40 |
#include "motion_est.h" |
#include "motion_est.h" |
41 |
#include "motion.h" |
#include "motion.h" |
42 |
#include "sad.h" |
#include "sad.h" |
43 |
|
#include "../utils/emms.h" |
44 |
|
|
45 |
#define INITIAL_SKIP_THRESH (10) |
#define INITIAL_SKIP_THRESH (10) |
46 |
#define FINAL_SKIP_THRESH (50) |
#define FINAL_SKIP_THRESH (50) |
47 |
#define MAX_SAD00_FOR_SKIP (20) |
#define MAX_SAD00_FOR_SKIP (20) |
48 |
#define MAX_CHROMA_SAD_FOR_SKIP (18) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
49 |
#define SKIP_THRESH_B (10) |
#define SKIP_THRESH_B (25) |
50 |
|
|
51 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
52 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
53 |
|
|
54 |
#define iDiamondSize 2 |
#define iDiamondSize 2 |
55 |
|
|
|
//FILE * debug; |
|
|
|
|
56 |
static __inline int |
static __inline int |
57 |
d_mv_bits(int x, int y, const uint32_t iFcode) |
d_mv_bits(int x, int y, const uint32_t iFcode) |
58 |
{ |
{ |
78 |
return xb + yb; |
return xb + yb; |
79 |
} |
} |
80 |
|
|
81 |
/* CHACK_CANDIATE FUNCTIONS START */ |
|
82 |
|
/* CHECK_CANDIATE FUNCTIONS START */ |
83 |
|
|
84 |
static void |
static void |
85 |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
86 |
{ |
{ |
87 |
int32_t * const sad = data->temp; |
int32_t * const sad = data->temp; |
|
// static int32_t sad[5]; |
|
88 |
int t; |
int t; |
89 |
const uint8_t * Reference; |
const uint8_t * Reference; |
90 |
|
|
148 |
} |
} |
149 |
|
|
150 |
static void |
static void |
151 |
|
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
152 |
|
{ |
153 |
|
int32_t sad; |
154 |
|
|
155 |
|
if (( x > data->max_dx) || ( x < data->min_dx) |
156 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
157 |
|
|
158 |
|
sad = lambda_vec16[data->iQuant] * |
159 |
|
d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
160 |
|
|
161 |
|
sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
162 |
|
data->iEdgedWidth, 256*4096); |
163 |
|
|
164 |
|
if (sad < *(data->iMinSAD)) { |
165 |
|
*(data->iMinSAD) = sad; |
166 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
167 |
|
*dir = Direction; } |
168 |
|
} |
169 |
|
|
170 |
|
|
171 |
|
static void |
172 |
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 int Direction, int * const dir, const SearchData * const data) |
173 |
{ |
{ |
174 |
int32_t sad; |
int32_t sad; |
269 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
270 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
271 |
|
|
272 |
if (( x > 31) || ( x < -31) || ( y > 31) || (y < -31)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
273 |
|
|
274 |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
275 |
|
|
336 |
*dir = Direction; } |
*dir = Direction; } |
337 |
} |
} |
338 |
|
|
339 |
/* CHACK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
340 |
|
|
341 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
342 |
|
|
517 |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
518 |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
519 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
520 |
sadC += sad8(current->v + x*8 + y*(iEdgedWidth/2)*8, |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
521 |
reference->v + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
522 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
523 |
|
|
524 |
return 1; |
return 1; |
528 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
529 |
{ |
{ |
530 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
531 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->mv16.x = 0; |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
532 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->mv16.y = 0; |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
533 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
534 |
} |
} |
535 |
|
|
550 |
|
|
551 |
uint32_t x, y; |
uint32_t x, y; |
552 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
553 |
int32_t InterBias; |
int32_t InterBias, quant = current->quant; |
554 |
|
|
555 |
|
// some pre-initialized thingies for SearchP |
556 |
|
int32_t temp[5]; |
557 |
|
VECTOR currentMV[5]; |
558 |
|
int32_t iMinSAD[5]; |
559 |
|
SearchData Data; |
560 |
|
Data.iEdgedWidth = pParam->edged_width; |
561 |
|
Data.currentMV = currentMV; |
562 |
|
Data.iMinSAD = iMinSAD; |
563 |
|
Data.temp = temp; |
564 |
|
Data.iFcode = current->fcode; |
565 |
|
|
566 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
567 |
|
|
577 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
578 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
579 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; } |
580 |
|
else |
581 |
|
if (pMB->dquant != NO_CHANGE) { |
582 |
|
quant += DQtab[pMB->dquant]; |
583 |
|
if (quant > 31) quant = 31; |
584 |
|
else if (quant < 1) quant = 1; |
585 |
|
pMB->quant = quant; |
586 |
|
} |
587 |
|
|
588 |
//initial skip decision |
//initial skip decision |
589 |
|
|
597 |
|
|
598 |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
599 |
y, current->motion_flags, pMB->quant, |
y, current->motion_flags, pMB->quant, |
600 |
current->fcode, pParam, pMBs, reference->mbs, |
&Data, pParam, pMBs, reference->mbs, |
601 |
current->global_flags & XVID_INTER4V, pMB); |
current->global_flags & XVID_INTER4V, pMB); |
602 |
|
|
603 |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
622 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
623 |
if (++iIntra >= iLimit) return 1; |
if (++iIntra >= iLimit) return 1; |
624 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
625 |
pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
626 |
pMB->mvs[3] = zeroMV; |
pMB->mvs[3] = zeroMV; |
627 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
628 |
pMB->sad8[3] = 0; |
pMB->sad8[3] = 0; |
639 |
static __inline int |
static __inline int |
640 |
make_mask(const VECTOR * const pmv, const int i) |
make_mask(const VECTOR * const pmv, const int i) |
641 |
{ |
{ |
642 |
int mask = 0xFF, j; |
int mask = 255, j; |
643 |
for (j = 0; j < i; j++) { |
for (j = 0; j < i; j++) { |
644 |
if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already |
if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already |
645 |
if (pmv[i].x == pmv[j].x) { |
if (pmv[i].x == pmv[j].x) { |
663 |
|
|
664 |
if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour |
if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour |
665 |
pmv[5].x = EVEN(pmv[3].x); |
pmv[5].x = EVEN(pmv[3].x); |
666 |
pmv[5].y = EVEN(pmv[3].y); } |
pmv[5].y = EVEN(pmv[3].y); |
667 |
else pmv[5].x = pmv[5].y = 0; |
} else pmv[5].x = pmv[5].y = 0; |
668 |
|
|
669 |
if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour |
if (x != 0) { pmv[3].x = EVEN(pmv[1].x); pmv[3].y = EVEN(pmv[1].y); }// pmv[3] is left neighbour |
670 |
else pmv[3].x = pmv[3].y = 0; |
else pmv[3].x = pmv[3].y = 0; |
682 |
|
|
683 |
if ((x != iWcount-1) && (y != iHcount-1)) { |
if ((x != iWcount-1) && (y != iHcount-1)) { |
684 |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); //[6] right-down neighbour in last frame |
685 |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); } |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
686 |
else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
687 |
} |
} |
688 |
|
|
689 |
static void |
static void |
696 |
const int y, |
const int y, |
