21 |
* along with this program ; if not, write to the Free Software |
* along with this program ; if not, write to the Free Software |
22 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
* |
* |
24 |
* $Id: motion_est.c,v 1.58.2.19 2003-06-26 11:50:37 syskin Exp $ |
* $Id: motion_est.c,v 1.58.2.26 2003-08-06 09:08:37 syskin Exp $ |
25 |
* |
* |
26 |
****************************************************************************/ |
****************************************************************************/ |
27 |
|
|
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 "gmc.h" |
44 |
#include "../utils/emms.h" |
#include "../utils/emms.h" |
45 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
46 |
|
|
72 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
73 |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
74 |
|
|
75 |
|
|
76 |
/***************************************************************************** |
/***************************************************************************** |
77 |
* Code |
* Code |
78 |
****************************************************************************/ |
****************************************************************************/ |
336 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
337 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
338 |
|
|
339 |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma && sad < data->iMinSAD[0]) |
340 |
|
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
341 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
342 |
|
|
343 |
if (sad < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
447 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
448 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
449 |
|
|
450 |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma && sad < *data->iMinSAD) |
451 |
|
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
452 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
453 |
|
|
454 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
459 |
} |
} |
460 |
|
|
461 |
static void |
static void |
462 |
|
CheckCandidate16I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
463 |
|
{ |
464 |
|
int sad; |
465 |
|
// int xc, yc; |
466 |
|
const uint8_t * Reference; |
467 |
|
// VECTOR * current; |
468 |
|
|
469 |
|
if ( (x > data->max_dx) || ( x < data->min_dx) |
470 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
471 |
|
|
472 |
|
Reference = GetReference(x, y, data); |
473 |
|
// xc = x; yc = y; |
474 |
|
|
475 |
|
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
476 |
|
// sad += d_mv_bits(x, y, data->predMV, data->iFcode, 0, 0); |
477 |
|
|
478 |
|
/* if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
479 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
480 |
|
*/ |
481 |
|
|
482 |
|
if (sad < data->iMinSAD[0]) { |
483 |
|
data->iMinSAD[0] = sad; |
484 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
485 |
|
*dir = Direction; |
486 |
|
} |
487 |
|
} |
488 |
|
|
489 |
|
static void |
490 |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
491 |
{ |
{ |
492 |
/* maximum speed - for P/B/I decision */ |
/* maximum speed - for P/B/I decision */ |
495 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
496 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
497 |
|
|
498 |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*(data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*((int)data->iEdgedWidth), |
499 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
500 |
|
|
501 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
548 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
549 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
550 |
|
|
551 |
if (data->chroma) sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], |
if (data->chroma && sad < *data->iMinSAD) |
552 |
|
sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], |
553 |
(ycf >> 1) + roundtab_79[ycf & 0x3], |
(ycf >> 1) + roundtab_79[ycf & 0x3], |
554 |
(xcb >> 1) + roundtab_79[xcb & 0x3], |
(xcb >> 1) + roundtab_79[xcb & 0x3], |
555 |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
609 |
|
|
610 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
611 |
|
|
612 |
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
if (data->chroma && sad < *data->iMinSAD) |
613 |
|
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
614 |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
615 |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
616 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
662 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
663 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
664 |
|
|
665 |
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
if (data->chroma && sad < *data->iMinSAD) |
666 |
|
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
667 |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
668 |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
669 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
677 |
|
|
678 |
|
|
679 |
static void |
static void |
680 |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
681 |
{ |
{ |
682 |
|
|
683 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
684 |
int32_t bits = 0; |
int32_t rd = 0; |
685 |
VECTOR * current; |
VECTOR * current; |
686 |
const uint8_t * ptr; |
const uint8_t * ptr; |
687 |
int i, cbp = 0, t, xc, yc; |
int i, cbp = 0, t, xc, yc; |
702 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
703 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
704 |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
705 |
bits += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i); |
rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i); |
706 |
} |
} |
707 |
|
|
708 |
bits += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
709 |
|
|
710 |
if (data->temp[0] + t < data->iMinSAD[1]) { |
if (data->temp[0] + t < data->iMinSAD[1]) { |
711 |
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; } |
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | cbp&32; } |
712 |
if (data->temp[1] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
713 |
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | cbp&16; } |
714 |
if (data->temp[2] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
715 |
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | cbp&8; } |
716 |
if (data->temp[3] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
717 |
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | cbp&4; } |
718 |
|
|
719 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
720 |
|
|
721 |
if (bits >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
722 |
|
|
723 |
/* chroma */ |
/* chroma */ |
724 |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
725 |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
726 |
|
|
727 |
/* chroma U */ |
/* chroma U */ |
728 |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
729 |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
730 |
bits += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4); |
731 |
if (bits >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
732 |
|
|
733 |
/* chroma V */ |
/* chroma V */ |
734 |
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
