37 |
#include "../prediction/mbprediction.h" |
#include "../prediction/mbprediction.h" |
38 |
#include "../global.h" |
#include "../global.h" |
39 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
40 |
|
#include "../image/interpolate8x8.h" |
41 |
#include "motion_est.h" |
#include "motion_est.h" |
42 |
#include "motion.h" |
#include "motion.h" |
43 |
#include "sad.h" |
#include "sad.h" |
52 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
53 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
54 |
|
|
55 |
|
#define GET_REFERENCE(X, Y, REF) { \ |
56 |
|
switch ( (((X)&1)<<1) + ((Y)&1) ) \ |
57 |
|
{ \ |
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 |
#define iDiamondSize 2 |
#define iDiamondSize 2 |
66 |
|
|
67 |
static __inline int |
static __inline int |
95 |
static void |
static void |
96 |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
97 |
{ |
{ |
|
int32_t * const sad = data->temp; |
|
98 |
int t; |
int t; |
99 |
const uint8_t * Reference; |
const uint8_t * Reference; |
100 |
|
|
108 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
109 |
} |
} |
110 |
|
|
111 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
112 |
|
|
113 |
|
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); |
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
117 |
|
|
118 |
data->temp[0] += lambda_vec16[data->iQuant] * t; |
data->temp[0] += lambda_vec16[data->iQuant] * t; |
119 |
data->temp[1] += lambda_vec8[data->iQuant] * t; |
data->temp[1] += lambda_vec8[data->iQuant] * t; |
120 |
|
|
151 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
152 |
} |
} |
153 |
|
|
154 |
|
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] * |
sad = lambda_vec16[data->iQuant] * |
159 |
d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
160 |
sad += sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
|
161 |
|
sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR); |
162 |
|
|
163 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
164 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
167 |
} |
} |
168 |
|
|
169 |
static void |
static void |
170 |
|
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 |
|
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1); |
228 |
|
|
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 |
|
/* *dir = Direction;*/ } |
237 |
|
|
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 |
|
// *dir = Direction; |
314 |
|
} |
315 |
|
} |
316 |
|
|
317 |
|
static void |
318 |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
319 |
{ |
{ |
320 |
int32_t sad; |
int32_t sad; |
436 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
437 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
438 |
|
|
439 |
if (( x > 31) || ( x < -31) || ( y > 31) || (y < -31)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
440 |
|
|
441 |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
442 |
|
|
495 |
} |
} |
496 |
|
|
497 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
498 |
|
|
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); |
sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
503 |
|
|
504 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
507 |
*dir = Direction; } |
*dir = Direction; } |
508 |
} |
} |
509 |
|
|
510 |
|
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 |
/* CHECK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
569 |
|
|
570 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
733 |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
734 |
} |
} |
735 |
|
|
736 |
|
|
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 |
static __inline int |
static __inline int |
759 |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
760 |
const int x, const int y, |
const int x, const int y, |
781 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
782 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
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; |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
784 |
|
|
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 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
789 |
} |
} |
790 |
|
|
805 |
|
|
806 |
uint32_t x, y; |
uint32_t x, y; |
807 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
808 |
int32_t InterBias; |
int32_t InterBias, quant = current->quant; |
809 |
|
uint8_t *qimage; |
810 |
|
|
811 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
812 |
int32_t temp[5]; |
int32_t temp[5]; |
813 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
814 |
|
VECTOR currentQMV[5]; |
815 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
816 |
SearchData Data; |
SearchData Data; |
817 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
818 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
819 |
|
Data.currentQMV = currentQMV; |
820 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
821 |
Data.temp = temp; |
Data.temp = temp; |
822 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
823 |
|
Data.rounding = pParam->m_rounding_type; |
824 |
|
Data.quarterpel = pParam->m_quarterpel; |
825 |
|
|
826 |
|
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
827 |
|
return 1; // allocate some mem for qpel interpolated blocks |
828 |
|
// somehow this is dirty since I think we shouldn't use malloc outside |
829 |
|
// encoder_create() - so please fix me! |
830 |
|
|
831 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
832 |
|
|
842 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
843 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
844 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; } |
845 |
|
