137 |
xcb = xb/2; ycb = yb/2; |
xcb = xb/2; ycb = yb/2; |
138 |
} |
} |
139 |
|
|
140 |
t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0) |
t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision) |
141 |
+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision); |
142 |
|
|
143 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
144 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
207 |
if (sad > *(data->iMinSAD)) return; |
if (sad > *(data->iMinSAD)) return; |
208 |
} |
} |
209 |
|
|
210 |
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) * sad)>>10; |
211 |
|
|
212 |
if (data->chroma && sad < *data->iMinSAD) |
if (data->chroma && sad < *data->iMinSAD) |
213 |
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
265 |
|
|
266 |
done: |
done: |
267 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
268 |
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) * sad)>>10; |
269 |
|
|
270 |
if (data->chroma && sad < *data->iMinSAD) |
if (data->chroma && sad < *data->iMinSAD) |
271 |
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
291 |
if ( (x > data->max_dx) || ( x < data->min_dx) |
if ( (x > data->max_dx) || ( x < data->min_dx) |
292 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
293 |
|
|
|
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ |
|
|
|
|
294 |
if (data->qpel_precision) { /* x and y are in 1/4 precision */ |
if (data->qpel_precision) { /* x and y are in 1/4 precision */ |
295 |
Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); |
Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); |
296 |
current = data->currentQMV; |
current = data->currentQMV; |
301 |
xc = x; yc = y; |
xc = x; yc = y; |
302 |
} |
} |
303 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
304 |
data->qpel^data->qpel_precision, data->rrv); |
data->qpel^data->qpel_precision); |
305 |
|
|
306 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
307 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
443 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
444 |
|
|
445 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
446 |
pParam->width, pParam->height, iFcode - Data->qpel, 1, 0); |
pParam->width, pParam->height, iFcode - Data->qpel, 1); |
447 |
|
|
448 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
449 |
if (Data->qpel) { |
if (Data->qpel) { |
490 |
if(MotionFlags & XVID_ME_FASTREFINE16) { |
if(MotionFlags & XVID_ME_FASTREFINE16) { |
491 |
/* fast */ |
/* fast */ |
492 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
493 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
494 |
FullRefine_Fast(Data, CheckCandidate16no4v, 0); |
FullRefine_Fast(Data, CheckCandidate16no4v, 0); |
495 |
|
|
496 |
} else { |
} else { |
505 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
506 |
} |
} |
507 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
508 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
509 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
510 |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate16no4v, 0); /* qpel part */ |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate16no4v, 0); /* qpel part */ |
511 |
} |
} |
539 |
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
540 |
|
|
541 |
sum = sad8bi(Data->CurU, |
sum = sad8bi(Data->CurU, |
542 |
Data->RefP[4] + (dy/2) * stride + dx/2, |
Data->RefP[4] + (dy/2) * (int)stride + dx/2, |
543 |
Data->b_RefP[4] + (b_dy/2) * stride + b_dx/2, |
Data->b_RefP[4] + (b_dy/2) * (int)stride + b_dx/2, |
544 |
stride); |
stride); |
545 |
|
|
546 |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
547 |
|
|
548 |
sum += sad8bi(Data->CurV, |
sum += sad8bi(Data->CurV, |
549 |
Data->RefP[5] + (dy/2) * stride + dx/2, |
Data->RefP[5] + (dy/2) * (int)stride + dx/2, |
550 |
Data->b_RefP[5] + (b_dy/2) * stride + b_dx/2, |
Data->b_RefP[5] + (b_dy/2) * (int)stride + b_dx/2, |
551 |
stride); |
stride); |
552 |
|
|
553 |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
711 |
Data->currentMV[0] = startF; |
Data->currentMV[0] = startF; |
712 |
Data->currentMV[1] = startB; |
Data->currentMV[1] = startB; |
713 |
|
|
714 |
get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0); |
get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); |
715 |
get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1, 0); |
get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1); |
716 |
|
|
717 |
if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; |
if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; |
718 |
if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; |
if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; |
741 |
int i, j; |
int i, j; |
742 |
