1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* motion estimation |
* - Motion Estimation module - |
5 |
|
* |
6 |
|
* Copyright(C) 2002 Christoph Lampert <gruel@web.de> |
7 |
|
* Copyright(C) 2002 Michael Militzer <michael@xvid.org> |
8 |
|
* Copyright(C) 2002 Edouard Gomez <ed.gomez@wanadoo.fr> |
9 |
|
* Copyright(C) 2002 chenm001 <chenm001@163.com> |
10 |
* |
* |
11 |
* This program is an implementation of a part of one or more MPEG-4 |
* This program is an implementation of a part of one or more MPEG-4 |
12 |
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
29 |
* |
* |
30 |
* You should have received a copy of the GNU General Public License |
* You should have received a copy of the GNU General Public License |
31 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
32 |
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
33 |
* |
* |
34 |
*************************************************************************/ |
*************************************************************************/ |
35 |
|
|
|
/************************************************************************** |
|
|
* |
|
|
* Modifications: |
|
|
* |
|
|
* 01.05.2002 updated MotionEstimationBVOP |
|
|
* 25.04.2002 partial prevMB conversion |
|
|
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
|
|
* 14.04.2002 added MotionEstimationBVOP() |
|
|
* 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
|
|
* EPZS and EPZS^2 |
|
|
* 08.02.2002 split up PMVfast into three routines: PMVFast, PMVFast_MainLoop |
|
|
* PMVFast_Refine to support multiple searches with different start points |
|
|
* 07.01.2002 uv-block-based interpolation |
|
|
* 06.01.2002 INTER/INTRA-decision is now done before any SEARCH8 (speedup) |
|
|
* changed INTER_BIAS to 150 (as suggested by suxen_drol) |
|
|
* removed halfpel refinement step in PMVfastSearch8 + quality=5 |
|
|
* added new quality mode = 6 which performs halfpel refinement |
|
|
* filesize difference between quality 5 and 6 is smaller than 1% |
|
|
* (Isibaar) |
|
|
* 31.12.2001 PMVfastSearch16 and PMVfastSearch8 (gruel) |
|
|
* 30.12.2001 get_range/MotionSearchX simplified; blue/green bug fix |
|
|
* 22.12.2001 commented best_point==99 check |
|
|
* 19.12.2001 modified get_range (purple bug fix) |
|
|
* 15.12.2001 moved pmv displacement from mbprediction |
|
|
* 02.12.2001 motion estimation/compensation split (Isibaar) |
|
|
* 16.11.2001 rewrote/tweaked search algorithms; pross@cs.rmit.edu.au |
|
|
* 10.11.2001 support for sad16/sad8 functions |
|
|
* 28.08.2001 reactivated MODE_INTER4V for EXT_MODE |
|
|
* 24.08.2001 removed MODE_INTER4V_Q, disabled MODE_INTER4V for EXT_MODE |
|
|
* 22.08.2001 added MODE_INTER4V_Q |
|
|
* 20.08.2001 added pragma to get rid of internal compiler error with VC6 |
|
|
* idea by Cyril. Thanks. |
|
|
* |
|
|
* Michael Militzer <isibaar@videocoding.de> |
|
|
* |
|
|
**************************************************************************/ |
|
|
|
|
36 |
#include <assert.h> |
#include <assert.h> |
37 |
#include <stdio.h> |
#include <stdio.h> |
38 |
#include <stdlib.h> |
#include <stdlib.h> |
154 |
|
|
155 |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
156 |
|
|
157 |
|
if (pMB->mode == MODE_NOT_CODED) |
158 |
|
continue; |
159 |
|
|
160 |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
161 |
|
|
162 |
pMB->sad16 = |
pMB->sad16 = |
429 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
430 |
|
|
431 |
int32_t iDirection = 0; |
int32_t iDirection = 0; |
432 |
|
int32_t iDirectionBackup; |
433 |
int32_t iSAD; |
int32_t iSAD; |
434 |
VECTOR backupMV; |
VECTOR backupMV; |
435 |
|
|
443 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
444 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
445 |
|
|
446 |
if (iDirection) |
if (iDirection) { |
447 |
while (!iFound) { |
while (!iFound) { |
448 |
iFound = 1; |
iFound = 1; |
449 |
backupMV = *currMV; |
backupMV = *currMV; |
450 |
|
iDirectionBackup = iDirection; |
451 |
|
|
452 |
if (iDirection != 2) |
if (iDirectionBackup != 2) |
453 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
454 |
backupMV.y, 1); |
backupMV.y, 1); |
455 |
if (iDirection != 1) |
if (iDirectionBackup != 1) |
456 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
457 |
backupMV.y, 2); |
backupMV.y, 2); |
458 |
if (iDirection != 4) |
if (iDirectionBackup != 4) |
459 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
460 |
backupMV.y - iDiamondSize, 3); |
backupMV.y - iDiamondSize, 3); |
461 |
if (iDirection != 3) |
if (iDirectionBackup != 3) |
462 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
463 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
464 |
|
} |
465 |
} else { |
} else { |
466 |
currMV->x = start_x; |
currMV->x = start_x; |
467 |
currMV->y = start_y; |
currMV->y = start_y; |
525 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
526 |
|
|
527 |
|
|
528 |
if (iDirection) |
if (iDirection) { |
529 |
while (!iFound) { |
while (!iFound) { |
530 |
iFound = 1; |
iFound = 1; |
531 |
backupMV = *currMV; |
backupMV = *currMV; |
534 |
case 1: |
case 1: |
535 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
536 |
backupMV.y, 1); |
backupMV.y, 1); |
537 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
538 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
539 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
540 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
541 |
break; |
break; |
542 |
case 2: |
case 2: |
543 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
544 |
2); |
2); |
545 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
546 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
547 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
548 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
549 |
break; |
break; |
550 |
|
|
551 |
case 3: |
case 3: |
552 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
553 |
4); |
4); |
554 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
555 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
556 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
557 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
558 |
break; |
break; |
559 |
