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; |
947 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
948 |
const int x, const int y, |
const int x, const int y, |
949 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
950 |
const uint32_t iFcode, |
const uint32_t iFcode, |
951 |
const MBParam * const pParam, |
const MBParam * const pParam, |
952 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
953 |
const VECTOR * const predMV, |
const VECTOR * const predMV, |
954 |
int32_t * const best_sad, |
int32_t * const best_sad, |
955 |
const int32_t mode_current) |
const int32_t mode_current, |
956 |
|
SearchData * const Data) |
957 |
{ |
{ |
958 |
|
|
959 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
960 |
|
|
961 |
int i, iDirection, mask; |
int i, iDirection, mask; |
962 |
VECTOR currentMV, pmv[7]; |
VECTOR pmv[7]; |
963 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
964 |
int32_t iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
965 |
SearchData Data; |
Data->iFcode = iFcode; |
966 |
|
|
967 |
Data.iMinSAD = &iMinSAD; |
Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
968 |
Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; |
969 |
Data.iEdgedWidth = iEdgedWidth; |
Data->RefV = pRefV + (x + y * iEdgedWidth) * 16; |
970 |
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; |
|
971 |
|
|
972 |
Data.iQuant = iQuant; |
Data->predMV = *predMV; |
|
Data.iFcode = iFcode; |
|
|
Data.predMV = *predMV; |
|
973 |
|
|
974 |
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, |
975 |
pParam->width, pParam->height, iFcode); |
pParam->width, pParam->height, iFcode); |
976 |
|
|
977 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
pmv[0] = Data->predMV; |
978 |
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? |
|
|
|
|
|
|
|
|
pmv[0] = Data.predMV; |
|
|
PreparePredictionsBF(pmv, x, y, pParam->mb_width, |
|
|
pMB, mode_current); |
|
979 |
|
|
980 |
currentMV.x = currentMV.y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
981 |
|
|
982 |
CheckCandidate = CheckCandidate16no4v; |
CheckCandidate = CheckCandidate16no4v; |
983 |
|
|
984 |
// main loop. checking all predictions |
// main loop. checking all predictions |
985 |
for (i = 0; i < 8; i++) { |
for (i = 0; i < 8; i++) { |
986 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
987 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, &Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
988 |
} |
} |
989 |
|
|
990 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) |
993 |
MainSearchPtr = AdvDiamondSearch; |
MainSearchPtr = AdvDiamondSearch; |
994 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
995 |
|
|
996 |
(*MainSearchPtr)(currentMV.x, currentMV.y, &Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
997 |
|
|
998 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data); |
HalfpelRefine(Data); |
999 |
|
|
1000 |
// three bits are needed to code backward mode. four for forward |
// three bits are needed to code backward mode. four for forward |
1001 |
// we treat the bits just like they were vector's |
// we treat the bits just like they were vector's |
1002 |
if (mode_current == MODE_FORWARD) iMinSAD += 4 * lambda_vec16[iQuant]; |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant]; |
1003 |
else iMinSAD += 3 * lambda_vec16[iQuant]; |
else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant]; |
1004 |
|
|
1005 |
|
|
1006 |
if (iMinSAD < *best_sad) { |
if (*Data->iMinSAD < *best_sad) { |
1007 |
*best_sad = iMinSAD; |
*best_sad = *Data->iMinSAD; |
1008 |
pMB->mode = mode_current; |
pMB->mode = mode_current; |
1009 |
pMB->pmvs[0].x = currentMV.x - predMV->x; |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1010 |
pMB->pmvs[0].y = currentMV.y - predMV->y; |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
1011 |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = currentMV; |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
1012 |
else pMB->b_mvs[0] = currentMV; |
else pMB->b_mvs[0] = *Data->currentMV; |
1013 |
} |
} |
1014 |
|
|
1015 |
} |
} |
1026 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1027 |
const int x, const int y, |
const int x, const int y, |
1028 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
1029 |
const int32_t TRB, const int32_t TRD, |