735 |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
736 |
bits += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
737 |
|
|
738 |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
739 |
|
|
740 |
if (bits < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
741 |
data->iMinSAD[0] = bits; |
data->iMinSAD[0] = rd; |
742 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
743 |
*dir = Direction; |
*dir = Direction; |
744 |
|
*data->cbp = cbp; |
745 |
} |
} |
746 |
} |
} |
747 |
|
|
748 |
static void |
static void |
749 |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
750 |
{ |
{ |
751 |
|
|
752 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
753 |
int32_t bits; |
int32_t rd; |
754 |
VECTOR * current; |
VECTOR * current; |
755 |
const uint8_t * ptr; |
const uint8_t * ptr; |
756 |
int cbp = 0; |
int cbp = 0; |
767 |
} |
} |
768 |
|
|
769 |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
770 |
bits = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
771 |
bits += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
772 |
|
|
773 |
if (bits < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
774 |
data->temp[0] = cbp; |
*data->cbp = cbp; |
775 |
data->iMinSAD[0] = bits; |
data->iMinSAD[0] = rd; |
776 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
777 |
*dir = Direction; |
*dir = Direction; |
778 |
} |
} |
970 |
} |
} |
971 |
|
|
972 |
static __inline void |
static __inline void |
973 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
ZeroMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
974 |
{ |
{ |
975 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_INTER; |
976 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
977 |
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV; |
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV; |
978 |
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; |
988 |
const uint32_t VopFlags, |
const uint32_t VopFlags, |
989 |
const uint32_t VolFlags, |
const uint32_t VolFlags, |
990 |
const IMAGE * const pCurrent, |
const IMAGE * const pCurrent, |
991 |
const IMAGE * const pRef) |
const IMAGE * const pRef, |
992 |
|
const IMAGE * const vGMC, |
993 |
|
const int coding_type) |
994 |
{ |
{ |
995 |
int mode = MODE_INTER; |
int mode = MODE_INTER; |
996 |
|
int mcsel = 0; |
997 |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
998 |
const uint32_t iQuant = pMB->quant; |
const uint32_t iQuant = pMB->quant; |
999 |
|
|
1000 |
const int skip_possible = (!(VolFlags & XVID_VOL_GMC)) && (pMB->dquant == 0); |
const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0); |
1001 |
|
|
1002 |
if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { /* normal, fast, SAD-based mode decision */ |
pMB->mcsel = 0; |
1003 |
|
|
1004 |
|
if (!(VopFlags & XVID_VOP_MODEDECISION_RD)) { /* normal, fast, SAD-based mode decision */ |
1005 |
int sad; |
int sad; |
1006 |
int InterBias = MV16_INTER_BIAS; |
int InterBias = MV16_INTER_BIAS; |
1007 |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1023 |
sad = 0; |
sad = 0; |
1024 |
} |
} |
1025 |
|
|
1026 |
|
/* mcsel */ |
1027 |
|
if (coding_type == S_VOP) { |
1028 |
|
|
1029 |
|
int32_t iSAD = sad16(Data->Cur, |
1030 |
|
vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); |
1031 |
|
|
1032 |
|
if (Data->chroma) { |
1033 |
|
iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1034 |
|
iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1035 |
|
} |
1036 |
|
|
1037 |
|
if (iSAD <= sad) { /* mode decision GMC */ |
1038 |
|
mode = MODE_INTER; |
1039 |
|
mcsel = 1; |
1040 |
|
sad = iSAD; |
1041 |
|
} |
1042 |
|
|
1043 |
|
} |
1044 |
|
|
1045 |
/* intra decision */ |
/* intra decision */ |
1046 |
|
|
1047 |
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
1050 |
if (x != 0) |
if (x != 0) |
1051 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
1052 |
|
|
1053 |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? */ |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
1054 |
if (Data->rrv) InterBias *= 4; |
if (Data->rrv) InterBias *= 4; |
1055 |
|
|
1056 |
if (InterBias < pMB->sad16) { |
if (InterBias < sad) { |
1057 |
int32_t deviation; |
int32_t deviation; |
1058 |
if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth); |
if (!Data->rrv) |
1059 |
else deviation = dev16(Data->Cur, Data->iEdgedWidth) + |
deviation = dev16(Data->Cur, Data->iEdgedWidth); |
1060 |
|
else |
1061 |
|
deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ |
1062 |
dev16(Data->Cur+16, Data->iEdgedWidth) + |
dev16(Data->Cur+16, Data->iEdgedWidth) + |
1063 |
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
1064 |
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
1066 |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
1067 |
} |
} |
1068 |
|
|
1069 |
} else { /* BITS */ |
pMB->cbp = 63; |
1070 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
1071 |
|
|
1072 |
|
} else { /* Rate-Distortion */ |
1073 |
|
|
1074 |
int bits, intra, i; |
int min_rd, intra_rd, i, cbp, c[2] = {0, 0}; |
1075 |
VECTOR backup[5], *v; |
VECTOR backup[5], *v; |
1076 |
Data->iQuant = iQuant; |
Data->iQuant = iQuant; |
1077 |
|
Data->cbp = c; |
1078 |
|
|
1079 |
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1080 |
for (i = 0; i < 5; i++) { |
for (i = 0; i < 5; i++) { |
1082 |
backup[i] = v[i]; |
backup[i] = v[i]; |
1083 |
} |
} |
1084 |
|
|
1085 |
bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags); |
min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags); |
1086 |
if (bits == 0) |
cbp = *Data->cbp; |
1087 |
mode = MODE_INTER; /* quick stop */ |
|
1088 |
else { |
if (coding_type == S_VOP) { |
1089 |
if (inter4v) { |
int gmc_rd; |
1090 |
int bits_inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
*Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ |
1091 |
if (bits_inter4v < bits) { Data->iMinSAD[0] = bits = bits_inter4v; mode = MODE_INTER4V; } |
gmc_rd = findRDgmc(Data, vGMC, x, y); |
1092 |
|
if (gmc_rd < min_rd) { |
1093 |
|
mcsel = 1; |
1094 |
|
*Data->iMinSAD = min_rd = gmc_rd; |
1095 |
|
mode = MODE_INTER; |
1096 |
|
cbp = *Data->cbp; |
1097 |
|
} |
1098 |
} |
} |
1099 |
|
|
1100 |
intra = CountMBBitsIntra(Data); |
if (inter4v) { |
1101 |
|
int v4_rd; |
1102 |
|
v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1103 |
|
if (v4_rd < min_rd) { |
1104 |
|
Data->iMinSAD[0] = min_rd = v4_rd; |
1105 |
|
mode = MODE_INTER4V; |
1106 |
|
cbp = *Data->cbp; |
1107 |
|
} |
1108 |
|
} |
1109 |
|
|
1110 |
if (intra < bits) { *Data->iMinSAD = bits = intra; mode = MODE_INTRA; } |
intra_rd = findRDintra(Data); |
1111 |
|
if (intra_rd < min_rd) { |
1112 |
|
*Data->iMinSAD = min_rd = intra_rd; |
1113 |
|
mode = MODE_INTRA; |
1114 |
} |
} |
1115 |
|
|
1116 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
1117 |
|
pMB->cbp = cbp; |
1118 |
} |
} |
1119 |
|
|
1120 |
if (Data->rrv) { |
if (Data->rrv) { |
1122 |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1123 |
} |
} |
1124 |
|
|
1125 |
if (mode == MODE_INTER) { |
if (mode == MODE_INTER && mcsel == 0) { |
1126 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
1127 |
|
|
1128 |
if(Data->qpel) { |
if(Data->qpel) { |
1129 |
pMB->qmvs[0] = pMB->qmvs[1] |
pMB->qmvs[0] = pMB->qmvs[1] |
1135 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1136 |