697 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
698 |
const uint32_t iQuant, |
const uint32_t iQuant, |
699 |
const uint32_t iFcode, |
SearchData * const Data, |
700 |
const MBParam * const pParam, |
const MBParam * const pParam, |
701 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
702 |
const MACROBLOCK * const prevMBs, |
const MACROBLOCK * const prevMBs, |
704 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB) |
705 |
{ |
{ |
706 |
|
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
|
|
|
707 |
int i, iDirection = 255, mask, threshA; |
int i, iDirection = 255, mask, threshA; |
708 |
int32_t temp[5]; |
VECTOR pmv[7]; |
|
VECTOR currentMV[5], pmv[7]; |
|
|
int32_t psad[4], iMinSAD[5]; |
|
|
MainSearchFunc * MainSearchPtr; |
|
|
SearchData Data; |
|
|
|
|
|
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, psad); //has to be changed to get_pmv(2)() |
|
|
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, iFcode); |
|
709 |
|
|
710 |
Data.predMV = pmv[0]; |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
711 |
Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
712 |
Data.iEdgedWidth = iEdgedWidth; |
pParam->width, pParam->height, Data->iFcode); |
713 |
Data.currentMV = currentMV; |
|
714 |
Data.iMinSAD = iMinSAD; |
Data->predMV = pmv[0]; |
715 |
Data.Ref = pRef + (x + iEdgedWidth*y)*16; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
716 |
Data.RefH = pRefH + (x + iEdgedWidth*y) * 16; |
Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; |
717 |
Data.RefV = pRefV + (x + iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
718 |
Data.RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
719 |
Data.temp = temp; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
720 |
|
|
721 |
Data.iQuant = iQuant; |
Data->iQuant = iQuant; |
|
Data.iFcode = iFcode; |
|
722 |
|
|
723 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
724 |
Data.min_dx = EVEN(Data.min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
725 |
Data.max_dx = EVEN(Data.max_dx); |
Data->max_dx = EVEN(Data->max_dx); |
726 |
Data.min_dy = EVEN(Data.min_dy); |
Data->min_dy = EVEN(Data->min_dy); |
727 |
Data.max_dy = EVEN(Data.max_dy); } |
Data->max_dy = EVEN(Data->max_dy); } |
|
|
|
|
for(i = 0; i < 5; i++) currentMV[i].x = currentMV[i].y = 0; |
|
|
|
|
|
i = d_mv_bits(pmv[0].x, pmv[0].y, iFcode); |
|
|
|
|
|
iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
|
|
iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
|
|
iMinSAD[2] = pMB->sad8[1]; |
|
|
iMinSAD[3] = pMB->sad8[2]; |
|
|
iMinSAD[4] = pMB->sad8[3]; |
|
728 |
|
|
729 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
730 |
|
|
731 |
|
if (inter4v) CheckCandidate = CheckCandidate16; |
732 |
|
else CheckCandidate = CheckCandidate16no4v; |
733 |
|
|
734 |
|
for(i = 0; i < 5; i++) Data->currentMV[i].x = Data->currentMV[i].y = 0; |
735 |
|
|
736 |
|
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
737 |
|
Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
738 |
|
Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
739 |
|
Data->iMinSAD[2] = pMB->sad8[1]; |
740 |
|
Data->iMinSAD[3] = pMB->sad8[2]; |
741 |
|
Data->iMinSAD[4] = pMB->sad8[3]; |
742 |
|
|
743 |
if ((x == 0) && (y == 0)) threshA = 512; |
if ((x == 0) && (y == 0)) threshA = 512; |
744 |
else { |
else { |
745 |
threshA = psad[0] + 20; |
threshA = Data->temp[0]; // that's when we keep this SAD atm |
746 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
747 |
if (threshA > 1024) threshA = 1024; } |
if (threshA > 1024) threshA = 1024; } |
748 |
|
|
752 |
if (inter4v) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
753 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
754 |
|
|
755 |
|
|
756 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
757 |
|
|
758 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) { |
759 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
760 |
CheckCandidate16(pmv[i].x, pmv[i].y, mask, &iDirection, &Data); |
(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
761 |
if (iMinSAD[0] < threshA) break; |
if (Data->iMinSAD[0] <= threshA) break; |
762 |
} |
} |
763 |
|
|
764 |
if ((iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
765 |
(MVequal(currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
766 |
(iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { |
767 |
inter4v = 0; |
inter4v = 0; |
768 |
if (MotionFlags & PMV_QUICKSTOP16) goto PMVfast16_Terminate_without_Refine; |
} else { |
|
if (MotionFlags & PMV_EARLYSTOP16) { |
|
|
CheckCandidate = CheckCandidate16no4v; // I sure hope it's faster |
|
|
goto PMVfast16_Terminate_with_Refine; |
|
|
} |
|
|
} |
|
769 |
|
|
770 |
if (MotionFlags & PMV_USESQUARES16) |
MainSearchFunc * MainSearchPtr; |
771 |
MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; |
772 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
|
MainSearchPtr = AdvDiamondSearch; |
|
773 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
774 |
|
|
775 |
(*MainSearchPtr)(currentMV->x, currentMV->y, &Data, iDirection); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
776 |
|
|
777 |
/* extended search, diamond starting in 0,0 and in prediction. |
/* extended search, diamond starting in 0,0 and in prediction. |
778 |
note that this search is/might be done in halfpel positions, |
note that this search is/might be done in halfpel positions, |
780 |
|
|
781 |
if (MotionFlags & PMV_EXTSEARCH16) { |
if (MotionFlags & PMV_EXTSEARCH16) { |
782 |
int32_t bSAD; |
int32_t bSAD; |
783 |
VECTOR startMV = Data.predMV, backupMV = currentMV[0]; |
VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; |
784 |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
785 |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
786 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
787 |
bSAD = iMinSAD[0]; iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
788 |
|
|
789 |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, &Data); |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
790 |
(*MainSearchPtr)(startMV.x, startMV.y, &Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
791 |
if (bSAD < iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
792 |
currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
793 |
iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
794 |
} |
} |
795 |
|
|
796 |
backupMV = currentMV[0]; |
backupMV = Data->currentMV[0]; |
797 |
if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1; |
if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1; |
798 |
else startMV.x = startMV.y = 0; |
else startMV.x = startMV.y = 0; |
799 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
800 |
bSAD = iMinSAD[0]; iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
801 |
|
|
802 |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, &Data); |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
803 |
(*MainSearchPtr)(startMV.x, startMV.y, &Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
804 |
if (bSAD < iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
805 |
currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
806 |
iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
807 |
|
} |
808 |
} |
} |
809 |
} |
} |
810 |
|
|
811 |
PMVfast16_Terminate_with_Refine: |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