const int32_t TRB, const int32_t TRD, |
1030 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1031 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1032 |
const MACROBLOCK * const b_mb, |
const MACROBLOCK * const b_mb, |
1033 |
int32_t * const best_sad) |
int32_t * const best_sad, |
1034 |
|
SearchData * const Data) |
1035 |
|
|
1036 |
{ |
{ |
1037 |
const uint32_t iEdgedWidth = pParam->edged_width; |
int32_t skip_sad; |
|
int32_t iMinSAD = 266*4096, skip_sad; |
|
1038 |
int k; |
int k; |
1039 |
VECTOR currentMV; |
|
1040 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
|
SearchData Data; |
|
1041 |
|
|
1042 |
Data.iMinSAD = &iMinSAD; |
*Data->iMinSAD = 256*4096; |
1043 |
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; |
|
1044 |
|
|
1045 |
Data.Ref= f_Ref->y + (x + iEdgedWidth*y) * 16; |
Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1046 |
Data.RefH = f_RefH + (x + iEdgedWidth*y) * 16; |
Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; |
1047 |
Data.RefV = f_RefV + (x + iEdgedWidth*y) * 16; |
Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16; |
1048 |
Data.RefHV = f_RefHV + (x + iEdgedWidth*y) * 16; |
Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16; |
1049 |
Data.bRef = b_Ref->y + (x + iEdgedWidth*y) * 16; |
Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1050 |
Data.bRefH = b_RefH + (x + iEdgedWidth*y) * 16; |
Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16; |
1051 |
Data.bRefV = b_RefV + (x + iEdgedWidth*y) * 16; |
Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16; |
1052 |
Data.bRefHV = b_RefHV + (x + iEdgedWidth*y) * 16; |
Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16; |
1053 |
/* |
|
1054 |
//What we do here is a complicated version of CheckCandidateDirect(0,0); |
Data->max_dx = 2 * pParam->width - 2 * (x) * 16; |
1055 |
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; |
1056 |
|
Data->min_dx = -(2 * 16 + 2 * (x) * 16); |
1057 |
*/ |
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); |
|
1058 |
|
|
1059 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1060 |
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); |
1061 |
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; |
1062 |
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); |
1063 |
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; |
1064 |
|
|
1065 |
if (( pMB->mvs[k].x > Data.max_dx ) || ( pMB->mvs[k].x < Data.min_dx ) |
if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx ) |
1066 |
|| ( pMB->mvs[k].y > Data.max_dy ) || ( pMB->mvs[k].y < Data.min_dy ) |
|| ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_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 )) { |
|
1067 |
|
|
1068 |
*best_sad = 256*4096; // in that case, we won't use direct mode |
*best_sad = 256*4096; // in that case, we won't use direct mode |
1069 |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
1073 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1074 |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
1075 |
pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0]; |
pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0]; |
1076 |
Data.directmvF[1] = Data.directmvF[2] = Data.directmvF[3] = Data.directmvF[0]; |
Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0]; |
1077 |
Data.directmvB[1] = Data.directmvB[2] = Data.directmvB[3] = Data.directmvB[0]; |
Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0]; |
1078 |
break; |
break; |
1079 |
} |
} |
1080 |
} |
} |
1083 |
CheckCandidate = CheckCandidateDirect; |
CheckCandidate = CheckCandidateDirect; |
1084 |
else CheckCandidate = CheckCandidateDirectno4v; |
else CheckCandidate = CheckCandidateDirectno4v; |
1085 |
|
|
1086 |
(*CheckCandidate)(0, 0, 255, &k, &Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1087 |
|
|
1088 |
// skip decision |
// skip decision |
1089 |
if (iMinSAD - 2 * lambda_vec16[iQuant] < (int32_t)iQuant * SKIP_THRESH_B) { |
if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
1090 |
//checking chroma. everything copied from MC |
//checking chroma. everything copied from MC |
1091 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1092 |
int sum, dx, dy, b_dx, b_dy; |
int sum, dx, dy, b_dx, b_dy; |
1103 |
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
