) +
365 :     d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) );
366 : chl 1.19
367 : chl 1.44.2.1 sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
368 : edgomez 1.20
369 : chl 1.44.2.1 if (sad < *(data->iMinSAD)) {
370 :     *(data->iMinSAD) = sad;
371 :     data->currentMV->x = xf; data->currentMV->y = yf;
372 :     *dir = Direction; }
373 :     }
374 : edgomez 1.20
375 : chl 1.44.2.1 static void
376 :     CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
377 :     {
378 :     int32_t sad;
379 :     int k;
380 :     const uint8_t *ReferenceF;
381 :     const uint8_t *ReferenceB;
382 :     VECTOR mvs, b_mvs;
383 : edgomez 1.20
384 : chl 1.44.2.1 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
385 : edgomez 1.20
386 : chl 1.44.2.1 sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
387 : edgomez 1.20
388 : chl 1.44.2.1 for (k = 0; k < 4; k++) {
389 :     mvs.x = data->directmvF[k].x + x;
390 :     b_mvs.x = ((x == 0) ?
391 :     data->directmvB[k].x
392 :     : mvs.x - data->referencemv[k].x);
393 : edgomez 1.20
394 : chl 1.44.2.1 mvs.y = data->directmvF[k].y + y;
395 :     b_mvs.y = ((y == 0) ?
396 :     data->directmvB[k].y
397 :     : mvs.y - data->referencemv[k].y);
398 :    
399 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
400 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
401 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
402 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
403 :    
404 :     switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
405 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
406 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
407 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
408 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
409 :     }
410 :    
411 :     switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
412 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
413 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
414 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
415 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
416 :     }
417 :    
418 :     sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
419 :     ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
420 :     ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth),
421 :     data->iEdgedWidth);
422 :     if (sad > *(data->iMinSAD)) return;
423 :     }
424 : chl 1.17
425 : chl 1.44.2.1 if (sad < *(data->iMinSAD)) {
426 :     *(data->iMinSAD) = sad;
427 :     data->currentMV->x = x; data->currentMV->y = y;
428 :     *dir = Direction; }
429 : Isibaar 1.1 }
430 :    
431 : chl 1.44.2.1 static void
432 :     CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
433 :     {
434 :     int32_t sad;
435 :     const uint8_t *ReferenceF;
436 :     const uint8_t *ReferenceB;
437 :     VECTOR mvs, b_mvs;
438 : chl 1.35
439 : syskin 1.44.2.6 if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
440 : suxen_drol 1.25
441 : chl 1.44.2.1 sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
442 : Isibaar 1.1
443 : chl 1.44.2.1 mvs.x = data->directmvF[0].x + x;
444 :     b_mvs.x = ((x == 0) ?
445 :     data->directmvB[0].x
446 :     : mvs.x - data->referencemv[0].x);
447 :    
448 :     mvs.y = data->directmvF[0].y + y;
449 :     b_mvs.y = ((y == 0) ?
450 :     data->directmvB[0].y
451 :     : mvs.y - data->referencemv[0].y);
452 :    
453 :     if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx )
454 :     || ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy )
455 :     || ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx )
456 :     || ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return;
457 :    
458 :     switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) {
459 :     case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break;
460 :     case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
461 :     case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break;
462 :     default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break;
463 :     }
464 :    
465 :     switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) {
466 :     case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
467 :     case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
468 :     case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break;
469 :     default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break;
470 :     }
471 :    
472 :     sad += sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth);
473 : Isibaar 1.1
474 : chl 1.44.2.1 if (sad < *(data->iMinSAD)) {
475 :     *(data->iMinSAD) = sad;
476 :     data->currentMV->x = x; data->currentMV->y = y;
477 :     *dir = Direction; }
478 :     }
479 : Isibaar 1.1
480 : chl 1.44.2.1 static void
481 :     CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
482 :     {
483 :     int32_t sad;
484 :     const uint8_t * Reference;
485 : Isibaar 1.1
486 : chl 1.44.2.1 if (( x > data->max_dx) || ( x < data->min_dx)
487 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
488 : Isibaar 1.1
489 : chl 1.44.2.1 switch ( ((x&1)<<1) + (y&1) )
490 :     {
491 :     case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
492 :     case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
493 :     case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
494 :     default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
495 : edgomez 1.2 }
496 : Isibaar 1.1
497 : chl 1.44.2.1 sad = sad8(data->Cur, Reference, data->iEdgedWidth);
498 : Isibaar 1.44.2.10
499 :     if(data->quarterpel)
500 :     sad += lambda_vec8[data->iQuant] * d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
501 :     else
502 :     sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
503 : chl 1.3
504 : chl 1.44.2.1 if (sad < *(data->iMinSAD)) {
505 :     *(data->iMinSAD) = sad;
506 :     data->currentMV->x = x; data->currentMV->y = y;
507 :     *dir = Direction; }
508 : chl 1.3 }
509 :    
510 : Isibaar 1.44.2.7 static void
511 :     CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
512 :     // CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
513 :     // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
514 :     // around currentMV!
515 :    
516 :     {
517 :     int32_t sad;
518 :     uint8_t *Reference = (uint8_t *) data->RefQ;
519 :     const uint8_t *ref1, *ref2, *ref3, *ref4;
520 :     VECTOR halfpelMV = *(data->currentMV);
521 :    
522 :     int32_t iEdgedWidth = data->iEdgedWidth;
523 :     uint32_t rounding = data->rounding;
524 :    
525 :     if (( x > data->max_dx) || ( x < data->min_dx)
526 :     || ( y > data->max_dy) || (y < data->min_dy)) return;
527 :    
528 :     switch( ((x&1)<<1) + (y&1) )
529 :     {
530 :     case 0: // pure halfpel position - shouldn't happen during a refinement step
531 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, (const uint8_t *) Reference);
532 :     break;
533 :    
534 :     case 1: // x halfpel, y qpel - top or bottom during qpel refinement
535 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
536 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
537 :    
538 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
539 :     break;
540 :    
541 :     case 2: // x qpel, y halfpel - left or right during qpel refinement
542 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
543 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
544 :    
545 :     interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
546 :     break;
547 :    
548 :     default: // x and y in qpel resolution - the "corners" (top left/right and
549 :     // bottom left/right) during qpel refinement
550 :     GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
551 :     GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
552 :     GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
553 :     GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
554 :    
555 :     interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
556 :     break;
557 :     }
558 :    
559 :     sad = sad8(data->Cur, Reference, data->iEdgedWidth);
560 :     sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
561 :    
562 :     if (sad < *(data->iMinSAD)) {
563 :     *(data->iMinSAD) = sad;
564 :     data->currentQMV->x = x; data->currentQMV->y = y;
565 :     *dir = Direction; }
566 :     }
567 :    
568 : Isibaar 1.44.2.2 /* CHECK_CANDIATE FUNCTIONS END */
569 : chl 1.3
570 : chl 1.44.2.1 /* MAINSEARCH FUNCTIONS START */
571 : chl 1.18
572 : chl 1.44.2.1 static void
573 :     AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
574 :     {
575 : chl 1.18
576 :     /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
577 :    
578 : chl 1.44.2.1 int iDirection;
579 : edgomez 1.20
580 :     do {
581 : chl 1.18 iDirection = 0;
582 : chl 1.44.2.1 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
583 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
584 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
585 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
586 : chl 1.18
587 :     /* now we're doing diagonal checks near our candidate */
588 :    
589 : chl 1.44.2.1 if (iDirection) { //checking if anything found
590 : chl 1.18 bDirection = iDirection;
591 :     iDirection = 0;
592 : chl 1.44.2.1 x = data->currentMV->x; y = data->currentMV->y;
593 :     if (bDirection & 3) { //our candidate is left or right
594 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
595 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
596 :     } else { // what remains here is up or down
597 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
598 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
599 : chl 1.18
600 : edgomez 1.20 if (iDirection) {
601 :     bDirection += iDirection;
602 : chl 1.44.2.1 x = data->currentMV->x; y = data->currentMV->y; }
603 :     } else { //about to quit, eh? not so fast....