|
|
|
|
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data); |
|
|
|
|
|
PMVfast16_Terminate_without_Refine: |
|
812 |
|
|
813 |
if (inter4v) |
if (inter4v) { |
814 |
for(i = 0; i < 4; i++) |
SearchData Data8; |
815 |
Search8(&Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
Data8.iFcode = Data->iFcode; |
816 |
|
Data8.iQuant = Data->iQuant; |
817 |
|
Data8.iEdgedWidth = Data->iEdgedWidth; |
818 |
|
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
819 |
|
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
820 |
|
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
821 |
|
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
822 |
|
} |
823 |
|
|
824 |
if (!(inter4v) || |
if (!(inter4v) || |
825 |
(iMinSAD[0] < iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
826 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
827 |
// INTER MODE |
// INTER MODE |
828 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
829 |
pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
830 |
= pMB->mvs[2] = pMB->mvs[3] = currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
831 |
|
|
832 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
833 |
pMB->sad8[2] = pMB->sad8[3] = iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
834 |
|
|
835 |
pMB->pmvs[0].x = currentMV[0].x - Data.predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
836 |
pMB->pmvs[0].y = currentMV[0].y - Data.predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
837 |
} else { |
} else { |
838 |
// INTER4V MODE; all other things are already set in Search8 |
// INTER4V MODE; all other things are already set in Search8 |
839 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
840 |
pMB->sad16 = iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * iQuant; |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
841 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
842 |
} |
} |
843 |
|
|
844 |
} |
} |
850 |
const MBParam * const pParam, |
const MBParam * const pParam, |
851 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
852 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
853 |
const int block) |
const int block, |
854 |
|
SearchData * const Data) |
855 |
{ |
{ |
856 |
SearchData Data; |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
857 |
|
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
858 |
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->currentMV = OldData->currentMV + 1 + block; |
|
Data.iMinSAD = OldData->iMinSAD + 1 + block; |
|
|
Data.currentMV = OldData->currentMV+1+block; |
|
|
Data.iFcode = OldData->iFcode; |
|
|
Data.iQuant = OldData->iQuant; |
|
859 |
|
|
860 |
if (block != 0) |
if (block != 0) |
861 |
*(Data.iMinSAD) += lambda_vec8[Data.iQuant] * |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
862 |
d_mv_bits( Data.currentMV->x - Data.predMV.x, |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
863 |
Data.currentMV->y - Data.predMV.y, |
Data->currentMV->y - Data->predMV.y, |
864 |
Data.iFcode); |
Data->iFcode); |
|
|
|
865 |
|
|
866 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
867 |
|
|
868 |
Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
869 |
Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
870 |
Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
871 |
Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
872 |
|
|
873 |
Data.iEdgedWidth = pParam->edged_width; |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
874 |
|
|
875 |
Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
|
|
|
|
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8, |
|
876 |
pParam->width, pParam->height, OldData->iFcode); |
pParam->width, pParam->height, OldData->iFcode); |
877 |
|
|
878 |
CheckCandidate = CheckCandidate8; |
CheckCandidate = CheckCandidate8; |
884 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
885 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
886 |
|
|
887 |
(*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255); } |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); } |
888 |
|
|
889 |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data); |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(Data); |
890 |
} |
} |
891 |
|
|
892 |
pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
893 |
pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
894 |
pMB->mvs[block] = *(Data.currentMV); |
pMB->mvs[block] = *(Data->currentMV); |
895 |
pMB->sad8[block] = 4 * (*(Data.iMinSAD)); |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
896 |
|
|
897 |
} |
} |
898 |
|
|
899 |
/* B-frames code starts here */ |
/* B-frames code starts here */ |
921 |
pmv[2] = ChoosePred(pMB, mode_curr); |
pmv[2] = ChoosePred(pMB, mode_curr); |
922 |
pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); |
pmv[2].x = EVEN(pmv[2].x); pmv[2].y = EVEN(pmv[2].y); |
923 |
|
|
|
pmv[3].x = pmv[3].y = 0; |
|
924 |
if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour |
if ((y != 0)&&(x != (int)(iWcount+1))) { // [3] top-right neighbour |
925 |
pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); |
pmv[3] = ChoosePred(pMB+1-iWcount, mode_curr); |
926 |
pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); } |
pmv[3].x = EVEN(pmv[3].x); pmv[3].y = EVEN(pmv[3].y); |
927 |
|
} else pmv[3].x = pmv[3].y = 0; |
928 |
|
|
929 |
if (y != 0) { |
if (y != 0) { |
930 |
pmv[4] = ChoosePred(pMB-iWcount, mode_curr); |
pmv[4] = ChoosePred(pMB-iWcount, mode_curr); |
954 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
955 |
const int x, const int y, |
const int x, const int y, |
956 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
957 |
const uint32_t iFcode, |
const uint32_t iFcode, |
958 |
const MBParam * const pParam, |
const MBParam * const pParam, |
959 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
960 |
const VECTOR * const predMV, |
const VECTOR * const predMV, |
961 |
int32_t * const best_sad, |
int32_t * const best_sad, |
962 |
const int32_t mode_current) |
const int32_t mode_current, |
963 |
|
SearchData * const Data) |
964 |
{ |
{ |
965 |
|
|
966 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
967 |
|
|
968 |
int i, iDirection, mask; |
int i, iDirection, mask; |
969 |
VECTOR currentMV, pmv[7]; |
VECTOR pmv[7]; |
970 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
971 |
int32_t iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
972 |
SearchData Data; |
Data->iFcode = iFcode; |
973 |
|
|
974 |
Data.iMinSAD = &iMinSAD; |
Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
975 |
Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; |
976 |
Data.iEdgedWidth = iEdgedWidth; |
Data->RefV = pRefV + (x + y * iEdgedWidth) * 16; |
977 |
Data.currentMV = ¤tMV; |
Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16; |
|
Data.iMinSAD = &iMinSAD; |
|
|
Data.Ref = pRef + (x + y * iEdgedWidth) * 16; |
|
|
Data.RefH = pRefH + (x + y * iEdgedWidth) * 16; |
|
|
Data.RefV = pRefV + (x + y * iEdgedWidth) * 16; |
|
|
Data.RefHV = pRefHV + (x + y * iEdgedWidth) * 16; |
|
978 |
|
|
979 |
Data.iQuant = iQuant; |
Data->predMV = *predMV; |
|
Data.iFcode = iFcode; |
|
|
Data.predMV = *predMV; |
|
980 |
|
|
981 |
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
982 |
pParam->width, pParam->height, iFcode); |
pParam->width, pParam->height, iFcode); |
983 |
|
|