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 |
|
|
853 |
//initial skip decision |
//initial skip decision |
854 |
|
|
857 |
if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) { |
if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) { |
858 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
859 |
continue; |
continue; |
860 |
|
sad00 = 256 * 4096; |
861 |
} |
} |
862 |
} else sad00 = 256*4096; // skip not allowed - for final skip decision |
} else sad00 = 256*4096; // skip not allowed - for final skip decision |
863 |
|
|
864 |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x, |
865 |
y, current->motion_flags, pMB->quant, |
y, current->motion_flags, pMB->quant, |
866 |
&Data, pParam, pMBs, reference->mbs, |
&Data, pParam, pMBs, reference->mbs, |
867 |
current->global_flags & XVID_INTER4V, pMB); |
current->global_flags & XVID_INTER4V, pMB); |
886 |
pParam->edged_width); |
pParam->edged_width); |
887 |
|
|
888 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
889 |
if (++iIntra >= iLimit) return 1; |
if (++iIntra >= iLimit) { free(qimage); return 1; } |
890 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
891 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
892 |
pMB->mvs[3] = zeroMV; |
pMB->mvs[3] = zeroMV; |
893 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = |
894 |
|
pMB->qmvs[3] = zeroMV; |
895 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
896 |
pMB->sad8[3] = 0; |
pMB->sad8[3] = 0; |
897 |
} |
} |
898 |
} |
} |
899 |
} |
} |
900 |
} |
} |
901 |
|
free(qimage); |
902 |
return 0; |
return 0; |
903 |
} |
} |
904 |
|
|
960 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
961 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
962 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
963 |
|
const uint8_t * const pRefQ, |
964 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
965 |
const int x, |
const int x, |
966 |
const int y, |
const int y, |
977 |
int i, iDirection = 255, mask, threshA; |
int i, iDirection = 255, mask, threshA; |
978 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
979 |
|
|
980 |
|
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
981 |
|
|
982 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(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, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
984 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
985 |
|
|
986 |
Data->predMV = pmv[0]; |
Data->predMV = pmv[0]; |
987 |
|
|
988 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
989 |
Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; |
Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; |
990 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
991 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
992 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
993 |
|
Data->RefQ = pRefQ; |
994 |
|
|
995 |
Data->iQuant = iQuant; |
Data->iQuant = iQuant; |
996 |
|
|
1005 |
if (inter4v) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
1006 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1007 |
|
|
1008 |
for(i = 0; i < 5; i++) Data->currentMV[i].x = Data->currentMV[i].y = 0; |
for(i = 0; i < 5; i++) |
1009 |
|
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
1010 |
|
|
1011 |
|
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); |
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
1015 |
|
|
1016 |
Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
1017 |
Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
1018 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
1089 |
|
|
1090 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1091 |
|
|
1092 |
|
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 |
|
|
1104 |
|
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 |
|
|
1107 |
|
QuarterpelRefine(Data); |
1108 |
|
} |
1109 |
|
|
1110 |
if (inter4v) { |
if (inter4v) { |
1111 |
SearchData Data8; |
SearchData Data8; |
1112 |
Data8.iFcode = Data->iFcode; |
Data8.iFcode = Data->iFcode; |
1126 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1127 |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1128 |
|
|
1129 |
|
pMB->qmvs[0] = pMB->qmvs[1] |
1130 |
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1131 |
|
|
1132 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1133 |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1134 |
|
|
1135 |
|
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; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1141 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1142 |
|
} |
1143 |
} else { |
} else { |
1144 |
// INTER4V MODE; all other things are already set in Search8 |
// INTER4V MODE; all other things are already set in Search8 |
1145 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1146 |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
1147 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
1148 |
} |
} |
|
|
|
1149 |
} |
} |
1150 |
|
|
1151 |
static void |
static void |
1159 |
SearchData * const Data) |
SearchData * const Data) |
1160 |
{ |
{ |
1161 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1162 |
|
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1163 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1164 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1165 |
|
Data->currentQMV = OldData->currentQMV + 1 + block; |
1166 |
|
Data->quarterpel = OldData->quarterpel; |
1167 |
|
|
1168 |
|
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 |
|
|
1176 |
|
} else // add d_mv_bits[hpel] everywhere but not in 0 (it's already there) |
1177 |
if (block != 0) |
if (block != 0) |