604 : edgomez 1.20 switch (bDirection) {
605 : chl 1.18 case 2:
606 : chl 1.44.2.1 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
607 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
608 : chl 1.18 break;
609 :     case 1:
610 : chl 1.44.2.1 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
611 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
612 : edgomez 1.20 break;
613 :     case 2 + 4:
614 : chl 1.44.2.1 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
615 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
616 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
617 : chl 1.18 break;
618 :     case 4:
619 : chl 1.44.2.1 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
620 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
621 : chl 1.18 break;
622 :     case 8:
623 : chl 1.44.2.1 CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
624 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
625 : edgomez 1.20 break;
626 :     case 1 + 4:
627 : chl 1.44.2.1 CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
628 :     CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
629 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
630 : edgomez 1.20 break;
631 :     case 2 + 8:
632 : chl 1.44.2.1 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
633 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
634 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
635 : edgomez 1.20 break;
636 :     case 1 + 8:
637 : chl 1.44.2.1 CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
638 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
639 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
640 : edgomez 1.20 break;
641 :     default: //1+2+4+8 == we didn't find anything at all
642 : chl 1.44.2.1 CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1 + 4);
643 :     CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1 + 8);
644 :     CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2 + 4);
645 :     CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
646 : chl 1.18 break;
647 :     }
648 : chl 1.44.2.1 if (!iDirection) break; //ok, the end. really
649 :     bDirection = iDirection;
650 :     x = data->currentMV->x; y = data->currentMV->y;
651 : chl 1.18 }
652 :     }
653 : edgomez 1.20 while (1); //forever
654 : chl 1.18 }
655 :    
656 : chl 1.44.2.1 static void
657 :     SquareSearch(int x, int y, const SearchData * const data, int bDirection)
658 :     {
659 :     int iDirection;
660 : chl 1.35
661 : chl 1.44.2.1 do {
662 :     iDirection = 0;
663 :     if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
664 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
665 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
666 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8+64+128);
667 :     if (bDirection & 16) CHECK_CANDIDATE(x - iDiamondSize, y - iDiamondSize, 1+4+16+32+64);
668 :     if (bDirection & 32) CHECK_CANDIDATE(x + iDiamondSize, y - iDiamondSize, 2+4+16+32+128);
669 :     if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
670 :     if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
671 :    
672 :     bDirection = iDirection;
673 :     x = data->currentMV->x; y = data->currentMV->y;
674 :     } while (iDirection);
675 : chl 1.35 }
676 :    
677 : chl 1.44.2.1 static void
678 :     DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
679 : chl 1.18 {
680 :    
681 :     /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
682 :    
683 : chl 1.44.2.1 int iDirection;
684 : edgomez 1.20
685 :     do {
686 : chl 1.18 iDirection = 0;
687 : chl 1.44.2.1 if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
688 :     if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
689 :     if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
690 :     if (bDirection & 8) CHECK_CANDIDATE(x, y + iDiamondSize, 8);
691 : chl 1.18
692 :     /* now we're doing diagonal checks near our candidate */
693 :    
694 : chl 1.44.2.1 if (iDirection) { //checking if anything found
695 : chl 1.18 bDirection = iDirection;
696 :     iDirection = 0;
697 : chl 1.44.2.1 x = data->currentMV->x; y = data->currentMV->y;
698 :     if (bDirection & 3) { //our candidate is left or right
699 :     CHECK_CANDIDATE(x, y + iDiamondSize, 8);
700 :     CHECK_CANDIDATE(x, y - iDiamondSize, 4);
701 :     } else { // what remains here is up or down
702 :     CHECK_CANDIDATE(x + iDiamondSize, y, 2);
703 :     CHECK_CANDIDATE(x - iDiamondSize, y, 1); }
704 : chl 1.18
705 : chl 1.44.2.1 bDirection += iDirection;
706 :     x = data->currentMV->x; y = data->currentMV->y;
707 : chl 1.18 }
708 :     }
709 : chl 1.44.2.1 while (iDirection);
710 : chl 1.18 }
711 :    
712 : chl 1.44.2.1 /* MAINSEARCH FUNCTIONS END */
713 :    
714 :     /* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */
715 : chl 1.18
716 : chl 1.44.2.1 static void
717 :     HalfpelRefine(const SearchData * const data)
718 : Isibaar 1.1 {
719 :     /* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */
720 :    
721 : chl 1.44.2.1 VECTOR backupMV = *(data->currentMV);
722 :     int iDirection; //not needed
723 : edgomez 1.20
724 : chl 1.44.2.1 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
725 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
726 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
727 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
728 : edgomez 1.20
729 : chl 1.44.2.1 CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
730 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
731 :    
732 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
733 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
734 : Isibaar 1.1 }
735 :    
736 : Isibaar 1.44.2.7
737 :     static void
738 :     QuarterpelRefine(const SearchData * const data)
739 :     {
740 :     /* Perform quarter pixel refinement*/
741 :    
742 :     VECTOR backupMV = *(data->currentQMV);
743 :     int iDirection; //not needed
744 :    
745 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
746 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
747 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
748 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
749 :    
750 :     CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
751 :     CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
752 :    
753 :     CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
754 :     CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
755 :    
756 :     }
757 :    
758 : chl 1.44.2.1 static __inline int
759 :     SkipDecisionP(const IMAGE * current, const IMAGE * reference,
760 :     const int x, const int y,
761 :     const uint32_t iEdgedWidth, const uint32_t iQuant)
762 : Isibaar 1.1
763 : chl 1.44.2.1 {
764 :     /* keep repeating checks for all b-frames before this P frame,
765 :     to make sure that SKIP is possible (todo)
766 :     how: if skip is not possible set sad00 to a very high value */
767 : chl 1.3
768 : chl 1.44.2.1 uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8,
769 :     reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
770 :     if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
771 : h 1.44.2.3 sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8,
772 :     reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
773 : chl 1.44.2.1 if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
774 : chl 1.35
775 : chl 1.44.2.1 return 1;
776 :     }
777 : Isibaar 1.1
778 : chl 1.44.2.1 static __inline void
779 :     SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
780 :     {
781 :     pMB->mode = MODE_NOT_CODED;
782 : Isibaar 1.44.2.2 pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
783 :     pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
784 : Isibaar 1.44.2.7
785 :     pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
786 :     pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
787 :    
788 : chl 1.44.2.1 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
789 :     }
790 :    
791 :     bool
792 :     MotionEstimation(MBParam * const pParam,
793 :     FRAMEINFO * const current,
794 :     FRAMEINFO * const reference,
795 :     const IMAGE * const pRefH,
796 :     const IMAGE * const pRefV,
797 :     const IMAGE * const pRefHV,
798 :     const uint32_t iLimit)
799 :     {
800 :     MACROBLOCK *const pMBs = current->mbs;
801 :     const IMAGE *const pCurrent = &current->image;
802 :     const IMAGE *const pRef = &reference->image;
803 : Isibaar 1.1
804 : chl 1.44.2.1 const VECTOR zeroMV = { 0, 0 };
805 : edgomez 1.20
806 : chl 1.44.2.1 uint32_t x, y;
807 :     uint32_t iIntra = 0;
808 : syskin 1.44.2.5 int32_t InterBias, quant = current->quant;
809 : Isibaar 1.44.2.7 uint8_t *qimage;
810 : chl 1.44.2.1
811 : Isibaar 1.44.2.2 // some pre-initialized thingies for SearchP
812 :     int32_t temp[5];
813 :     VECTOR currentMV[5];
814 : Isibaar 1.44.2.7 VECTOR currentQMV[5];
815 : Isibaar 1.44.2.2 int32_t iMinSAD[5];
816 :     SearchData Data;
817 :     Data.iEdgedWidth = pParam->edged_width;
818 :     Data.currentMV = currentMV;
819 : Isibaar 1.44.2.7 Data.currentQMV = currentQMV;
820 : Isibaar 1.44.2.2 Data.iMinSAD = iMinSAD;
821 :     Data.temp = temp;
822 :     Data.iFcode = current->fcode;
823 : Isibaar 1.44.2.7 Data.rounding = pParam->m_rounding_type;
824 : Isibaar 1.44.2.10 Data.quarterpel = pParam->m_quarterpel;
825 : Isibaar 1.44.2.7
826 :     if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
827 : syskin 1.44.2.8 return 1; // allocate some mem for qpel interpolated blocks
828 : Isibaar 1.44.2.7 // somehow this is dirty since I think we shouldn't use malloc outside
829 :     // encoder_create() - so please fix me!