int b_range[4], f_range[4]; |
int b_range[4], f_range[4]; |
743 |
|
|
744 |
get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0); |
get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); |
745 |
get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1, 0); |
get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1); |
746 |
|
|
747 |
/* diamond */ |
/* diamond */ |
748 |
do { |
do { |
772 |
if (Data->qpel) { |
if (Data->qpel) { |
773 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
774 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
775 |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 2, 0); |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 2); |
776 |
|
|
777 |
Data->currentQMV[0].x = 2 * Data->currentMV[0].x; |
Data->currentQMV[0].x = 2 * Data->currentMV[0].x; |
778 |
Data->currentQMV[0].y = 2 * Data->currentMV[0].y; |
Data->currentQMV[0].y = 2 * Data->currentMV[0].y; |
783 |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateInt, 1); |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateInt, 1); |
784 |
|
|
785 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
786 |
x, y, 4, pParam->width, pParam->height, Data->bFcode, 2, 0); |
x, y, 4, pParam->width, pParam->height, Data->bFcode, 2); |
787 |
|
|
788 |
xvid_me_SubpelRefine(Data->currentQMV[1], Data, CheckCandidateInt, 2); |
xvid_me_SubpelRefine(Data->currentQMV[1], Data, CheckCandidateInt, 2); |
789 |
} |
} |
829 |
|
|
830 |
pMB->sad16 = best_sad; |
pMB->sad16 = best_sad; |
831 |
pMB->mode = mode; |
pMB->mode = mode; |
832 |
|
pMB->cbp = 63; |
833 |
|
|
834 |
switch (mode) { |
switch (mode) { |
835 |
|
|
919 |
} |
} |
920 |
} |
} |
921 |
|
|
922 |
|
static __inline void |
923 |
|
maxMotionBVOP(int * const MVmaxF, int * const MVmaxB, const MACROBLOCK * const pMB, const int qpel) |
924 |
|
{ |
925 |
|
if (pMB->mode == MODE_FORWARD || pMB->mode == MODE_INTERPOLATE) { |
926 |
|
const VECTOR * const mv = qpel ? pMB->qmvs : pMB->mvs; |
927 |
|
int max = *MVmaxF; |
928 |
|
if (mv[0].x > max) max = mv[0].x; |
929 |
|
else if (-mv[0].x - 1 > max) max = -mv[0].x - 1; |
930 |
|
if (mv[0].y > max) max = mv[0].y; |
931 |
|
else if (-mv[0].y - 1 > max) max = -mv[0].y - 1; |
932 |
|
|
933 |
|
*MVmaxF = max; |
934 |
|
} |
935 |
|
|
936 |
|
if (pMB->mode == MODE_BACKWARD || pMB->mode == MODE_INTERPOLATE) { |
937 |
|
const VECTOR * const mv = qpel ? pMB->b_qmvs : pMB->b_mvs; |
938 |
|
int max = *MVmaxB; |
939 |
|
if (mv[0].x > max) max = mv[0].x; |
940 |
|
else if (-mv[0].x - 1 > max) max = -mv[0].x - 1; |
941 |
|
if (mv[0].y > max) max = mv[0].y; |
942 |
|
else if (-mv[0].y - 1 > max) max = -mv[0].y - 1; |
943 |
|
*MVmaxB = max; |
944 |
|
} |
945 |
|
} |
946 |
|
|
947 |
|
|
948 |
void |
void |
949 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
950 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
972 |
|
|
973 |
VECTOR f_predMV, b_predMV; |
VECTOR f_predMV, b_predMV; |
974 |
|
|
975 |
|
int MVmaxF = 0, MVmaxB = 0; |
976 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
977 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
978 |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
996 |
memcpy(&Data_b, &Data_d, sizeof(SearchData)); |
memcpy(&Data_b, &Data_d, sizeof(SearchData)); |
997 |
memcpy(&Data_i, &Data_d, sizeof(SearchData)); |
memcpy(&Data_i, &Data_d, sizeof(SearchData)); |
998 |
|
|
999 |
Data_f.iFcode = Data_i.iFcode = frame->fcode; |
Data_f.iFcode = Data_i.iFcode = frame->fcode = b_reference->fcode; |
1000 |
Data_b.iFcode = Data_i.bFcode = frame->bcode; |
Data_b.iFcode = Data_i.bFcode = frame->bcode = b_reference->fcode; |
|
|
|
1001 |
|
|
1002 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
1003 |
|
|
1029 |
|
|
1030 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { |
if (pMB->mode == MODE_DIRECT_NONE_MV) { |
1031 |
pMB->sad16 = best_sad; |
pMB->sad16 = best_sad; |
1032 |
|
pMB->cbp = 0; |
1033 |
continue; |
continue; |
1034 |
} |
} |
1035 |
|
|
1068 |
|
|
1069 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { /* skipped? */ |
if (pMB->mode == MODE_DIRECT_NONE_MV) { /* skipped? */ |
1070 |
pMB->sad16 = skip_sad; |
pMB->sad16 = skip_sad; |
1071 |
|
pMB->cbp = 0; |
1072 |
continue; |
continue; |
1073 |
} |
} |
1074 |
} |
} |
1079 |
else |
else |
1080 |
ModeDecision_BVOP_SAD(&Data_d, &Data_b, &Data_f, &Data_i, pMB, b_mb, &f_predMV, &b_predMV); |
ModeDecision_BVOP_SAD(&Data_d, &Data_b, &Data_f, &Data_i, pMB, b_mb, &f_predMV, &b_predMV); |
1081 |
|
|
1082 |
|
maxMotionBVOP(&MVmaxF, &MVmaxB, pMB, Data_d.qpel); |
1083 |
|
|
1084 |
} |
} |
1085 |
} |
} |
1086 |
|
|
1087 |
|
frame->fcode = getMinFcode(MVmaxF); |
1088 |
|
frame->bcode = getMinFcode(MVmaxB); |
1089 |
} |
} |