|
|
560 |
case 4: |
case 4: |
561 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
562 |
3); |
3); |
563 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
564 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
565 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
566 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
567 |
break; |
break; |
568 |
|
|
569 |
case 5: |
case 5: |
570 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
571 |
1); |
1); |
572 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
573 |
3); |
3); |
574 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
575 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
576 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
577 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
578 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
579 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
580 |
break; |
break; |
581 |
|
|
582 |
case 6: |
case 6: |
583 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
584 |
2); |
2); |
585 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
586 |
3); |
3); |
587 |
|
|
588 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
589 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
590 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
591 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
592 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
593 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
594 |
|
|
595 |
break; |
break; |
597 |
case 7: |
case 7: |
598 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
599 |
backupMV.y, 1); |
backupMV.y, 1); |
600 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
601 |
4); |
4); |
602 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
603 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
604 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
605 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
606 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
607 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
608 |
break; |
break; |
609 |
|
|
610 |
case 8: |
case 8: |
611 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
612 |
2); |
2); |
613 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
614 |
4); |
4); |
615 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
616 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
617 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
618 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
619 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
620 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
621 |
break; |
break; |
622 |
default: |
default: |
623 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
624 |
1); |
1); |
625 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
626 |
2); |
2); |
627 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
628 |
3); |
3); |
629 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
630 |
4); |
4); |
631 |
|
|
632 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
633 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
634 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
635 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
636 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
637 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
638 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
639 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
640 |
break; |
break; |
641 |
} |
} |
642 |
|
} |
643 |
} else { |
} else { |
644 |
currMV->x = start_x; |
currMV->x = start_x; |
645 |
currMV->y = start_y; |
currMV->y = start_y; |
845 |
return iMinSAD; |
return iMinSAD; |
846 |
} |
} |
847 |
|
|
848 |
|
#define CHECK_MV16_F_INTERPOL(X,Y) { \ |
849 |
#define CHECK_MV16_F_INTERPOL(X,Y,BX,BY) { \ |
if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ |
850 |
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
&& ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
|
851 |
{ \ |
{ \ |
852 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16bi( cur, \ |
853 |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
854 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ |
855 |
|
iEdgedWidth); \ |
856 |
|
iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
857 |
|
iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
858 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
859 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); } } \ |
860 |
} |
} |
861 |
|
|
862 |
#define CHECK_MV16_F_INTERPOL_DIR(X,Y,BX,BY,D) { \ |
#define CHECK_MV16_F_INTERPOL_FOUND(X,Y) { \ |
863 |
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ |
864 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ |
865 |
{ \ |
{ \ |
866 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16bi( cur, \ |
867 |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
868 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ |
869 |
|
iEdgedWidth); \ |
870 |
|
iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
871 |
|
iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
872 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
873 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); iFound=0;} } \ |
874 |
} |
} |
875 |
|
|
876 |
#define CHECK_MV16_F_INTERPOL_FOUND(X,Y,BX,BY,D) { \ |
#define CHECK_MV16_B_INTERPOL(X,Y) { \ |
877 |
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ |
878 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ |
879 |
{ \ |
{ \ |
880 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16bi( cur, \ |
881 |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ |
882 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
883 |
|
iEdgedWidth); \ |
884 |
|
iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
885 |
|
iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
886 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
887 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); } } \ |
888 |
} |
} |
889 |
|
|
890 |
|
#define CHECK_MV16_B_INTERPOL_FOUND(X,Y) { \ |
891 |
#define CHECK_MV16_B_INTERPOL(FX,FY,X,Y) { \ |
if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ |
892 |