830 : Isibaar 1.44.2.2
831 : chl 1.44.2.1 if (sadInit) (*sadInit) ();
832 :    
833 :     for (y = 0; y < pParam->mb_height; y++) {
834 :     for (x = 0; x < pParam->mb_width; x++) {
835 :    
836 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
837 :     int32_t sad00 = pMB->sad16
838 :     = sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16,
839 :     pRef->y + (x + y * pParam->edged_width) * 16,
840 :     pParam->edged_width, pMB->sad8 );
841 :    
842 :     if (!(current->global_flags & XVID_LUMIMASKING)) {
843 :     pMB->dquant = NO_CHANGE;
844 :     pMB->quant = current->quant; }
845 : syskin 1.44.2.5 else
846 :     if (pMB->dquant != NO_CHANGE) {
847 :     quant += DQtab[pMB->dquant];
848 :     if (quant > 31) quant = 31;
849 :     else if (quant < 1) quant = 1;
850 :     pMB->quant = quant;
851 :     }
852 : chl 1.44.2.1
853 :     //initial skip decision
854 :    
855 :     if ((pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant)
856 :     && (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) ) {
857 :     if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) {
858 :     SkipMacroblockP(pMB, sad00);
859 :     continue;
860 : Isibaar 1.44.2.7 sad00 = 256 * 4096;
861 : chl 1.44.2.1 }
862 :     } else sad00 = 256*4096; // skip not allowed - for final skip decision
863 : edgomez 1.20
864 : Isibaar 1.44.2.7 SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x,
865 : chl 1.44.2.1 y, current->motion_flags, pMB->quant,
866 : Isibaar 1.44.2.2 &Data, pParam, pMBs, reference->mbs,
867 : chl 1.44.2.1 current->global_flags & XVID_INTER4V, pMB);
868 :    
869 :     /* final skip decision, a.k.a. "the vector you found, really that good?" */
870 :     if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
871 :     if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH)
872 :     { SkipMacroblockP(pMB, sad00); continue; }
873 :    
874 :     /* finally, intra decision */
875 : Isibaar 1.1
876 : chl 1.44.2.1 InterBias = MV16_INTER_BIAS;
877 :     if (pMB->quant > 8) InterBias += 50 * (pMB->quant - 8); // to make high quants work
878 :     if (y != 0)
879 :     if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 50;
880 :     if (x != 0)
881 :     if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 50;
882 :    
883 :     if (InterBias < pMB->sad16) {
884 :     const int32_t deviation =
885 :     dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
886 :     pParam->edged_width);
887 : edgomez 1.20
888 : chl 1.44.2.1 if (deviation < (pMB->sad16 - InterBias)) {
889 : Isibaar 1.44.2.7 if (++iIntra >= iLimit) { free(qimage); return 1; }
890 : chl 1.44.2.1 pMB->mode = MODE_INTRA;
891 : Isibaar 1.44.2.2 pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
892 :     pMB->mvs[3] = zeroMV;
893 : Isibaar 1.44.2.7 pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
894 :     pMB->qmvs[3] = zeroMV;
895 : chl 1.44.2.1 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
896 :     pMB->sad8[3] = 0;
897 :     }
898 :     }
899 :     }
900 :     }
901 : Isibaar 1.44.2.7 free(qimage);
902 : chl 1.44.2.1 return 0;
903 :     }
904 : chl 1.18
905 : edgomez 1.20
906 : chl 1.44.2.1 #define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16)
907 : edgomez 1.20
908 : chl 1.44.2.1 static __inline int
909 :     make_mask(const VECTOR * const pmv, const int i)
910 :     {
911 : Isibaar 1.44.2.2 int mask = 255, j;
912 : chl 1.44.2.1 for (j = 0; j < i; j++) {
913 :     if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already
914 :     if (pmv[i].x == pmv[j].x) {
915 :     if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; }
916 :     if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; }
917 :     } else
918 :     if (pmv[i].y == pmv[j].y) {
919 :     if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; }
920 :     if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; }
921 :     }
922 :     }
923 :     return mask;
924 :     }
925 :    
926 :     static __inline void
927 :     PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount,
928 :     const int iHcount, const MACROBLOCK * const prevMB)
929 :     {
930 :    
931 :     //this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself
932 :    
933 :     if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour
934 :     pmv[5].x = EVEN(pmv[3].x);
935 : Isibaar 1.44.2.2 pmv[5].y = EVEN(pmv[3].y);
936 :     } else pmv[5].x = pmv[5].y = 0;
937 : chl 1.44.2.1
938 :     if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour
939 :     else pmv[3].x = pmv[3].y = 0;
940 :    
941 :     if (y != 0) { pmv[4].x = EVEN(pmv[2].x); pmv[4].y = EVEN(pmv[2].y); }// [4] top neighbour
942 :     else pmv[4].x = pmv[4].y = 0;
943 :    
944 :     // [1] median prediction
945 :     pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y);
946 :    
947 :     pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask
948 :    
949 :     pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame
950 :     pmv[2].y = EVEN(prevMB->mvs[0].y);
951 :    
952 :     if ((x != iWcount-1) && (y != iHcount-1)) {
953 :     pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame
954 : Isibaar 1.44.2.2 pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y);
955 :     } else pmv[6].x = pmv[6].y = 0;
956 : chl 1.44.2.1 }
957 :    
958 :     static void
959 :     SearchP(const uint8_t * const pRef,
960 :     const uint8_t * const pRefH,
961 :     const uint8_t * const pRefV,
962 :     const uint8_t * const pRefHV,
963 : Isibaar 1.44.2.7 const uint8_t * const pRefQ,
964 : chl 1.44.2.1 const IMAGE * const pCur,
965 :     const int x,
966 :     const int y,
967 :     const uint32_t MotionFlags,
968 :     const uint32_t iQuant,
969 : Isibaar 1.44.2.2 SearchData * const Data,
970 : chl 1.44.2.1 const MBParam * const pParam,
971 :     const MACROBLOCK * const pMBs,
972 :     const MACROBLOCK * const prevMBs,
973 :     int inter4v,
974 :     MACROBLOCK * const pMB)
975 :     {
976 : Isibaar 1.1
977 : chl 1.44.2.1 int i, iDirection = 255, mask, threshA;
978 : Isibaar 1.44.2.2 VECTOR pmv[7];
979 : chl 1.44.2.1
980 : Isibaar 1.44.2.7 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
981 :    
982 : Isibaar 1.44.2.2 get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)()
983 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
984 : Isibaar 1.44.2.7 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
985 : Isibaar 1.44.2.2
986 :     Data->predMV = pmv[0];
987 : Isibaar 1.44.2.7
988 : Isibaar 1.44.2.2 Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
989 :     Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16;
990 :     Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
991 :     Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
992 :     Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
993 : Isibaar 1.44.2.7 Data->RefQ = pRefQ;
994 : Isibaar 1.1
995 : Isibaar 1.44.2.2 Data->iQuant = iQuant;
996 : Isibaar 1.1
997 : edgomez 1.20 if (!(MotionFlags & PMV_HALFPEL16)) {
998 : Isibaar 1.44.2.2 Data->min_dx = EVEN(Data->min_dx);
999 :     Data->max_dx = EVEN(Data->max_dx);
1000 :     Data->min_dy = EVEN(Data->min_dy);
1001 :     Data->max_dy = EVEN(Data->max_dy); }
1002 : edgomez 1.44.2.4
1003 :     if (pMB->dquant != NO_CHANGE) inter4v = 0;
1004 : Isibaar 1.44.2.2
1005 :     if (inter4v) CheckCandidate = CheckCandidate16;
1006 :     else CheckCandidate = CheckCandidate16no4v;
1007 : chl 1.44.2.1
1008 : Isibaar 1.44.2.7 for(i = 0; i < 5; i++)
1009 :     Data->currentMV[i].x = Data->currentMV[i].y = 0;
1010 : chl 1.44.2.1
1011 : Isibaar 1.44.2.10 if(Data->quarterpel)
1012 :     i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode);
1013 :     else
1014 :     i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
1015 :    
1016 : Isibaar 1.44.2.2 Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i;
1017 :     Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i;
1018 :     Data->iMinSAD[2] = pMB->sad8[1];
1019 :     Data->iMinSAD[3] = pMB->sad8[2];
1020 :     Data->iMinSAD[4] = pMB->sad8[3];
1021 : chl 1.44.2.1
1022 :     if ((x == 0) && (y == 0)) threshA = 512;
1023 :     else {
1024 : edgomez 1.44.2.4 threshA = Data->temp[0]; // that's when we keep this SAD atm
1025 : chl 1.44.2.1 if (threshA < 512) threshA = 512;
1026 :     if (threshA > 1024) threshA = 1024; }
1027 :    
1028 :     PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
1029 :     prevMBs + x + y * pParam->mb_width);
1030 :    
1031 : edgomez 1.44.2.4 if (inter4v) CheckCandidate = CheckCandidate16;
1032 :     else CheckCandidate = CheckCandidate16no4v;
1033 :    
1034 :    
1035 : chl 1.44.2.1 /* main loop. checking all predictions */
1036 :    
1037 :     for (i = 1; i < 7; i++) {
1038 :     if (!(mask = make_mask(pmv, i)) ) continue;
1039 : edgomez 1.44.2.4 (*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1040 : Isibaar 1.44.2.2 if (Data->iMinSAD[0] <= threshA) break;
1041 : chl 1.44.2.1 }
1042 :    
1043 : Isibaar 1.44.2.2 if ((Data->iMinSAD[0] <= threshA) ||
1044 :     (MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
1045 :     (Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
1046 : chl 1.44.2.1 inter4v = 0;
1047 : Isibaar 1.44.2.2 } else {
1048 : edgomez 1.20
1049 : Isibaar 1.44.2.2 MainSearchFunc * MainSearchPtr;
1050 :     if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1051 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1052 :     else MainSearchPtr = DiamondSearch;
1053 : chl 1.35
1054 : Isibaar 1.44.2.2 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
1055 : chl 1.13
1056 : chl 1.44.2.1 /* extended search, diamond starting in 0,0 and in prediction.
1057 :     note that this search is/might be done in halfpel positions,
1058 :     which makes it more different than the diamond above */
1059 : Isibaar 1.1
1060 : Isibaar 1.44.2.2 if (MotionFlags & PMV_EXTSEARCH16) {
1061 :     int32_t bSAD;
1062 :     VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
1063 :     if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel?