} |
} |
1137 |
|
|
1138 |
} else if (mode == MODE_INTER4V) |
} else if (mode == MODE_INTER ) { // but mcsel == 1 |
1139 |
pMB->sad16 = Data->iMinSAD[0]; |
|
1140 |
|
pMB->mcsel = 1; |
1141 |
|
if (Data->qpel) { |
1142 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; |
1143 |
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; |
1144 |
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; |
1145 |
|
} else |
1146 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; |
1147 |
|
|
1148 |
|
} else |
1149 |
|
if (mode == MODE_INTER4V) ; /* anything here? */ |
1150 |
else /* INTRA, NOT_CODED */ |
else /* INTRA, NOT_CODED */ |
1151 |
SkipMacroblockP(pMB, 0); |
ZeroMacroblockP(pMB, 0); |
1152 |
|
|
1153 |
pMB->mode = mode; |
pMB->mode = mode; |
1154 |
} |
} |
1160 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
1161 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
1162 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
1163 |
|
const IMAGE * const pGMC, |
1164 |
const uint32_t iLimit) |
const uint32_t iLimit) |
1165 |
{ |
{ |
1166 |
MACROBLOCK *const pMBs = current->mbs; |
MACROBLOCK *const pMBs = current->mbs; |
1174 |
|
|
1175 |
uint32_t x, y; |
uint32_t x, y; |
1176 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
1177 |
int32_t quant = current->quant, sad00; |
int32_t sad00; |
1178 |
int skip_thresh = \ |
int skip_thresh = INITIAL_SKIP_THRESH * \ |
|
INITIAL_SKIP_THRESH * \ |
|
1179 |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
1180 |
(current->vop_flags & XVID_VOP_MODEDECISION_BITS ? 2:1); |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
1181 |
|
|
1182 |
/* some pre-initialized thingies for SearchP */ |
/* some pre-initialized thingies for SearchP */ |
1183 |
int32_t temp[8]; |
int32_t temp[8]; |
1195 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
1196 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
1197 |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
1198 |
Data.chroma = MotionFlags & XVID_ME_CHROMA16; |
Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP; |
1199 |
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED ? 1:0); |
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED) ? 1:0; |
1200 |
Data.dctSpace = dct_space; |
Data.dctSpace = dct_space; |
1201 |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
1202 |
|
|
1233 |
|
|
1234 |
sad00 = pMB->sad16; |
sad00 = pMB->sad16; |
1235 |
|
|
|
if (pMB->dquant != 0) { |
|
|
quant += DQtab[pMB->dquant]; |
|
|
if (quant > 31) quant = 31; |
|
|
else if (quant < 1) quant = 1; |
|
|
} |
|
|
pMB->quant = quant; |
|
|
|
|
1236 |
/* initial skip decision */ |
/* initial skip decision */ |
1237 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
1238 |
if (!(current->vol_flags & XVID_VOL_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
1239 |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
1240 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
1241 |
SkipMacroblockP(pMB, sad00); |
ZeroMacroblockP(pMB, sad00); |
1242 |
|
pMB->mode = MODE_NOT_CODED; |
1243 |
continue; |
continue; |
1244 |
} |
} |
1245 |
} |
} |
1246 |
|
|
1247 |
|
if ((current->vop_flags & XVID_VOP_CARTOON) && |
1248 |
|
(sad00 < pMB->quant * 4 * skip_thresh)) { /* favorize (0,0) vector for cartoons */ |
1249 |
|
ZeroMacroblockP(pMB, sad00); |
1250 |
|
continue; |
1251 |
|
} |
1252 |
|
|
1253 |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1254 |
y, MotionFlags, current->vop_flags, current->vol_flags, |
y, MotionFlags, current->vop_flags, current->vol_flags, |
1255 |
&Data, pParam, pMBs, reference->mbs, pMB); |
&Data, pParam, pMBs, reference->mbs, pMB); |
1256 |
|
|
1257 |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
1258 |
MotionFlags, current->vop_flags, current->vol_flags, |
MotionFlags, current->vop_flags, current->vol_flags, |
1259 |
pCurrent, pRef); |
pCurrent, pRef, pGMC, current->coding_type); |
1260 |
|
|
1261 |
if (pMB->mode == MODE_INTRA) |
if (pMB->mode == MODE_INTRA) |
1262 |
if (++iIntra > iLimit) return 1; |
if (++iIntra > iLimit) return 1; |
1263 |
} |
} |
1264 |
} |
} |
1265 |
|
|
|
if (current->vol_flags & XVID_VOL_GMC ) /* GMC only for S(GMC)-VOPs */ |
|
|
{ |
|
|
current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV); |
|
|
} |
|
1266 |
return 0; |
return 0; |
1267 |
} |
} |
1268 |
|
|
1381 |
Data->iMinSAD[3] = pMB->sad8[2]; |
Data->iMinSAD[3] = pMB->sad8[2]; |
1382 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
1383 |
|
|
1384 |
if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) && (x | y)) { |
if ((!(VopFlags & XVID_VOP_MODEDECISION_RD)) && (x | y)) { |
1385 |
threshA = Data->temp[0]; /* that's where we keep this SAD atm */ |
threshA = Data->temp[0]; /* that's where we keep this SAD atm */ |
1386 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
1387 |
else if (threshA > 1024) threshA = 1024; |
else if (threshA > 1024) threshA = 1024; |
1480 |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
1481 |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1482 |
|
|
1483 |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_BITS))) { |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_RD))) { |
1484 |
/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */ |
/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */ |
1485 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
1486 |
|
|
1546 |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
1547 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1548 |
|
|
1549 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_BITS))) { |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) { |
1550 |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
1551 |
|
|
1552 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1773 |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1774 |
stride); |
stride); |
1775 |
|
|
1776 |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
1777 |
|
|
1778 |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
1779 |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
1780 |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1781 |
stride); |
stride); |
1782 |
|
|
1783 |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
1784 |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
1785 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1786 |
pMB->qmvs[k] = pMB->mvs[k]; |
pMB->qmvs[k] = pMB->mvs[k]; |
2093 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
2094 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2095 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant]; |
2096 |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL; |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
2097 |
Data.rounding = 0; |
Data.rounding = 0; |
2098 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA8; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
2099 |
Data.temp = temp; |
Data.temp = temp; |
2100 |
|
|
2101 |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
2270 |
MACROBLOCK * const pMBs = Current->mbs; |
MACROBLOCK * const pMBs = Current->mbs; |
2271 |
const IMAGE * const pCurrent = &Current->image; |
const IMAGE * const pCurrent = &Current->image; |
2272 |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + b_thresh; |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + b_thresh; |