984 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
pmv[0] = Data->predMV; |
985 |
Data.min_dx = EVEN(Data.min_dx); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
|
Data.max_dx = EVEN(Data.max_dx); |
|
|
Data.min_dy = EVEN(Data.min_dy); |
|
|
Data.max_dy = EVEN(Data.max_dy); } // no-halpel and b-frames. do we need it? |
|
|
|
|
986 |
|
|
987 |
pmv[0] = Data.predMV; |
Data->currentMV->x = Data->currentMV->y = 0; |
|
PreparePredictionsBF(pmv, x, y, pParam->mb_width, |
|
|
pMB, mode_current); |
|
|
|
|
|
currentMV.x = currentMV.y = 0; |
|
988 |
|
|
989 |
CheckCandidate = CheckCandidate16no4v; |
CheckCandidate = CheckCandidate16no4v; |
990 |
|
|
991 |
// main loop. checking all predictions |
// main loop. checking all predictions |
992 |
for (i = 0; i < 8; i++) { |
for (i = 0; i < 8; i++) { |
993 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
994 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, &Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
995 |
} |
} |
996 |
|
|
997 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) |
1000 |
MainSearchPtr = AdvDiamondSearch; |
MainSearchPtr = AdvDiamondSearch; |
1001 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1002 |
|
|
1003 |
(*MainSearchPtr)(currentMV.x, currentMV.y, &Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1004 |
|
|
1005 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data); |
HalfpelRefine(Data); |
1006 |
|
|
1007 |
// three bits are needed to code backward mode. four for forward |
// three bits are needed to code backward mode. four for forward |
1008 |
// we treat the bits just like they were vector's |
// we treat the bits just like they were vector's |
1009 |
if (mode_current == MODE_FORWARD) iMinSAD += 4 * lambda_vec16[iQuant]; |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant]; |
1010 |
else iMinSAD += 3 * lambda_vec16[iQuant]; |
else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant]; |
1011 |
|
|
1012 |
|
|
1013 |
if (iMinSAD < *best_sad) { |
if (*Data->iMinSAD < *best_sad) { |
1014 |
*best_sad = iMinSAD; |
*best_sad = *Data->iMinSAD; |
1015 |
pMB->mode = mode_current; |
pMB->mode = mode_current; |
1016 |
pMB->pmvs[0].x = currentMV.x - predMV->x; |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1017 |
pMB->pmvs[0].y = currentMV.y - predMV->y; |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
1018 |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = currentMV; |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
1019 |
else pMB->b_mvs[0] = currentMV; |
else pMB->b_mvs[0] = *Data->currentMV; |
1020 |
} |
} |
1021 |
|
|
1022 |
} |
} |
1023 |
|
|
1024 |
static int32_t |
static int32_t |
1025 |
SearchDirect(const uint8_t * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
1026 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1027 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1028 |
const uint8_t * const f_RefHV, |
const uint8_t * const f_RefHV, |
1029 |
const uint8_t * const b_Ref, |
const IMAGE * const b_Ref, |
1030 |
const uint8_t * const b_RefH, |
const uint8_t * const b_RefH, |
1031 |
const uint8_t * const b_RefV, |
const uint8_t * const b_RefV, |
1032 |
const uint8_t * const b_RefHV, |
const uint8_t * const b_RefHV, |
1033 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1034 |
const int x, const int y, |
const int x, const int y, |
1035 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
1036 |
const int32_t TRB, const int32_t TRD, |
const int32_t TRB, const int32_t TRD, |
1037 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1038 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1039 |
const MACROBLOCK * const b_mb, |
const MACROBLOCK * const b_mb, |
1040 |
int32_t * const best_sad) |
int32_t * const best_sad, |
1041 |
|
SearchData * const Data) |
1042 |
|
|
1043 |
{ |
{ |
1044 |
const uint32_t iEdgedWidth = pParam->edged_width; |
int32_t skip_sad; |
|
int32_t iMinSAD = 0, skip_sad; |
|
1045 |
int k; |
int k; |
1046 |
VECTOR currentMV; |
|
1047 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
|
SearchData Data; |
|
1048 |
|
|
1049 |
Data.iMinSAD = &iMinSAD; |
*Data->iMinSAD = 256*4096; |
1050 |
Data.Cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
Data->referencemv = b_mb->mvs; |
|
Data.iEdgedWidth = iEdgedWidth; |
|
|
Data.currentMV = ¤tMV; |
|
|
Data.iQuant = iQuant; |
|
|
Data.referencemv = b_mb->mvs; |
|
1051 |
|
|
1052 |
Data.Ref= f_Ref + (x + iEdgedWidth*y) * 16; |
Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1053 |
Data.RefH = f_RefH + (x + iEdgedWidth*y) * 16; |
Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; |
1054 |
Data.RefV = f_RefV + (x + iEdgedWidth*y) * 16; |
Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16; |
1055 |
Data.RefHV = f_RefHV + (x + iEdgedWidth*y) * 16; |
Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16; |
1056 |
Data.bRef = b_Ref + (x + iEdgedWidth*y) * 16; |
Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1057 |
Data.bRefH = b_RefH + (x + iEdgedWidth*y) * 16; |
Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16; |
1058 |
Data.bRefV = b_RefV + (x + iEdgedWidth*y) * 16; |
Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16; |
1059 |
Data.bRefHV = b_RefHV + (x + iEdgedWidth*y) * 16; |
Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16; |
1060 |
/* |
|
1061 |
//What we do here is a complicated version of CheckCandidateDirect(0,0); |
Data->max_dx = 2 * pParam->width - 2 * (x) * 16; |
1062 |
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, pParam->width, pParam->height, 19); |
Data->max_dy = 2 * pParam->height - 2 * (y) * 16; |
1063 |
|
Data->min_dx = -(2 * 16 + 2 * (x) * 16); |
1064 |
*/ |
Data->min_dy = -(2 * 16 + 2 * (y) * 16); |
|
Data.max_dx = 2 * pParam->width - 2 * (x) * 16; |
|
|
Data.max_dy = 2 * pParam->height - 2 * (y) * 16; |
|
|
Data.min_dx = -(2 * 16 + 2 * (x) * 16); |
|
|
Data.min_dy = -(2 * 16 + 2 * (y) * 16); |
|
1065 |
|
|
1066 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1067 |
pMB->mvs[k].x = Data.directmvF[k].x = ((TRB * Data.referencemv[k].x) / TRD); |
pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
1068 |
pMB->b_mvs[k].x = Data.directmvB[k].x = ((TRB - TRD) * Data.referencemv[k].x) / TRD; |
pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD; |
1069 |
pMB->mvs[k].y = Data.directmvF[k].y = ((TRB * Data.referencemv[k].y) / TRD); |
pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD); |
1070 |
pMB->b_mvs[k].y = Data.directmvB[k].y = ((TRB - TRD) * Data.referencemv[k].y) / TRD; |
pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD; |
|
|
|
|
if (( pMB->mvs[k].x > Data.max_dx ) || ( pMB->mvs[k].x < Data.min_dx ) |
|
|
|| ( pMB->mvs[k].y > Data.max_dy ) || ( pMB->mvs[k].y < Data.min_dy ) |
|
|
|| ( pMB->b_mvs[k].x > Data.max_dx ) || ( pMB->b_mvs[k].x < Data.min_dx ) |
|
|
|| ( pMB->b_mvs[k].y > Data.max_dy ) || ( pMB->b_mvs[k].y < Data.min_dy )) { |
|
|
/* |
|
|
fprintf(debug, "\nERROR - out of range : vector %d,%d and %d,%d\n", pMB->mvs[k].x, pMB->mvs[k].y,pMB->b_mvs[k].x,pMB->b_mvs[k].y ); |
|
|
fprintf(debug, " range is x: %d..%d y: %d..%d \n", Data.min_dx,Data.max_dx,Data.min_dy,Data.max_dy); |
|
|
fprintf(debug,"macroblock %d, %d \n", x, y); |
|
|
fprintf(debug, "direct MV is %d,%d \n", directmv[k].x, directmv[k].y); |
|
|
*/ |
|
|