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
&& ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ |
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
|
893 |
{ \ |
{ \ |
894 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16bi( cur, \ |
895 |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ |
896 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
897 |
|
iEdgedWidth); \ |
898 |
|
iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
899 |
|
iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
900 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
901 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); iFound=0;} } \ |
902 |
} |
} |
903 |
|
|
|
|
|
|
#define CHECK_MV16_B_INTERPOL_DIR(FX,FY,X,Y,D) { \ |
|
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
|
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
|
|
{ \ |
|
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
|
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
|
|
if (iSAD < iMinSAD) \ |
|
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
|
|
} |
|
|
|
|
|
|
|
|
#define CHECK_MV16_B_INTERPOL_FOUND(FX,FY,X,Y,D) { \ |
|
|
if ( ((X) <= max_dx) && ((X) >= min_dx) \ |
|
|
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
|
|
{ \ |
|
|
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
|
|
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
|
|
if (iSAD < iMinSAD) \ |
|
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
|
|
} |
|
|
|
|
|
|
|
|
#if (0==1) |
|
904 |
int32_t |
int32_t |
905 |
Diamond16_InterpolMainSearch( |
Diamond16_InterpolMainSearch( |
906 |
const uint8_t * const f_pRef, |
const uint8_t * const f_pRef, |
907 |
const uint8_t * const f_pRefH, |
const uint8_t * const f_pRefH, |
908 |
const uint8_t * const f_pRefV, |
const uint8_t * const f_pRefV, |
909 |
const uint8_t * const f_pRefHV, |
const uint8_t * const f_pRefHV, |
910 |
|
|
911 |
const uint8_t * const cur, |
const uint8_t * const cur, |
912 |
|
|
913 |
const uint8_t * const b_pRef, |
const uint8_t * const b_pRef, |
932 |
const int b_center_x, |
const int b_center_x, |
933 |
const int b_center_y, |
const int b_center_y, |
934 |
|
|
935 |
const int32_t min_dx, |
const int32_t f_min_dx, |
936 |
const int32_t max_dx, |
const int32_t f_max_dx, |
937 |
const int32_t min_dy, |
const int32_t f_min_dy, |
938 |
const int32_t max_dy, |
const int32_t f_max_dy, |
939 |
|
|
940 |
|
const int32_t b_min_dx, |
941 |
|
const int32_t b_max_dx, |
942 |
|
const int32_t b_min_dy, |
943 |
|
const int32_t b_max_dy, |
944 |
|
|
945 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
946 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
947 |
|
|
953 |
{ |
{ |
954 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
955 |
|
|
|
int32_t f_iDirection = 0; |
|
|
int32_t b_iDirection = 0; |
|
956 |
int32_t iSAD; |
int32_t iSAD; |
957 |
|
|
958 |
VECTOR f_backupMV; |
VECTOR f_backupMV; |
959 |
VECTOR b_backupMV; |
VECTOR b_backupMV; |
960 |
|
|
961 |
f_backupMV.x = start_x; |
f_currMV->x = f_start_x; |
962 |
f_backupMV.y = start_y; |
f_currMV->y = f_start_y; |
963 |
b_backupMV.x = start_x; |
b_currMV->x = b_start_x; |
964 |
b_backupMV.y = start_y; |
b_currMV->y = b_start_y; |
965 |
|
|
966 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
do |
967 |
|
{ |
968 |
|
iFound = 1; |
969 |
|
|
970 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
f_backupMV = *f_currMV; |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); |
|
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
|
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
|
971 |
|
|
972 |
if (iDirection) |
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x - iDiamondSize, f_backupMV.y); |
973 |
while (!iFound) { |
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x + iDiamondSize, f_backupMV.y); |
974 |
iFound = 1; |
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y - iDiamondSize); |
975 |
backupMV = *currMV; |
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y + iDiamondSize); |
976 |
|
|
977 |
|
b_backupMV = *b_currMV; |
978 |
|
|
979 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x - iDiamondSize, b_backupMV.y); |
980 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x + iDiamondSize, b_backupMV.y); |
981 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y - iDiamondSize); |
982 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y + iDiamondSize); |
983 |
|
|
984 |
|
} while (!iFound); |
985 |
|
|
986 |
|
return iMinSAD; |
987 |
|
} |
988 |
|
|
989 |
|
/* Sorry, these MACROS really got too large... I'll turn them into function soon! */ |
990 |
|
|
991 |
|
#define CHECK_MV16_DIRECT_FOUND(X,Y) \ |
992 |
|
if ( (X)>=(-32) && (X)<=(31) && ((Y)>=-32) && ((Y)<=31) ) \ |
993 |
|
{ int k;\ |
994 |
|
VECTOR mvs,b_mvs; \ |
995 |
|
iSAD = 0;\ |
996 |
|
for (k = 0; k < 4; k++) { \ |
997 |
|
mvs.x = (int32_t) ((TRB * directmv[k].x) / TRD + (X)); \ |
998 |
|
b_mvs.x = (int32_t) (((X) == 0) \ |
999 |
|
? ((TRB - TRD) * directmv[k].x) / TRD \ |
1000 |
|
: mvs.x - directmv[k].x); \ |
1001 |
|
\ |
1002 |
|
mvs.y = (int32_t) ((TRB * directmv[k].y) / TRD + (Y)); \ |
1003 |
|
b_mvs.y = (int32_t) (((Y) == 0) \ |
1004 |
|
? ((TRB - TRD) * directmv[k].y) / TRD \ |
1005 |
|
: mvs.y - directmv[k].y); \ |
1006 |
|
\ |
1007 |
|
if ( (mvs.x <= max_dx) && (mvs.x >= min_dx) \ |
1008 |
|
&& (mvs.y <= max_dy) && (mvs.y >= min_dy) \ |
1009 |
|
&& (b_mvs.x <= max_dx) && (b_mvs.x >= min_dx) \ |
1010 |
|
&& (b_mvs.y <= max_dy) && (b_mvs.y >= min_dy) ) { \ |
1011 |
|
iSAD += sad8bi( cur + 8*(k&1) + 8*(k>>1)*iEdgedWidth, \ |
1012 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ |
1013 |
|
mvs.x, mvs.y, iEdgedWidth), \ |
1014 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ |
1015 |
|
b_mvs.x, b_mvs.y, iEdgedWidth), \ |
1016 |
|
iEdgedWidth); \ |
1017 |
|
} \ |
1018 |
|
else \ |
1019 |
|
iSAD = 65535; \ |
1020 |
|
} \ |
1021 |
|
iSAD += calc_delta_16((X),(Y), 1, iQuant);\ |
1022 |
|
if (iSAD < iMinSAD) \ |
1023 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iFound=0; } \ |
1024 |
|
} |
1025 |
|
|
1026 |
|
|
1027 |
|
|
1028 |
|
int32_t |
1029 |
|
Diamond16_DirectMainSearch( |
1030 |
|
const uint8_t * const f_pRef, |