2273 |
int s = 0, blocks = 0; |
int blocks = 0; |
2274 |
int complexity = 0; |
int complexity = 0; |
2275 |
|
|
2276 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
2283 |
Data.temp = temp; |
Data.temp = temp; |
2284 |
CheckCandidate = CheckCandidate32I; |
CheckCandidate = CheckCandidate32I; |
2285 |
|
|
|
|
|
2286 |
if (intraCount != 0) { |
if (intraCount != 0) { |
2287 |
if (intraCount < 10) // we're right after an I frame |
if (intraCount < 10) // we're right after an I frame |
2288 |
IntraThresh += 15* (intraCount - 10) * (intraCount - 10); |
IntraThresh += 15* (intraCount - 10) * (intraCount - 10); |
2317 |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16, |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16, |
2318 |
pParam->edged_width); |
pParam->edged_width); |
2319 |
|
|
2320 |
complexity += dev; |
complexity += MAX(dev, 300); |
2321 |
if (dev + IntraThresh < pMB->sad16) { |
if (dev + IntraThresh < pMB->sad16) { |
2322 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
2323 |
if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP; |
if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP; |
2327 |
if (dev > 500 && pMB->sad16 < 1000) |
if (dev > 500 && pMB->sad16 < 1000) |
2328 |
sSAD += 1000; |
sSAD += 1000; |
2329 |
|
|
2330 |
sSAD += pMB->sad16; |
sSAD += (dev < 3000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */ |
2331 |
} |
} |
2332 |
} |
} |
2333 |
} |
} |
2335 |
|
|
2336 |
sSAD /= complexity + 4*blocks; |
sSAD /= complexity + 4*blocks; |
2337 |
|
|
2338 |
if (intraCount > 12 && sSAD > INTRA_THRESH2 ) return I_VOP; |
if (intraCount > 80 && sSAD > INTRA_THRESH2 ) return I_VOP; |
2339 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2340 |
emms(); |
emms(); |
2341 |
return B_VOP; |
return B_VOP; |
2342 |
} |
} |
2343 |
|
|
|
static WARPPOINTS |
|
|
GlobalMotionEst(const MACROBLOCK * const pMBs, |
|
|
const MBParam * const pParam, |
|
|
const FRAMEINFO * const current, |
|
|
const FRAMEINFO * const reference, |
|
|
const IMAGE * const pRefH, |
|
|
const IMAGE * const pRefV, |
|
|
const IMAGE * const pRefHV ) |
|
|
{ |
|
|
|
|
|
const int deltax=8; /* upper bound for difference between a MV and it's neighbour MVs */ |
|
|
const int deltay=8; |
|
|
const int grad=512; /* lower bound for deviation in MB */ |
|
|
|
|
|
WARPPOINTS gmc; |
|
|
|
|
|
uint32_t mx, my; |
|
|
|
|
|
int MBh = pParam->mb_height; |
|
|
int MBw = pParam->mb_width; |
|
|
|
|
|
int *MBmask= calloc(MBh*MBw,sizeof(int)); |
|
|
double DtimesF[4] = { 0.,0., 0., 0. }; |
|
|
double sol[4] = { 0., 0., 0., 0. }; |
|
|
double a,b,c,n,denom; |
|
|
double meanx,meany; |
|
|
int num,oldnum; |
|
|
|
|
|
if (!MBmask) { fprintf(stderr,"Mem error\n"); |
|
|
gmc.duv[0].x= gmc.duv[0].y = |
|
|
gmc.duv[1].x= gmc.duv[1].y = |
|
|
gmc.duv[2].x= gmc.duv[2].y = 0; |
|
|
return gmc; } |
|
|
|
|
|
/* filter mask of all blocks */ |
|
|
|
|
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
|
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
|
|
{ |
|
|
const int mbnum = mx + my * MBw; |
|
|
const MACROBLOCK *pMB = &pMBs[mbnum]; |
|
|
const VECTOR mv = pMB->mvs[0]; |
|
|
|
|
|
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) |
|
|
continue; |
|
|
|
|
|
if ( ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) ) |
|
|
&& ( (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) ) |
|
|
&& ( (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) ) |
|
|
&& ( (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) ) |
|
|
MBmask[mbnum]=1; |
|
|
} |
|
|
|
|
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
|
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
|
|
{ |
|
|
const uint8_t *const pCur = current->image.y + 16*my*pParam->edged_width + 16*mx; |
|
|
|
|
|
const int mbnum = mx + my * MBw; |
|
|
if (!MBmask[mbnum]) |
|
|
continue; |
|
|
|
|
|
if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= (uint32_t)grad ) |
|
|
MBmask[mbnum] = 0; |
|
|
if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= (uint32_t)grad ) |
|
|
MBmask[mbnum] = 0; |
|
|
|
|
|
} |
|
|
|
|
|
emms(); |
|
|
|
|
|
do { /* until convergence */ |
|
|
|
|
|
a = b = c = n = 0; |
|
|
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
|
|
for (my = 0; my < (uint32_t)MBh; my++) |
|
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
|
|
{ |
|
|
const int mbnum = mx + my * MBw; |
|
|
const MACROBLOCK *pMB = &pMBs[mbnum]; |
|
|
const VECTOR mv = pMB->mvs[0]; |
|
|
|
|
|
if (!MBmask[mbnum]) |
|
|
continue; |
|
|
|
|
|
n++; |
|
|
a += 16*mx+8; |
|
|
b += 16*my+8; |
|
|
c += (16*mx+8)*(16*mx+8)+(16*my+8)*(16*my+8); |
|
|
|
|
|
DtimesF[0] += (double)mv.x; |
|
|
DtimesF[1] += (double)mv.x*(16*mx+8) + (double)mv.y*(16*my+8); |
|
|
DtimesF[2] += (double)mv.x*(16*my+8) - (double)mv.y*(16*mx+8); |
|
|
DtimesF[3] += (double)mv.y; |
|
|
} |
|
|
|
|
|
denom = a*a+b*b-c*n; |
|
|
|
|
|
/* Solve the system: sol = (D'*E*D)^{-1} D'*E*F */ |
|
|
/* D'*E*F has been calculated in the same loop as matrix */ |
|
|
|
|
|
sol[0] = -c*DtimesF[0] + a*DtimesF[1] + b*DtimesF[2]; |
|
|
sol[1] = a*DtimesF[0] - n*DtimesF[1] + b*DtimesF[3]; |
|
|
sol[2] = b*DtimesF[0] - n*DtimesF[2] - a*DtimesF[3]; |
|
|
sol[3] = b*DtimesF[1] - a*DtimesF[2] - c*DtimesF[3]; |
|
|
|
|
|
sol[0] /= denom; |
|
|
sol[1] /= denom; |
|
|
sol[2] /= denom; |
|
|
sol[3] /= denom; |
|
|
|
|
|
meanx = meany = 0.; |
|
|
oldnum = 0; |
|
|
for (my = 0; my < (uint32_t)MBh; my++) |
|
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
|
|
{ |
|
|
const int mbnum = mx + my * MBw; |
|
|
const MACROBLOCK *pMB = &pMBs[mbnum]; |
|
|
const VECTOR mv = pMB->mvs[0]; |
|
|
|
|
|
if (!MBmask[mbnum]) |
|
|
continue; |
|
|
|
|
|
oldnum++; |
|
|
meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ); |
|
|
meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ); |
|
|
} |
|
|
|
|
|
if (4*meanx > oldnum) /* better fit than 0.25 is useless */ |
|
|
meanx /= oldnum; |
|
|
else |
|
|
meanx = 0.25; |
|
|
|
|
|
if (4*meany > oldnum) |
|
|
meany /= oldnum; |
|
|
else |
|
|
meany = 0.25; |
|
|
|
|
|
/* fprintf(stderr,"sol = (%8.5f, %8.5f, %8.5f, %8.5f)\n",sol[0],sol[1],sol[2],sol[3]); |
|
|
fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum); |
|
|
*/ |
|
|
num = 0; |
|
|
for (my = 0; my < (uint32_t)MBh; my++) |
|
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
|
|
{ |
|
|
const int mbnum = mx + my * MBw; |
|
|
const MACROBLOCK *pMB = &pMBs[mbnum]; |
|
|
const VECTOR mv = pMB->mvs[0]; |
|
|
|
|
|
if (!MBmask[mbnum]) |
|
|
continue; |
|
|
|
|
|
if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx ) |
|
|
|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) ) |
|
|
MBmask[mbnum]=0; |
|
|
else |
|
|
num++; |
|
|
} |
|
|
|
|
|
} while ( (oldnum != num) && (num>=4) ); |
|
|
|
|
|
if (num < 4) |
|
|
{ |
|
|
gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0; |
|
|
} else { |
|
|
|
|
|
gmc.duv[0].x=(int)(sol[0]+0.5); |
|
|
gmc.duv[0].y=(int)(sol[3]+0.5); |
|
|
|
|
|
gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5); |
|
|
gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5); |
|
|
|
|
|
gmc.duv[2].x=0; |
|
|
gmc.duv[2].y=0; |
|
|
} |
|
|
/* fprintf(stderr,"wp1 = ( %4d, %4d) wp2 = ( %4d, %4d) \n", gmc.duv[0].x, gmc.duv[0].y, gmc.duv[1].x, gmc.duv[1].y); */ |
