21 |
* along with this program ; if not, write to the Free Software |
* along with this program ; if not, write to the Free Software |
22 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
* |
* |
24 |
* $Id: motion_est.c,v 1.58.2.28 2003-08-25 15:10:13 syskin Exp $ |
* $Id: motion_est.c,v 1.58.2.33 2003-09-04 18:40:02 Isibaar Exp $ |
25 |
* |
* |
26 |
****************************************************************************/ |
****************************************************************************/ |
27 |
|
|
70 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
71 |
|
|
72 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
73 |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
CheckCandidate((X),(Y), data, (D) ); } |
74 |
|
|
75 |
|
|
76 |
/***************************************************************************** |
/***************************************************************************** |
156 |
int sad; |
int sad; |
157 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
158 |
int offset = (dx>>1) + (dy>>1)*stride; |
int offset = (dx>>1) + (dy>>1)*stride; |
159 |
|
int next = 1; |
160 |
|
|
161 |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */ |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */ |
162 |
data->temp[5] = dx; data->temp[6] = dy; /* backup */ |
data->temp[5] = dx; data->temp[6] = dy; /* backup */ |
167 |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
168 |
break; |
break; |
169 |
case 1: |
case 1: |
170 |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + stride, stride); |
next = stride; |
|
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + stride, stride); |
|
|
break; |
|
171 |
case 2: |
case 2: |
172 |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + 1, stride); |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + next, stride); |
173 |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + 1, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + next, stride); |
174 |
break; |
break; |
175 |
default: |
default: |
176 |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
300 |
/* CHECK_CANDIATE FUNCTIONS START */ |
/* CHECK_CANDIATE FUNCTIONS START */ |
301 |
|
|
302 |
static void |
static void |
303 |
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 SearchData * const data, const int Direction) |
304 |
{ |
{ |
305 |
int xc, yc; |
int xc, yc; |
306 |
const uint8_t * Reference; |
const uint8_t * Reference; |
320 |
current = data->currentQMV; |
current = data->currentQMV; |
321 |
} |
} |
322 |
|
|
323 |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp); |
324 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
325 |
|
|
326 |
|
sad += (data->lambda16 * t * sad)>>10; |
327 |
|
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
328 |
|
|
329 |
|
if (data->chroma && sad < data->iMinSAD[0]) |
330 |
|
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
331 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
332 |
|
|
333 |
|
if (sad < data->iMinSAD[0]) { |
334 |
|
data->iMinSAD[0] = sad; |
335 |
|
current[0].x = x; current[0].y = y; |
336 |
|
*data->dir = Direction; |
337 |
|
} |
338 |
|
|
339 |
|
if (data->temp[0] < data->iMinSAD[1]) { |
340 |
|
data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; } |
341 |
|
if (data->temp[1] < data->iMinSAD[2]) { |
342 |
|
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
343 |
|
if (data->temp[2] < data->iMinSAD[3]) { |
344 |
|
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
345 |
|
if (data->temp[3] < data->iMinSAD[4]) { |
346 |
|
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
347 |
|
} |
348 |
|
|
349 |
|
static void |
350 |
|
CheckCandidate16_subpel(const int x, const int y, const SearchData * const data, const int Direction) |
351 |
|
{ |
352 |
|
int xc, yc; |
353 |
|
const uint8_t *Reference; |
354 |
|
VECTOR *current, *current2; |
355 |
|
int32_t sad; uint32_t t; |
356 |
|
|
357 |
|
if ( (x > data->max_dx) || (x < data->min_dx) |
358 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
359 |
|
|
360 |
|
if (!data->qpel_precision) { |
361 |
|
Reference = GetReference(x, y, data); |
362 |
|
current = data->currentMV; |
363 |
|
current2 = data->currentMV2; |
364 |
|
xc = x; yc = y; |
365 |
|
} else { /* x and y are in 1/4 precision */ |
366 |
|
Reference = Interpolate16x16qpel(x, y, 0, data); |
367 |
|
xc = x/2; yc = y/2; /* for chroma sad */ |
368 |
|
current = data->currentQMV; |
369 |
|
current2 = data->currentQMV2; |
370 |
|
} |
371 |
|
|
372 |
|
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp); |
373 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
374 |
|
|
375 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
376 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
377 |
|
|
378 |
if (data->chroma && sad < data->iMinSAD[0]) |
if (data->chroma && sad < data->iMinSAD[0]) |
379 |
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
380 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
381 |
|
|
382 |
|
if (data->temp[0] < data->iMinSAD[1]) { |
383 |
|
data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; } |
384 |
|
if (data->temp[1] < data->iMinSAD[2]) { |
385 |
|
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
386 |
|
if (data->temp[2] < data->iMinSAD[3]) { |
387 |
|
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
388 |
|
if (data->temp[3] < data->iMinSAD[4]) { |
389 |
|
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
390 |
|
|
391 |
if (sad < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
392 |
|
*(data->iMinSAD2) = *(data->iMinSAD); |
393 |
|
current2->x = current->x; current2->y = current->y; |
394 |
|
|
395 |
data->iMinSAD[0] = sad; |
data->iMinSAD[0] = sad; |
396 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
397 |
*dir = Direction; |
*data->dir = Direction; |
398 |
|
return; |
399 |
} |
} |
400 |
|
|
401 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (sad < *(data->iMinSAD2)) { |
402 |
data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y = y; } |
*(data->iMinSAD2) = sad; |
403 |
if (data->temp[2] < data->iMinSAD[2]) { |
current2->x = x; current2->y = y; |
404 |
data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; } |
*data->dir = Direction; |
405 |
if (data->temp[3] < data->iMinSAD[3]) { |
} |
|
data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; } |
|
|
if (data->temp[4] < data->iMinSAD[4]) { |
|
|
data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; } |
|
406 |
} |
} |
407 |
|
|
408 |
static void |
static void |
409 |
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate8(const int x, const int y, const SearchData * const data, const int Direction) |
410 |
{ |
{ |
411 |
int32_t sad; uint32_t t; |
int32_t sad; uint32_t t; |
412 |
const uint8_t * Reference; |
const uint8_t * Reference; |
431 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
432 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
433 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
434 |
*dir = Direction; |
*data->dir = Direction; |
435 |
} |
} |
436 |
} |
} |
437 |
|
|
438 |
static void |
static void |
439 |
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32(const int x, const int y, const SearchData * const data, const int Direction) |
440 |
{ |
{ |
441 |
uint32_t t; |
uint32_t t; |
442 |
const uint8_t * Reference; |
const uint8_t * Reference; |
443 |
|
int sad; |
444 |
|
|
445 |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
446 |
(x > data->max_dx) || (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
449 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
450 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
451 |
|
|
452 |
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp); |
453 |
|
|
454 |
data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10; |
sad += (data->lambda16 * t * sad) >> 10; |
455 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
456 |
|
|
457 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
458 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = sad; |
459 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
460 |
*dir = Direction; } |
*data->dir = Direction; |
461 |
|
} |
462 |
|
|
463 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[0] < data->iMinSAD[1]) { |
464 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
465 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
466 |