1104 |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1105 |
|
|
1106 |
sum = sad8bi(pCur->u + 8*x + 8*y*(iEdgedWidth/2), |
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1107 |
f_Ref->u + (y*8 + dy/2) * (iEdgedWidth/2) + x*8 + dx/2, |
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1108 |
b_Ref->u + (y*8 + b_dy/2) * (iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1109 |
iEdgedWidth/2); |
Data->iEdgedWidth/2); |
1110 |
sum += sad8bi(pCur->v + 8*x + 8*y*(iEdgedWidth/2), |
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
1111 |
f_Ref->v + (y*8 + dy/2) * (iEdgedWidth/2) + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1112 |
b_Ref->v + (y*8 + b_dy/2) * (iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1113 |
iEdgedWidth/2); |
Data->iEdgedWidth/2); |
1114 |
|
|
1115 |
if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data.iQuant) { |
if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) { |
1116 |
pMB->mode = MODE_DIRECT_NONE_MV; |
pMB->mode = MODE_DIRECT_NONE_MV; |
1117 |
return iMinSAD; |
return *Data->iMinSAD; |
1118 |
} |
} |
1119 |
} |
} |
1120 |
|
|
1121 |
skip_sad = iMinSAD; |
skip_sad = *Data->iMinSAD; |
1122 |
|
|
1123 |
// DIRECT MODE DELTA VECTOR SEARCH. |
// DIRECT MODE DELTA VECTOR SEARCH. |
1124 |
// 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 |
1127 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1128 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1129 |
|
|
1130 |
(*MainSearchPtr)(0, 0, &Data, 255); |
(*MainSearchPtr)(0, 0, Data, 255); |
1131 |
|
|
1132 |
HalfpelRefine(&Data); |
HalfpelRefine(Data); |
1133 |
|
|
1134 |
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 |
1135 |
*best_sad = iMinSAD; |
*best_sad = *Data->iMinSAD; |
1136 |
|
|
1137 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1138 |
pMB->mode = MODE_DIRECT; |
pMB->mode = MODE_DIRECT; |
1139 |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
1140 |
|
|
1141 |
pMB->pmvs[3] = currentMV; |
pMB->pmvs[3] = *Data->currentMV; |
1142 |
|
|
1143 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1144 |
pMB->mvs[k].x = Data.directmvF[k].x + currentMV.x; |
pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x; |
1145 |
pMB->b_mvs[k].x = ((currentMV.x == 0) |
pMB->b_mvs[k].x = ((Data->currentMV->x == 0) |
1146 |
? Data.directmvB[k].x |
? Data->directmvB[k].x |
1147 |
: pMB->mvs[k].x - Data.referencemv[k].x); |
: pMB->mvs[k].x - Data->referencemv[k].x); |
1148 |
pMB->mvs[k].y = (Data.directmvF[k].y + currentMV.y); |
pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y); |
1149 |
pMB->b_mvs[k].y = ((currentMV.y == 0) |
pMB->b_mvs[k].y = ((Data->currentMV->y == 0) |
1150 |
? Data.directmvB[k].y |
? Data->directmvB[k].y |
1151 |
: pMB->mvs[k].y - Data.referencemv[k].y); |
: pMB->mvs[k].y - Data->referencemv[k].y); |
1152 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1153 |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
1154 |
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]; |
1155 |
break; |
break; |
1156 |
} |
} |
1157 |
} |
} |
1158 |
return 0;//skip_sad; |
return skip_sad; |
1159 |
} |
} |
1160 |
|
|
1161 |
|
|
1162 |
static __inline void |
static __inline void |
1163 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const uint8_t * const f_Ref, |
1164 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1173 |
const uint32_t fcode, |
const uint32_t fcode, |
1174 |
const uint32_t bcode, |
const uint32_t bcode, |
1175 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
|
const uint32_t iQuant, |
|
1176 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1177 |
const VECTOR * const f_predMV, |
const VECTOR * const f_predMV, |
1178 |
const VECTOR * const b_predMV, |
const VECTOR * const b_predMV, |
1179 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1180 |
int32_t * const best_sad) |
int32_t * const best_sad, |
1181 |
|
SearchData * const fData) |
1182 |
|
|
1183 |
{ |
{ |
|
/* 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". */ |
|
1184 |
|
|
1185 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1186 |
|
|
1187 |
int iDirection, i, j; |
int iDirection, i, j; |
1188 |
int32_t iMinSAD = 256*4096; |
SearchData bData; |
|
VECTOR currentMV[3]; |
|
|
SearchData fData, bData; |
|
|
|
|
|
fData.iMinSAD = bData.iMinSAD = &iMinSAD; |