|
|
|
|
|
free(MBmask); |
|
|
|
|
|
return gmc; |
|
|
} |
|
2344 |
|
|
2345 |
/* functions which perform BITS-based search/bitcount */ |
/* functions which perform BITS-based search/bitcount */ |
2346 |
|
|
2347 |
static int |
static int |
2348 |
CountMBBitsInter(SearchData * const Data, |
findRDinter(SearchData * const Data, |
2349 |
const MACROBLOCK * const pMBs, const int x, const int y, |
const MACROBLOCK * const pMBs, const int x, const int y, |
2350 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2351 |
const uint32_t MotionFlags) |
const uint32_t MotionFlags) |
2353 |
int i, iDirection; |
int i, iDirection; |
2354 |
int32_t bsad[5]; |
int32_t bsad[5]; |
2355 |
|
|
2356 |
CheckCandidate = CheckCandidateBits16; |
CheckCandidate = CheckCandidateRD16; |
2357 |
|
|
2358 |
if (Data->qpel) { |
if (Data->qpel) { |
2359 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
2361 |
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
2362 |
} |
} |
2363 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
2364 |
CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
2365 |
|
|
2366 |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_BITS | XVID_ME_EXTSEARCH_BITS)) { /* we have to prepare for halfpixel-precision search */ |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
2367 |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
2368 |
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, |
2369 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
2370 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
2371 |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
2372 |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
2373 |
} |
} |
2374 |
|
|
2375 |
} else { /* not qpel */ |
} else { /* not qpel */ |
2376 |
|
|
2377 |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
2378 |
} |
} |
2379 |
|
|
2380 |
if (MotionFlags&XVID_ME_EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
if (MotionFlags&XVID_ME_EXTSEARCH_RD) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
2381 |
|
|
2382 |
if (MotionFlags&XVID_ME_HALFPELREFINE16_BITS) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) SubpelRefine(Data); |
2383 |
|
|
2384 |
if (Data->qpel) { |
if (Data->qpel) { |
2385 |
if (MotionFlags&(XVID_ME_EXTSEARCH_BITS | XVID_ME_HALFPELREFINE16_BITS)) { /* there was halfpel-precision search */ |
if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ |
2386 |
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
2387 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ |
2388 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
2393 |
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, |
2394 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
2395 |
} |
} |
2396 |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_BITS) SubpelRefine(Data); |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) SubpelRefine(Data); |
2397 |
} |
} |
2398 |
|
|
2399 |
if (MotionFlags&XVID_ME_CHECKPREDICTION_BITS) { /* let's check vector equal to prediction */ |
if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ |
2400 |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
2401 |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
2402 |
CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
CheckCandidateRD16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
2403 |
} |
} |
2404 |
return Data->iMinSAD[0]; |
return Data->iMinSAD[0]; |
2405 |
} |
} |
2406 |
|
|
2407 |
static int |
static int |
2408 |
CountMBBitsInter4v(const SearchData * const Data, |
findRDinter4v(const SearchData * const Data, |
2409 |
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
2410 |
const int x, const int y, |
const int x, const int y, |
2411 |
const MBParam * const pParam, const uint32_t MotionFlags, |
const MBParam * const pParam, const uint32_t MotionFlags, |
2419 |
uint8_t * ptr; |
uint8_t * ptr; |
2420 |
|
|
2421 |
memcpy(Data8, Data, sizeof(SearchData)); |
memcpy(Data8, Data, sizeof(SearchData)); |
2422 |
CheckCandidate = CheckCandidateBits8; |
CheckCandidate = CheckCandidateRD8; |
2423 |
|
|
2424 |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
2425 |
|
|
2431 |
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2432 |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2433 |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2434 |
|
*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; // copy corresponding cbp bit |
2435 |
|
|
2436 |
if(Data->qpel) { |
if(Data->qpel) { |
2437 |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2453 |
{ |
{ |
2454 |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
2455 |
if (!MVequal (*v, backup[i+1]) ) |
if (!MVequal (*v, backup[i+1]) ) |
2456 |
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
CheckCandidateRD8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
2457 |
} |
} |
2458 |
|
|
2459 |
if (Data8->qpel) { |
if (Data8->qpel) { |
2460 |
if (MotionFlags&XVID_ME_HALFPELREFINE8_BITS || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_BITS)) { /* halfpixel motion search follows */ |
if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ |
2461 |
int32_t s = *Data8->iMinSAD; |
int32_t s = *Data8->iMinSAD; |
2462 |
Data8->currentMV->x = Data8->currentQMV->x/2; |
Data8->currentMV->x = Data8->currentQMV->x/2; |
2463 |
Data8->currentMV->y = Data8->currentQMV->y/2; |
Data8->currentMV->y = Data8->currentQMV->y/2; |
2466 |
pParam->width, pParam->height, Data8->iFcode - 1, 0, 0); |
pParam->width, pParam->height, Data8->iFcode - 1, 0, 0); |
2467 |
|
|
2468 |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
2469 |
CheckCandidateBits8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8); |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8); |
2470 |
|
|
2471 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) |
2472 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
2473 |
|
|
2474 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2475 |
SubpelRefine(Data8); |
SubpelRefine(Data8); |
2476 |
|
|
2477 |
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
2484 |
pParam->width, pParam->height, Data8->iFcode, 1, 0); |
pParam->width, pParam->height, Data8->iFcode, 1, 0); |
2485 |
|
|
2486 |
} |
} |
2487 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8); |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) SubpelRefine(Data8); |
2488 |
|
|
2489 |
} else { /* not qpel */ |
} else { /* not qpel */ |
2490 |
|
|
2491 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) /* extsearch */ |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */ |
2492 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
2493 |
|
|
2494 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2495 |
SubpelRefine(Data8); /* halfpel refinement */ |
SubpelRefine(Data8); /* halfpel refinement */ |
2496 |
} |
} |
2497 |
|
|
2498 |
/* checking vector equal to predicion */ |
/* checking vector equal to predicion */ |
2499 |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_BITS) { |
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) { |
2500 |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
2501 |
if (!MVequal(*v, Data8->predMV)) |
if (!MVequal(*v, Data8->predMV)) |
2502 |
CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
2503 |
} |
} |
2504 |
|
|
2505 |
bits += *Data8->iMinSAD; |
bits += *Data8->iMinSAD; |
2520 |
} |
} |
2521 |
pMB->mvs[i] = *Data8->currentMV; |
pMB->mvs[i] = *Data8->currentMV; |
2522 |
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
2523 |
if (Data8->temp[0]) cbp |= 1 << (5 - i); |
if (Data8->cbp[0]) cbp |= 1 << (5 - i); |
2524 |
|
|
2525 |
} /* /for all luma blocks */ |
} /* end - for all luma blocks */ |
2526 |
|
|
2527 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
2528 |
|
|
2544 |
|
|
2545 |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
2546 |
|
|
2547 |
|
*Data->cbp = cbp; |
2548 |
return bits; |
return bits; |
2549 |
} |
} |
2550 |
|
|
2551 |
static int |
static int |
2552 |
CountMBBitsIntra(const SearchData * const Data) |
findRDintra(const SearchData * const Data) |
2553 |
{ |
{ |
2554 |
int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
2555 |
int cbp = 0, i, dc = 0; |
int cbp = 0, i, dc = 0; |
2579 |
|
|
2580 |
return bits; |
return bits; |
2581 |
} |
} |
2582 |
|
|
2583 |
|
static int |
2584 |
|
findRDgmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) |
2585 |
|
{ |
2586 |
|
int bits = BITS_MULT*1; /* this one is mcsel */ |
2587 |
|
int cbp = 0, i; |
2588 |
|
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
2589 |
|
|
2590 |
|
for(i = 0; i < 4; i++) { |
2591 |
|
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2592 |
|
transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); |
2593 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i); |
2594 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2595 |
|
} |
2596 |
|
|
2597 |
|
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
2598 |
|
|
2599 |
|
/*chroma U */ |
2600 |
|
transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
2601 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4); |
2602 |
|
|
2603 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2604 |
|
|
2605 |
|
/* chroma V */ |
2606 |
|
transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
2607 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5); |
2608 |
|
|
2609 |
|
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
2610 |
|
|
2611 |
|
*Data->cbp = cbp; |
2612 |
|
|
2613 |
|
return bits; |
2614 |
|
} |
2615 |
|
|
2616 |
|
|
2617 |
|
|
2618 |
|
|
2619 |
|
static __inline void |
2620 |
|
GMEanalyzeMB ( const uint8_t * const pCur, |
2621 |
|
const uint8_t * const pRef, |
2622 |
|
const uint8_t * const pRefH, |
2623 |
|
const uint8_t * const pRefV, |
2624 |
|
const uint8_t * const pRefHV, |
2625 |
|
const int x, |
2626 |
|
const int y, |
2627 |
|
const MBParam * const pParam, |
2628 |
|
MACROBLOCK * const pMBs, |
2629 |
|
SearchData * const Data) |
2630 |
|
{ |
2631 |
|
|
2632 |
|
int i=0; |
2633 |
|
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
2634 |
|
|
2635 |
|
Data->iMinSAD[0] = MV_MAX_ERROR; |
2636 |
|
|
2637 |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
2638 |
|
|
2639 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2640 |
|
pParam->width, pParam->height, 16, 0, 0); |
2641 |
|
|
2642 |
|
Data->Cur = pCur + 16*(x + y * pParam->edged_width); |
2643 |
|
Data->RefP[0] = pRef + 16*(x + y * pParam->edged_width); |
2644 |
|
Data->RefP[1] = pRefV + 16*(x + y * pParam->edged_width); |
2645 |
|
Data->RefP[2] = pRefH + 16*(x + y * pParam->edged_width); |
2646 |
|
Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width); |
2647 |
|
|
2648 |
|
Data->currentMV[0].x = Data->currentMV[0].y = 0; |
2649 |
|
CheckCandidate16I(0, 0, 255, &i, Data); |
2650 |
|
|
2651 |
|
if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) ) |
2652 |
|
CheckCandidate16I(Data->predMV.x, Data->predMV.y, 255, &i, Data); |
2653 |
|
|
2654 |
|
AdvDiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
2655 |
|
|
2656 |
|
SubpelRefine(Data); |
2657 |
|
|
2658 |
|
|
2659 |
|
/* for QPel halfpel positions are worse than in halfpel mode :( */ |
2660 |
|
/* if (Data->qpel) { |
2661 |
|
Data->currentQMV->x = 2*Data->currentMV->x; |
2662 |
|
Data->currentQMV->y = 2*Data->currentMV->y; |
2663 |
|
Data->qpel_precision = 1; |
2664 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2665 |
|
pParam->width, pParam->height, iFcode, 1, 0); |
2666 |
|
SubpelRefine(Data); |
2667 |
|
} |
2668 |
|
*/ |
2669 |
|
|
2670 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
2671 |
|
pMB->sad16 = Data->iMinSAD[0]; |
2672 |
|
pMB->mode = MODE_INTER; |
2673 |
|
pMB->sad16 += 10*d_mv_bits(pMB->mvs[0].x, pMB->mvs[0].y, Data->predMV, Data->iFcode, 0, 0); |
2674 |
|
return; |
2675 |
|
} |
2676 |
|
|
2677 |
|
void |
2678 |
|
GMEanalysis(const MBParam * const pParam, |
2679 |
|
const FRAMEINFO * const current, |
2680 |
|
const FRAMEINFO * const reference, |
2681 |
|
const IMAGE * const pRefH, |
2682 |
|
const IMAGE * const pRefV, |
2683 |
|
const IMAGE * const pRefHV) |
2684 |
|
{ |
2685 |
|
uint32_t x, y; |
2686 |
|
MACROBLOCK * const pMBs = current->mbs; |
2687 |
|
const IMAGE * const pCurrent = ¤t->image; |
2688 |
|
const IMAGE * const pReference = &reference->image; |
2689 |
|
|
2690 |
|
int32_t iMinSAD[5], temp[5]; |
2691 |
|
VECTOR currentMV[5]; |
2692 |
|
SearchData Data; |
2693 |
|
memset(&Data, 0, sizeof(SearchData)); |
2694 |
|
|
2695 |
|
Data.iEdgedWidth = pParam->edged_width; |
2696 |
|
Data.rounding = pParam->m_rounding_type; |
2697 |
|
|
2698 |
|
Data.currentMV = ¤tMV[0]; |
2699 |
|
Data.iMinSAD = &iMinSAD[0]; |
2700 |
|
Data.iFcode = current->fcode; |
2701 |
|
Data.temp = temp; |
2702 |
|
|
2703 |
|
CheckCandidate = CheckCandidate16I; |
2704 |
|
|
2705 |
|
if (sadInit) (*sadInit) (); |
2706 |
|
|
2707 |
|
for (y = 0; y < pParam->mb_height; y ++) { |
2708 |
|
for (x = 0; x < pParam->mb_width; x ++) { |
2709 |
|
|
2710 |
|
GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data); |
2711 |
|
} |
2712 |
|
} |
2713 |
|
return; |
2714 |
|
} |
2715 |
|
|
2716 |
|
|
2717 |
|
WARPPOINTS |
2718 |
|
GlobalMotionEst(MACROBLOCK * const pMBs, |
2719 |
|
const MBParam * const pParam, |
2720 |
|
const FRAMEINFO * const current, |
2721 |
|
const FRAMEINFO * const reference, |
2722 |
|
const IMAGE * const pRefH, |
2723 |
|
const IMAGE * const pRefV, |
2724 |
|
const IMAGE * const pRefHV) |
2725 |
|
{ |
2726 |
|
|
2727 |
|
const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs |
2728 |
|
const int deltay=8; |
2729 |
|
const unsigned int gradx=512; // lower bound for gradient in MB (ignore "flat" blocks) |
2730 |
|
const unsigned int grady=512; |
2731 |
|
|
2732 |
|
double sol[4] = { 0., 0., 0., 0. }; |
2733 |
|
|
2734 |
|
WARPPOINTS gmc; |
2735 |
|
|
2736 |
|
uint32_t mx, my; |
2737 |
|
|
2738 |
|
int MBh = pParam->mb_height; |
2739 |
|
int MBw = pParam->mb_width; |
2740 |
|
const int minblocks = 9; //MBh*MBw/32+3; /* just some reasonable number 3% + 3 */ |
2741 |
|
const int maxblocks = MBh*MBw/4; /* just some reasonable number 3% + 3 */ |
2742 |
|
|
2743 |
|
int num=0; |
2744 |
|
int oldnum; |
2745 |
|
|
2746 |
|
gmc.duv[0].x = gmc.duv[0].y = gmc.duv[1].x = gmc.duv[1].y = gmc.duv[2].x = gmc.duv[2].y = 0; |
2747 |
|
|
2748 |
|
GMEanalysis(pParam,current, reference, pRefH, pRefV, pRefHV); |
2749 |
|
|
2750 |
|
/* block based ME isn't done, yet, so do a quick presearch */ |
2751 |
|
|
2752 |
|
// filter mask of all blocks |