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
467 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
468 |
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
469 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
470 |
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
471 |
} |
} |
472 |
|
|
473 |
static void |
static void |
474 |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4v(const int x, const int y, const SearchData * const data, const int Direction) |
475 |
{ |
{ |
476 |
int32_t sad, xc, yc; |
int32_t sad, xc, yc; |
477 |
const uint8_t * Reference; |
const uint8_t * Reference; |
505 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
506 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
507 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
508 |
*dir = Direction; |
*data->dir = Direction; |
509 |
} |
} |
510 |
} |
} |
511 |
|
|
512 |
static void |
static void |
513 |
CheckCandidate16I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16I(const int x, const int y, const SearchData * const data, const int Direction) |
514 |
{ |
{ |
515 |
int sad; |
int sad; |
516 |
// int xc, yc; |
// int xc, yc; |
533 |
if (sad < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
534 |
data->iMinSAD[0] = sad; |
data->iMinSAD[0] = sad; |
535 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
536 |
*dir = Direction; |
*data->dir = Direction; |
537 |
} |
} |
538 |
} |
} |
539 |
|
|
540 |
static void |
static void |
541 |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32I(const int x, const int y, const SearchData * const data, const int Direction) |
542 |
{ |
{ |
543 |
/* maximum speed - for P/B/I decision */ |
/* maximum speed - for P/B/I decision */ |
544 |
int32_t sad; |
int32_t sad; |
546 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
547 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
548 |
|
|
549 |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*((int)data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->RefP[0] + x + y*((int)data->iEdgedWidth), |
550 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp); |
551 |
|
|
552 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
553 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
554 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
555 |
*dir = Direction; |
*data->dir = Direction; |
556 |
} |
} |
557 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[0] < data->iMinSAD[1]) { |
558 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
559 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
560 |
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
561 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
562 |
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
563 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
564 |
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
565 |
|
|
566 |
} |
} |
567 |
|
|
568 |
static void |
static void |
569 |
CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateInt(const int xf, const int yf, const SearchData * const data, const int Direction) |
570 |
{ |
{ |
571 |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
572 |
uint32_t t; |
uint32_t t; |
608 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
609 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
610 |
current->x = xf; current->y = yf; |
current->x = xf; current->y = yf; |
611 |
*dir = Direction; |
*data->dir = Direction; |
612 |
} |
} |
613 |
} |
} |
614 |
|
|
615 |
static void |
static void |
616 |
CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateDirect(const int x, const int y, const SearchData * const data, const int Direction) |
617 |
{ |
{ |
618 |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
619 |
uint32_t k; |
uint32_t k; |
669 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
670 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
671 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
672 |
*dir = Direction; |
*data->dir = Direction; |
673 |
} |
} |
674 |
} |
} |
675 |
|
|
676 |
static void |
static void |
677 |
CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateDirectno4v(const int x, const int y, const SearchData * const data, const int Direction) |
678 |
{ |
{ |
679 |
int32_t sad, xcf, ycf, xcb, ycb; |
int32_t sad, xcf, ycf, xcb, ycb; |
680 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
722 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
723 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
724 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
725 |
*dir = Direction; |
*data->dir = Direction; |
726 |
} |
} |
727 |
} |
} |
728 |
|
|
729 |
|
|
730 |
static void |
static void |
731 |
CheckCandidateRD16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD16(const int x, const int y, const SearchData * const data, const int Direction) |
732 |
{ |
{ |
733 |
|
|
734 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
791 |
if (rd < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
792 |
data->iMinSAD[0] = rd; |
data->iMinSAD[0] = rd; |
793 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
794 |
*dir = Direction; |
*data->dir = Direction; |
795 |
*data->cbp = cbp; |
*data->cbp = cbp; |
796 |
} |
} |
797 |
} |
} |
798 |
|
|
799 |
static void |
static void |
800 |
CheckCandidateRD8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD8(const int x, const int y, const SearchData * const data, const int Direction) |
801 |
{ |
{ |
802 |
|
|
803 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
825 |
*data->cbp = cbp; |
*data->cbp = cbp; |
826 |
data->iMinSAD[0] = rd; |
data->iMinSAD[0] = rd; |
827 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
828 |
*dir = Direction; |
*data->dir = Direction; |
829 |
} |
} |
830 |
} |
} |
831 |
|
|
834 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
835 |
|
|
836 |
static void |
static void |
837 |
AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection) |
AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
838 |
{ |
{ |
839 |
|
|
840 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
841 |
|
|
842 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
843 |
|
|
844 |
for(;;) { /* forever */ |
for(;;) { /* forever */ |
845 |
iDirection = 0; |
*iDirection = 0; |
846 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
847 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
848 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
850 |
|
|
851 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
852 |
|
|
853 |
if (iDirection) { /* if anything found */ |
if (*iDirection) { /* if anything found */ |
854 |
bDirection = iDirection; |
bDirection = *iDirection; |
855 |
iDirection = 0; |
*iDirection = 0; |
856 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
857 |
if (bDirection & 3) { /* our candidate is left or right */ |
if (bDirection & 3) { /* our candidate is left or right */ |
858 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
862 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
863 |
} |
} |
864 |
|
|
865 |
if (iDirection) { |
if (*iDirection) { |
866 |
bDirection += iDirection; |
bDirection += *iDirection; |
867 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
868 |
} |
} |
869 |
} else { /* about to quit, eh? not so fast.... */ |
} else { /* about to quit, eh? not so fast.... */ |
911 |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
912 |
break; |
break; |
913 |
} |
} |
914 |
if (!iDirection) break; /* ok, the end. really */ |
if (!*iDirection) break; /* ok, the end. really */ |
915 |
bDirection = iDirection; |
bDirection = *iDirection; |
916 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
917 |
} |
} |
918 |
} |
} |
919 |
} |
} |
920 |
|
|
921 |
static void |
static void |
922 |
SquareSearch(int x, int y, const SearchData * const data, int bDirection) |
SquareSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
923 |
{ |
{ |
924 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
925 |
|
|
926 |
do { |
do { |
927 |
iDirection = 0; |
*iDirection = 0; |
928 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64); |
929 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128); |
930 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32); |
934 |
if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128); |
if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128); |
935 |