1064 :     startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y);
1065 :     if (!(MVequal(startMV, backupMV))) {
1066 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1067 :    
1068 :     CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
1069 :     (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1070 :     if (bSAD < Data->iMinSAD[0]) {
1071 :     Data->currentMV[0] = backupMV;
1072 :     Data->iMinSAD[0] = bSAD; }
1073 :     }
1074 : chl 1.13
1075 : Isibaar 1.44.2.2 backupMV = Data->currentMV[0];
1076 :     if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1;
1077 :     else startMV.x = startMV.y = 0;
1078 :     if (!(MVequal(startMV, backupMV))) {
1079 :     bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
1080 :    
1081 :     CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
1082 :     (*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
1083 :     if (bSAD < Data->iMinSAD[0]) {
1084 :     Data->currentMV[0] = backupMV;
1085 :     Data->iMinSAD[0] = bSAD; }
1086 :     }
1087 : chl 1.44.2.1 }
1088 :     }
1089 : chl 1.13
1090 : Isibaar 1.44.2.2 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1091 : Isibaar 1.1
1092 : Isibaar 1.44.2.7 for(i = 0; i < 5; i++) {
1093 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1094 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1095 :     }
1096 :    
1097 :     if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1098 :    
1099 :     if(inter4v)
1100 :     CheckCandidate = CheckCandidate16_qpel;
1101 :     else
1102 :     CheckCandidate = CheckCandidate16no4v_qpel;
1103 : syskin 1.44.2.8
1104 : Isibaar 1.44.2.10 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1105 :     pParam->width, pParam->height, Data->iFcode, 0); // get real range
1106 : syskin 1.44.2.8
1107 : Isibaar 1.44.2.7 QuarterpelRefine(Data);
1108 :     }
1109 :    
1110 : Isibaar 1.44.2.2 if (inter4v) {
1111 :     SearchData Data8;
1112 :     Data8.iFcode = Data->iFcode;
1113 :     Data8.iQuant = Data->iQuant;
1114 :     Data8.iEdgedWidth = Data->iEdgedWidth;
1115 :     Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1116 :     Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1117 :     Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1118 :     Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1119 :     }
1120 : chl 1.13
1121 : chl 1.44.2.1 if (!(inter4v) ||
1122 : Isibaar 1.44.2.2 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
1123 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1124 : chl 1.44.2.1 // INTER MODE
1125 :     pMB->mode = MODE_INTER;
1126 : Isibaar 1.44.2.2 pMB->mvs[0] = pMB->mvs[1]
1127 :     = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1128 : edgomez 1.20
1129 : Isibaar 1.44.2.7 pMB->qmvs[0] = pMB->qmvs[1]
1130 :     = pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
1131 :    
1132 : chl 1.44.2.1 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1133 : Isibaar 1.44.2.2 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1134 : chl 1.13
1135 : Isibaar 1.44.2.7 if(pParam->m_quarterpel) {
1136 :     pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
1137 :     pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y;
1138 :     }
1139 :     else {
1140 :     pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1141 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1142 :     }
1143 : chl 1.44.2.1 } else {
1144 :     // INTER4V MODE; all other things are already set in Search8
1145 :     pMB->mode = MODE_INTER4V;
1146 : Isibaar 1.44.2.2 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] +
1147 :     Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
1148 : chl 1.44.2.1 }
1149 :     }
1150 : edgomez 1.20
1151 : chl 1.44.2.1 static void
1152 :     Search8(const SearchData * const OldData,
1153 :     const int x, const int y,
1154 :     const uint32_t MotionFlags,
1155 :     const MBParam * const pParam,
1156 :     MACROBLOCK * const pMB,
1157 :     const MACROBLOCK * const pMBs,
1158 : Isibaar 1.44.2.2 const int block,
1159 :     SearchData * const Data)
1160 : chl 1.44.2.1 {
1161 : Isibaar 1.44.2.2 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1162 : Isibaar 1.44.2.7 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1163 : Isibaar 1.44.2.2 Data->iMinSAD = OldData->iMinSAD + 1 + block;
1164 :     Data->currentMV = OldData->currentMV + 1 + block;
1165 : Isibaar 1.44.2.7 Data->currentQMV = OldData->currentQMV + 1 + block;
1166 : Isibaar 1.44.2.10 Data->quarterpel = OldData->quarterpel;
1167 : Isibaar 1.44.2.7
1168 : Isibaar 1.44.2.10 if(Data->quarterpel) // add d_mv_bits[qpel] everywhere but not in 0 (it's already there)
1169 :     {
1170 :     if (block != 0)
1171 :     *(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
1172 :     d_mv_bits(Data->currentQMV->x - Data->predQMV.x,
1173 :     Data->currentQMV->y - Data->predQMV.y,
1174 :     Data->iFcode);
1175 : chl 1.44.2.1
1176 : Isibaar 1.44.2.10 } else // add d_mv_bits[hpel] everywhere but not in 0 (it's already there)
1177 :     if (block != 0)
1178 :     *(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
1179 :     d_mv_bits(Data->currentMV->x - Data->predMV.x,
1180 :     Data->currentMV->y - Data->predMV.y,
1181 :     Data->iFcode);
1182 : Isibaar 1.1
1183 : chl 1.44.2.1 if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
1184 : edgomez 1.20
1185 : Isibaar 1.44.2.2 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1186 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1187 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1188 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1189 : Isibaar 1.44.2.7 Data->RefQ = OldData->RefQ;
1190 : Isibaar 1.1
1191 : Isibaar 1.44.2.2 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1192 : chl 1.44.2.1
1193 : Isibaar 1.44.2.2 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1194 : Isibaar 1.44.2.7 pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1195 : Isibaar 1.1
1196 : chl 1.44.2.1 CheckCandidate = CheckCandidate8;
1197 : Isibaar 1.1
1198 : chl 1.44.2.1 if (MotionFlags & PMV_EXTSEARCH8) {
1199 : Isibaar 1.44.2.7 int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1200 :    
1201 : chl 1.44.2.1 MainSearchFunc *MainSearchPtr;
1202 :     if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1203 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1204 :     else MainSearchPtr = DiamondSearch;
1205 : Isibaar 1.1
1206 : Isibaar 1.44.2.7 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1207 : Isibaar 1.1
1208 : Isibaar 1.44.2.10 if(*(Data->iMinSAD) < temp_sad) { //found a better match?
1209 : Isibaar 1.44.2.7 Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1210 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1211 :     }
1212 :     }
1213 :    
1214 :     if (MotionFlags & PMV_HALFPELREFINE8) {
1215 :     int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
1216 :    
1217 :     HalfpelRefine(Data); // perform halfpel refine of current best vector
1218 :    
1219 :     if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1220 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1221 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1222 :     }
1223 :     }
1224 :    
1225 : Isibaar 1.44.2.10 if((Data->quarterpel) && (!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1226 :     (MotionFlags & PMV_QUARTERPELREFINE8)) {
1227 :    
1228 :     CheckCandidate = CheckCandidate8_qpel;
1229 :    
1230 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1231 :     pParam->width, pParam->height, OldData->iFcode, 0); // get real range
1232 : syskin 1.44.2.8
1233 : Isibaar 1.44.2.10 QuarterpelRefine(Data);
1234 : Isibaar 1.44.2.7 }
1235 :     }
1236 :    
1237 :     if(pParam->m_quarterpel) {
1238 :     pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1239 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1240 :     }
1241 :     else {
1242 :     pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1243 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1244 : Isibaar 1.1 }
1245 :    
1246 : Isibaar 1.44.2.2 pMB->mvs[block] = *(Data->currentMV);
1247 : Isibaar 1.44.2.7 pMB->qmvs[block] = *(Data->currentQMV);
1248 : Isibaar 1.44.2.2
1249 : Isibaar 1.44.2.10 pMB->sad8[block] = 4 * (*Data->iMinSAD); // Isibaar: why?
1250 : Isibaar 1.1 }
1251 :    
1252 : chl 1.44.2.1 /* B-frames code starts here */
1253 : Isibaar 1.1
1254 : chl 1.44.2.1 static __inline VECTOR
1255 :     ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode)
1256 :     {
1257 :     /* the stupidiest function ever */
1258 :     if (mode == MODE_FORWARD) return pMB->mvs[0];
1259 :     else return pMB->b_mvs[0];
1260 :     }
1261 : Isibaar 1.1
1262 : chl 1.44.2.1 static void __inline
1263 :     PreparePredictionsBF(VECTOR * const pmv, const int x, const int y,
1264 :     const uint32_t iWcount,
1265 :     const MACROBLOCK * const pMB,
1266 :     const uint32_t mode_curr)
1267 :     {
1268 : Isibaar 1.1
1269 : chl 1.44.2.1 // [0] is prediction
1270 :     pmv[0].x = EVEN(pmv[0].x); pmv[0].y = EVEN(pmv[0].y);
1271 : Isibaar 1.1
1272 : chl 1.44.2.1 pmv[1].x = pmv[1].y = 0; // [1] is zero
1273 : Isibaar 1.1
1274 : chl 1.44.2.1 pmv[2] = ChoosePred(pMB, mode_curr);
1275 :     pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y);
1276 : Isibaar 1.1
1277 : chl 1.44.2.1 if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour
1278 :     pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr);
1279 : Isibaar 1.44.2.2 pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y);
1280 :     } else pmv[3].x = pmv[3].y = 0;
1281 : chl 1.39
1282 : chl 1.44.2.1 if (y != 0) {
1283 :     pmv[4] = ChoosePred(pMB-iWcount, mode_curr);
1284 :     pmv[4].x = EVEN(pmv[4].x); pmv[4].y = EVEN(pmv[4].y);
1285 :     } else pmv[4].x = pmv[4].y = 0;
1286 : chl 1.39
1287 : chl 1.44.2.1 if (x != 0) {
1288 :     pmv[5] = ChoosePred(pMB-1, mode_curr);
1289 :     pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1290 :     } else pmv[5].x = pmv[5].y = 0;
1291 : chl 1.39
1292 : chl 1.44.2.1 if ((x != 0)&&(y != 0)) {
1293 :     pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr);
1294 :     pmv[6].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y);
1295 :     } else pmv[6].x = pmv[6].y = 0;
1296 : chl 1.39
1297 : chl 1.44.2.1 // more?