2753 |
|
|
2754 |
|
for (my = 0; my < (uint32_t)MBh; my++) |
2755 |
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2756 |
|
{ |
2757 |
|
const int mbnum = mx + my * MBw; |
2758 |
|
pMBs[mbnum].mcsel = 0; |
2759 |
|
} |
2760 |
|
|
2761 |
|
|
2762 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ |
2763 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ |
2764 |
|
{ |
2765 |
|
const int mbnum = mx + my * MBw; |
2766 |
|
MACROBLOCK *const pMB = &pMBs[mbnum]; |
2767 |
|
const VECTOR mv = pMB->mvs[0]; |
2768 |
|
|
2769 |
|
/* don't use object boundaries */ |
2770 |
|
if ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) |
2771 |
|
&& (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) |
2772 |
|
&& (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) |
2773 |
|
&& (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) |
2774 |
|
&& (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) |
2775 |
|
&& (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) |
2776 |
|
&& (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) |
2777 |
|
&& (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) |
2778 |
|
{ const int iEdgedWidth = pParam->edged_width; |
2779 |
|
const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); |
2780 |
|
if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) |
2781 |
|
&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) |
2782 |
|
{ pMB->mcsel = 1; |
2783 |
|
num++; |
2784 |
|
} |
2785 |
|
|
2786 |
|
/* only use "structured" blocks */ |
2787 |
|
} |
2788 |
|
} |
2789 |
|
emms(); |
2790 |
|
|
2791 |
|
/* further filtering would be possible, but during iteration, remaining |
2792 |
|
outliers usually are removed, too */ |
2793 |
|
|
2794 |
|
if (num>= minblocks) |
2795 |
|
do { /* until convergence */ |
2796 |
|
double DtimesF[4]; |
2797 |
|
double a,b,c,n,invdenom; |
2798 |
|
double meanx,meany; |
2799 |
|
|
2800 |
|
a = b = c = n = 0; |
2801 |
|
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
2802 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
2803 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2804 |
|
{ |
2805 |
|
const int mbnum = mx + my * MBw; |
2806 |
|
const VECTOR mv = pMBs[mbnum].mvs[0]; |
2807 |
|
|
2808 |
|
if (!pMBs[mbnum].mcsel) |
2809 |
|
continue; |
2810 |
|
|
2811 |
|
n++; |
2812 |
|
a += 16*mx+8; |
2813 |
|
b += 16*my+8; |
2814 |
|
c += (16*mx+8)*(16*mx+8)+(16*my+8)*(16*my+8); |
2815 |
|
|
2816 |
|
DtimesF[0] += (double)mv.x; |
2817 |
|
DtimesF[1] += (double)mv.x*(16*mx+8) + (double)mv.y*(16*my+8); |
2818 |
|
DtimesF[2] += (double)mv.x*(16*my+8) - (double)mv.y*(16*mx+8); |
2819 |
|
DtimesF[3] += (double)mv.y; |
2820 |
|
} |
2821 |
|
|
2822 |
|
invdenom = a*a+b*b-c*n; |
2823 |
|
|
2824 |
|
/* Solve the system: sol = (D'*E*D)^{-1} D'*E*F */ |
2825 |
|
/* D'*E*F has been calculated in the same loop as matrix */ |
2826 |
|
|
2827 |
|
sol[0] = -c*DtimesF[0] + a*DtimesF[1] + b*DtimesF[2]; |
2828 |
|
sol[1] = a*DtimesF[0] - n*DtimesF[1] + b*DtimesF[3]; |
2829 |
|
sol[2] = b*DtimesF[0] - n*DtimesF[2] - a*DtimesF[3]; |
2830 |
|
sol[3] = b*DtimesF[1] - a*DtimesF[2] - c*DtimesF[3]; |
2831 |
|
|
2832 |
|
sol[0] /= invdenom; |
2833 |
|
sol[1] /= invdenom; |
2834 |
|
sol[2] /= invdenom; |
2835 |
|
sol[3] /= invdenom; |
2836 |
|
|
2837 |
|
meanx = meany = 0.; |
2838 |
|
oldnum = 0; |
2839 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
2840 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2841 |
|
{ |
2842 |
|
const int mbnum = mx + my * MBw; |
2843 |
|
const VECTOR mv = pMBs[mbnum].mvs[0]; |
2844 |
|
|
2845 |
|
if (!pMBs[mbnum].mcsel) |
2846 |
|
continue; |
2847 |
|
|
2848 |
|
oldnum++; |
2849 |
|
meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ); |
2850 |
|
meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ); |
2851 |
|
} |
2852 |
|
|
2853 |
|
if (4*meanx > oldnum) /* better fit than 0.25 (=1/4pel) is useless */ |
2854 |
|
meanx /= oldnum; |
2855 |
|
else |
2856 |
|
meanx = 0.25; |
2857 |
|
|
2858 |
|
if (4*meany > oldnum) |
2859 |
|
meany /= oldnum; |
2860 |
|
else |
2861 |
|
meany = 0.25; |
2862 |
|
|
2863 |
|
num = 0; |
2864 |
|
for (my = 0; my < (uint32_t)MBh; my++) |
2865 |
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2866 |
|
{ |
2867 |
|
const int mbnum = mx + my * MBw; |
2868 |
|
const VECTOR mv = pMBs[mbnum].mvs[0]; |
2869 |
|
|
2870 |
|
if (!pMBs[mbnum].mcsel) |
2871 |
|
continue; |
2872 |
|
|
2873 |
|
if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ) > meanx ) |
2874 |
|
|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ) > meany ) ) |
2875 |
|
pMBs[mbnum].mcsel=0; |
2876 |
|
else |
2877 |
|
num++; |
2878 |
|
} |
2879 |
|
|
2880 |
|
} while ( (oldnum != num) && (num>= minblocks) ); |
2881 |
|
|
2882 |
|
if (num < minblocks) |
2883 |
|
{ |
2884 |
|
const int iEdgedWidth = pParam->edged_width; |
2885 |
|
num = 0; |
2886 |
|
|
2887 |
|
/* fprintf(stderr,"Warning! Unreliable GME (%d/%d blocks), falling back to translation.\n",num,MBh*MBw); |
2888 |
|
*/ |
2889 |
|
gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0; |
2890 |
|
|
2891 |
|
if (!(current->motion_flags & XVID_ME_GME_REFINE)) |
2892 |
|
return gmc; |
2893 |
|
|
2894 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ |
2895 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ |
2896 |
|
{ |
2897 |
|
const int mbnum = mx + my * MBw; |
2898 |
|
MACROBLOCK *const pMB = &pMBs[mbnum]; |
2899 |
|
const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); |
2900 |
|
if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) |
2901 |
|
&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) |
2902 |
|
{ pMB->mcsel = 1; |
2903 |
|
gmc.duv[0].x += pMB->mvs[0].x; |
2904 |
|
gmc.duv[0].y += pMB->mvs[0].y; |
2905 |
|
num++; |
2906 |
|
} |
2907 |
|
} |
2908 |
|
|
2909 |
|
if (gmc.duv[0].x) |
2910 |
|
gmc.duv[0].x /= num; |
2911 |
|
if (gmc.duv[0].y) |
2912 |
|
gmc.duv[0].y /= num; |
2913 |
|
} else { |
2914 |
|
|
2915 |
|
gmc.duv[0].x=(int)(sol[0]+0.5); |
2916 |
|
gmc.duv[0].y=(int)(sol[3]+0.5); |
2917 |
|
|
2918 |
|
gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5); |
2919 |
|
gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5); |
2920 |
|
|
2921 |
|
gmc.duv[2].x=-gmc.duv[1].y; /* two warp points only */ |
2922 |
|
gmc.duv[2].y=gmc.duv[1].x; |
2923 |
|
} |
2924 |
|
if (num>maxblocks) |
2925 |
|
{ for (my = 1; my < (uint32_t)MBh-1; my++) |
2926 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2927 |
|
{ |
2928 |
|
const int mbnum = mx + my * MBw; |
2929 |
|
if (pMBs[mbnum-1].mcsel) |
2930 |
|
pMBs[mbnum].mcsel=0; |
2931 |
|
else |
2932 |
|
if (pMBs[mbnum-MBw].mcsel) |
2933 |
|
pMBs[mbnum].mcsel=0; |
2934 |
|
} |
2935 |
|
} |
2936 |
|
return gmc; |
2937 |
|
} |
2938 |
|
|
2939 |
|
int |
2940 |
|
GlobalMotionEstRefine( |
2941 |
|
WARPPOINTS *const startwp, |
2942 |
|
MACROBLOCK * const pMBs, |
2943 |
|
const MBParam * const pParam, |
2944 |
|
const FRAMEINFO * const current, |
2945 |
|
const FRAMEINFO * const reference, |
2946 |
|
const IMAGE * const pCurr, |
2947 |
|
const IMAGE * const pRef, |
2948 |
|
const IMAGE * const pRefH, |
2949 |
|
const IMAGE * const pRefV, |
2950 |
|
const IMAGE * const pRefHV) |
2951 |
|
{ |
2952 |
|
uint8_t* GMCblock = (uint8_t*)malloc(16*pParam->edged_width); |
2953 |