|
1189 |
|
|
1190 |
fData.Cur = bData.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
bData.iMinSAD = fData->iMinSAD; |
1191 |
fData.iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
*bData.iMinSAD = 4096*256; |
1192 |
fData.currentMV = currentMV; bData.currentMV = currentMV + 1; |
bData.Cur = fData->Cur; |
1193 |
fData.iQuant = bData.iQuant = iQuant; |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
1194 |
fData.iFcode = bData.bFcode = fcode; fData.bFcode = bData.iFcode = bcode; |
bData.currentMV = fData->currentMV + 1; |
1195 |
|
bData.iQuant = fData->iQuant; |
1196 |
bData.bRef = fData.Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1197 |
bData.bRefH = fData.RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
|
1198 |
bData.bRefV = fData.RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
1199 |
bData.bRefHV = fData.RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
1200 |
bData.Ref = fData.bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
1201 |
bData.RefH = fData.bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
1202 |
bData.RefV = fData.bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
1203 |
bData.RefHV = fData.bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
1204 |
|
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
1205 |
bData.bpredMV = fData.predMV = *f_predMV; |
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
1206 |
fData.bpredMV = bData.predMV = *b_predMV; |
|
1207 |
|
bData.bpredMV = fData->predMV = *f_predMV; |
1208 |
currentMV[0] = pMB->mvs[0]; |
fData->bpredMV = bData.predMV = *b_predMV; |
1209 |
currentMV[1] = pMB->b_mvs[0]; |
|
1210 |
get_range(&fData.min_dx, &fData.max_dx, &fData.min_dy, &fData.max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
fData->currentMV[0] = pMB->mvs[0]; |
1211 |
|
fData->currentMV[1] = pMB->b_mvs[0]; |
1212 |
|
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
1213 |
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); |
1214 |
|
|
1215 |
if (currentMV[0].x > fData.max_dx) currentMV[0].x = fData.max_dx; |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1216 |
if (currentMV[0].x < fData.min_dx) currentMV[0].x = fData.min_dy; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
1217 |
if (currentMV[0].y > fData.max_dy) currentMV[0].y = fData.max_dx; |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; |
1218 |
if (currentMV[0].y > fData.min_dy) currentMV[0].y = fData.min_dy; |
if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
1219 |
|
|
1220 |
if (currentMV[1].x > bData.max_dx) currentMV[1].x = bData.max_dx; |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1221 |
if (currentMV[1].x < bData.min_dx) currentMV[1].x = bData.min_dy; |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; |
1222 |
if (currentMV[1].y > bData.max_dy) currentMV[1].y = bData.max_dx; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; |
1223 |
if (currentMV[1].y > bData.min_dy) currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
1224 |
|
|
1225 |
CheckCandidateInt(currentMV[0].x, currentMV[0].y, 255, &iDirection, &fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1226 |
|
|
1227 |
//diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
//diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
1228 |
|
|
1229 |
do { |
do { |
1230 |
iDirection = 255; |
iDirection = 255; |
1231 |
// forward MV moves |
// forward MV moves |
1232 |
i = currentMV[0].x; j = currentMV[0].y; |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
1233 |
|
|
1234 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &fData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
1235 |
CheckCandidateInt(i, j + 1, 0, &iDirection, &fData); |
CheckCandidateInt(i, j + 1, 0, &iDirection, fData); |
1236 |
CheckCandidateInt(i - 1, j, 0, &iDirection, &fData); |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
1237 |
CheckCandidateInt(i, j - 1, 0, &iDirection, &fData); |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
1238 |
|
|
1239 |
// backward MV moves |
// backward MV moves |
1240 |
i = currentMV[1].x; j = currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
1241 |
currentMV[2] = currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
1242 |
|
|
1243 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