*best_sad = 256*4096; // in that case, we won't use direct mode |
|
|
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
|
|
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; /* because backwards and interpol might rely on this */ |
|
|
return 0; } |
|
1071 |
|
|
1072 |
|
if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx ) |
1073 |
|
|| ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) { |
1074 |
|
|
1075 |
|
*best_sad = 256*4096; // in that case, we won't use direct mode |
1076 |
|
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
1077 |
|
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
1078 |
|
return 0; |
1079 |
|
} |
1080 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1081 |
iMinSAD = sad16bi(Data.Cur, |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
1082 |
get_ref_mv(f_Ref, f_RefH, f_RefV, f_RefHV, |
pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0]; |
1083 |
x, y, 16, &pMB->mvs[0], iEdgedWidth), |
Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0]; |
1084 |
get_ref_mv(b_Ref, b_RefH, b_RefV, b_RefHV, |
Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0]; |
|
x, y, 16, &pMB->b_mvs[0], iEdgedWidth), iEdgedWidth); |
|
|
|
|
|
Data.directmvF[1] = Data.directmvF[2] = Data.directmvF[3] = Data.directmvF[0]; |
|
|
Data.directmvB[1] = Data.directmvB[2] = Data.directmvB[3] = Data.directmvB[0]; |
|
1085 |
break; |
break; |
1086 |
} |
} |
|
iMinSAD += sad8bi(Data.Cur + (k&1)*8 + (k>>1)* 8 * iEdgedWidth, |
|
|
get_ref_mv(f_Ref, f_RefH, f_RefV, f_RefHV, |
|
|
(2*x+(k&1)), (2*y+(k>>1)), 8, &pMB->mvs[k], iEdgedWidth), |
|
|
get_ref_mv(b_Ref, b_RefH, b_RefV, b_RefHV, |
|
|
(2*x+(k&1)), (2*y+(k>>1)), 8, &pMB->b_mvs[k], iEdgedWidth), |
|
|
iEdgedWidth); |
|
1087 |
} |
} |
1088 |
|
|
|
// skip decision |
|
|
if (iMinSAD < (int32_t)iQuant * SKIP_THRESH_B) { |
|
|
pMB->mode = MODE_DIRECT_NONE_MV; |
|
|
return iMinSAD; } |
|
|
|
|
|
skip_sad = iMinSAD; |
|
|
iMinSAD += 2 * lambda_vec16[iQuant]; // 2 bits needed to code vector 0,0 |
|
|
currentMV.x = currentMV.y = 0; |
|
1089 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1090 |
CheckCandidate = CheckCandidateDirect; |
CheckCandidate = CheckCandidateDirect; |
1091 |
else CheckCandidate = CheckCandidateDirectno4v; |
else CheckCandidate = CheckCandidateDirectno4v; |
1092 |
|
|
1093 |
|
(*CheckCandidate)(0, 0, 255, &k, Data); |
1094 |
|
|
1095 |
|
// skip decision |
1096 |
|
if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
1097 |
|
//checking chroma. everything copied from MC |
1098 |
|
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1099 |
|
int sum, dx, dy, b_dx, b_dy; |
1100 |
|
|
1101 |
|
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1102 |
|
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1103 |
|
|
1104 |
|
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1105 |
|
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1106 |
|
|
1107 |
|
sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
1108 |
|
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1109 |
|
|
1110 |
|
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
1111 |
|
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1112 |
|
|
1113 |
|
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1114 |
|
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1115 |
|
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1116 |
|
Data->iEdgedWidth/2); |
1117 |
|
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
1118 |
|
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1119 |
|
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1120 |
|
Data->iEdgedWidth/2); |
1121 |
|
|
1122 |
|
if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) { |
1123 |
|
pMB->mode = MODE_DIRECT_NONE_MV; |
1124 |
|
return *Data->iMinSAD; |
1125 |
|
} |
1126 |
|
} |
1127 |
|
|
1128 |
|
skip_sad = *Data->iMinSAD; |
1129 |
|
|
1130 |
// DIRECT MODE DELTA VECTOR SEARCH. |
// DIRECT MODE DELTA VECTOR SEARCH. |
1131 |
// This has to be made more effective, but at the moment I'm happy it's running at all |
// This has to be made more effective, but at the moment I'm happy it's running at all |
1132 |
|
|
1134 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1135 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1136 |
|
|
1137 |
(*MainSearchPtr)(0, 0, &Data, 255); |
(*MainSearchPtr)(0, 0, Data, 255); |
1138 |
|
|
1139 |
HalfpelRefine(&Data); |
HalfpelRefine(Data); |
1140 |
|
|
1141 |
iMinSAD += 1 * lambda_vec16[iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's |
*Data->iMinSAD += 1 * lambda_vec16[Data->iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's |
1142 |
*best_sad = iMinSAD; |
*best_sad = *Data->iMinSAD; |
1143 |
|
|
1144 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1145 |
pMB->mode = MODE_DIRECT; |
pMB->mode = MODE_DIRECT; |
1146 |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
1147 |
|
|
1148 |
pMB->pmvs[3] = currentMV; |
pMB->pmvs[3] = *Data->currentMV; |
1149 |
|
|
1150 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1151 |
pMB->mvs[k].x = Data.directmvF[k].x + currentMV.x; |
pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x; |
1152 |
pMB->b_mvs[k].x = ((currentMV.x == 0) |
pMB->b_mvs[k].x = ((Data->currentMV->x == 0) |
1153 |
? Data.directmvB[k].x |
? Data->directmvB[k].x |
1154 |
: pMB->mvs[k].x - Data.referencemv[k].x); |
: pMB->mvs[k].x - Data->referencemv[k].x); |
1155 |
pMB->mvs[k].y = (Data.directmvF[k].y + currentMV.y); |
pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y); |
1156 |
pMB->b_mvs[k].y = ((currentMV.y == 0) |
pMB->b_mvs[k].y = ((Data->currentMV->y == 0) |
1157 |
? Data.directmvB[k].y |
? Data->directmvB[k].y |
1158 |
: pMB->mvs[k].y - Data.referencemv[k].y); |
: pMB->mvs[k].y - Data->referencemv[k].y); |
1159 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1160 |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
1161 |
pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; |
pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; |
1162 |
break; |
break; |
1163 |
} |
} |
1164 |
} |
} |
1165 |
return 0;//skip_sad; |
return skip_sad; |
1166 |
} |
} |
1167 |
|
|
1168 |
|
|
1169 |
static __inline void |
static __inline void |
1170 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const uint8_t * const f_Ref, |
1171 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1180 |
const uint32_t fcode, |
const uint32_t fcode, |
1181 |
const uint32_t bcode, |
const uint32_t bcode, |
1182 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
1183 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1184 |
const VECTOR * const f_predMV, |
const VECTOR * const f_predMV, |
1185 |
const VECTOR * const b_predMV, |
const VECTOR * const b_predMV, |
1186 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1187 |
int32_t * const best_sad) |
int32_t * const best_sad, |
1188 |
|
SearchData * const fData) |
1189 |
|
|
1190 |
{ |
{ |
|
/* Interpolated MC motion vector search, this is tedious and more complicated because there are |
|
|
two values for everything, always one for backward and one for forward ME. Still, we don't gain |
|
|
much from this search, maybe it should simply be skipped and simply current i_sad16 value used |
|
|
as "optimal". */ |
|
1191 |
|
|
1192 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1193 |
|
|
1194 |
int iDirection, i, j; |