1298 : chl 1.39 }
1299 :    
1300 :    
1301 : chl 1.44.2.1 /* search backward or forward, for b-frames */
1302 :     static void
1303 :     SearchBF( const uint8_t * const pRef,
1304 :     const uint8_t * const pRefH,
1305 :     const uint8_t * const pRefV,
1306 :     const uint8_t * const pRefHV,
1307 :     const IMAGE * const pCur,
1308 :     const int x, const int y,
1309 :     const uint32_t MotionFlags,
1310 :     const uint32_t iFcode,
1311 :     const MBParam * const pParam,
1312 :     MACROBLOCK * const pMB,
1313 :     const VECTOR * const predMV,
1314 :     int32_t * const best_sad,
1315 : h 1.44.2.3 const int32_t mode_current,
1316 :     SearchData * const Data)
1317 : Isibaar 1.1 {
1318 :    
1319 : chl 1.44.2.1 const int32_t iEdgedWidth = pParam->edged_width;
1320 :    
1321 :     int i, iDirection, mask;
1322 : h 1.44.2.3 VECTOR pmv[7];
1323 : chl 1.44.2.1 MainSearchFunc *MainSearchPtr;
1324 : h 1.44.2.3 *Data->iMinSAD = MV_MAX_ERROR;
1325 :     Data->iFcode = iFcode;
1326 : chl 1.44.2.1
1327 : h 1.44.2.3 Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
1328 :     Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
1329 :     Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
1330 :     Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
1331 : chl 1.44.2.1
1332 : h 1.44.2.3 Data->predMV = *predMV;
1333 : edgomez 1.20
1334 : h 1.44.2.3 get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1335 : Isibaar 1.44.2.7 pParam->width, pParam->height, iFcode, pParam->m_quarterpel);
1336 : Isibaar 1.1
1337 : h 1.44.2.3 pmv[0] = Data->predMV;
1338 :     PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
1339 : Isibaar 1.1
1340 : h 1.44.2.3 Data->currentMV->x = Data->currentMV->y = 0;
1341 : Isibaar 1.1
1342 : chl 1.44.2.1 CheckCandidate = CheckCandidate16no4v;
1343 : edgomez 1.20
1344 : chl 1.44.2.1 // main loop. checking all predictions
1345 :     for (i = 0; i < 8; i++) {
1346 :     if (!(mask = make_mask(pmv, i)) ) continue;
1347 : h 1.44.2.3 CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
1348 : edgomez 1.20 }
1349 :    
1350 : chl 1.44.2.1 if (MotionFlags & PMV_USESQUARES16)
1351 :     MainSearchPtr = SquareSearch;
1352 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1353 :     MainSearchPtr = AdvDiamondSearch;
1354 :     else MainSearchPtr = DiamondSearch;
1355 : Isibaar 1.1
1356 : h 1.44.2.3 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1357 : Isibaar 1.1
1358 : h 1.44.2.3 HalfpelRefine(Data);
1359 : Isibaar 1.1
1360 : chl 1.44.2.1 // three bits are needed to code backward mode. four for forward
1361 :     // we treat the bits just like they were vector's
1362 : h 1.44.2.3 if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant];
1363 :     else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant];
1364 : chl 1.44.2.1
1365 :    
1366 : h 1.44.2.3 if (*Data->iMinSAD < *best_sad) {
1367 :     *best_sad = *Data->iMinSAD;
1368 : chl 1.44.2.1 pMB->mode = mode_current;
1369 : h 1.44.2.3 pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
1370 :     pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
1371 :     if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV;
1372 :     else pMB->b_mvs[0] = *Data->currentMV;
1373 : Isibaar 1.1 }
1374 : chl 1.44.2.1
1375 :     }
1376 :    
1377 :     static int32_t
1378 : Isibaar 1.44.2.2 SearchDirect(const IMAGE * const f_Ref,
1379 : chl 1.44.2.1 const uint8_t * const f_RefH,
1380 :     const uint8_t * const f_RefV,
1381 :     const uint8_t * const f_RefHV,
1382 : Isibaar 1.44.2.2 const IMAGE * const b_Ref,
1383 : chl 1.44.2.1 const uint8_t * const b_RefH,
1384 :     const uint8_t * const b_RefV,
1385 :     const uint8_t * const b_RefHV,
1386 :     const IMAGE * const pCur,
1387 :     const int x, const int y,
1388 :     const uint32_t MotionFlags,
1389 :     const int32_t TRB, const int32_t TRD,
1390 :     const MBParam * const pParam,
1391 :     MACROBLOCK * const pMB,
1392 :     const MACROBLOCK * const b_mb,
1393 : h 1.44.2.3 int32_t * const best_sad,
1394 :     SearchData * const Data)
1395 : chl 1.44.2.1
1396 :     {
1397 : h 1.44.2.3 int32_t skip_sad;
1398 : chl 1.44.2.1 int k;
1399 : h 1.44.2.3
1400 : chl 1.44.2.1 MainSearchFunc *MainSearchPtr;
1401 :    
1402 : h 1.44.2.3 *Data->iMinSAD = 256*4096;
1403 :     Data->referencemv = b_mb->mvs;
1404 : chl 1.44.2.1
1405 : h 1.44.2.3 Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1406 :     Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16;
1407 :     Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16;
1408 :     Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16;
1409 :     Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16;
1410 :     Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16;
1411 :     Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16;
1412 :     Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16;
1413 :    
1414 :     Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
1415 :     Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
1416 :     Data->min_dx = -(2 * 16 + 2 * (x) * 16);
1417 :     Data->min_dy = -(2 * 16 + 2 * (y) * 16);
1418 : chl 1.44.2.1
1419 :     for (k = 0; k < 4; k++) {
1420 : h 1.44.2.3 pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
1421 :     pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
1422 :     pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
1423 :     pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
1424 :    
1425 :     if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx )
1426 :     || ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) {
1427 : chl 1.44.2.1
1428 : Isibaar 1.44.2.2 *best_sad = 256*4096; // in that case, we won't use direct mode
1429 :     pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
1430 :     pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0;
1431 :     return 0;
1432 :     }
1433 :     if (b_mb->mode != MODE_INTER4V) {
1434 :     pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
1435 :     pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
1436 : h 1.44.2.3 Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
1437 :     Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
1438 : Isibaar 1.44.2.2 break;
1439 :     }
1440 :     }
1441 : chl 1.44.2.1
1442 :     if (b_mb->mode == MODE_INTER4V)
1443 :     CheckCandidate = CheckCandidateDirect;
1444 :     else CheckCandidate = CheckCandidateDirectno4v;
1445 :    
1446 : h 1.44.2.3 (*CheckCandidate)(0, 0, 255, &k, Data);
1447 : Isibaar 1.44.2.2
1448 :     // skip decision
1449 : h 1.44.2.3 if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) {
1450 : syskin 1.44.2.8 //possible skip - checking chroma. everything copied from MC
1451 : Isibaar 1.44.2.2 //this is not full chroma compensation, only it's fullpel approximation. should work though
1452 :     int sum, dx, dy, b_dx, b_dy;
1453 :    
1454 :     sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
1455 :     dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1456 :    
1457 :     sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
1458 :     dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1459 :    
1460 :     sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
1461 :     b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1462 :    
1463 :     sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
1464 :     b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
1465 :    
1466 : h 1.44.2.3 sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2),
1467 :     f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1468 :     b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1469 :     Data->iEdgedWidth/2);
1470 :     sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2),
1471 :     f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
1472 :     b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
1473 :     Data->iEdgedWidth/2);
1474 : Isibaar 1.44.2.2
1475 : h 1.44.2.3 if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) {
1476 : Isibaar 1.44.2.2 pMB->mode = MODE_DIRECT_NONE_MV;
1477 : h 1.44.2.3 return *Data->iMinSAD;
1478 : Isibaar 1.44.2.2 }
1479 :     }
1480 :    
1481 : h 1.44.2.3 skip_sad = *Data->iMinSAD;
1482 : Isibaar 1.44.2.2
1483 : chl 1.44.2.1 // DIRECT MODE DELTA VECTOR SEARCH.