1244 |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
1248 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1249 |
|
|
1250 |
// 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 |
1251 |
iMinSAD += 2 * lambda_vec16[iQuant]; |
*fData->iMinSAD += 2 * lambda_vec16[fData->iQuant]; |
1252 |
if (iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1253 |
*best_sad = iMinSAD; |
*best_sad = *fData->iMinSAD; |
1254 |
pMB->mvs[0] = currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1255 |
pMB->b_mvs[0] = currentMV[1]; |
pMB->b_mvs[0] = fData->currentMV[1]; |
1256 |
pMB->mode = MODE_INTERPOLATE; |
pMB->mode = MODE_INTERPOLATE; |
1257 |
|
|
1258 |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
1262 |
} |
} |
1263 |
} |
} |
1264 |
|
|
1265 |
|
|
1266 |
void |
void |
1267 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
1268 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
1291 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
1292 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
1293 |
|
|
1294 |
|
// some pre-inintialized data for the rest of the search |
1295 |
|
|
1296 |
|
SearchData Data; |
1297 |
|
int32_t iMinSAD; |
1298 |
|
VECTOR currentMV[3]; |
1299 |
|
Data.iEdgedWidth = pParam->edged_width; |
1300 |
|
Data.currentMV = currentMV; |
1301 |
|
Data.iMinSAD = &iMinSAD; |
1302 |
|
Data.iQuant = frame->quant; |
1303 |
|
|
1304 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
1305 |
|
|
1306 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
1317 |
continue; |
continue; |
1318 |
} |
} |
1319 |
|
|
1320 |
|
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
1321 |
/* direct search comes first, because it (1) checks for SKIP-mode |
/* direct search comes first, because it (1) checks for SKIP-mode |
1322 |
and (2) sets very good predictions for forward and backward search */ |
and (2) sets very good predictions for forward and backward search */ |
1323 |
|
|
1326 |
&frame->image, |
&frame->image, |
1327 |
i, j, |
i, j, |
1328 |
frame->motion_flags, |
frame->motion_flags, |
|
frame->quant, |
|
1329 |
TRB, TRD, |
TRB, TRD, |
1330 |
pParam, |
pParam, |
1331 |
pMB, b_mb, |
pMB, b_mb, |
1332 |
&best_sad); |
&best_sad, |
1333 |
|
&Data); |
1334 |
|
|
1335 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
1336 |
|
|
1341 |
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, |
1342 |
&frame->image, i, j, |
&frame->image, i, j, |
1343 |
frame->motion_flags, |
frame->motion_flags, |
1344 |
frame->quant, frame->fcode, pParam, |
frame->fcode, pParam, |
1345 |
pMB, &f_predMV, &best_sad, |
pMB, &f_predMV, &best_sad, |
1346 |
MODE_FORWARD); |
MODE_FORWARD, &Data); |
1347 |
|
|
1348 |
// backward search |
// backward search |
1349 |
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, |
1350 |
&frame->image, i, j, |
&frame->image, i, j, |
1351 |
frame->motion_flags, |
frame->motion_flags, |
1352 |
frame->quant, frame->bcode, pParam, |
frame->bcode, pParam, |
1353 |
pMB, &b_predMV, &best_sad, |
pMB, &b_predMV, &best_sad, |
1354 |
MODE_BACKWARD); |
MODE_BACKWARD, &Data); |
1355 |
|
|
1356 |
// 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 |
1357 |
|
|
1361 |
i, j, |
i, j, |
1362 |
frame->fcode, frame->bcode, |
frame->fcode, frame->bcode, |
1363 |
frame->motion_flags, |
frame->motion_flags, |
1364 |
frame->quant, pParam, |
pParam, |
1365 |
&f_predMV, &b_predMV, |
&f_predMV, &b_predMV, |
1366 |
pMB, &best_sad); |
pMB, &best_sad, |
1367 |
|
&Data); |
1368 |
|
|
1369 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1370 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1463 |
const int y, |
const int y, |
1464 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1465 |
const uint32_t iQuant, |
const uint32_t iQuant, |
|
const uint32_t iFcode, |
|
1466 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1467 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1468 |
int inter4v, |
int inter4v, |
1469 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB, |
1470 |
|
SearchData * Data) |
1471 |
{ |
{ |
1472 |
|
|
1473 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1474 |
|
|
1475 |
int i; |
int i; |
|
VECTOR currentMV[5]; |
|
|
int32_t iMinSAD[5]; |
|
|
int32_t temp[5]; |
|
1476 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
|
SearchData Data; |
|
1477 |