int iDirection, i, j; |
1195 |
int32_t iMinSAD = 256*4096; |
SearchData bData; |
|
VECTOR currentMV[3]; |
|
|
SearchData fData, bData; |
|
|
|
|
|
|
|
|
fData.iMinSAD = bData.iMinSAD = &iMinSAD; |
|
|
|
|
|
fData.Cur = bData.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
|
|
fData.iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
|
|
fData.currentMV = currentMV; bData.currentMV = currentMV + 1; |
|
|
fData.iQuant = bData.iQuant = iQuant; |
|
|
fData.iFcode = bData.bFcode = fcode; fData.bFcode = bData.iFcode = bcode; |
|
|
|
|
|
|
|
|
bData.bRef = fData.Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
|
|
bData.bRefH = fData.RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
|
|
bData.bRefV = fData.RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
|
|
bData.bRefHV = fData.RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
|
|
bData.Ref = fData.bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
|
|
bData.RefH = fData.bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
|
|
bData.RefV = fData.bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
|
|
bData.RefHV = fData.bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
|
1196 |
|
|
1197 |
bData.bpredMV = fData.predMV = *f_predMV; |
bData.iMinSAD = fData->iMinSAD; |
1198 |
fData.bpredMV = bData.predMV = *b_predMV; |
*bData.iMinSAD = 4096*256; |
1199 |
|
bData.Cur = fData->Cur; |
1200 |
|
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
1201 |
currentMV[0] = pMB->mvs[0]; |
bData.currentMV = fData->currentMV + 1; |
1202 |
currentMV[1] = pMB->b_mvs[0]; |
bData.iQuant = fData->iQuant; |
1203 |
get_range(&fData.min_dx, &fData.max_dx, &fData.min_dy, &fData.max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1204 |
|
|
1205 |
|
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
1206 |
|
bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
1207 |
|
bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
1208 |
|
bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
1209 |
|
bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
1210 |
|
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
1211 |
|
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
1212 |
|
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
1213 |
|
|
1214 |
|
bData.bpredMV = fData->predMV = *f_predMV; |
1215 |
|
fData->bpredMV = bData.predMV = *b_predMV; |
1216 |
|
|
1217 |
|
fData->currentMV[0] = pMB->mvs[0]; |
1218 |
|
fData->currentMV[1] = pMB->b_mvs[0]; |
1219 |
|
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
1220 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode); |
1221 |
|
|
1222 |
CheckCandidateInt(currentMV[0].x, currentMV[0].y, 255, &iDirection, &fData); |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1223 |
|
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
1224 |
|
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; |
1225 |
|
if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
1226 |
|
|
1227 |
|
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1228 |
|
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; |
1229 |
|
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; |
1230 |
|
if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
1231 |
|
|
1232 |
|
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1233 |
|
|
1234 |
//diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
//diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
1235 |
|
|
1236 |
do { |
do { |
1237 |
iDirection = 255; |
iDirection = 255; |
1238 |
// forward MV moves |
// forward MV moves |
1239 |
i = currentMV[0].x; j = currentMV[0].y; |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
1240 |
|
|
1241 |
CheckCandidateInt(i + 2, j, 0, &iDirection, &fData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
1242 |
CheckCandidateInt(i, j + 2, 0, &iDirection, &fData); |
CheckCandidateInt(i, j + 1, 0, &iDirection, fData); |
1243 |
CheckCandidateInt(i - 2, j, 0, &iDirection, &fData); |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
1244 |
CheckCandidateInt(i, j - 2, 0, &iDirection, &fData); |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
1245 |
|
|
1246 |
// backward MV moves |
// backward MV moves |
1247 |
i = currentMV[1].x; j = currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
1248 |
currentMV[2] = currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
1249 |
|
|
1250 |
CheckCandidateInt(i + 2, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
1251 |
CheckCandidateInt(i, j + 2, 0, &iDirection, &bData); |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
1252 |
CheckCandidateInt(i - 2, j, 0, &iDirection, &bData); |
CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); |
1253 |
CheckCandidateInt(i, j - 2, 0, &iDirection, &bData); |
CheckCandidateInt(i, j - 1, 0, &iDirection, &bData); |
1254 |
|
|
1255 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1256 |
|
|
|
/* halfpel refinement. luckly we can use normal halfpel function for it */ |
|
|
|
|
|
if (MotionFlags & PMV_HALFPELREFINE16) { |
|
|
CheckCandidate = CheckCandidateInt; |
|
|
HalfpelRefine(&fData); |
|
|
currentMV[2] = currentMV[0]; |
|
|
HalfpelRefine(&bData); |
|
|
} |
|
|
|
|
1257 |
// two bits are needed to code interpolate mode. we treat the bits just like they were vector's |
// two bits are needed to code interpolate mode. we treat the bits just like they were vector's |
1258 |
iMinSAD += 2 * lambda_vec16[iQuant]; |
*fData->iMinSAD += 2 * lambda_vec16[fData->iQuant]; |
1259 |
if (iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1260 |
*best_sad = iMinSAD; |
*best_sad = *fData->iMinSAD; |
1261 |
pMB->mvs[0] = currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1262 |
pMB->b_mvs[0] = currentMV[1]; |
pMB->b_mvs[0] = fData->currentMV[1]; |
1263 |
pMB->mode = MODE_INTERPOLATE; |
pMB->mode = MODE_INTERPOLATE; |
1264 |
|
|
1265 |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
1269 |
} |
} |
1270 |
} |
} |
1271 |
|
|
1272 |
|
|
1273 |
void |
void |
1274 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
1275 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
1298 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
1299 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
1300 |
|
|
1301 |
|
// some pre-inintialized data for the rest of the search |
1302 |
|
|
1303 |
|
SearchData Data; |
1304 |
|
int32_t iMinSAD; |
1305 |
|
VECTOR currentMV[3]; |
1306 |
|
Data.iEdgedWidth = pParam->edged_width; |
1307 |
|
Data.currentMV = currentMV; |
1308 |
|
Data.iMinSAD = &iMinSAD; |
1309 |
|
Data.iQuant = frame->quant; |
1310 |
|
|
1311 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
1312 |
|
|
1313 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
1324 |
continue; |
continue; |
1325 |
} |
} |
1326 |
|
|
1327 |
|
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
1328 |
/* direct search comes first, because it (1) checks for SKIP-mode |
/* direct search comes first, because it (1) checks for SKIP-mode |
1329 |
and (2) sets very good predictions for forward and backward search */ |
and (2) sets very good predictions for forward and backward search */ |
1330 |
|
|
1331 |
skip_sad = SearchDirect(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
skip_sad = SearchDirect(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
1332 |