if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128); |
if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128); |
936 |
|
|
937 |
bDirection = iDirection; |
bDirection = *iDirection; |
938 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
939 |
} while (iDirection); |
} while (*iDirection); |
940 |
} |
} |
941 |
|
|
942 |
static void |
static void |
943 |
DiamondSearch(int x, int y, const SearchData * const data, int bDirection) |
DiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
944 |
{ |
{ |
945 |
|
|
946 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
947 |
|
|
948 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
949 |
|
|
950 |
do { |
do { |
951 |
iDirection = 0; |
*iDirection = 0; |
952 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
953 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
954 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
956 |
|
|
957 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
958 |
|
|
959 |
if (iDirection) { /* checking if anything found */ |
if (*iDirection) { /* checking if anything found */ |
960 |
bDirection = iDirection; |
bDirection = *iDirection; |
961 |
iDirection = 0; |
*iDirection = 0; |
962 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
963 |
if (bDirection & 3) { /* our candidate is left or right */ |
if (bDirection & 3) { /* our candidate is left or right */ |
964 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
967 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
968 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
969 |
} |
} |
970 |
bDirection += iDirection; |
bDirection += *iDirection; |
971 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
972 |
} |
} |
973 |
} |
} |
974 |
while (iDirection); |
while (*iDirection); |
975 |
} |
} |
976 |
|
|
977 |
/* MAINSEARCH FUNCTIONS END */ |
/* MAINSEARCH FUNCTIONS END */ |
978 |
|
|
979 |
static void |
static void |
980 |
SubpelRefine(const SearchData * const data) |
SubpelRefine_Fast(SearchData * data, CheckFunc * CheckCandidate) |
981 |
|
{ |
982 |
|
/* Do a half-pel or q-pel refinement */ |
983 |
|
VECTOR centerMV; |
984 |
|
VECTOR second_best; |
985 |
|
int best_sad = *data->iMinSAD; |
986 |
|
int xo, yo, xo2, yo2; |
987 |
|
int size = 2; |
988 |
|
CheckFunc *backupFunc = CheckCandidate; |
989 |
|
|
990 |
|
if(data->qpel_precision) |
991 |
|
size = 1; |
992 |
|
|
993 |
|
centerMV = *data->currentMV; |
994 |
|
*data->iMinSAD = 256 * 4096; |
995 |
|
|
996 |
|
CHECK_CANDIDATE(centerMV.x, centerMV.y - size, 0); |
997 |
|
CHECK_CANDIDATE(centerMV.x + size, centerMV.y - size, 0); |
998 |
|
CHECK_CANDIDATE(centerMV.x + size, centerMV.y, 0); |
999 |
|
CHECK_CANDIDATE(centerMV.x + size, centerMV.y + size, 0); |
1000 |
|
|
1001 |
|
CHECK_CANDIDATE(centerMV.x, centerMV.y + size, 0); |
1002 |
|
CHECK_CANDIDATE(centerMV.x - size, centerMV.y + size, 0); |
1003 |
|
CHECK_CANDIDATE(centerMV.x - size, centerMV.y, 0); |
1004 |
|
CHECK_CANDIDATE(centerMV.x - size, centerMV.y - size, 0); |
1005 |
|
|
1006 |
|
second_best = *data->currentMV; |
1007 |
|
|
1008 |
|
if(data->qpel_precision) |
1009 |
|
second_best.x *= 2; second_best.y *= 2; |
1010 |
|
|
1011 |
|
data->currentMV[0] = centerMV; |
1012 |
|
*data->iMinSAD = best_sad; |
1013 |
|
|
1014 |
|
centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
1015 |
|
|
1016 |
|
xo = centerMV.x; |
1017 |
|
yo = centerMV.y; |
1018 |
|
xo2 = second_best.x; |
1019 |
|
yo2 = second_best.y; |
1020 |
|
|
1021 |
|
CheckCandidate = CheckCandidate16_subpel; |
1022 |
|
*data->iMinSAD2 = 256 * 4096; |
1023 |
|
|
1024 |
|
if (yo == yo2) |
1025 |
|
{ |
1026 |
|
CHECK_CANDIDATE((xo+xo2)>>1, yo, 0); |
1027 |
|
CHECK_CANDIDATE(xo, yo-1, 0); |
1028 |
|
CHECK_CANDIDATE(xo, yo+1, 0); |
1029 |
|
|
1030 |
|
if(best_sad <= *data->iMinSAD2) |
1031 |
|
goto ende; |
1032 |
|
|
1033 |
|
if(data->currentQMV[0].x == data->currentQMV2[0].x) { |
1034 |
|
CHECK_CANDIDATE((xo+xo2)>>1, yo-1, 0); |
1035 |
|
CHECK_CANDIDATE((xo+xo2)>>1, yo+1, 0); |
1036 |
|
goto ende; |
1037 |
|
} |
1038 |
|
else { |
1039 |
|
CHECK_CANDIDATE((xo+xo2)>>1, |
1040 |
|
(data->currentQMV[0].x == xo) ? data->currentQMV[0].y : data->currentQMV2[0].y, |
1041 |
|
0); |
1042 |
|
goto ende; |
1043 |
|
} |
1044 |
|
} |
1045 |
|
|
1046 |
|
if (xo == xo2) |
1047 |
|
{ |
1048 |
|
CHECK_CANDIDATE(xo, (yo+yo2)>>1, 0); |
1049 |
|
CHECK_CANDIDATE(xo-1, yo, 0); |
1050 |
|
CHECK_CANDIDATE(xo+1, yo, 0); |
1051 |
|
|
1052 |
|
if(best_sad < *data->iMinSAD2) |
1053 |
|
goto ende; |
1054 |
|
|
1055 |
|
if(data->currentQMV[0].y == data->currentQMV2[0].y) { |
1056 |
|
CHECK_CANDIDATE(xo-1, (yo+yo2)>>1, 0); |
1057 |
|
CHECK_CANDIDATE(xo+1, (yo+yo2)>>1, 0); |
1058 |
|
goto ende; |
1059 |
|
} |
1060 |
|
else { |
1061 |
|
CHECK_CANDIDATE((data->currentQMV[0].y == yo) ? data->currentQMV[0].x : data->currentQMV2[0].x, (yo+yo2)>>1, 0); |
1062 |
|
goto ende; |
1063 |
|
} |
1064 |
|
} |
1065 |
|
|
1066 |
|
CHECK_CANDIDATE(xo, (yo+yo2)>>1, 0); |
1067 |
|
CHECK_CANDIDATE((xo+xo2)>>1, yo, 0); |
1068 |
|
|
1069 |
|
if(best_sad <= *data->iMinSAD2) |
1070 |
|
goto ende; |
1071 |
|
|
1072 |
|
CHECK_CANDIDATE((xo+xo2)>>1, (yo+yo2)>>1, 0); |
1073 |
|
|
1074 |
|
ende: |
1075 |
|
CheckCandidate = backupFunc; |
1076 |
|
} |
1077 |
|
|
1078 |
|
static void |
1079 |
|
SubpelRefine(const SearchData * const data, CheckFunc * const CheckCandidate) |
1080 |
{ |
{ |
1081 |
/* Do a half-pel or q-pel refinement */ |
/* Do a half-pel or q-pel refinement */ |
1082 |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
|
int iDirection; /* only needed because macro expects it */ |
|
1083 |
|
|
1084 |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
1085 |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
1128 |
} |
} |
1129 |
|
|
1130 |
static __inline void |
static __inline void |
1131 |
|
ModeDecision_Fast(SearchData * const Data, |
1132 |
|
MACROBLOCK * const pMB, |
1133 |
|
const MACROBLOCK * const pMBs, |
1134 |
|
const int x, const int y, |
1135 |
|
const MBParam * const pParam, |
1136 |
|
const uint32_t MotionFlags, |
1137 |
|
const uint32_t VopFlags, |
1138 |
|
const uint32_t VolFlags, |
1139 |
|
const IMAGE * const pCurrent, |
1140 |
|
const IMAGE * const pRef, |
1141 |
|
const IMAGE * const vGMC, |
1142 |
|
const int coding_type) |
1143 |
|
{ |
1144 |
|
int mode = MODE_INTER; |
1145 |
|
int mcsel = 0; |
1146 |
|
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
1147 |
|
const uint32_t iQuant = pMB->quant; |
1148 |
|
const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0); |
1149 |
|
int sad; |
1150 |
|
int min_rd = -1, intra_rd, i, cbp, c[2] = {0, 0}; |
1151 |
|
VECTOR backup[5], *v; |
1152 |
|
int sad_backup[5]; |
1153 |
|
int InterBias = MV16_INTER_BIAS; |
1154 |
|
int thresh = 0; |
1155 |
|
int count = 0; |
1156 |
|
int top = 0, top_right = 0, left = 0; |
1157 |
|
|
1158 |
|
pMB->mcsel = 0; |
1159 |
|
|
1160 |
|
/* INTER <-> INTER4V decision */ |
1161 |
|
if ((Data->iMinSAD[0] + 125 < Data->iMinSAD[1] + |
1162 |
|
Data->iMinSAD[2] + Data->iMinSAD[3] + Data->iMinSAD[4])) { /* normal, fast, SAD-based mode decision */ |
1163 |
|
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1164 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
1165 |
|
mode = MODE_INTER; |
1166 |
|
sad = Data->iMinSAD[0]; |
1167 |
|
} else { |
1168 |
|
mode = MODE_INTER4V; |
1169 |
|
sad = Data->iMinSAD[1] + Data->iMinSAD[2] + |
1170 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant; |
1171 |
|
Data->iMinSAD[0] = sad; |
1172 |
|
} |
1173 |
|
|
1174 |
|
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
1175 |
|
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
1176 |
|
if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
1177 |
|
if (Data->chroma || SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
1178 |
|
mode = MODE_NOT_CODED; |
1179 |
|
goto early_out; |
1180 |
|
} |
1181 |
|
|
1182 |
|
/* mcsel */ |
1183 |
|
if (coding_type == S_VOP) { |
1184 |
|
|
1185 |
|
int32_t iSAD = sad16(Data->Cur, |
1186 |
|
vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); |
1187 |
|
|
1188 |
|
if (Data->chroma) { |
1189 |
|
iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1190 |
|
iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1191 |
|
} |
1192 |
|
|
1193 |
|
if (iSAD <= sad) { /* mode decision GMC */ |
1194 |
|
mode = MODE_INTER; |
1195 |
|
mcsel = 1; |
1196 |
|
sad = iSAD; |
1197 |
|
} |
1198 |
|
|
1199 |
|
} |
1200 |