|
WARPPOINTS bestwp=*startwp; |
2954 |
|
WARPPOINTS centerwp,currwp; |
2955 |
|
int gmcminSAD=0; |
2956 |
|
int gmcSAD=0; |
2957 |
|
int direction; |
2958 |
|
// int mx,my; |
2959 |
|
|
2960 |
|
/* use many blocks... */ |
2961 |
|
/* for (my = 0; my < (uint32_t)pParam->mb_height; my++) |
2962 |
|
for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) |
2963 |
|
{ |
2964 |
|
const int mbnum = mx + my * pParam->mb_width; |
2965 |
|
pMBs[mbnum].mcsel=1; |
2966 |
|
} |
2967 |
|
*/ |
2968 |
|
|
2969 |
|
/* or rather don't use too many blocks... */ |
2970 |
|
/* |
2971 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
2972 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2973 |
|
{ |
2974 |
|
const int mbnum = mx + my * MBw; |
2975 |
|
if (MBmask[mbnum-1]) |
2976 |
|
MBmask[mbnum-1]=0; |
2977 |
|
else |
2978 |
|
if (MBmask[mbnum-MBw]) |
2979 |
|
MBmask[mbnum-1]=0; |
2980 |
|
|
2981 |
|
} |
2982 |
|
*/ |
2983 |
|
gmcminSAD = globalSAD(&bestwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
2984 |
|
|
2985 |
|
if ( (reference->coding_type == S_VOP) |
2986 |
|
&& ( (reference->warp.duv[1].x != bestwp.duv[1].x) |
2987 |
|
|| (reference->warp.duv[1].y != bestwp.duv[1].y) |
2988 |
|
|| (reference->warp.duv[0].x != bestwp.duv[0].x) |
2989 |
|
|| (reference->warp.duv[0].y != bestwp.duv[0].y) |
2990 |
|
|| (reference->warp.duv[2].x != bestwp.duv[2].x) |
2991 |
|
|| (reference->warp.duv[2].y != bestwp.duv[2].y) ) ) |
2992 |
|
{ |
2993 |
|
gmcSAD = globalSAD(&reference->warp, pParam, pMBs, |
2994 |
|
current, pRef, pCurr, GMCblock); |
2995 |
|
|
2996 |
|
if (gmcSAD < gmcminSAD) |
2997 |
|
{ bestwp = reference->warp; |
2998 |
|
gmcminSAD = gmcSAD; |
2999 |
|
} |
3000 |
|
} |
3001 |
|
|
3002 |
|
do { |
3003 |
|
direction = 0; |
3004 |
|
centerwp = bestwp; |
3005 |
|
|
3006 |
|
currwp = centerwp; |
3007 |
|
|
3008 |
|
currwp.duv[0].x--; |
3009 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3010 |
|
if (gmcSAD < gmcminSAD) |
3011 |
|
{ bestwp = currwp; |
3012 |
|
gmcminSAD = gmcSAD; |
3013 |
|
direction = 1; |
3014 |
|
} |
3015 |
|
else |
3016 |
|
{ |
3017 |
|
currwp = centerwp; currwp.duv[0].x++; |
3018 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3019 |
|
if (gmcSAD < gmcminSAD) |
3020 |
|
{ bestwp = currwp; |
3021 |
|
gmcminSAD = gmcSAD; |
3022 |
|
direction = 2; |
3023 |
|
} |
3024 |
|
} |
3025 |
|
if (direction) continue; |
3026 |
|
|
3027 |
|
currwp = centerwp; currwp.duv[0].y--; |
3028 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3029 |
|
if (gmcSAD < gmcminSAD) |
3030 |
|
{ bestwp = currwp; |
3031 |
|
gmcminSAD = gmcSAD; |
3032 |
|
direction = 4; |
3033 |
|
} |
3034 |
|
else |
3035 |
|
{ |
3036 |
|
currwp = centerwp; currwp.duv[0].y++; |
3037 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3038 |
|
if (gmcSAD < gmcminSAD) |
3039 |
|
{ bestwp = currwp; |
3040 |
|
gmcminSAD = gmcSAD; |
3041 |
|
direction = 8; |
3042 |
|
} |
3043 |
|
} |
3044 |
|
if (direction) continue; |
3045 |
|
|
3046 |
|
currwp = centerwp; currwp.duv[1].x++; |
3047 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3048 |
|
if (gmcSAD < gmcminSAD) |
3049 |
|
{ bestwp = currwp; |
3050 |
|
gmcminSAD = gmcSAD; |
3051 |
|
direction = 32; |
3052 |
|
} |
3053 |
|
currwp.duv[2].y++; |
3054 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3055 |
|
if (gmcSAD < gmcminSAD) |
3056 |
|
{ bestwp = currwp; |
3057 |
|
gmcminSAD = gmcSAD; |
3058 |
|
direction = 1024; |
3059 |
|
} |
3060 |
|
|
3061 |
|
currwp = centerwp; currwp.duv[1].x--; |
3062 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3063 |
|
if (gmcSAD < gmcminSAD) |
3064 |
|
{ bestwp = currwp; |
3065 |
|
gmcminSAD = gmcSAD; |
3066 |
|
direction = 16; |
3067 |
|
} |
3068 |
|
else |
3069 |
|
{ |
3070 |
|
currwp = centerwp; currwp.duv[1].x++; |
3071 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3072 |
|
if (gmcSAD < gmcminSAD) |
3073 |
|
{ bestwp = currwp; |
3074 |
|
gmcminSAD = gmcSAD; |
3075 |
|
direction = 32; |
3076 |
|
} |
3077 |
|
} |
3078 |
|
if (direction) continue; |
3079 |
|
|
3080 |
|
|
3081 |
|
currwp = centerwp; currwp.duv[1].y--; |
3082 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3083 |
|
if (gmcSAD < gmcminSAD) |
3084 |
|
{ bestwp = currwp; |
3085 |
|
gmcminSAD = gmcSAD; |
3086 |
|
direction = 64; |
3087 |
|
} |
3088 |
|
else |
3089 |
|
{ |
3090 |
|
currwp = centerwp; currwp.duv[1].y++; |
3091 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3092 |
|
if (gmcSAD < gmcminSAD) |
3093 |
|
{ bestwp = currwp; |
3094 |
|
gmcminSAD = gmcSAD; |
3095 |
|
direction = 128; |
3096 |
|
} |
3097 |
|
} |
3098 |
|
if (direction) continue; |
3099 |
|
|
3100 |
|
currwp = centerwp; currwp.duv[2].x--; |
3101 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3102 |
|
if (gmcSAD < gmcminSAD) |
3103 |
|
{ bestwp = currwp; |
3104 |
|
gmcminSAD = gmcSAD; |
3105 |
|
direction = 256; |
3106 |
|
} |
3107 |
|
else |
3108 |
|
{ |
3109 |
|
currwp = centerwp; currwp.duv[2].x++; |
3110 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3111 |
|
if (gmcSAD < gmcminSAD) |
3112 |
|
{ bestwp = currwp; |
3113 |
|
gmcminSAD = gmcSAD; |
3114 |
|
direction = 512; |
3115 |
|
} |
3116 |
|
} |
3117 |
|
if (direction) continue; |
3118 |
|
|
3119 |
|
currwp = centerwp; currwp.duv[2].y--; |
3120 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3121 |
|
if (gmcSAD < gmcminSAD) |
3122 |
|
{ bestwp = currwp; |
3123 |
|
gmcminSAD = gmcSAD; |
3124 |
|
direction = 1024; |
3125 |
|
} |
3126 |
|
else |
3127 |
|
{ |
3128 |
|
currwp = centerwp; currwp.duv[2].y++; |
3129 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3130 |
|
if (gmcSAD < gmcminSAD) |
3131 |
|
{ bestwp = currwp; |
3132 |
|
gmcminSAD = gmcSAD; |
3133 |
|
direction = 2048; |
3134 |
|
} |
3135 |
|
} |
3136 |
|
} while (direction); |
3137 |
|
free(GMCblock); |
3138 |
|
|
3139 |
|
*startwp = bestwp; |
3140 |
|
|
3141 |
|
return gmcminSAD; |
3142 |
|
} |
3143 |
|
|
3144 |
|
int |
3145 |
|
globalSAD(const WARPPOINTS *const wp, |
3146 |
|
const MBParam * const pParam, |
3147 |
|
const MACROBLOCK * const pMBs, |
3148 |
|
const FRAMEINFO * const current, |
3149 |
|
const IMAGE * const pRef, |
3150 |
|
const IMAGE * const pCurr, |
3151 |
|
uint8_t *const GMCblock) |
3152 |
|
{ |
3153 |
|
NEW_GMC_DATA gmc_data; |
3154 |
|
int iSAD, gmcSAD=0; |
3155 |
|
int num=0; |
3156 |
|
unsigned int mx, my; |
3157 |
|
|
3158 |
|
generate_GMCparameters( 3, 3, wp, pParam->width, pParam->height, &gmc_data); |
3159 |
|
|
3160 |
|
for (my = 0; my < (uint32_t)pParam->mb_height; my++) |
3161 |
|
for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) { |
3162 |
|
|
3163 |
|
const int mbnum = mx + my * pParam->mb_width; |
3164 |
|
const int iEdgedWidth = pParam->edged_width; |
3165 |
|
|
3166 |
|
if (!pMBs[mbnum].mcsel) |
3167 |
|
continue; |
3168 |
|
|
3169 |
|
gmc_data.predict_16x16(&gmc_data, GMCblock, |
3170 |
|
pRef->y, |
3171 |
|
iEdgedWidth, |
3172 |
|
iEdgedWidth, |
3173 |
|
mx, my, |
3174 |
|
pParam->m_rounding_type); |
3175 |
|
|
3176 |
|
iSAD = sad16 ( pCurr->y + 16*(my*iEdgedWidth + mx), |
3177 |
|
GMCblock , iEdgedWidth, 65536); |
3178 |
|
iSAD -= pMBs[mbnum].sad16; |
3179 |
|
|
3180 |
|
if (iSAD<0) |
3181 |
|
gmcSAD += iSAD; |
3182 |
|
num++; |
3183 |
|
} |
3184 |
|
return gmcSAD; |
3185 |
|
} |
3186 |
|
|