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
1333 |
&frame->image, |
&frame->image, |
1334 |
i, j, |
i, j, |
1335 |
frame->motion_flags, |
frame->motion_flags, |
|
frame->quant, |
|
1336 |
TRB, TRD, |
TRB, TRD, |
1337 |
pParam, |
pParam, |
1338 |
pMB, b_mb, |
pMB, b_mb, |
1339 |
&best_sad); |
&best_sad, |
1340 |
|
&Data); |
1341 |
|
|
|
if (!(frame->global_flags & XVID_HALFPEL)) best_sad = skip_sad = 256*4096; |
|
|
else |
|
1342 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
1343 |
|
|
1344 |
// best_sad = 256*4096; //uncomment to disable Directsearch. |
// best_sad = 256*4096; //uncomment to disable Directsearch. |
1348 |
SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
1349 |
&frame->image, i, j, |
&frame->image, i, j, |
1350 |
frame->motion_flags, |
frame->motion_flags, |
1351 |
frame->quant, frame->fcode, pParam, |
frame->fcode, pParam, |
1352 |
pMB, &f_predMV, &best_sad, |
pMB, &f_predMV, &best_sad, |
1353 |
MODE_FORWARD); |
MODE_FORWARD, &Data); |
1354 |
|
|
1355 |
// backward search |
// backward search |
1356 |
SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
1357 |
&frame->image, i, j, |
&frame->image, i, j, |
1358 |
frame->motion_flags, |
frame->motion_flags, |
1359 |
frame->quant, frame->bcode, pParam, |
frame->bcode, pParam, |
1360 |
pMB, &b_predMV, &best_sad, |
pMB, &b_predMV, &best_sad, |
1361 |
MODE_BACKWARD); |
MODE_BACKWARD, &Data); |
1362 |
|
|
1363 |
// interpolate search comes last, because it uses data from forward and backward as prediction |
// interpolate search comes last, because it uses data from forward and backward as prediction |
1364 |
|
|
1368 |
i, j, |
i, j, |
1369 |
frame->fcode, frame->bcode, |
frame->fcode, frame->bcode, |
1370 |
frame->motion_flags, |
frame->motion_flags, |
1371 |
frame->quant, pParam, |
pParam, |
1372 |
&f_predMV, &b_predMV, |
&f_predMV, &b_predMV, |
1373 |
pMB, &best_sad); |
pMB, &best_sad, |
1374 |
|
&Data); |
1375 |
|
|
1376 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1377 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1470 |
const int y, |
const int y, |
1471 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1472 |
const uint32_t iQuant, |
const uint32_t iQuant, |
|
const uint32_t iFcode, |
|
1473 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1474 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1475 |
int inter4v, |
int inter4v, |
1476 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB, |
1477 |
|
SearchData * const Data) |
1478 |
{ |
{ |
1479 |
|
|
1480 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1481 |
|
|
1482 |
int i; |
int i, t; |
|
VECTOR currentMV[5]; |
|
|
int32_t iMinSAD[5]; |
|
|
int32_t temp[5]; |
|
1483 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
|
SearchData Data; |
|
|
|
|
|
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, iFcode); |
|
1484 |
|
|
1485 |
Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1486 |
Data.iEdgedWidth = iEdgedWidth; |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1487 |
Data.currentMV = currentMV; |
pParam->width, pParam->height, Data->iFcode); |
1488 |
Data.iMinSAD = iMinSAD; |
|
1489 |
Data.Ref = pRef + (x + iEdgedWidth*y)*16; |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
1490 |
Data.RefH = pRefH + (x + iEdgedWidth*y) * 16; |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
1491 |
Data.RefV = pRefV + (x + iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
1492 |
Data.RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
1493 |
Data.temp = temp; |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
1494 |
Data.iQuant = iQuant; |
Data->iQuant = iQuant; |
|
Data.iFcode = iFcode; |
|
1495 |
|
|
1496 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1497 |
Data.min_dx = EVEN(Data.min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
1498 |
Data.max_dx = EVEN(Data.max_dx); |
Data->max_dx = EVEN(Data->max_dx); |
1499 |
Data.min_dy = EVEN(Data.min_dy); |
Data->min_dy = EVEN(Data->min_dy); |
1500 |
Data.max_dy = EVEN(Data.max_dy); |
Data->max_dy = EVEN(Data->max_dy); |
1501 |
} |
} |
1502 |
|
|
1503 |
for(i = 0; i < 5; i++) iMinSAD[i] = MV_MAX_ERROR; |
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
1504 |
|
|
1505 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1506 |
|
|
1511 |
|
|
1512 |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
1513 |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
1514 |
if (pMB->mvs[0].x > Data.max_dx) pMB->mvs[0].x = Data.max_dx; // this is in case iFcode changed |
if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed |
1515 |
if (pMB->mvs[0].x < Data.min_dx) pMB->mvs[0].x = Data.min_dx; |
if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx; |
1516 |
if (pMB->mvs[0].y > Data.max_dy) pMB->mvs[0].y = Data.max_dy; |
if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy; |
1517 |
if (pMB->mvs[0].y < Data.min_dy) pMB->mvs[0].y = Data.min_dy; |
if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy; |
1518 |
|
|
1519 |
CheckCandidate16(pMB->mvs[0].x, pMB->mvs[0].y, 0, &i, &Data); |
(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data); |
1520 |
|
|
1521 |
if (pMB->mode == MODE_INTER4V) |
if (pMB->mode == MODE_INTER4V) |
1522 |
for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search |
for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search |
1523 |
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
1524 |
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
1525 |
if (!(make_mask(pMB->mvs, i))) |
if (!(make_mask(pMB->mvs, i))) |
1526 |
CheckCandidate16(pMB->mvs[i].x, pMB->mvs[i].y, 0, &i, &Data); |
(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data); |
1527 |
} |
} |
1528 |
|
|
1529 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) |
1532 |
MainSearchPtr = AdvDiamondSearch; |
MainSearchPtr = AdvDiamondSearch; |
1533 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1534 |
|
|
1535 |
(*MainSearchPtr)(currentMV->x, currentMV->y, &Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1536 |
|
|
1537 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1538 |
|
|
1539 |
if (inter4v) |
if (inter4v) |
1540 |
for(i = 0; i < 4; i++) |
for(i = 0; i < 4; i++) |
1541 |
Search8hinted(&Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
1542 |
|
|
1543 |
if (!(inter4v) || |
if (!(inter4v) || |
1544 |
(iMinSAD[0] < iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
1545 |
|
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1546 |
// INTER MODE |
// INTER MODE |
1547 |
|
|
1548 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
1549 |
pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1550 |
= pMB->mvs[2] = pMB->mvs[3] = currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1551 |
|
|
1552 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1553 |
pMB->sad8[2] = pMB->sad8[3] = iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1554 |
|
|
1555 |
pMB->pmvs[0].x = currentMV[0].x - Data.predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1556 |
pMB->pmvs[0].y = currentMV[0].y - Data.predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1557 |
} else { |
} else { |
1558 |
// INTER4V MODE; all other things are already set in Search8hinted |
// INTER4V MODE; all other things are already set in Search8hinted |
1559 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1560 |