|
} else { /* Rate-Distortion INTER<->INTER4V */ |
1201 |
|
Data->iQuant = iQuant; |
1202 |
|
Data->cbp = c; |
1203 |
|
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1204 |
|
|
1205 |
|
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
1206 |
|
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
1207 |
|
if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
1208 |
|
if (Data->chroma || SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
1209 |
|
mode = MODE_NOT_CODED; |
1210 |
|
goto early_out; |
1211 |
|
} |
1212 |
|
|
1213 |
|
for (i = 0; i < 5; i++) { |
1214 |
|
sad_backup[i] = Data->iMinSAD[i]; |
1215 |
|
Data->iMinSAD[i] = 256*4096; |
1216 |
|
backup[i] = v[i]; |
1217 |
|
} |
1218 |
|
|
1219 |
|
min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags); |
1220 |
|
cbp = *Data->cbp; |
1221 |
|
sad = sad_backup[0]; |
1222 |
|
|
1223 |
|
if (coding_type == S_VOP) { |
1224 |
|
int gmc_rd; |
1225 |
|
int32_t iSAD = sad16(Data->Cur, |
1226 |
|
vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); |
1227 |
|
|
1228 |
|
if (Data->chroma) { |
1229 |
|
iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1230 |
|
iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1231 |
|
} |
1232 |
|
|
1233 |
|
*Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ |
1234 |
|
gmc_rd = findRDgmc(Data, vGMC, x, y); |
1235 |
|
if (gmc_rd < min_rd) { |
1236 |
|
mcsel = 1; |
1237 |
|
*Data->iMinSAD = min_rd = gmc_rd; |
1238 |
|
mode = MODE_INTER; |
1239 |
|
cbp = *Data->cbp; |
1240 |
|
sad = iSAD; |
1241 |
|
} |
1242 |
|
} |
1243 |
|
|
1244 |
|
if (inter4v) { |
1245 |
|
int v4_rd; |
1246 |
|
v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1247 |
|
if (v4_rd < min_rd) { |
1248 |
|
Data->iMinSAD[0] = min_rd = v4_rd; |
1249 |
|
mode = MODE_INTER4V; |
1250 |
|
cbp = *Data->cbp; |
1251 |
|
sad = sad_backup[1] + sad_backup[2] + |
1252 |
|
sad_backup[3] + sad_backup[4] + IMV16X16 * (int32_t)iQuant; |
1253 |
|
} |
1254 |
|
} |
1255 |
|
} |
1256 |
|
|
1257 |
|
left = top = top_right = -1; |
1258 |
|
thresh = 0; |
1259 |
|
|
1260 |
|
if(x > 0 && y > 0 && x < pParam->mb_width) { |
1261 |
|
left = (&pMBs[(x-1) + y * pParam->mb_width])->sad16; // left |
1262 |
|
top = (&pMBs[x + (y-1) * pParam->mb_width])->sad16; // top |
1263 |
|
top_right = (&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16; // top right |
1264 |
|
|
1265 |
|
if(((&pMBs[(x-1) + y * pParam->mb_width])->mode != MODE_INTRA) && |
1266 |
|
((&pMBs[x + (y-1) * pParam->mb_width])->mode != MODE_INTRA) && |
1267 |
|
((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mode != MODE_INTRA)) |
1268 |
|
thresh = MAX(MAX(top, left), top_right); |
1269 |
|
else |
1270 |
|
thresh = MIN(MIN(top, left), top_right); |
1271 |
|
} |
1272 |
|
|
1273 |
|
/* INTRA <-> INTER decision */ |
1274 |
|
if (sad < thresh) { /* normal, fast, SAD-based mode decision */ |
1275 |
|
/* intra decision */ |
1276 |
|
|
1277 |
|
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
1278 |
|
if (y != 0) |
1279 |
|
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
1280 |
|
if (x != 0) |
1281 |
|
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
1282 |
|
|
1283 |
|
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
1284 |
|
if (Data->rrv) InterBias *= 4; |
1285 |
|
|
1286 |
|
if (InterBias < sad) { |
1287 |
|
int32_t deviation; |
1288 |
|
if (!Data->rrv) |
1289 |
|
deviation = dev16(Data->Cur, Data->iEdgedWidth); |
1290 |
|
else |
1291 |
|
deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ |
1292 |
|
dev16(Data->Cur+16, Data->iEdgedWidth) + |
1293 |
|
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
1294 |
|
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
1295 |
|
|
1296 |
|
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
1297 |
|
} |
1298 |
|
|
1299 |
|
pMB->cbp = 63; |
1300 |
|
} else { /* Rate-Distortion INTRA<->INTER */ |
1301 |
|
if(min_rd < 0) { |
1302 |
|
Data->iQuant = iQuant; |
1303 |
|
Data->cbp = c; |
1304 |
|
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1305 |
|
|
1306 |
|
for (i = 0; i < 5; i++) { |
1307 |
|
Data->iMinSAD[i] = 256*4096; |
1308 |
|
backup[i] = v[i]; |
1309 |
|
} |
1310 |
|
|
1311 |
|
if(mode == MODE_INTER) { |
1312 |
|
min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags); |
1313 |
|
cbp = *Data->cbp; |
1314 |
|
|
1315 |
|
if (coding_type == S_VOP) { |
1316 |
|
int gmc_rd; |
1317 |
|
|
1318 |
|
*Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ |
1319 |
|
gmc_rd = findRDgmc(Data, vGMC, x, y); |
1320 |
|
if (gmc_rd < min_rd) { |
1321 |
|
mcsel = 1; |
1322 |
|
*Data->iMinSAD = min_rd = gmc_rd; |
1323 |
|
mode = MODE_INTER; |
1324 |
|
cbp = *Data->cbp; |
1325 |
|
} |
1326 |
|
} |
1327 |
|
} |
1328 |
|
|
1329 |
|
if(mode == MODE_INTER4V) { |
1330 |
|
int v4_rd; |
1331 |
|
v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1332 |
|
if (v4_rd < min_rd) { |
1333 |
|
Data->iMinSAD[0] = min_rd = v4_rd; |
1334 |
|
mode = MODE_INTER4V; |
1335 |
|
cbp = *Data->cbp; |
1336 |
|
} |
1337 |
|
} |
1338 |
|
} |
1339 |
|
|
1340 |
|
intra_rd = findRDintra(Data); |
1341 |
|
if (intra_rd < min_rd) { |
1342 |
|
*Data->iMinSAD = min_rd = intra_rd; |
1343 |
|
mode = MODE_INTRA; |
1344 |
|
} |
1345 |
|
|
1346 |
|
pMB->cbp = cbp; |
1347 |
|
} |
1348 |
|
|
1349 |
|
early_out: |
1350 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
1351 |
|
|
1352 |
|
if (Data->rrv) { |
1353 |
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
1354 |
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1355 |
|
} |
1356 |
|
|
1357 |
|
if (mode == MODE_INTER && mcsel == 0) { |
1358 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1359 |
|
|
1360 |
|
if(Data->qpel) { |
1361 |
|
pMB->qmvs[0] = pMB->qmvs[1] |
1362 |
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1363 |
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
1364 |
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
1365 |
|
} else { |
1366 |
|
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1367 |
|
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1368 |
|
} |
1369 |
|
|
1370 |
|
} else if (mode == MODE_INTER ) { // but mcsel == 1 |
1371 |
|
|
1372 |
|
pMB->mcsel = 1; |
1373 |
|
if (Data->qpel) { |
1374 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; |
1375 |
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; |
1376 |
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; |
1377 |
|
} else |
1378 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; |
1379 |
|
|
1380 |
|
} else |
1381 |
|
if (mode == MODE_INTER4V) ; /* anything here? */ |
1382 |
|
else /* INTRA, NOT_CODED */ |
1383 |
|
ZeroMacroblockP(pMB, 0); |
1384 |
|
|
1385 |
|
pMB->mode = mode; |
1386 |
|
} |
1387 |
|
|
1388 |
|
static __inline void |
1389 |
ModeDecision(SearchData * const Data, |
ModeDecision(SearchData * const Data, |
1390 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1391 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1578 |
uint32_t mb_height = pParam->mb_height; |
uint32_t mb_height = pParam->mb_height; |
1579 |
const uint32_t iEdgedWidth = pParam->edged_width; |
const uint32_t iEdgedWidth = pParam->edged_width; |
1580 |
const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags); |
const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags); |
1581 |
|
int stat_thresh = 0; |
1582 |
|
|
1583 |
uint32_t x, y; |
uint32_t x, y; |
1584 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
1588 |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
1589 |
|
|
1590 |
/* some pre-initialized thingies for SearchP */ |
/* some pre-initialized thingies for SearchP */ |
1591 |
int32_t temp[8]; |
int32_t temp[8]; uint32_t dir; |
1592 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
1593 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
1594 |
|
VECTOR currentMV2[5]; |
1595 |
|
VECTOR currentQMV2[5]; |
1596 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
1597 |
|
int32_t iMinSAD2[5]; |
1598 |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
1599 |
SearchData Data; |
SearchData Data; |
1600 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
1601 |
Data.iEdgedWidth = iEdgedWidth; |
Data.iEdgedWidth = iEdgedWidth; |
1602 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
1603 |
Data.currentQMV = currentQMV; |
Data.currentQMV = currentQMV; |
1604 |
|
Data.currentMV2 = currentMV2; |
1605 |
|
Data.currentQMV2 = currentQMV2; |