1178 |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
1179 |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
1186 |
Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
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)); |
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)); |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1189 |
|
Data->RefQ = OldData->RefQ; |
1190 |
|
|
1191 |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1192 |
|
|
1193 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1194 |
pParam->width, pParam->height, OldData->iFcode); |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
1195 |
|
|
1196 |
CheckCandidate = CheckCandidate8; |
CheckCandidate = CheckCandidate8; |
1197 |
|
|
1198 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1199 |
|
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1200 |
|
|
1201 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1202 |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
1203 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1204 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1205 |
|
|
1206 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); } |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1207 |
|
|
1208 |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(Data); |
if(*(Data->iMinSAD) < temp_sad) { //found a better match? |
1209 |
|
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 |
|
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 |
|
|
1233 |
|
QuarterpelRefine(Data); |
1234 |
|
} |
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; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1243 |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1244 |
|
} |
1245 |
|
|
1246 |
pMB->mvs[block] = *(Data->currentMV); |
pMB->mvs[block] = *(Data->currentMV); |
1247 |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
pMB->qmvs[block] = *(Data->currentQMV); |
1248 |
|
|
1249 |
|
pMB->sad8[block] = 4 * (*Data->iMinSAD); // Isibaar: why? |
1250 |
} |
} |
1251 |
|
|
1252 |
/* B-frames code starts here */ |
/* B-frames code starts here */ |
1332 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1333 |
|
|
1334 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1335 |
pParam->width, pParam->height, iFcode); |
pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
1336 |
|
|
1337 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1338 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
1447 |
|
|
1448 |
// skip decision |
// skip decision |
1449 |
if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
1450 |
//checking chroma. everything copied from MC |
//possible skip - checking chroma. everything copied from MC |
1451 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1452 |
int sum, dx, dy, b_dx, b_dy; |
int sum, dx, dy, b_dx, b_dy; |
1453 |
|
|
1569 |
|
|
1570 |
fData->currentMV[0] = pMB->mvs[0]; |
fData->currentMV[0] = pMB->mvs[0]; |
1571 |
fData->currentMV[1] = pMB->b_mvs[0]; |
fData->currentMV[1] = pMB->b_mvs[0]; |
1572 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
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); |
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 |
|
|
1575 |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
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; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
1757 |
|
|
1758 |
/* Hinted ME starts here */ |
/* Hinted ME starts here */ |
1759 |
|
|
1760 |
static __inline void |
static void |
1761 |
Search8hinted( const SearchData * const OldData, |
Search8hinted( const SearchData * const OldData, |
1762 |
const int x, const int y, |
const int x, const int y, |
1763 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1764 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1765 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1766 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1767 |
const int block) |
const int block, |
1768 |
|
SearchData * const Data) |
1769 |
{ |
{ |
1770 |
SearchData Data; |
int32_t temp_sad; |
1771 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1772 |
|
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 |
|
Data->quarterpel = OldData->quarterpel; |
1778 |
|
|
1779 |
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
if (block != 0) { |
1780 |
Data.iMinSAD = OldData->iMinSAD + 1 + block; |
if(pParam->m_quarterpel) { |
1781 |
Data.currentMV = OldData->currentMV+1+block; |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
1782 |
Data.iFcode = OldData->iFcode; |
d_mv_bits( Data->currentQMV->x - Data->predQMV.x, |
1783 |
Data.iQuant = OldData->iQuant; |
Data->currentQMV->y - Data->predQMV.y, |
1784 |
|
Data->iFcode); |
1785 |
Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
} |
1786 |
Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
else { |
1787 |
Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
1788 |
Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
1789 |
Data.iEdgedWidth = pParam->edged_width; |
Data->currentMV->y - Data->predMV.y, |
1790 |
Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->iFcode); |
1791 |
|
} |
1792 |
|
} |
1793 |
|
|
1794 |
CheckCandidate = CheckCandidate8; |
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 |
|
|
1800 |
if (block != 0) |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1801 |
*(Data.iMinSAD) += lambda_vec8[Data.iQuant] * |
|
1802 |
d_mv_bits( Data.currentMV->x - Data.predMV.x, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1803 |
Data.currentMV->y - Data.predMV.y, |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
|