1484 :     // This has to be made more effective, but at the moment I'm happy it's running at all
1485 :    
1486 :     if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
1487 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
1488 :     else MainSearchPtr = DiamondSearch;
1489 :    
1490 : h 1.44.2.3 (*MainSearchPtr)(0, 0, Data, 255);
1491 : chl 1.44.2.1
1492 : h 1.44.2.3 HalfpelRefine(Data);
1493 : chl 1.44.2.1
1494 : h 1.44.2.3 *Data->iMinSAD += 1 * lambda_vec16[Data->iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's
1495 :     *best_sad = *Data->iMinSAD;
1496 : chl 1.44.2.1
1497 :     if (b_mb->mode == MODE_INTER4V)
1498 :     pMB->mode = MODE_DIRECT;
1499 :     else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
1500 :    
1501 : h 1.44.2.3 pMB->pmvs[3] = *Data->currentMV;
1502 : chl 1.44.2.1
1503 :     for (k = 0; k < 4; k++) {
1504 : h 1.44.2.3 pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
1505 :     pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
1506 :     ? Data->directmvB[k].x
1507 :     : pMB->mvs[k].x - Data->referencemv[k].x);
1508 :     pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
1509 :     pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
1510 :     ? Data->directmvB[k].y
1511 :     : pMB->mvs[k].y - Data->referencemv[k].y);
1512 : chl 1.44.2.1 if (b_mb->mode != MODE_INTER4V) {
1513 :     pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
1514 :     pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
1515 :     break;
1516 :     }
1517 :     }
1518 : h 1.44.2.3 return skip_sad;
1519 : chl 1.44.2.1 }
1520 :    
1521 : h 1.44.2.3
1522 : chl 1.44.2.1 static __inline void
1523 :     SearchInterpolate(const uint8_t * const f_Ref,
1524 :     const uint8_t * const f_RefH,
1525 :     const uint8_t * const f_RefV,
1526 :     const uint8_t * const f_RefHV,
1527 :     const uint8_t * const b_Ref,
1528 :     const uint8_t * const b_RefH,
1529 :     const uint8_t * const b_RefV,
1530 :     const uint8_t * const b_RefHV,
1531 :     const IMAGE * const pCur,
1532 :     const int x, const int y,
1533 :     const uint32_t fcode,
1534 :     const uint32_t bcode,
1535 :     const uint32_t MotionFlags,
1536 :     const MBParam * const pParam,
1537 :     const VECTOR * const f_predMV,
1538 :     const VECTOR * const b_predMV,
1539 :     MACROBLOCK * const pMB,
1540 : h 1.44.2.3 int32_t * const best_sad,
1541 :     SearchData * const fData)
1542 : edgomez 1.20
1543 : chl 1.3 {
1544 :    
1545 : edgomez 1.20 const int32_t iEdgedWidth = pParam->edged_width;
1546 : chl 1.3
1547 : chl 1.44.2.1 int iDirection, i, j;
1548 : h 1.44.2.3 SearchData bData;
1549 : edgomez 1.20
1550 : h 1.44.2.3 bData.iMinSAD = fData->iMinSAD;
1551 :     *bData.iMinSAD = 4096*256;
1552 :     bData.Cur = fData->Cur;
1553 :     fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
1554 :     bData.currentMV = fData->currentMV + 1;
1555 :     bData.iQuant = fData->iQuant;
1556 :     fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
1557 :    
1558 :     bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
1559 :     bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
1560 :     bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
1561 :     bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
1562 :     bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
1563 :     bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
1564 :     bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
1565 :     bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
1566 :    
1567 :     bData.bpredMV = fData->predMV = *f_predMV;
1568 :     fData->bpredMV = bData.predMV = *b_predMV;
1569 :    
1570 :     fData->currentMV[0] = pMB->mvs[0];
1571 :     fData->currentMV[1] = pMB->b_mvs[0];
1572 : Isibaar 1.44.2.7 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);
1573 :     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);
1574 : chl 1.3
1575 : h 1.44.2.3 if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
1576 :     if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
1577 :     if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
1578 :     if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
1579 :    
1580 :     if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
1581 :     if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
1582 :     if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
1583 :     if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
1584 : Isibaar 1.44.2.2
1585 : h 1.44.2.3 CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
1586 : chl 1.3
1587 : chl 1.44.2.1 //diamond. I wish we could use normal mainsearch functions (square, advdiamond)
1588 : chl 1.3
1589 : chl 1.44.2.1 do {
1590 :     iDirection = 255;
1591 :     // forward MV moves
1592 : h 1.44.2.3 i = fData->currentMV[0].x; j = fData->currentMV[0].y;
1593 : chl 1.35
1594 : h 1.44.2.3 CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
1595 :     CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
1596 :     CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
1597 :     CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
1598 : edgomez 1.20
1599 : chl 1.44.2.1 // backward MV moves
1600 : h 1.44.2.3 i = fData->currentMV[1].x; j = fData->currentMV[1].y;
1601 :     fData->currentMV[2] = fData->currentMV[0];
1602 : chl 1.3
1603 : Isibaar 1.44.2.2 CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
1604 :     CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
1605 :     CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
1606 :     CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
1607 : chl 1.3
1608 : chl 1.44.2.1 } while (!(iDirection));
1609 : chl 1.3
1610 : chl 1.44.2.1 // two bits are needed to code interpolate mode. we treat the bits just like they were vector's
1611 : h 1.44.2.3 *fData->iMinSAD += 2 * lambda_vec16[fData->iQuant];
1612 :     if (*fData->iMinSAD < *best_sad) {
1613 :     *best_sad = *fData->iMinSAD;
1614 :     pMB->mvs[0] = fData->currentMV[0];
1615 :     pMB->b_mvs[0] = fData->currentMV[1];
1616 : chl 1.44.2.1 pMB->mode = MODE_INTERPOLATE;
1617 : edgomez 1.20
1618 : chl 1.44.2.1 pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
1619 :     pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y;
1620 :     pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x;
1621 :     pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y;
1622 : edgomez 1.20 }
1623 : chl 1.3 }
1624 :    
1625 : h 1.44.2.3
1626 : chl 1.44.2.1 void
1627 :     MotionEstimationBVOP(MBParam * const pParam,
1628 :     FRAMEINFO * const frame,
1629 :     const int32_t time_bp,
1630 :     const int32_t time_pp,
1631 :     // forward (past) reference
1632 :     const MACROBLOCK * const f_mbs,
1633 :     const IMAGE * const f_ref,
1634 :     const IMAGE * const f_refH,
1635 :     const IMAGE * const f_refV,
1636 :     const IMAGE * const f_refHV,
1637 :     // backward (future) reference
1638 :     const MACROBLOCK * const b_mbs,
1639 :     const IMAGE * const b_ref,
1640 :     const IMAGE * const b_refH,
1641 :     const IMAGE * const b_refV,
1642 :     const IMAGE * const b_refHV)
1643 :     {
1644 :     uint32_t i, j;
1645 :     int32_t best_sad, skip_sad;
1646 :     int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0;
1647 :     static const VECTOR zeroMV={0,0};
1648 : chl 1.3
1649 : chl 1.44.2.1 VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/
1650 : chl 1.3
1651 : chl 1.44.2.1 const int32_t TRB = time_pp - time_bp;
1652 :     const int32_t TRD = time_pp;
1653 : chl 1.3
1654 : h 1.44.2.3 // some pre-inintialized data for the rest of the search
1655 :    
1656 :     SearchData Data;
1657 :     int32_t iMinSAD;
1658 :     VECTOR currentMV[3];
1659 :     Data.iEdgedWidth = pParam->edged_width;
1660 :     Data.currentMV = currentMV;
1661 :     Data.iMinSAD = &iMinSAD;
1662 :     Data.iQuant = frame->quant;
1663 :    
1664 : chl 1.44.2.1 // note: i==horizontal, j==vertical
1665 : edgomez 1.20
1666 : chl 1.44.2.1 for (j = 0; j < pParam->mb_height; j++) {
1667 : chl 1.3
1668 : chl 1.44.2.1 f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */
1669 : edgomez 1.20
1670 : chl 1.44.2.1 for (i = 0; i < pParam->mb_width; i++) {
1671 :     MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width;
1672 :     const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width;
1673 : edgomez 1.20
1674 : chl 1.44.2.1 /* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */
1675 :     if (b_mb->mode == MODE_NOT_CODED) {
1676 :     pMB->mode = MODE_NOT_CODED;
1677 :     continue;
1678 :     }
1679 : edgomez 1.20
1680 : h 1.44.2.3 Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
1681 : chl 1.44.2.1 /* direct search comes first, because it (1) checks for SKIP-mode
1682 :     and (2) sets very good predictions for forward and backward search */
1683 : edgomez 1.20
1684 : Isibaar 1.44.2.2 skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y,
1685 :     b_ref, b_refH->y, b_refV->y, b_refHV->y,
1686 : chl 1.44.2.1 &frame->image,
1687 :     i, j,
1688 :     frame->motion_flags,
1689 :     TRB, TRD,
1690 :     pParam,
1691 :     pMB, b_mb,
1692 : h 1.44.2.3 &best_sad,
1693 :     &Data);
1694 : edgomez 1.20
1695 : Isibaar 1.44.2.2 if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
1696 : chl 1.3
1697 : chl 1.44.2.1 // best_sad = 256*4096; //uncomment to disable Directsearch.