1606 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
1607 |
|
Data.iMinSAD2 = iMinSAD2; |
1608 |
Data.temp = temp; |
Data.temp = temp; |
1609 |
|
Data.dir = &dir; |
1610 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
1611 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
1612 |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
1627 |
for (y = 0; y < mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
1628 |
for (x = 0; x < mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
1629 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
1630 |
|
MACROBLOCK *prevMB = &reference->mbs[x + y * pParam->mb_width]; |
1631 |
|
|
1632 |
if (!Data.rrv) pMB->sad16 = |
if (!Data.rrv) pMB->sad16 = |
1633 |
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
1660 |
} |
} |
1661 |
} |
} |
1662 |
|
|
1663 |
|
if(MotionFlags & XVID_ME_DETECT_STATIC_MOTION) { |
1664 |
|
if(x > 0 && y > 0 && x < pParam->mb_width) |
1665 |
|
if(MVequal((&pMBs[(x-1) + y * pParam->mb_width])->mvs[0], zeroMV) && |
1666 |
|
MVequal((&pMBs[x + (y-1) * pParam->mb_width])->mvs[0], zeroMV) && |
1667 |
|
MVequal((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mvs[0], zeroMV) && |
1668 |
|
MVequal(prevMB->mvs[0], zeroMV)) { |
1669 |
|
stat_thresh = MAX((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
1670 |
|
MAX((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
1671 |
|
MAX((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
1672 |
|
prevMB->sad16))); |
1673 |
|
} |
1674 |
|
else |
1675 |
|
stat_thresh = MIN((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
1676 |
|
MIN((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
1677 |
|
MIN((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
1678 |
|
prevMB->sad16))); |
1679 |
|
} |
1680 |
|
|
1681 |
if ((current->vop_flags & XVID_VOP_CARTOON) && |
if ((current->vop_flags & XVID_VOP_CARTOON) && |
1682 |
(sad00 < pMB->quant * 4 * skip_thresh)) { /* favorize (0,0) vector for cartoons */ |
(sad00 < pMB->quant * 4 * skip_thresh) || (sad00 < stat_thresh)) { /* favorize (0,0) vector for cartoons */ |
1683 |
ZeroMacroblockP(pMB, sad00); |
ZeroMacroblockP(pMB, sad00); |
1684 |
continue; |
continue; |
1685 |
} |
} |
1688 |
y, MotionFlags, current->vop_flags, current->vol_flags, |
y, MotionFlags, current->vop_flags, current->vol_flags, |
1689 |
&Data, pParam, pMBs, reference->mbs, pMB); |
&Data, pParam, pMBs, reference->mbs, pMB); |
1690 |
|
|
1691 |
|
if(current->vop_flags & XVID_VOP_FAST_MODEDECISION_RD) { |
1692 |
|
ModeDecision_Fast(&Data, pMB, pMBs, x, y, pParam, |
1693 |
|
MotionFlags, current->vop_flags, current->vol_flags, |
1694 |
|
pCurrent, pRef, pGMC, current->coding_type); |
1695 |
|
} |
1696 |
|
else { |
1697 |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
1698 |
MotionFlags, current->vop_flags, current->vol_flags, |
MotionFlags, current->vop_flags, current->vol_flags, |
1699 |
pCurrent, pRef, pGMC, current->coding_type); |
pCurrent, pRef, pGMC, current->coding_type); |
1700 |
|
} |
1701 |
|
|
1702 |
if (pMB->mode == MODE_INTRA) |
if (pMB->mode == MODE_INTRA) |
1703 |
if (++iIntra > iLimit) return 1; |
if (++iIntra > iLimit) return 1; |
1707 |
return 0; |
return 0; |
1708 |
} |
} |
1709 |
|
|
1710 |
|
/* check if given vector is equal to any vector checked before */ |
1711 |
|
static __inline int |
1712 |
|
vector_repeats(const VECTOR * const pmv, const int i) |
1713 |
|
{ |
1714 |
|
unsigned int j; |
1715 |
|
for (j = 0; j < i; j++) |
1716 |
|
if (MVequal(pmv[i], pmv[j])) return 1; /* same vector has been checked already */ |
1717 |
|
return 0; |
1718 |
|
} |
1719 |
|
|
1720 |
|
/* make a binary mask that prevents diamonds/squares |
1721 |
|
from checking a vector which has been checked as a prediction */ |
1722 |
static __inline int |
static __inline int |
1723 |
make_mask(const VECTOR * const pmv, const int i) |
make_mask(const VECTOR * const pmv, const int i, const int current) |
1724 |
{ |
{ |
1725 |
int mask = 255, j; |
unsigned int mask = 255, j; |
1726 |
for (j = 0; j < i; j++) { |
for (j = 0; j < i; j++) { |
1727 |
if (MVequal(pmv[i], pmv[j])) return 0; /* same vector has been checked already */ |
if (pmv[current].x == pmv[j].x) { |
1728 |
if (pmv[i].x == pmv[j].x) { |
if (pmv[current].y == pmv[j].y + iDiamondSize) mask &= ~4; |
1729 |
if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4; |
else if (pmv[current].y == pmv[j].y - iDiamondSize) mask &= ~8; |
|
else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8; |
|
1730 |
} else |
} else |
1731 |
if (pmv[i].y == pmv[j].y) { |
if (pmv[current].y == pmv[j].y) { |
1732 |
if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1; |
if (pmv[current].x == pmv[j].x + iDiamondSize) mask &= ~1; |
1733 |
else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2; |
else if (pmv[current].x == pmv[j].x - iDiamondSize) mask &= ~2; |
1734 |
} |
} |
1735 |
} |
} |
1736 |
return mask; |
return mask; |
1794 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB) |
1795 |
{ |
{ |
1796 |
|
|
1797 |
int i, iDirection = 255, mask, threshA; |
int i, threshA; |
1798 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1799 |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
1800 |
|
CheckFunc * CheckCandidate; |
1801 |
|
|
1802 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
1803 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
1820 |
Data->lambda16 = lambda_vec16[pMB->quant]; |
Data->lambda16 = lambda_vec16[pMB->quant]; |
1821 |
Data->lambda8 = lambda_vec8[pMB->quant]; |
Data->lambda8 = lambda_vec8[pMB->quant]; |
1822 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1823 |
|
*Data->dir = 0; |
1824 |
|
|
1825 |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
1826 |
|
|
1845 |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
1846 |
|
|
1847 |
if (!Data->rrv) { |
if (!Data->rrv) { |
1848 |
if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
1849 |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
1850 |
} else CheckCandidate = CheckCandidate32; |
} else CheckCandidate = CheckCandidate32; |
1851 |
|
|
1852 |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
1853 |
|
|
1854 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) |
1855 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!vector_repeats(pmv, i)) { |
1856 |
CheckCandidate(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate(pmv[i].x, pmv[i].y, Data, i); |
1857 |
if (Data->iMinSAD[0] <= threshA) break; |
if (Data->iMinSAD[0] <= threshA) { i++; break; } |
1858 |
} |
} |
1859 |
|
|
1860 |
if ((Data->iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
1864 |
else { |
else { |
1865 |
|
|
1866 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1867 |
|
int mask = make_mask(pmv, i, *Data->dir); // all vectors pmv[0..i-1] have been checked |
1868 |
|
|
1869 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
1870 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1871 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1872 |
|
|
1873 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate); |
1874 |
|
|
1875 |
/* extended search, diamond starting in 0,0 and in prediction. |
/* extended search, diamond starting in 0,0 and in prediction. |
1876 |
note that this search is/might be done in halfpel positions, |
note that this search is/might be done in halfpel positions, |
1886 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1887 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1888 |
|
|
1889 |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, Data, 255); |
1890 |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate); |
1891 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1892 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1893 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
1898 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1899 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1900 |
|
|
1901 |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, Data, 255); |
1902 |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate); |
1903 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1904 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1905 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; |
1906 |
|
} |
1907 |
} |
} |
1908 |
} |
} |
1909 |
} |
} |
1910 |
|
|
1911 |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