Data.iFcode); |
|
1804 |
|
|
1805 |
|
CheckCandidate = CheckCandidate8; |
1806 |
|
|
1807 |
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8, |
temp_sad = *(Data->iMinSAD); // store current MinSAD |
|
pParam->width, pParam->height, OldData->iFcode); |
|
|
|
|
|
if (pMB->mode == MODE_INTER4V) { |
|
|
int dummy; |
|
|
CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); } |
|
1808 |
|
|
1809 |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
1810 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1811 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1812 |
|
|
1813 |
(*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255); |
(*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 |
|
|
1823 |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data); |
HalfpelRefine(Data); // perform halfpel refine of current best vector |
1824 |
|
|
1825 |
pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x; |
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
1826 |
pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y; |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1827 |
pMB->mvs[block] = *(Data.currentMV); |
Data->currentQMV->y = 2 * Data->currentMV->y; |
1828 |
pMB->sad8[block] = 4 * (*(Data.iMinSAD)); |
} |
1829 |
|
} |
1830 |
|
|
1831 |
|
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 |
|
} |
1841 |
|
|
1842 |
|
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 |
} |
} |
1856 |
|
|
1857 |
|
|
1860 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1861 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1862 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1863 |
|
const uint8_t * const pRefQ, |
1864 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1865 |
const int x, |
const int x, |
1866 |
const int y, |
const int y, |
1878 |
int i, t; |
int i, t; |
1879 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1880 |
|
|
1881 |
|
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1882 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
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, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1884 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
1885 |
|
|
1886 |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
1887 |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
1888 |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
1889 |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
1890 |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
1891 |
|
Data->RefQ = pRefQ; |
1892 |
|
|
1893 |
Data->iQuant = iQuant; |
Data->iQuant = iQuant; |
1894 |
|
|
1895 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1935 |
|
|
1936 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1937 |
|
|
1938 |
|
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) |
if (inter4v) |
1945 |
for(i = 0; i < 4; i++) |
CheckCandidate = CheckCandidate16_qpel; |
1946 |
Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
else |
1947 |
|
CheckCandidate = CheckCandidate16no4v_qpel; |
1948 |
|
|
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 |
|
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 |
|
|
1966 |
if (!(inter4v) || |
if (!(inter4v) || |
1967 |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
1999 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
2000 |
|
|
2001 |
uint32_t x, y; |
uint32_t x, y; |
2002 |
int32_t temp[5]; |
uint8_t *qimage; |
2003 |
|
int32_t temp[5], quant = current->quant; |
2004 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
2005 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
2006 |
|
VECTOR currentQMV[5]; |
2007 |
SearchData Data; |
SearchData Data; |
2008 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2009 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
2010 |
|
Data.currentQMV = currentQMV; |
2011 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
2012 |
Data.temp = temp; |
Data.temp = temp; |
2013 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
2014 |
|
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 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2022 |
|
|
2028 |
//intra mode is copied from the first pass. At least for the time being |
//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; |
if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; |
2030 |
|
|
2031 |
|
|
2032 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
2033 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
2034 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; } |
2035 |
|
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 |
|
|
2043 |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x, |
2044 |
y, current->motion_flags, pMB->quant, |
y, current->motion_flags, pMB->quant, |
2045 |
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
2046 |
&Data); |
&Data); |
2047 |
|
|
2048 |
} |
} |
2049 |
} |
} |
2050 |
|
free(qimage); |
2051 |
} |
} |
2052 |
|
|
2053 |
static __inline int |
static __inline int |
2067 |
*(Data->iMinSAD) = MV_MAX_ERROR; |
*(Data->iMinSAD) = MV_MAX_ERROR; |
2068 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
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, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2070 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
2071 |
|
|
2072 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2073 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
2144 |
return 0; // B frame |
return 0; // B frame |
2145 |
|
|
2146 |
} |
} |
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 = ¤t->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 |
|
} |