1031 |
|
const uint8_t * const f_pRefH, |
1032 |
|
const uint8_t * const f_pRefV, |
1033 |
|
const uint8_t * const f_pRefHV, |
1034 |
|
|
1035 |
|
const uint8_t * const cur, |
1036 |
|
|
1037 |
|
const uint8_t * const b_pRef, |
1038 |
|
const uint8_t * const b_pRefH, |
1039 |
|
const uint8_t * const b_pRefV, |
1040 |
|
const uint8_t * const b_pRefHV, |
1041 |
|
|
1042 |
|
const int x, |
1043 |
|
const int y, |
1044 |
|
|
1045 |
|
const int TRB, |
1046 |
|
const int TRD, |
1047 |
|
|
1048 |
|
const int start_x, |
1049 |
|
const int start_y, |
1050 |
|
|
1051 |
|
int iMinSAD, |
1052 |
|
VECTOR * const currMV, |
1053 |
|
const VECTOR * const directmv, |
1054 |
|
|
1055 |
|
const int32_t min_dx, |
1056 |
|
const int32_t max_dx, |
1057 |
|
const int32_t min_dy, |
1058 |
|
const int32_t max_dy, |
1059 |
|
|
1060 |
|
const int32_t iEdgedWidth, |
1061 |
|
const int32_t iDiamondSize, |
1062 |
|
|
1063 |
|
const int32_t iQuant, |
1064 |
|
int iFound) |
1065 |
|
{ |
1066 |
|
/* Do a diamond search around given starting point, return SAD of best */ |
1067 |
|
|
1068 |
|
int32_t iSAD; |
1069 |
|
|
1070 |
|
VECTOR backupMV; |
1071 |
|
|
|
if (iDirection != 2) |
|
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
|
|
backupMV.y, 1); |
|
|
if (iDirection != 1) |
|
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
|
|
backupMV.y, 2); |
|
|
if (iDirection != 4) |
|
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
|
|
backupMV.y - iDiamondSize, 3); |
|
|
if (iDirection != 3) |
|
|
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
|
|
backupMV.y + iDiamondSize, 4); |
|
|
} else { |
|
1072 |
currMV->x = start_x; |
currMV->x = start_x; |
1073 |
currMV->y = start_y; |
currMV->y = start_y; |
1074 |
} |
|
1075 |
|
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1076 |
|
|
1077 |
|
do |
1078 |
|
{ |
1079 |
|
iFound = 1; |
1080 |
|
|
1081 |
|
backupMV = *currMV; |
1082 |
|
|
1083 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x - iDiamondSize, backupMV.y); |
1084 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x + iDiamondSize, backupMV.y); |
1085 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y - iDiamondSize); |
1086 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y + iDiamondSize); |
1087 |
|
|
1088 |
|
} while (!iFound); |
1089 |
|
|
1090 |
return iMinSAD; |
return iMinSAD; |
1091 |
} |
} |
|
#endif |
|
1092 |
|
|
1093 |
|
|
1094 |
int32_t |
int32_t |
1339 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1340 |
const int x, |
const int x, |
1341 |
const int y, |
const int y, |
1342 |
const int start_x, |
const int start_x, /* start is searched first, so it should contain the most */ |
1343 |
const int start_y, |
const int start_y, /* likely motion vector for this block */ |
1344 |
const int center_x, |
const int center_x, /* center is from where length of MVs is measured */ |
1345 |
const int center_y, |
const int center_y, |
1346 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1347 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1678 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
1679 |
|
|
1680 |
int32_t iDirection = 0; |
int32_t iDirection = 0; |
1681 |
|
int32_t iDirectionBackup; |
1682 |
int32_t iSAD; |
int32_t iSAD; |
1683 |
VECTOR backupMV; |
VECTOR backupMV; |
1684 |
|
|
1692 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1693 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
1694 |
|
|
1695 |
if (iDirection) |
if (iDirection) { |
1696 |
while (!iFound) { |
while (!iFound) { |
1697 |
iFound = 1; |
iFound = 1; |
1698 |
backupMV = *currMV; // since iDirection!=0, this is well defined! |
backupMV = *currMV; // since iDirection!=0, this is well defined! |
1699 |
|
iDirectionBackup = iDirection; |
1700 |
|
|
1701 |
if (iDirection != 2) |
if (iDirectionBackup != 2) |
1702 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1703 |
backupMV.y, 1); |
backupMV.y, 1); |
1704 |
if (iDirection != 1) |
if (iDirectionBackup != 1) |
1705 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1706 |
backupMV.y, 2); |
backupMV.y, 2); |
1707 |
if (iDirection != 4) |
if (iDirectionBackup != 4) |
1708 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1709 |
backupMV.y - iDiamondSize, 3); |
backupMV.y - iDiamondSize, 3); |
1710 |
if (iDirection != 3) |
if (iDirectionBackup != 3) |
1711 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1712 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
1713 |
|
} |
1714 |
} else { |
} else { |
1715 |
currMV->x = start_x; |
currMV->x = start_x; |
1716 |
currMV->y = start_y; |
currMV->y = start_y; |
1718 |
return iMinSAD; |
return iMinSAD; |
1719 |
} |
} |
1720 |
|
|
1721 |
|
|
1722 |
|
|
1723 |
|
|
1724 |
|
int32_t |
1725 |
|
Square8_MainSearch(const uint8_t * const pRef, |
1726 |
|
const uint8_t * const pRefH, |
1727 |
|
const uint8_t * const pRefV, |
1728 |
|
const uint8_t * const pRefHV, |
1729 |
|
const uint8_t * const cur, |
1730 |
|
const int x, |
1731 |
|
const int y, |
1732 |
|
int32_t start_x, |
1733 |
|
int32_t start_y, |
1734 |
|
int32_t iMinSAD, |
1735 |
|
VECTOR * const currMV, |
1736 |
|
const int center_x, |
1737 |
|
const int center_y, |
1738 |
|
const int32_t min_dx, |
1739 |
|
const int32_t max_dx, |
1740 |
|
const int32_t min_dy, |
1741 |
|
const int32_t max_dy, |
1742 |
|
const int32_t iEdgedWidth, |
1743 |
|
const int32_t iDiamondSize, |
1744 |
|
const int32_t iFcode, |
1745 |
|
const int32_t iQuant, |
1746 |
|
int iFound) |
1747 |
|
{ |
1748 |
|
/* Do a square search around given starting point, return SAD of best */ |
1749 |
|
|
1750 |
|
int32_t iDirection = 0; |
1751 |
|
int32_t iSAD; |
1752 |
|
VECTOR backupMV; |
1753 |
|
|
1754 |
|
backupMV.x = start_x; |
1755 |
|
backupMV.y = start_y; |
1756 |
|
|
1757 |
|
/* It's one search with full square pattern, and new parts for all following diamonds */ |
1758 |
|
|
1759 |
|
/* new direction are extra, so 1-4 is normal diamond |
1760 |
|
537 |
1761 |
|
1*2 |
1762 |
|
648 |
1763 |
|
*/ |
1764 |
|
|
1765 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
1766 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); |