1698 :     // To disable any other mode, just comment the function call
1699 : chl 1.3
1700 : chl 1.44.2.1 // forward search
1701 :     SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1702 :     &frame->image, i, j,
1703 :     frame->motion_flags,
1704 : h 1.44.2.3 frame->fcode, pParam,
1705 : chl 1.44.2.1 pMB, &f_predMV, &best_sad,
1706 : h 1.44.2.3 MODE_FORWARD, &Data);
1707 : chl 1.10
1708 : chl 1.44.2.1 // backward search
1709 :     SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1710 :     &frame->image, i, j,
1711 :     frame->motion_flags,
1712 : h 1.44.2.3 frame->bcode, pParam,
1713 : chl 1.44.2.1 pMB, &b_predMV, &best_sad,
1714 : h 1.44.2.3 MODE_BACKWARD, &Data);
1715 : chl 1.10
1716 : chl 1.44.2.1 // interpolate search comes last, because it uses data from forward and backward as prediction
1717 : edgomez 1.20
1718 : chl 1.44.2.1 SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
1719 :     b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
1720 :     &frame->image,
1721 :     i, j,
1722 :     frame->fcode, frame->bcode,
1723 :     frame->motion_flags,
1724 : h 1.44.2.3 pParam,
1725 : chl 1.44.2.1 &f_predMV, &b_predMV,
1726 : h 1.44.2.3 pMB, &best_sad,
1727 :     &Data);
1728 : edgomez 1.20
1729 : chl 1.44.2.1 switch (pMB->mode) {
1730 :     case MODE_FORWARD:
1731 :     f_count++;
1732 :     f_predMV = pMB->mvs[0];
1733 :     break;
1734 :     case MODE_BACKWARD:
1735 :     b_count++;
1736 :     b_predMV = pMB->b_mvs[0];
1737 :     break;
1738 :     case MODE_INTERPOLATE:
1739 :     i_count++;
1740 :     f_predMV = pMB->mvs[0];
1741 :     b_predMV = pMB->b_mvs[0];
1742 :     break;
1743 :     case MODE_DIRECT:
1744 :     case MODE_DIRECT_NO4V:
1745 :     d_count++;
1746 :     break;
1747 :     default:
1748 :     break;
1749 : chl 1.10 }
1750 :     }
1751 :     }
1752 :    
1753 : chl 1.44.2.1 // fprintf(debug,"B-Stat: F: %04d B: %04d I: %04d D: %04d, N: %04d\n",
1754 :     // f_count,b_count,i_count,d_count,n_count);
1755 : chl 1.3
1756 : chl 1.44.2.1 }
1757 : chl 1.3
1758 : chl 1.44.2.1 /* Hinted ME starts here */
1759 : chl 1.18
1760 : syskin 1.44.2.8 static void
1761 :     Search8hinted(const SearchData * const OldData,
1762 :     const int x, const int y,
1763 :     const uint32_t MotionFlags,
1764 :     const MBParam * const pParam,
1765 :     MACROBLOCK * const pMB,
1766 :     const MACROBLOCK * const pMBs,
1767 :     const int block,
1768 :     SearchData * const Data)
1769 : chl 1.44.2.1 {
1770 : syskin 1.44.2.8 int32_t temp_sad;
1771 : chl 1.44.2.1 MainSearchFunc *MainSearchPtr;
1772 : syskin 1.44.2.8 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1773 :     Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
1774 :     Data->iMinSAD = OldData->iMinSAD + 1 + block;
1775 :     Data->currentMV = OldData->currentMV + 1 + block;
1776 :     Data->currentQMV = OldData->currentQMV + 1 + block;
1777 : Isibaar 1.44.2.10 Data->quarterpel = OldData->quarterpel;
1778 : chl 1.18
1779 : syskin 1.44.2.8 if (block != 0) {
1780 :     if(pParam->m_quarterpel) {
1781 :     *(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
1782 :     d_mv_bits( Data->currentQMV->x - Data->predQMV.x,
1783 :     Data->currentQMV->y - Data->predQMV.y,
1784 :     Data->iFcode);
1785 :     }
1786 :     else {
1787 :     *(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
1788 :     d_mv_bits( Data->currentMV->x - Data->predMV.x,
1789 :     Data->currentMV->y - Data->predMV.y,
1790 :     Data->iFcode);
1791 :     }
1792 :     }
1793 : chl 1.3
1794 : syskin 1.44.2.8 Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
1795 :     Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
1796 :     Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1797 :     Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
1798 :     Data->RefQ = OldData->RefQ;
1799 : edgomez 1.20
1800 : syskin 1.44.2.8 Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
1801 :    
1802 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1803 : Isibaar 1.44.2.7 pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
1804 : chl 1.3
1805 : syskin 1.44.2.8 CheckCandidate = CheckCandidate8;
1806 : chl 1.3
1807 : syskin 1.44.2.8 temp_sad = *(Data->iMinSAD); // store current MinSAD
1808 :    
1809 : chl 1.44.2.1 if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
1810 :     else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
1811 :     else MainSearchPtr = DiamondSearch;
1812 : chl 1.3
1813 : syskin 1.44.2.8 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1814 :    
1815 :     if(*(Data->iMinSAD) < temp_sad) {
1816 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1817 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1818 :     }
1819 :    
1820 :     if (MotionFlags & PMV_HALFPELREFINE8) {
1821 :     temp_sad = *(Data->iMinSAD); // store current MinSAD
1822 : chl 1.3
1823 : syskin 1.44.2.8 HalfpelRefine(Data); // perform halfpel refine of current best vector
1824 :    
1825 :     if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
1826 :     Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
1827 :     Data->currentQMV->y = 2 * Data->currentMV->y;
1828 :     }
1829 :     }
1830 :    
1831 : Isibaar 1.44.2.10 if((Data->quarterpel) && (!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
1832 :     (MotionFlags & PMV_QUARTERPELREFINE8)) {
1833 :    
1834 :     CheckCandidate = CheckCandidate8_qpel;
1835 :    
1836 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
1837 :     pParam->width, pParam->height, OldData->iFcode, 0); // get real range
1838 :    
1839 :     QuarterpelRefine(Data);
1840 : syskin 1.44.2.8 }
1841 : chl 1.3
1842 : syskin 1.44.2.8 if(pParam->m_quarterpel) {
1843 :     pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
1844 :     pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
1845 :     }
1846 :     else {
1847 :     pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
1848 :     pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
1849 :     }
1850 :    
1851 :     pMB->mvs[block] = *(Data->currentMV);
1852 :     pMB->qmvs[block] = *(Data->currentQMV);
1853 :    
1854 :     pMB->sad8[block] = 4 * (*Data->iMinSAD);
1855 : chl 1.3 }
1856 :    
1857 : suxen_drol 1.6
1858 : chl 1.44.2.1 static void
1859 :     SearchPhinted ( const uint8_t * const pRef,
1860 : chl 1.30 const uint8_t * const pRefH,
1861 :     const uint8_t * const pRefV,
1862 :     const uint8_t * const pRefHV,
1863 : Isibaar 1.44.2.10 const uint8_t * const pRefQ,
1864 : chl 1.30 const IMAGE * const pCur,
1865 :     const int x,
1866 :     const int y,
1867 :     const uint32_t MotionFlags,
1868 :     const uint32_t iQuant,
1869 :     const MBParam * const pParam,
1870 :     const MACROBLOCK * const pMBs,
1871 : chl 1.44.2.1 int inter4v,
1872 : h 1.44.2.3 MACROBLOCK * const pMB,
1873 : edgomez 1.44.2.4 SearchData * const Data)
1874 : chl 1.44.2.1 {
1875 : chl 1.30
1876 : chl 1.44.2.1 const int32_t iEdgedWidth = pParam->edged_width;
1877 :    
1878 : edgomez 1.44.2.4 int i, t;
1879 : chl 1.44.2.1 MainSearchFunc * MainSearchPtr;
1880 :    
1881 : Isibaar 1.44.2.10 Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1882 : h 1.44.2.3 Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
1883 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1884 : Isibaar 1.44.2.7 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
1885 : chl 1.44.2.1
1886 : h 1.44.2.3 Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16;
1887 :     Data->Ref = pRef + (x + iEdgedWidth*y)*16;
1888 :     Data->RefH = pRefH + (x + iEdgedWidth*y) * 16;
1889 :     Data->RefV = pRefV + (x + iEdgedWidth*y) * 16;
1890 :     Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
1891 : Isibaar 1.44.2.10 Data->RefQ = pRefQ;
1892 :    
1893 : h 1.44.2.3 Data->iQuant = iQuant;
1894 : edgomez 1.44.2.4
1895 : chl 1.44.2.1 if (!(MotionFlags & PMV_HALFPEL16)) {
1896 : h 1.44.2.3 Data->min_dx = EVEN(Data->min_dx);
1897 :     Data->max_dx = EVEN(Data->max_dx);
1898 :     Data->min_dy = EVEN(Data->min_dy);
1899 :     Data->max_dy = EVEN(Data->max_dy);
1900 : chl 1.30 }
1901 :    
1902 : h 1.44.2.3 for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
1903 : chl 1.30
1904 : chl 1.44.2.1 if (pMB->dquant != NO_CHANGE) inter4v = 0;
1905 : chl 1.30
1906 : chl 1.44.2.1 if (inter4v)
1907 :     CheckCandidate = CheckCandidate16;
1908 :     else CheckCandidate = CheckCandidate16no4v;
1909 : chl 1.30
1910 :    
1911 : chl 1.44.2.1 pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
1912 :     pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
1913 : h 1.44.2.3 if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
1914 :     if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
1915 :     if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
1916 :     if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
1917 : chl 1.30
1918 : edgomez 1.44.2.4 (*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
1919 : chl 1.30
1920 : chl 1.44.2.1 if (pMB->mode == MODE_INTER4V)
1921 :     for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
1922 :     pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
1923 :     pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
1924 :     if (!(make_mask(pMB->mvs, i)))
1925 : edgomez 1.44.2.4 (*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
1926 : chl 1.44.2.1 }
1927 : chl 1.30
1928 :     if (MotionFlags & PMV_USESQUARES16)
1929 : chl 1.44.2.1 MainSearchPtr = SquareSearch;
1930 : chl 1.30 else if (MotionFlags & PMV_ADVANCEDDIAMOND16)
1931 : chl 1.44.2.1 MainSearchPtr = AdvDiamondSearch;
1932 :     else MainSearchPtr = DiamondSearch;
1933 : chl 1.30
1934 : h 1.44.2.3 (*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
1935 : chl 1.30
1936 : h 1.44.2.3 if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
1937 : chl 1.30
1938 : syskin 1.44.2.8 for(i = 0; i < 5; i++) {
1939 :     Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
1940 :     Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
1941 :     }
1942 :    
1943 :     if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
1944 :     if(inter4v)
1945 :     CheckCandidate = CheckCandidate16_qpel;
1946 :     else
1947 :     CheckCandidate = CheckCandidate16no4v_qpel;
1948 : Isibaar 1.44.2.10
1949 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
1950 :     pParam->width, pParam->height, Data->iFcode, 0); // get real range
1951 :    
1952 : syskin 1.44.2.8 QuarterpelRefine(Data);
1953 :     }
1954 :    
1955 :     if (inter4v) {
1956 :     SearchData Data8;
1957 :     Data8.iFcode = Data->iFcode;
1958 :     Data8.iQuant = Data->iQuant;
1959 :     Data8.iEdgedWidth = Data->iEdgedWidth;
1960 :     Search8hinted(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
1961 :     Search8hinted(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
1962 :     Search8hinted(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
1963 :     Search8hinted(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
1964 :     }
1965 : chl 1.30
1966 : chl 1.44.2.1 if (!(inter4v) ||
1967 : h 1.44.2.3 (Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] +
1968 :     Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
1969 : chl 1.44.2.1 // INTER MODE
1970 : chl 1.30
1971 : chl 1.44.2.1 pMB->mode = MODE_INTER;
1972 : Isibaar 1.44.2.2 pMB->mvs[0] = pMB->mvs[1]
1973 : h 1.44.2.3 = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
1974 : chl 1.30
1975 : chl 1.44.2.1 pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
1976 : h 1.44.2.3 pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
1977 : chl 1.30
1978 : h 1.44.2.3 pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
1979 :     pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
1980 : chl 1.44.2.1 } else {
1981 :     // INTER4V MODE; all other things are already set in Search8hinted
1982 :     pMB->mode = MODE_INTER4V;
1983 : h 1.44.2.3 pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3]
1984 :     + Data->iMinSAD[4] + IMV16X16 * iQuant;
1985 : chl 1.44.2.1 }
1986 : chl 1.30
1987 :     }
1988 : suxen_drol 1.6
1989 : edgomez 1.20 void
1990 : chl 1.44.2.1 MotionEstimationHinted( MBParam * const pParam,
1991 :     FRAMEINFO * const current,
1992 :     FRAMEINFO * const reference,
1993 :     const IMAGE * const pRefH,
1994 :     const IMAGE * const pRefV,
1995 :     const IMAGE * const pRefHV)
1996 : suxen_drol 1.6 {
1997 : chl 1.44.2.1 MACROBLOCK *const pMBs = current->mbs;
1998 :     const IMAGE *const pCurrent = &current->image;
1999 :     const IMAGE *const pRef = &reference->image;
2000 : chl 1.32
2001 : chl 1.44.2.1 uint32_t x, y;
2002 : Isibaar 1.44.2.10 uint8_t *qimage;
2003 : syskin 1.44.2.5 int32_t temp[5], quant = current->quant;
2004 : h 1.44.2.3 int32_t iMinSAD[5];
2005 :     VECTOR currentMV[5];
2006 : Isibaar 1.44.2.10 VECTOR currentQMV[5];
2007 : h 1.44.2.3 SearchData Data;
2008 :     Data.iEdgedWidth = pParam->edged_width;
2009 :     Data.currentMV = currentMV;
2010 : Isibaar 1.44.2.10 Data.currentQMV = currentQMV;
2011 : h 1.44.2.3 Data.iMinSAD = iMinSAD;
2012 :     Data.temp = temp;
2013 :     Data.iFcode = current->fcode;
2014 : syskin 1.44.2.8 Data.rounding = pParam->m_rounding_type;
2015 :    
2016 :     if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
2017 :     return; // allocate some mem for qpel interpolated blocks
2018 :     // somehow this is dirty since I think we shouldn't use malloc outside
2019 :     // encoder_create() - so please fix me!