1912 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
1913 |
|
|
1914 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1915 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */ |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */ |
1921 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
1922 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1923 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
1924 |
SubpelRefine(Data); |
if(MotionFlags & XVID_ME_FASTREFINE16) |
1925 |
|
SubpelRefine_Fast(Data, CheckCandidate); |
1926 |
|
else |
1927 |
|
SubpelRefine(Data, CheckCandidate); |
1928 |
} |
} |
1929 |
|
|
1930 |
if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30) |
if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30) |
1971 |
SearchData * const Data) |
SearchData * const Data) |
1972 |
{ |
{ |
1973 |
int i = 0; |
int i = 0; |
1974 |
|
CheckFunc * CheckCandidate; |
1975 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1976 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1977 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
2014 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
2015 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
2016 |
|
|
2017 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate); |
2018 |
|
|
2019 |
if(*(Data->iMinSAD) < temp_sad) { |
if(*(Data->iMinSAD) < temp_sad) { |
2020 |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
2025 |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
2026 |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
2027 |
|
|
2028 |
SubpelRefine(Data); /* perform halfpel refine of current best vector */ |
SubpelRefine(Data, CheckCandidate); /* perform halfpel refine of current best vector */ |
2029 |
|
|
2030 |
if(*(Data->iMinSAD) < temp_sad) { /* we have found a better match */ |
if(*(Data->iMinSAD) < temp_sad) { /* we have found a better match */ |
2031 |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
2037 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2038 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
2039 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
2040 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
2041 |
} |
} |
2042 |
} |
} |
2043 |
|
|
2123 |
SearchData * const Data) |
SearchData * const Data) |
2124 |
{ |
{ |
2125 |
|
|
2126 |
int i, iDirection = 255, mask; |
int i; |
2127 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
2128 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
2129 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
2149 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
2150 |
|
|
2151 |
Data->currentMV->x = Data->currentMV->y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
|
CheckCandidate = CheckCandidate16no4v; |
|
2152 |
|
|
2153 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
2154 |
for (i = 0; i < 7; i++) { |
for (i = 0; i < 7; i++) |
2155 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!vector_repeats(pmv, i) ) |
2156 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
|
} |
|
2157 |
|
|
2158 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
2159 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
2160 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
2161 |
|
|
2162 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
if (*Data->iMinSAD > 512) { |
2163 |
|
unsigned int mask = make_mask(pmv, 7, *Data->dir); |
2164 |
|
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
2165 |
|
} |
2166 |
|
|
2167 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate16no4v); |
2168 |
|
|
2169 |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
2170 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
2172 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2173 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2174 |
pParam->width, pParam->height, iFcode, 2, 0); |
pParam->width, pParam->height, iFcode, 2, 0); |
2175 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate16no4v); |
2176 |
} |
} |
2177 |
|
|
2178 |
/* three bits are needed to code backward mode. four for forward */ |
/* three bits are needed to code backward mode. four for forward */ |
2212 |
{ |
{ |
2213 |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
2214 |
int32_t sum; |
int32_t sum; |
|
const int div = 1 + Data->qpel; |
|
2215 |
int k; |
int k; |
2216 |
const uint32_t stride = Data->iEdgedWidth/2; |
const uint32_t stride = Data->iEdgedWidth/2; |
2217 |
/* 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 */ |
2218 |
|
|
2219 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
2220 |
dy += Data->directmvF[k].y / div; |
dy += Data->directmvF[k].y >> Data->qpel; |
2221 |
dx += Data->directmvF[k].x / div; |
dx += Data->directmvF[k].x >> Data->qpel; |
2222 |
b_dy += Data->directmvB[k].y / div; |
b_dy += Data->directmvB[k].y >> Data->qpel; |
2223 |
b_dx += Data->directmvB[k].x / div; |
b_dx += Data->directmvB[k].x >> Data->qpel; |
2224 |
} |
} |
2225 |
|
|
2226 |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
2233 |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
2234 |
stride); |
stride); |
2235 |
|
|
2236 |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) return; /* no skip */ |
2237 |
|
|
2238 |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
2239 |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
2240 |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
2241 |
stride); |
stride); |
2242 |
|
|
2243 |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) { |
2244 |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
2245 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
2246 |
pMB->qmvs[k] = pMB->mvs[k]; |
pMB->qmvs[k] = pMB->mvs[k]; |
2272 |
int32_t skip_sad; |
int32_t skip_sad; |
2273 |
int k = (x + Data->iEdgedWidth*y) * 16; |
int k = (x + Data->iEdgedWidth*y) * 16; |
2274 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
2275 |
|
CheckFunc * CheckCandidate; |
2276 |
|
|
2277 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
2278 |
Data->RefP[0] = f_Ref->y + k; |
Data->RefP[0] = f_Ref->y + k; |
2322 |
|
|
2323 |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
2324 |
|
|
2325 |
CheckCandidate(0, 0, 255, &k, Data); |
CheckCandidate(0, 0, Data, 255); |
2326 |
|
|
2327 |
/* initial (fast) skip decision */ |
/* initial (fast) skip decision */ |
2328 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
if (*Data->iMinSAD < Data->iQuant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
2329 |
/* possible skip */ |
/* possible skip */ |
2330 |
if (Data->chroma) { |
if (Data->chroma) { |
2331 |
pMB->mode = MODE_DIRECT_NONE_MV; |
pMB->mode = MODE_DIRECT_NONE_MV; |
2348 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
2349 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
2350 |
|
|
2351 |
MainSearchPtr(0, 0, Data, 255); |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
2352 |
|
|
2353 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
2354 |
|
|
2355 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
2356 |
|
|
2409 |
|
|
2410 |
{ |
{ |
2411 |
|
|
2412 |
int iDirection, i, j; |
int i, j; |
2413 |
SearchData bData; |
SearchData bData; |
2414 |
|
|
2415 |
fData->qpel_precision = 0; |
fData->qpel_precision = 0; |
2432 |
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
2433 |
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
2434 |
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
2435 |
|
bData.dir = fData->dir; |
2436 |
|
|
2437 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
2438 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
2451 |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
2452 |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
2453 |
|
|
2454 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255); |
2455 |
|
|
2456 |
/* diamond */ |
/* diamond */ |
2457 |
do { |
do { |
2458 |
iDirection = 255; |
*fData->dir = 255; |
2459 |
/* forward MV moves */ |
/* forward MV moves */ |
2460 |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
2461 |
|
|
2462 |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i + 1, j, fData, 0); |
2463 |
CheckCandidateInt(i, j + 1, 0, &iDirection, fData); |
CheckCandidateInt(i, j + 1, fData, 0); |
2464 |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i - 1, j, fData, 0); |
2465 |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
CheckCandidateInt(i, j - 1, fData, 0); |
2466 |
|
|