pMB->sad16 = iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * iQuant; |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] |
1561 |
|
+ Data->iMinSAD[4] + IMV16X16 * iQuant; |
1562 |
} |
} |
1563 |
|
|
1564 |
} |
} |
1576 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
1577 |
|
|
1578 |
uint32_t x, y; |
uint32_t x, y; |
1579 |
|
int32_t temp[5], quant = current->quant; |
1580 |
|
int32_t iMinSAD[5]; |
1581 |
|
VECTOR currentMV[5]; |
1582 |
|
SearchData Data; |
1583 |
|
Data.iEdgedWidth = pParam->edged_width; |
1584 |
|
Data.currentMV = currentMV; |
1585 |
|
Data.iMinSAD = iMinSAD; |
1586 |
|
Data.temp = temp; |
1587 |
|
Data.iFcode = current->fcode; |
1588 |
|
|
1589 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1590 |
|
|
1591 |
for (y = 0; y < pParam->mb_height; y++) { |
for (y = 0; y < pParam->mb_height; y++) { |
1592 |
for (x = 0; x < pParam->mb_width; x++) { |
for (x = 0; x < pParam->mb_width; x++) { |
|
int32_t sad00; |
|
1593 |
|
|
1594 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
1595 |
|
|
1596 |
//intra mode is copied from the first pass. At least for the time being |
//intra mode is copied from the first pass. At least for the time being |
1597 |
if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; |
if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; |
1598 |
|
|
1599 |
|
|
1600 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
1601 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
1602 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; } |
1603 |
|
else |
1604 |
|
if (pMB->dquant != NO_CHANGE) { |
1605 |
|
quant += DQtab[pMB->dquant]; |
1606 |
|
if (quant > 31) quant = 31; |
1607 |
|
else if (quant < 1) quant = 1; |
1608 |
|
pMB->quant = quant; |
1609 |
|
} |
1610 |
|
|
1611 |
if (pMB->dquant == NO_CHANGE) //no skip otherwise, anyway |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1612 |
sad00 = pMB->sad16 |
y, current->motion_flags, pMB->quant, |
1613 |
= sad16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
1614 |
pRef->y + (x + y * pParam->edged_width) * 16, |
&Data); |
1615 |
pParam->edged_width, 256*4096 ); |
|
1616 |
else sad00 = 256*4096; |
} |
1617 |
|
} |
1618 |
|
} |
1619 |
|
|
1620 |
|
static __inline int |
1621 |
|
MEanalyzeMB ( const uint8_t * const pRef, |
1622 |
|
const uint8_t * const pCur, |
1623 |
|
const int x, |
1624 |
|
const int y, |
1625 |
|
const MBParam * const pParam, |
1626 |
|
const MACROBLOCK * const pMBs, |
1627 |
|
MACROBLOCK * const pMB, |
1628 |
|
SearchData * const Data) |
1629 |
|
{ |
1630 |
|
|
1631 |
|
int i, mask; |
1632 |
|
VECTOR pmv[3]; |
1633 |
|
|
1634 |
//initial skip decision |
*(Data->iMinSAD) = MV_MAX_ERROR; |
1635 |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1636 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1637 |
|
pParam->width, pParam->height, Data->iFcode); |
1638 |
|
|
1639 |
if ( (pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant) |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
1640 |
&& ( //(pMB->mode == MODE_NOT_CODED) || |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
1641 |
(SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant) )) ) { |
|
1642 |
if (sad00 < pMB->quant * INITIAL_SKIP_THRESH) { |
CheckCandidate = CheckCandidate16no4vI; |
1643 |
SkipMacroblockP(pMB, sad00); |
|
1644 |
continue; } //skipped |
pmv[1].x = EVEN(pMB->mvs[0].x); |
1645 |
|
pmv[1].y = EVEN(pMB->mvs[0].y); |
1646 |
|
pmv[0].x = EVEN(Data->predMV.x); |
1647 |
|
pmv[0].y = EVEN(Data->predMV.y); |
1648 |
|
pmv[2].x = pmv[2].y = 0; |
1649 |
|
|
1650 |
|
CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data); |
1651 |
|
if (!(mask = make_mask(pmv, 1))) |
1652 |
|
CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data); |
1653 |
|
if (!(mask = make_mask(pmv, 2))) |
1654 |
|
CheckCandidate16no4vI(0, 0, mask, &i, Data); |
1655 |
|
|
1656 |
|
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1657 |
|
|
1658 |
|
pMB->mvs[0] = pMB->mvs[1] |
1659 |
|
= pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv() |
1660 |
|
|
1661 |
|
return *(Data->iMinSAD); |
1662 |
} |
} |
|
else sad00 = 256*4096; |
|
1663 |
|
|
1664 |
if (pMB->mode == MODE_NOT_CODED) |
#define INTRA_THRESH 1350 |
1665 |
SearchP( pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
#define INTER_THRESH 900 |
|
y, current->motion_flags, pMB->quant, |
|
|
current->fcode, pParam, pMBs, reference->mbs, |
|
|
current->global_flags & XVID_INTER4V, pMB); |
|
1666 |
|
|
1667 |
else |
int |
1668 |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
MEanalysis( const IMAGE * const pRef, |
1669 |
y, current->motion_flags, pMB->quant, |
const IMAGE * const pCurrent, |
1670 |
current->fcode, pParam, pMBs, |
MBParam * const pParam, |
1671 |
current->global_flags & XVID_INTER4V, pMB); |
MACROBLOCK * const pMBs, |
1672 |
|
const uint32_t iFcode) |
1673 |
|
{ |
1674 |
|
uint32_t x, y, intra = 0; |
1675 |
|
int sSAD = 0; |
1676 |
|
|
1677 |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
VECTOR currentMV; |
1678 |
if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
int32_t iMinSAD; |
1679 |
if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) |
SearchData Data; |
1680 |
SkipMacroblockP(pMB, sad00); |
Data.iEdgedWidth = pParam->edged_width; |
1681 |
|
Data.currentMV = ¤tMV; |
1682 |
|
Data.iMinSAD = &iMinSAD; |
1683 |
|
Data.iFcode = iFcode; |
1684 |
|
Data.iQuant = 2; |
1685 |
|
|
1686 |
|
if (sadInit) (*sadInit) (); |
1687 |
|
|
1688 |
|
for (y = 0; y < pParam->mb_height-1; y++) { |
1689 |
|
for (x = 0; x < pParam->mb_width; x++) { |
1690 |
|
int sad, dev; |
1691 |
|
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
1692 |
|
|
1693 |
|
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
1694 |
|
pParam, pMBs, pMB, &Data); |
1695 |
|
|
1696 |
|
if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work |
1697 |
|
if (sad > INTRA_THRESH) { |
1698 |
|
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
1699 |
|
pParam->edged_width); |
1700 |
|
if (dev + INTRA_THRESH < sad) intra++; |
1701 |
|
if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame |
1702 |
|
} |
1703 |
|
sSAD += sad; |
1704 |
|
} |
1705 |
|
|
1706 |
} |
} |
1707 |
} |
} |
1708 |
|
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
1709 |
|
if (sSAD > INTER_THRESH ) return 1; //P frame |
1710 |
|
emms(); |
1711 |
|
return 0; // B frame |
1712 |
|
|
1713 |
} |
} |
1714 |
|
|
1715 |
|
int |
1716 |
|
FindFcode( const MBParam * const pParam, |
1717 |
|
const FRAMEINFO * const current) |
1718 |
|
{ |
1719 |
|
uint32_t x, y; |
1720 |
|
int max = 0, min = 0, i; |
1721 |
|
|
1722 |
|
for (y = 0; y < pParam->mb_height; y++) { |
1723 |
|
for (x = 0; x < pParam->mb_width; x++) { |
1724 |
|
|
1725 |
|
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
1726 |
|
for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) { |
1727 |
|
if (pMB->mvs[i].x > max) max = pMB->mvs[i].x; |
1728 |
|
if (pMB->mvs[i].y > max) max = pMB->mvs[i].y; |
1729 |
|
|
1730 |
|
if (pMB->mvs[i].x < min) min = pMB->mvs[i].x; |
1731 |
|
if (pMB->mvs[i].y < min) min = pMB->mvs[i].y; |
1732 |
|
} |
1733 |
|
} |
1734 |
|
} |
1735 |
|
|
1736 |
|
min = -min; |
1737 |
|
max += 1; |
1738 |
|
if (min > max) max = min; |
1739 |
|
|
1740 |
|
for (i = 1; (max > 32 << (i - 1)); i++); |
1741 |
|
return i; |
1742 |
|
} |