2020 :    
2021 : chl 1.44.2.1 if (sadInit) (*sadInit) ();
2022 : chl 1.32
2023 : chl 1.44.2.1 for (y = 0; y < pParam->mb_height; y++) {
2024 :     for (x = 0; x < pParam->mb_width; x++) {
2025 : suxen_drol 1.6
2026 : chl 1.44.2.1 MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
2027 : suxen_drol 1.6
2028 : chl 1.44.2.1 //intra mode is copied from the first pass. At least for the time being
2029 :     if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue;
2030 : chl 1.37
2031 : syskin 1.44.2.5
2032 : chl 1.44.2.1 if (!(current->global_flags & XVID_LUMIMASKING)) {
2033 :     pMB->dquant = NO_CHANGE;
2034 :     pMB->quant = current->quant; }
2035 : syskin 1.44.2.5 else
2036 :     if (pMB->dquant != NO_CHANGE) {
2037 :     quant += DQtab[pMB->dquant];
2038 :     if (quant > 31) quant = 31;
2039 :     else if (quant < 1) quant = 1;
2040 :     pMB->quant = quant;
2041 :     }
2042 : chl 1.33
2043 : Isibaar 1.44.2.10 SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x,
2044 : Isibaar 1.44.2.2 y, current->motion_flags, pMB->quant,
2045 : h 1.44.2.3 pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
2046 :     &Data);
2047 : edgomez 1.20
2048 : Isibaar 1.44.2.2 }
2049 :     }
2050 : syskin 1.44.2.8 free(qimage);
2051 : Isibaar 1.44.2.2 }
2052 : chl 1.33
2053 : Isibaar 1.44.2.2 static __inline int
2054 :     MEanalyzeMB ( const uint8_t * const pRef,
2055 :     const uint8_t * const pCur,
2056 :     const int x,
2057 :     const int y,
2058 :     const MBParam * const pParam,
2059 :     const MACROBLOCK * const pMBs,
2060 : h 1.44.2.3 MACROBLOCK * const pMB,
2061 :     SearchData * const Data)
2062 : Isibaar 1.44.2.2 {
2063 : chl 1.32
2064 : Isibaar 1.44.2.2 int i, mask;
2065 : h 1.44.2.3 VECTOR pmv[3];
2066 : suxen_drol 1.22
2067 : h 1.44.2.3 *(Data->iMinSAD) = MV_MAX_ERROR;
2068 :     Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
2069 :     get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
2070 : Isibaar 1.44.2.7 pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
2071 : chl 1.35
2072 : h 1.44.2.3 Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
2073 :     Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
2074 : Isibaar 1.44.2.2
2075 :     CheckCandidate = CheckCandidate16no4vI;
2076 : chl 1.44.2.1
2077 : Isibaar 1.44.2.2 pmv[1].x = EVEN(pMB->mvs[0].x);
2078 :     pmv[1].y = EVEN(pMB->mvs[0].y);
2079 : h 1.44.2.3 pmv[0].x = EVEN(Data->predMV.x);
2080 :     pmv[0].y = EVEN(Data->predMV.y);
2081 : Isibaar 1.44.2.2 pmv[2].x = pmv[2].y = 0;
2082 :    
2083 : h 1.44.2.3 CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
2084 : Isibaar 1.44.2.2 if (!(mask = make_mask(pmv, 1)))
2085 : h 1.44.2.3 CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
2086 : Isibaar 1.44.2.2 if (!(mask = make_mask(pmv, 2)))
2087 : h 1.44.2.3 CheckCandidate16no4vI(0, 0, mask, &i, Data);
2088 : Isibaar 1.44.2.2
2089 : h 1.44.2.3 DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
2090 : Isibaar 1.44.2.2
2091 :     pMB->mvs[0] = pMB->mvs[1]
2092 : h 1.44.2.3 = pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv()
2093 : Isibaar 1.44.2.2
2094 : h 1.44.2.3 return *(Data->iMinSAD);
2095 : Isibaar 1.44.2.2 }
2096 :    
2097 :     #define INTRA_THRESH 1350
2098 :     #define INTER_THRESH 900
2099 :    
2100 :     int
2101 :     MEanalysis( const IMAGE * const pRef,
2102 :     const IMAGE * const pCurrent,
2103 :     MBParam * const pParam,
2104 :     MACROBLOCK * const pMBs,
2105 :     const uint32_t iFcode)
2106 :     {
2107 :     uint32_t x, y, intra = 0;
2108 :     int sSAD = 0;
2109 : h 1.44.2.3
2110 :     VECTOR currentMV;
2111 :     int32_t iMinSAD;
2112 :     SearchData Data;
2113 :     Data.iEdgedWidth = pParam->edged_width;
2114 :     Data.currentMV = &currentMV;
2115 :     Data.iMinSAD = &iMinSAD;
2116 :     Data.iFcode = iFcode;
2117 :     Data.iQuant = 2;
2118 :    
2119 : Isibaar 1.44.2.2 if (sadInit) (*sadInit) ();
2120 :    
2121 :     for (y = 0; y < pParam->mb_height-1; y++) {
2122 :     for (x = 0; x < pParam->mb_width; x++) {
2123 :     int sad, dev;
2124 :     MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
2125 :    
2126 :     sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
2127 : h 1.44.2.3 pParam, pMBs, pMB, &Data);
2128 : Isibaar 1.44.2.2
2129 :     if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work
2130 :     if (sad > INTRA_THRESH) {
2131 :     dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
2132 :     pParam->edged_width);
2133 :     if (dev + INTRA_THRESH < sad) intra++;
2134 :     if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame
2135 :     }
2136 :     sSAD += sad;
2137 :     }
2138 : chl 1.35
2139 : suxen_drol 1.6 }
2140 :     }
2141 : Isibaar 1.44.2.2 sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
2142 :     if (sSAD > INTER_THRESH ) return 1; //P frame
2143 :     emms();
2144 :     return 0; // B frame
2145 : chl 1.44.2.1
2146 : syskin 1.44.2.6 }
2147 :    
2148 :     int
2149 :     FindFcode( const MBParam * const pParam,
2150 :     const FRAMEINFO * const current)
2151 :     {
2152 :     uint32_t x, y;
2153 :     int max = 0, min = 0, i;
2154 :    
2155 :     for (y = 0; y < pParam->mb_height; y++) {
2156 :     for (x = 0; x < pParam->mb_width; x++) {
2157 :    
2158 :     MACROBLOCK *pMB = &current->mbs[x + y * pParam->mb_width];
2159 :     for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) {
2160 :     if (pMB->mvs[i].x > max) max = pMB->mvs[i].x;
2161 :     if (pMB->mvs[i].y > max) max = pMB->mvs[i].y;
2162 :    
2163 :     if (pMB->mvs[i].x < min) min = pMB->mvs[i].x;
2164 :     if (pMB->mvs[i].y < min) min = pMB->mvs[i].y;
2165 :     }
2166 :     }
2167 :     }
2168 :    
2169 :     min = -min;
2170 :     max += 1;
2171 :     if (min > max) max = min;
2172 :    
2173 :     for (i = 1; (max > 32 << (i - 1)); i++);
2174 :     return i;
2175 : Isibaar 1.44.2.2 }
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4