2467 |
/* backward MV moves */ |
/* backward MV moves */ |
2468 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
2469 |
fData->currentMV[2] = fData->currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
2470 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, &bData, 0); |
2471 |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j + 1, &bData, 0); |
2472 |
CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i - 1, j, &bData, 0); |
2473 |
CheckCandidateInt(i, j - 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j - 1, &bData, 0); |
2474 |
|
|
2475 |
} while (!(iDirection)); |
} while (!(*fData->dir)); |
2476 |
|
|
2477 |
/* qpel refinement */ |
/* qpel refinement */ |
2478 |
if (fData->qpel) { |
if (fData->qpel) { |
2479 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*fData->iMinSAD > *best_sad + 500) return; |
|
CheckCandidate = CheckCandidateInt; |
|
2480 |
fData->qpel_precision = bData.qpel_precision = 1; |
fData->qpel_precision = bData.qpel_precision = 1; |
2481 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); |
2482 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
2484 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
2485 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
2486 |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
2487 |
SubpelRefine(fData); |
SubpelRefine(fData, CheckCandidateInt); |
2488 |
if (*fData->iMinSAD > *best_sad + 300) return; |
if (*fData->iMinSAD > *best_sad + 300) return; |
2489 |
fData->currentQMV[2] = fData->currentQMV[0]; |
fData->currentQMV[2] = fData->currentQMV[0]; |
2490 |
SubpelRefine(&bData); |
SubpelRefine(&bData, CheckCandidateInt); |
2491 |
} |
} |
2492 |
|
|
2493 |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
2546 |
|
|
2547 |
SearchData Data; |
SearchData Data; |
2548 |
int32_t iMinSAD; |
int32_t iMinSAD; |
2549 |
|
uint32_t dir; |
2550 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
2551 |
VECTOR currentQMV[3]; |
VECTOR currentQMV[3]; |
2552 |
int32_t temp[8]; |
int32_t temp[8]; |
2554 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2555 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
2556 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2557 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[MAX(frame->quant-2, 2)]; |
2558 |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
2559 |
Data.rounding = 0; |
Data.rounding = 0; |
2560 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
2561 |
Data.temp = temp; |
Data.temp = temp; |
2562 |
|
Data.dir = &dir; |
2563 |
|
Data.iQuant = frame->quant; |
2564 |
|
|
2565 |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
2566 |
|
|
2583 |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
2584 |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
2585 |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
|
pMB->quant = frame->quant; |
|
2586 |
|
|
2587 |
/* direct search comes first, because it (1) checks for SKIP-mode |
/* direct search comes first, because it (1) checks for SKIP-mode |
2588 |
and (2) sets very good predictions for forward and backward search */ |
and (2) sets very good predictions for forward and backward search */ |
2628 |
&Data); |
&Data); |
2629 |
|
|
2630 |
/* final skip decision */ |
/* final skip decision */ |
2631 |
if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP * 2) |
if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * 2) |
2632 |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
2633 |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
2634 |
|
|
2666 |
SearchData * const Data) |
SearchData * const Data) |
2667 |
{ |
{ |
2668 |
|
|
2669 |
int i, mask; |
int i; |
|
int quarterpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0; |
|
2670 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
2671 |
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
2672 |
|
|
2673 |
|
unsigned int simplicity = 0; |
2674 |
|
|
2675 |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
2676 |
|
|
2677 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2678 |
|
pParam->width, pParam->height, Data->iFcode - Data->qpel - 1, 0, 0); |
2679 |
|
|
2680 |
|
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2681 |
|
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
2682 |
|
|
2683 |
|
pmv[0].x = pMB->mvs[0].x; |
2684 |
|
pmv[0].y = pMB->mvs[0].y; |
2685 |
|
|
2686 |
|
CheckCandidate32I(pmv[0].x, pmv[0].y, Data, 0); |
2687 |
|
|
2688 |
|
if (*Data->iMinSAD > 200) { |
2689 |
|
|
2690 |
|
pmv[1].x = pmv[1].y = 0; |
2691 |
|
|
2692 |
/* median is only used as prediction. it doesn't have to be real */ |
/* median is only used as prediction. it doesn't have to be real */ |
2693 |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
2694 |
else |
else |
2698 |
Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */ |
Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */ |
2699 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */ |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */ |
2700 |
|
|
2701 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
pmv[2].x = Data->predMV.x; |
2702 |
pParam->width, pParam->height, Data->iFcode - quarterpel, 1, 0); |
pmv[2].y = Data->predMV.y; |
|
|
|
|
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
|
|
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
|
|
|
|
|
pmv[1].x = EVEN(pMB->mvs[0].x); |
|
|
pmv[1].y = EVEN(pMB->mvs[0].y); |
|
|
pmv[2].x = EVEN(Data->predMV.x); |
|
|
pmv[2].y = EVEN(Data->predMV.y); |
|
|
pmv[0].x = pmv[0].y = 0; |
|
|
|
|
|
CheckCandidate32I(0, 0, 255, &i, Data); |
|
|
|
|
|
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
|
|
|
|
|
if (!(mask = make_mask(pmv, 1))) |
|
|
CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data); |
|
|
if (!(mask = make_mask(pmv, 2))) |
|
|
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
|
2703 |
|
|
2704 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) /* diamond only if needed */ |
if (!vector_repeats(pmv, 1)) |
2705 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
CheckCandidate32I(pmv[1].x, pmv[1].y, Data, 1); |
2706 |
} |
if (!vector_repeats(pmv, 2)) |
2707 |
|
CheckCandidate32I(pmv[2].x, pmv[2].y, Data, 2); |
2708 |
|
|
2709 |
|
if (*Data->iMinSAD > 500) { /* diamond only if needed */ |
2710 |
|
unsigned int mask = make_mask(pmv, 3, *Data->dir); |
2711 |
|
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate32I); |
2712 |
|
} else simplicity++; |
2713 |
|
|
2714 |
|
if (*Data->iMinSAD > 500) /* refinement from 2-pixel to 1-pixel */ |
2715 |
|
SubpelRefine(Data, CheckCandidate32I); |
2716 |
|
else simplicity++; |
2717 |
|
} else simplicity++; |
2718 |
|
|
2719 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
2720 |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width]; |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width]; |
2721 |
MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i]; |
MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i]; |
2722 |
MB->mode = MODE_INTER; |
MB->mode = MODE_INTER; |
2723 |
MB->sad16 = Data->iMinSAD[i+1]; |
/* if we skipped some search steps, we have to assume that SAD would be lower with them */ |
2724 |
|
MB->sad16 = Data->iMinSAD[i+1] - (simplicity<<7); |
2725 |
} |
} |
2726 |
} |
} |
2727 |
|
|
2728 |
#define INTRA_THRESH 2200 |
#define INTRA_THRESH 2200 |
2729 |
#define INTER_THRESH 50 |
#define INTER_THRESH 40 |
2730 |
#define INTRA_THRESH2 95 |
#define INTRA_THRESH2 95 |
2731 |
|
|
2732 |
int |
int |
2747 |
int complexity = 0; |
int complexity = 0; |
2748 |
|
|
2749 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
2750 |
|
uint32_t dir; |
2751 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
2752 |
SearchData Data; |
SearchData Data; |
2753 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2755 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
2756 |
Data.iFcode = Current->fcode; |
Data.iFcode = Current->fcode; |
2757 |
Data.temp = temp; |
Data.temp = temp; |
2758 |
CheckCandidate = CheckCandidate32I; |
Data.dir = &dir; |
2759 |
|
Data.qpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0; |
2760 |
|
Data.qpel_precision = 0; |
2761 |
|
|
2762 |
if (intraCount != 0) { |
if (intraCount != 0) { |
2763 |
if (intraCount < 10) // we're right after an I frame |
if (intraCount < 10) // we're right after an I frame |
2767 |
IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra; |
IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra; |
2768 |
} |
} |
2769 |
|
|
2770 |
InterThresh -= 12 * bCount; |
InterThresh -= 20 * bCount; |
2771 |
if (InterThresh < 15 + b_thresh) InterThresh = 15 + b_thresh; |
if (InterThresh < 10 + b_thresh) InterThresh = 10 + b_thresh; |
2772 |
|
|
2773 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2774 |
|
|
2800 |
} |
} |
2801 |
|
|
2802 |
if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) |
if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) |
2803 |
if (dev > 500 && pMB->sad16 < 1000) |
if (dev > 1000 && pMB->sad16 < 1000) |
2804 |
sSAD += 1000; |
sSAD += 1000; |
2805 |
|
|
2806 |
sSAD += (dev < 3000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */ |
sSAD += (dev < 4000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */ |
2807 |
} |
} |
2808 |
} |
} |
2809 |
} |
} |
2826 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2827 |
const uint32_t MotionFlags) |
const uint32_t MotionFlags) |
2828 |
{ |
{ |
2829 |
int i, iDirection; |
int i; |
2830 |
int32_t bsad[5]; |
int32_t bsad[5]; |
2831 |
|
|
|
CheckCandidate = CheckCandidateRD16; |
|
|
|
|
2832 |
if (Data->qpel) { |
if (Data->qpel) { |
2833 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
2834 |
Data->currentMV[i].x = Data->currentQMV[i].x/2; |
Data->currentMV[i].x = Data->currentQMV[i].x/2; |
2835 |
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
2836 |
} |
} |
2837 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2838 |
CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255); |
2839 |
|
|
2840 |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
2841 |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
2843 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
2844 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
2845 |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
2846 |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
2847 |
} |
} |
2848 |
|
|
2849 |
} else { /* not qpel */ |
} else { /* not qpel */ |
2850 |
|
|
2851 |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
2852 |
} |
} |
2853 |
|
|
2854 |
if (MotionFlags&XVID_ME_EXTSEARCH_RD) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
if (MotionFlags&XVID_ME_EXTSEARCH_RD) |
2855 |
|
SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16); |
2856 |
|
|
2857 |
if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) |
2858 |
|
SubpelRefine(Data, CheckCandidateRD16); |
2859 |
|
|
2860 |
if (Data->qpel) { |
if (Data->qpel) { |
2861 |
if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ |
if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ |
2869 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2870 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
2871 |
} |
} |
2872 |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) |
2873 |
|
SubpelRefine(Data, CheckCandidateRD16); |
2874 |
} |
} |
2875 |
|
|
2876 |
if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ |
if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ |
2877 |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
2878 |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
2879 |
CheckCandidateRD16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255); |
2880 |
} |
} |
2881 |
return Data->iMinSAD[0]; |
return Data->iMinSAD[0]; |
2882 |
} |
} |
2889 |
const VECTOR * const backup) |
const VECTOR * const backup) |
2890 |
{ |
{ |
2891 |
|
|
2892 |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
int cbp = 0, bits = 0, t = 0, i; |
2893 |
SearchData Data2, *Data8 = &Data2; |
SearchData Data2, *Data8 = &Data2; |
2894 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
2895 |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
2896 |
uint8_t * ptr; |
uint8_t * ptr; |
2897 |
|
|
2898 |
memcpy(Data8, Data, sizeof(SearchData)); |
memcpy(Data8, Data, sizeof(SearchData)); |
|
CheckCandidate = CheckCandidateRD8; |
|
2899 |
|
|
2900 |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
2901 |
|
|
2929 |
{ |
{ |
2930 |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
2931 |
if (!MVequal (*v, backup[i+1]) ) |
if (!MVequal (*v, backup[i+1]) ) |
2932 |
CheckCandidateRD8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255); |
2933 |
} |
} |
2934 |
|
|
2935 |
if (Data8->qpel) { |
if (Data8->qpel) { |
2942 |
pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); |
pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); |
2943 |
|
|
2944 |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
2945 |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8); |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255); |
2946 |
|
|
2947 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) |
2948 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); |
2949 |
|
|
2950 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2951 |
SubpelRefine(Data8); |
SubpelRefine(Data8, CheckCandidateRD8); |
2952 |
|
|
2953 |
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
2954 |
Data8->currentQMV->x = 2*Data8->currentMV->x; |
Data8->currentQMV->x = 2*Data8->currentMV->x; |
2960 |
pParam->width, pParam->height, Data8->iFcode, 2, 0); |
pParam->width, pParam->height, Data8->iFcode, 2, 0); |
2961 |
|
|
2962 |
} |
} |
2963 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) SubpelRefine(Data8); |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) |
2964 |
|
SubpelRefine(Data8, CheckCandidateRD8); |
2965 |
|
|
2966 |
} else { /* not qpel */ |
} else { /* not qpel */ |
2967 |
|
|
2968 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */ |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */ |
2969 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); |
2970 |
|
|
2971 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2972 |
SubpelRefine(Data8); /* halfpel refinement */ |
SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */ |
2973 |
} |
} |
2974 |
|
|
2975 |
/* checking vector equal to predicion */ |
/* checking vector equal to predicion */ |
2976 |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) { |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) { |
2977 |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
2978 |
if (!MVequal(*v, Data8->predMV)) |
if (!MVequal(*v, Data8->predMV)) |
2979 |
CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255); |
2980 |
} |
} |
2981 |
|
|
2982 |
bits += *Data8->iMinSAD; |
bits += *Data8->iMinSAD; |
3123 |
Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width); |
Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width); |
3124 |
|
|
3125 |
Data->currentMV[0].x = Data->currentMV[0].y = 0; |
Data->currentMV[0].x = Data->currentMV[0].y = 0; |
3126 |
CheckCandidate16I(0, 0, 255, &i, Data); |
CheckCandidate16I(0, 0, Data, 255); |
3127 |
|
|
3128 |
if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) ) |
if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) ) |
3129 |
CheckCandidate16I(Data->predMV.x, Data->predMV.y, 255, &i, Data); |
CheckCandidate16I(Data->predMV.x, Data->predMV.y, Data, 255); |
3130 |
|
|
3131 |
AdvDiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
DiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255, CheckCandidate16I); |
3132 |
|
|
3133 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate16I); |
3134 |
|
|
3135 |
|
|
3136 |
/* for QPel halfpel positions are worse than in halfpel mode :( */ |
/* for QPel halfpel positions are worse than in halfpel mode :( */ |
3166 |
|
|
3167 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
3168 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
3169 |
|
uint32_t dir; |
3170 |
SearchData Data; |
SearchData Data; |
3171 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
3172 |
|
|
3177 |
Data.iMinSAD = &iMinSAD[0]; |
Data.iMinSAD = &iMinSAD[0]; |
3178 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
3179 |
Data.temp = temp; |
Data.temp = temp; |
3180 |
|
Data.dir = &dir; |
|
CheckCandidate = CheckCandidate16I; |
|
3181 |
|
|
3182 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
3183 |
|
|
3184 |
for (y = 0; y < pParam->mb_height; y ++) { |
for (y = 0; y < pParam->mb_height; y ++) { |
3185 |
for (x = 0; x < pParam->mb_width; x ++) { |
for (x = 0; x < pParam->mb_width; x ++) { |
|
|
|
3186 |
GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data); |
GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data); |
3187 |
} |
} |
3188 |
} |
} |