1767 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1768 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
1769 |
|
|
1770 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1771 |
|
backupMV.y - iDiamondSize, 5); |
1772 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1773 |
|
backupMV.y + iDiamondSize, 6); |
1774 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1775 |
|
backupMV.y - iDiamondSize, 7); |
1776 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1777 |
|
backupMV.y + iDiamondSize, 8); |
1778 |
|
|
1779 |
|
|
1780 |
|
if (iDirection) { |
1781 |
|
while (!iFound) { |
1782 |
|
iFound = 1; |
1783 |
|
backupMV = *currMV; |
1784 |
|
|
1785 |
|
switch (iDirection) { |
1786 |
|
case 1: |
1787 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1788 |
|
backupMV.y, 1); |
1789 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1790 |
|
backupMV.y - iDiamondSize, 5); |
1791 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1792 |
|
backupMV.y - iDiamondSize, 7); |
1793 |
|
break; |
1794 |
|
case 2: |
1795 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1796 |
|
2); |
1797 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1798 |
|
backupMV.y + iDiamondSize, 6); |
1799 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1800 |
|
backupMV.y + iDiamondSize, 8); |
1801 |
|
break; |
1802 |
|
|
1803 |
|
case 3: |
1804 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1805 |
|
4); |
1806 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1807 |
|
backupMV.y - iDiamondSize, 7); |
1808 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1809 |
|
backupMV.y + iDiamondSize, 8); |
1810 |
|
break; |
1811 |
|
|
1812 |
|
case 4: |
1813 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1814 |
|
3); |
1815 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1816 |
|
backupMV.y - iDiamondSize, 5); |
1817 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1818 |
|
backupMV.y + iDiamondSize, 6); |
1819 |
|
break; |
1820 |
|
|
1821 |
|
case 5: |
1822 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1823 |
|
1); |
1824 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1825 |
|
3); |
1826 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1827 |
|
backupMV.y - iDiamondSize, 5); |
1828 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1829 |
|
backupMV.y + iDiamondSize, 6); |
1830 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1831 |
|
backupMV.y - iDiamondSize, 7); |
1832 |
|
break; |
1833 |
|
|
1834 |
|
case 6: |
1835 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1836 |
|
2); |
1837 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1838 |
|
3); |
1839 |
|
|
1840 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1841 |
|
backupMV.y - iDiamondSize, 5); |
1842 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1843 |
|
backupMV.y + iDiamondSize, 6); |
1844 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1845 |
|
backupMV.y + iDiamondSize, 8); |
1846 |
|
|
1847 |
|
break; |
1848 |
|
|
1849 |
|
case 7: |
1850 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1851 |
|
backupMV.y, 1); |
1852 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1853 |
|
4); |
1854 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1855 |
|
backupMV.y - iDiamondSize, 5); |
1856 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1857 |
|
backupMV.y - iDiamondSize, 7); |
1858 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1859 |
|
backupMV.y + iDiamondSize, 8); |
1860 |
|
break; |
1861 |
|
|
1862 |
|
case 8: |
1863 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1864 |
|
2); |
1865 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1866 |
|
4); |
1867 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1868 |
|
backupMV.y + iDiamondSize, 6); |
1869 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1870 |
|
backupMV.y - iDiamondSize, 7); |
1871 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1872 |
|
backupMV.y + iDiamondSize, 8); |
1873 |
|
break; |
1874 |
|
default: |
1875 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1876 |
|
1); |
1877 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1878 |
|
2); |
1879 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1880 |
|
3); |
1881 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1882 |
|
4); |
1883 |
|
|
1884 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1885 |
|
backupMV.y - iDiamondSize, 5); |
1886 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1887 |
|
backupMV.y + iDiamondSize, 6); |
1888 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1889 |
|
backupMV.y - iDiamondSize, 7); |
1890 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1891 |
|
backupMV.y + iDiamondSize, 8); |
1892 |
|
break; |
1893 |
|
} |
1894 |
|
} |
1895 |
|
} else { |
1896 |
|
currMV->x = start_x; |
1897 |
|
currMV->y = start_y; |
1898 |
|
} |
1899 |
|
return iMinSAD; |
1900 |
|
} |
1901 |
|
|
1902 |
|
|
1903 |
|
|
1904 |
|
|
1905 |
|
|
1906 |
int32_t |
int32_t |
1907 |
Halfpel8_Refine_c(const uint8_t * const pRef, |
Halfpel8_Refine_c(const uint8_t * const pRef, |
1908 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
2041 |
|
|
2042 |
// Prepare for main loop |
// Prepare for main loop |
2043 |
|
|
2044 |
// if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES8) |
2045 |
// MainSearchPtr = Square8_MainSearch; |
MainSearchPtr = Square8_MainSearch; |
2046 |
// else |
else |
2047 |
|
|
2048 |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2049 |
MainSearchPtr = AdvDiamond8_MainSearch; |
MainSearchPtr = AdvDiamond8_MainSearch; |
2721 |
|
|
2722 |
// there is no EPZS^2 for inter4v at the moment |
// there is no EPZS^2 for inter4v at the moment |
2723 |
|
|
2724 |
// if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES8) |
2725 |
// MainSearchPtr = Square8_MainSearch; |
MainSearchPtr = Square8_MainSearch; |
2726 |
// else |
else |
2727 |
|
|
2728 |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2729 |
MainSearchPtr = AdvDiamond8_MainSearch; |
MainSearchPtr = AdvDiamond8_MainSearch; |
2798 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2799 |
const int x, |
const int x, |
2800 |
const int y, |
const int y, |
2801 |
const int start_x, |
const int start_x, /* start should be most likely vector */ |
2802 |
const int start_y, |
const int start_y, |
2803 |
const int center_x, |
const int center_x, /* center is from where length of MVs is measured */ |
2804 |
const int center_y, |
const int center_y, |
2805 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2806 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2829 |
int32_t iFound; |
int32_t iFound; |
2830 |
|
|
2831 |
VECTOR newMV; |
VECTOR newMV; |
2832 |
VECTOR backupMV; /* just for PMVFAST */ |
VECTOR backupMV; |
2833 |
|
|
2834 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
2835 |
int32_t psad[4]; |
int32_t psad[4]; |
2859 |
*currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; |
*currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; |
2860 |
|
|
2861 |
} else { |
} else { |
2862 |
|
|
2863 |
|
bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2864 |
|
|
2865 |
threshA = psad[0]; |
threshA = psad[0]; |
2866 |
threshB = threshA + 256; |
threshB = threshA + 256; |
2867 |
if (threshA < 512) |
if (threshA < 512) |
2871 |
if (threshB > 1792) |
if (threshB > 1792) |
2872 |
threshB = 1792; |
threshB = 1792; |
2873 |
|
|
|
bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
|
2874 |
*currMV = pmv[0]; /* current best := prediction */ |
*currMV = pmv[0]; /* current best := prediction */ |
2875 |
} |
} |
2876 |
|
|
3086 |
|
|
3087 |
/* *********************************************************** |
/* *********************************************************** |
3088 |
bvop motion estimation |
bvop motion estimation |
|
// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) |
|
3089 |
***************************************************************/ |
***************************************************************/ |
3090 |
|
|
|
|
|
|
#define DIRECT_PENALTY 0 |
|
|
#define DIRECT_UPPERLIMIT 256 // never use direct mode if SAD is larger than this |
|
|
|
|
3091 |
void |
void |
3092 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
3093 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
3110 |
const int mb_height = pParam->mb_height; |
const int mb_height = pParam->mb_height; |
3111 |
const int edged_width = pParam->edged_width; |
const int edged_width = pParam->edged_width; |
3112 |
|
|
3113 |
|
const int32_t iWidth = pParam->width; |
3114 |
|
const int32_t iHeight = pParam->height; |
3115 |
|
|
3116 |
int i, j, k; |
int i, j, k; |
3117 |
|
|
3118 |
static const VECTOR zeroMV={0,0}; |
static const VECTOR zeroMV={0,0}; |
3120 |
int f_sad16; /* forward (as usual) search */ |
int f_sad16; /* forward (as usual) search */ |
3121 |
int b_sad16; /* backward (only in b-frames) search */ |
int b_sad16; /* backward (only in b-frames) search */ |
3122 |
int i_sad16; /* interpolated (both direction, b-frames only) */ |
int i_sad16; /* interpolated (both direction, b-frames only) */ |
3123 |
int d_sad16; /* direct mode (assume linear motion) */ |
int d_sad16; /* direct mode (assume almost linear motion) */ |
|
int dnv_sad16; /* direct mode (assume linear motion) without correction vector */ |
|
3124 |
|
|
3125 |
int best_sad; |
int best_sad; |
3126 |
|
|
3127 |
VECTOR f_predMV, b_predMV; /* there is no direct prediction */ |
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
3128 |
|
VECTOR f_interpolMV, b_interpolMV; |
3129 |
VECTOR pmv_dontcare; |
VECTOR pmv_dontcare; |
3130 |
|
|
3131 |
|
int min_dx, max_dx, min_dy, max_dy; |
3132 |
|
int f_min_dx, f_max_dx, f_min_dy, f_max_dy; |
3133 |
|
int b_min_dx, b_max_dx, b_min_dy, b_max_dy; |
3134 |
|
|
3135 |
int f_count=0; |
int f_count=0; |
3136 |
int b_count=0; |
int b_count=0; |
3137 |
int i_count=0; |
int i_count=0; |
3138 |
int d_count=0; |
int d_count=0; |
3139 |
int dnv_count=0; |
|
|
int s_count=0; |
|
3140 |
const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; |
const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; |
3141 |
const int64_t TRD = (int32_t)time_pp; |
const int64_t TRD = (int32_t)time_pp; |
3142 |
|
|
3152 |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
3153 |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
3154 |
|
|
3155 |
VECTOR directMV; |
mb->deltamv=zeroMV; |
|
VECTOR deltaMV=zeroMV; |
|
3156 |
|
|
3157 |
/* special case, if collocated block is SKIPed: encoding is forward(0,0) */ |
/* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */ |
3158 |
|
|
3159 |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
3160 |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
3161 |
mb->mode = MODE_NOT_CODED; |
mb->mode = MODE_NOT_CODED; |
3162 |
mb->mvs[0].x = 0; |
mb->b_mvs[0] = mb->mvs[0] = zeroMV; |
|
mb->mvs[0].y = 0; |
|
|
mb->b_mvs[0].x = 0; |
|
|
mb->b_mvs[0].y = 0; |
|
3163 |
continue; |
continue; |
3164 |
} |
} |
3165 |
|
|
|
dnv_sad16 = DIRECT_PENALTY; |
|
|
|
|
3166 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
3167 |
{ |
{ |
3168 |
|
d_sad16 = 0; |
3169 |
/* same method of scaling as in decoder.c, so we copy from there */ |
/* same method of scaling as in decoder.c, so we copy from there */ |
3170 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
3171 |
|
|
3172 |
directMV = b_mb->mvs[k]; |
mb->directmv[k] = b_mb->mvs[k]; |
3173 |
|
|
3174 |
mb->mvs[k].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); |
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3175 |
mb->b_mvs[k].x = (int32_t) ((deltaMV.x == 0) |
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
3176 |
? ((TRB - TRD) * directMV.x) / TRD |
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
3177 |
: mb->mvs[k].x - directMV.x); |
: mb->mvs[k].x - mb->directmv[k].x); |
3178 |
mb->mvs[k].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); |
|
3179 |
mb->b_mvs[k].y = (int32_t) ((deltaMV.y == 0) |
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3180 |
? ((TRB - TRD) * directMV.y) / TRD |
mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) |
3181 |
: mb->mvs[k].y - directMV.y); |
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
3182 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
3183 |
|
|
3184 |
dnv_sad16 += |
d_sad16 += |
3185 |
sad8bi(frame->image.y + 2*(i+(k&1))*8 + 2*(j+(k>>1))*8*edged_width, |
sad8bi(frame->image.y + (2*i+(k&1))*8 + (2*j+(k>>1))*8*edged_width, |
3186 |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3187 |
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->mvs[k], edged_width), |
(2*i+(k&1)), (2*j+(k>>1)), 8, &mb->mvs[k], edged_width), |
3188 |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3189 |
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->b_mvs[k], edged_width), |
(2*i+(k&1)), (2*j+(k>>1)), 8, &mb->b_mvs[k], edged_width), |
3190 |
edged_width); |
edged_width); |
3191 |
} |
} |
3192 |
} |
} |
3193 |
else |
else |
3194 |
{ |
{ |
3195 |
directMV = b_mb->mvs[0]; |
mb->directmv[3] = mb->directmv[2] = mb->directmv[1] = |
3196 |
|
mb->directmv[0] = b_mb->mvs[0]; |
3197 |
|
|
3198 |
mb->mvs[0].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); |
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3199 |
mb->b_mvs[0].x = (int32_t) ((deltaMV.x == 0) |
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
3200 |
? ((TRB - TRD) * directMV.x) / TRD |
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
3201 |
: mb->mvs[0].x - directMV.x); |
: mb->mvs[0].x - mb->directmv[0].x); |
3202 |
mb->mvs[0].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); |
|
3203 |
mb->b_mvs[0].y = (int32_t) ((deltaMV.y == 0) |
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
3204 |
? ((TRB - TRD) * directMV.y) / TRD |
mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) |
3205 |
: mb->mvs[0].y - directMV.y); |
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
3206 |
|
: mb->mvs[0].y - mb->directmv[0].y); |
3207 |
|
|
3208 |
dnv_sad16 = DIRECT_PENALTY + |
d_sad16 = sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
|
sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
|
3209 |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3210 |
i, j, 16, &mb->mvs[0], edged_width), |
i, j, 16, &mb->mvs[0], edged_width), |
3211 |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3212 |
i, j, 16, &mb->b_mvs[0], edged_width), |
i, j, 16, &mb->b_mvs[0], edged_width), |
3213 |
edged_width); |
edged_width); |
3214 |
|
|
|
|
|
3215 |
} |
} |
3216 |
|
d_sad16 += calc_delta_16(mb->deltamv.x, mb->deltamv.y, 1, frame->quant); |
3217 |
|
|
3218 |
// forward search |
// forward search |
3219 |
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3220 |
&frame->image, i, j, |
&frame->image, i, j, |
3221 |
mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ |
mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ |
3222 |
f_predMV.x, f_predMV.y, /* center is f-prediction */ |
f_predMV.x, f_predMV.y, /* center is f-prediction */ |
3223 |
frame->motion_flags & (~(PMV_EARLYSTOP16|PMV_QUICKSTOP16)), |
frame->motion_flags, |
3224 |
frame->quant, frame->fcode, pParam, |
frame->quant, frame->fcode, pParam, |
3225 |
f_mbs, f_mbs, |
f_mbs, f_mbs, |
3226 |
&mb->mvs[0], &pmv_dontcare); |
&mb->mvs[0], &pmv_dontcare); |
3231 |
&frame->image, i, j, |
&frame->image, i, j, |
3232 |
mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ |
mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ |
3233 |
b_predMV.x, b_predMV.y, /* center is b-prediction */ |
b_predMV.x, b_predMV.y, /* center is b-prediction */ |
3234 |
frame->motion_flags & (~(PMV_EARLYSTOP16|PMV_QUICKSTOP16)), |
frame->motion_flags, |
3235 |
frame->quant, frame->bcode, pParam, |
frame->quant, frame->bcode, pParam, |
3236 |
b_mbs, b_mbs, |
b_mbs, b_mbs, |
3237 |
&mb->b_mvs[0], &pmv_dontcare); |
&mb->b_mvs[0], &pmv_dontcare); |
3245 |
edged_width); |
edged_width); |
3246 |
i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
3247 |
frame->fcode, frame->quant); |
frame->fcode, frame->quant); |
3248 |
i_sad16 += calc_delta_16(mb->b_mvs[0].y-b_predMV.y, mb->b_mvs[0].y-b_predMV.y, |
i_sad16 += calc_delta_16(mb->b_mvs[0].x-b_predMV.x, mb->b_mvs[0].y-b_predMV.y, |
3249 |
frame->bcode, frame->quant); |
frame->bcode, frame->quant); |
3250 |
|
|
3251 |
// TODO: direct search |
get_range(&f_min_dx, &f_max_dx, &f_min_dy, &f_max_dy, i, j, 16, iWidth, iHeight, |
3252 |
// predictor + delta vector in range [-32,32] (fcode=1) |
frame->fcode); |
3253 |
|
get_range(&b_min_dx, &b_max_dx, &b_min_dy, &b_max_dy, i, j, 16, iWidth, iHeight, |
3254 |
|
frame->bcode); |
3255 |
|
|
3256 |
|
/* Interpolated MC motion vector search, this is tedious and more complicated because there are |
3257 |
|
two values for everything, always one for backward and one for forward ME. Still, we don't gain |
3258 |
|
much from this search, maybe it should simply be skipped and simply current i_sad16 value used |
3259 |
|
as "optimal". */ |
3260 |
|
|
3261 |
|
i_sad16 = Diamond16_InterpolMainSearch( |
3262 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3263 |
|
frame->image.y + i * 16 + j * 16 * edged_width, |
3264 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3265 |
|
i, j, |
3266 |
|
mb->mvs[0].x, mb->mvs[0].y, |
3267 |
|
mb->b_mvs[0].x, mb->b_mvs[0].y, |
3268 |
|
i_sad16, |
3269 |
|
&f_interpolMV, &b_interpolMV, |
3270 |
|
f_predMV.x, f_predMV.y, b_predMV.x, b_predMV.y, |
3271 |
|
f_min_dx, f_max_dx, f_min_dy, f_max_dy, |
3272 |
|
b_min_dx, b_max_dx, b_min_dy, b_max_dy, |
3273 |
|
edged_width, 2, |
3274 |
|
frame->fcode, frame->bcode,frame->quant,0); |
3275 |
|
|
3276 |
|
i_sad16 = Diamond16_InterpolMainSearch( |
3277 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3278 |
|
frame->image.y + i * 16 + j * 16 * edged_width, |
3279 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3280 |
|
i, j, |
3281 |
|
f_interpolMV.x, f_interpolMV.y, |
3282 |
|
b_interpolMV.x, b_interpolMV.y, |
3283 |
|
i_sad16, |
3284 |
|
&f_interpolMV, &b_interpolMV, |
3285 |
|
f_predMV.x, f_predMV.y, b_predMV.x, b_predMV.y, |
3286 |
|
f_min_dx, f_max_dx, f_min_dy, f_max_dy, |
3287 |
|
b_min_dx, b_max_dx, b_min_dy, b_max_dy, |
3288 |
|
edged_width, 1, |
3289 |
|
frame->fcode, frame->bcode,frame->quant,0); // equiv to halfpel refine |
3290 |
|
|
3291 |
|
|
3292 |
|
/* DIRECT MODE DELTA VECTOR SEARCH. |
3293 |
|
This has to be made more effective, but at the moment I'm happy it's running at all */ |
3294 |
|
|
3295 |
|
/* There are two range restrictions for direct mode: deltaMV is limited to [-32,31] in halfpel units, and |
3296 |
|
absolute vector must not lie outside of image dimensions. Constraint one is dealt with by CHECK_MV16_DIRECT |
3297 |
|
and for constraint two we need distance to boundary. This is done by get_range very large fcode (hack!) */ |
3298 |
|
|
3299 |
|
get_range(&min_dx, &max_dx, &min_dy, &max_dy, i, j, 16, iWidth, iHeight, 19); |
3300 |
|
|
3301 |
|
d_sad16 = Diamond16_DirectMainSearch( |
3302 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3303 |
|
frame->image.y + i*16 + j*16*edged_width, |
3304 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3305 |
|
i, j, |
3306 |
|
TRB,TRD, |
3307 |
|
0,0, |
3308 |
|
d_sad16, |
3309 |
|
&mb->deltamv, |
3310 |
|
mb->directmv, // this has to be pre-initialized with b_mb->mvs[] |
3311 |
|
min_dx, max_dx, min_dy, max_dy, |
3312 |
|
edged_width, 2, frame->quant, 0); |
3313 |
|
|
3314 |
// i_sad16 = 65535; |
d_sad16 = Diamond16_DirectMainSearch( |
3315 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3316 |
|
frame->image.y + i*16 + j*16*edged_width, |
3317 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3318 |
|
i, j, |
3319 |
|
TRB,TRD, |
3320 |
|
mb->deltamv.x, mb->deltamv.y, |
3321 |
|
d_sad16, |
3322 |
|
&mb->deltamv, |
3323 |
|
mb->directmv, // this has to be pre-initialized with b_mb->mvs[] |
3324 |
|
min_dx, max_dx, min_dy, max_dy, |
3325 |
|
edged_width, 1, frame->quant, 0); // equiv to halfpel refine |
3326 |
|
|
3327 |
|
|
3328 |
|
// i_sad16 = 65535; /* remove the comment to disable any of the MODEs */ |
3329 |
// f_sad16 = 65535; |
// f_sad16 = 65535; |
3330 |
// b_sad16 = 65535; |
// b_sad16 = 65535; |
3331 |
|
// d_sad16 = 65535; |
3332 |
|
|
3333 |
if (f_sad16 < b_sad16) { |
if (f_sad16 < b_sad16) { |
3334 |
best_sad = f_sad16; |
best_sad = f_sad16; |
3343 |
mb->mode = MODE_INTERPOLATE; |
mb->mode = MODE_INTERPOLATE; |
3344 |
} |
} |
3345 |
|
|
3346 |
if (dnv_sad16 < best_sad) { |
if (d_sad16 < best_sad) { |
3347 |
|
|
3348 |
if (dnv_sad16 > DIRECT_UPPERLIMIT) |
if (b_mb->mode == MODE_INTER4V) |
3349 |
{ |
{ |
|
/* if SAD value is too large, try same vector with MODE_INTERPOLATE |
|
|
instead (interpolate has residue encoding, direct mode without MV |
|
|
doesn't) |
|
3350 |
|
|
3351 |
This has to be replaced later by "real" direct mode, including delta |
/* how to calc vectors is defined in standard. mvs[] and b_mvs[] are only for motion compensation */ |
3352 |
vector and (if needed) residue encoding |
/* for the bitstream, the value mb->deltamv is read directly */ |
3353 |
|
|
3354 |
*/ |
for (k = 0; k < 4; k++) { |
3355 |
|
|
3356 |
directMV = b_mb->mvs[0]; |
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3357 |
|
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
3358 |
|
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
3359 |
|
: mb->mvs[k].x - mb->directmv[k].x); |
3360 |
|
|
3361 |
|
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3362 |
|
mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) |
3363 |
|
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
3364 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
3365 |
|
} |
3366 |
|
} |
3367 |
|
else |
3368 |
|
{ |
3369 |
|
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3370 |
|
|
3371 |
mb->mvs[0].x = (int32_t) ((TRB * directMV.x) / TRD + deltaMV.x); |
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
3372 |
mb->b_mvs[0].x = (int32_t) ((deltaMV.x == 0) |
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
3373 |
? ((TRB - TRD) * directMV.x) / TRD |
: mb->mvs[0].x - mb->directmv[0].x); |
|
: mb->mvs[0].x - directMV.x); |
|
|
mb->mvs[0].y = (int32_t) ((TRB * directMV.y) / TRD + deltaMV.y); |
|
|
mb->b_mvs[0].y = (int32_t) ((deltaMV.y == 0) |
|
|
? ((TRB - TRD) * directMV.y) / TRD |
|
|
: mb->mvs[0].y - directMV.y); |
|
3374 |
|
|
3375 |
dnv_sad16 = |
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
|
sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
|
|
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
|
|
i, j, 16, &mb->mvs[0], edged_width), |
|
|
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
|
|
i, j, 16, &mb->b_mvs[0], edged_width), |
|
|
edged_width); |
|
|
dnv_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
|
|
frame->fcode, frame->quant); |
|
|
dnv_sad16 += calc_delta_16(mb->b_mvs[0].y-b_predMV.y, mb->b_mvs[0].y-b_predMV.y, |
|
|
frame->bcode, frame->quant); |
|
3376 |
|
|
3377 |
if (dnv_sad16 < best_sad) |
mb->b_mvs[0].y = (int32_t) ((mb->deltamv.y == 0) |
3378 |
{ |
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
3379 |
best_sad = dnv_sad16; |
: mb->mvs[0].y - mb->directmv[0].y); |
|
mb->mode = MODE_INTERPOLATE; |
|
3380 |
|
|
3381 |
/* fprintf(stderr,"f_sad16 = %d, b_sad16 = %d, i_sad16 = %d, dnv_sad16 = %d\n", |
mb->mvs[3] = mb->mvs[2] = mb->mvs[1] = mb->mvs[0]; |
3382 |
f_sad16,b_sad16,i_sad16,dnv_sad16); |
mb->b_mvs[3] = mb->b_mvs[2] = mb->b_mvs[1] = mb->b_mvs[0]; |
|
*/ } |
|
|
} |
|
|
else |
|
|
{ |
|
|
best_sad = dnv_sad16; |
|
|
mb->mode = MODE_DIRECT_NONE_MV; |
|
3383 |
} |
} |
3384 |
|
|
3385 |
|
best_sad = d_sad16; |
3386 |
|
mb->mode = MODE_DIRECT; |
3387 |
} |
} |
3388 |
|
|
3389 |
switch (mb->mode) |
switch (mb->mode) |
3399 |
break; |
break; |
3400 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
3401 |
i_count++; |
i_count++; |
3402 |
|
mb->mvs[0] = f_interpolMV; |
3403 |
|
mb->b_mvs[0] = b_interpolMV; |
3404 |
f_predMV = mb->mvs[0]; |
f_predMV = mb->mvs[0]; |
3405 |
b_predMV = mb->b_mvs[0]; |
b_predMV = mb->b_mvs[0]; |
3406 |
break; |
break; |
3407 |
case MODE_DIRECT: |
case MODE_DIRECT: |
3408 |
d_count++; break; |
d_count++; |
3409 |
case MODE_DIRECT_NONE_MV: |
break; |
|
dnv_count++; break; |
|
3410 |
default: |
default: |
3411 |
s_count++; break; |
break; |
3412 |
} |
} |
3413 |
|
|
3414 |
} |
} |
3415 |
} |
} |
3416 |
|
|
3417 |
#ifdef _DEBUG_BFRAME_STAT |
#ifdef _DEBUG_BFRAME_STAT |
3418 |
fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D0: %04d D: %04d S: %04d\n", |
fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D: %04d\n", |
3419 |
f_count,b_count,i_count,dnv_count,d_count,s_count); |
f_count,b_count,i_count,d_count); |
3420 |
#endif |
#endif |
3421 |
|
|
3422 |
} |
} |