|
|
1478 |
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1479 |
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, |
1480 |
pParam->width, pParam->height, iFcode); |
pParam->width, pParam->height, Data->iFcode); |
1481 |
|
|
1482 |
Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
1483 |
Data.iEdgedWidth = iEdgedWidth; |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
1484 |
Data.currentMV = currentMV; |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
1485 |
Data.iMinSAD = iMinSAD; |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
1486 |
Data.Ref = pRef + (x + iEdgedWidth*y)*16; |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
1487 |
Data.RefH = pRefH + (x + iEdgedWidth*y) * 16; |
Data->iQuant = iQuant; |
|
Data.RefV = pRefV + (x + iEdgedWidth*y) * 16; |
|
|
Data.RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
|
|
Data.temp = temp; |
|
|
Data.iQuant = iQuant; |
|
|
Data.iFcode = iFcode; |
|
1488 |
|
|
1489 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1490 |
Data.min_dx = EVEN(Data.min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
1491 |
Data.max_dx = EVEN(Data.max_dx); |
Data->max_dx = EVEN(Data->max_dx); |
1492 |
Data.min_dy = EVEN(Data.min_dy); |
Data->min_dy = EVEN(Data->min_dy); |
1493 |
Data.max_dy = EVEN(Data.max_dy); |
Data->max_dy = EVEN(Data->max_dy); |
1494 |
} |
} |
1495 |
|
|
1496 |
for(i = 0; i < 5; i++) iMinSAD[i] = MV_MAX_ERROR; |
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
1497 |
|
|
1498 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1499 |
|
|
1504 |
|
|
1505 |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
1506 |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
1507 |
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 |
1508 |
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; |
1509 |
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; |
1510 |
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; |
1511 |
|
|
1512 |
CheckCandidate16(pMB->mvs[0].x, pMB->mvs[0].y, 0, &i, &Data); |
(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &i, Data); |
1513 |
|
|
1514 |
if (pMB->mode == MODE_INTER4V) |
if (pMB->mode == MODE_INTER4V) |
1515 |
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 |
1516 |
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
1517 |
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
1518 |
if (!(make_mask(pMB->mvs, i))) |
if (!(make_mask(pMB->mvs, i))) |
1519 |
CheckCandidate16(pMB->mvs[i].x, pMB->mvs[i].y, 0, &i, &Data); |
(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &i, Data); |
1520 |
} |
} |
1521 |
|
|
1522 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) |
1525 |
MainSearchPtr = AdvDiamondSearch; |
MainSearchPtr = AdvDiamondSearch; |
1526 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1527 |
|
|
1528 |
(*MainSearchPtr)(currentMV->x, currentMV->y, &Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1529 |
|
|
1530 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1531 |
|
|
1532 |
if (inter4v) |
if (inter4v) |
1533 |
for(i = 0; i < 4; i++) |
for(i = 0; i < 4; i++) |
1534 |
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); |
1535 |
|
|
1536 |
if (!(inter4v) || |
if (!(inter4v) || |
1537 |
(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] + |
1538 |
|
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1539 |
// INTER MODE |
// INTER MODE |
1540 |
|
|
1541 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
1542 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1543 |
= pMB->mvs[2] = pMB->mvs[3] = currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1544 |
|
|
1545 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1546 |
pMB->sad8[2] = pMB->sad8[3] = iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1547 |
|
|
1548 |
pMB->pmvs[0].x = currentMV[0].x - Data.predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1549 |
pMB->pmvs[0].y = currentMV[0].y - Data.predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1550 |
} else { |
} else { |
1551 |
// INTER4V MODE; all other things are already set in Search8hinted |
// INTER4V MODE; all other things are already set in Search8hinted |
1552 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1553 |
pMB->sad16 = iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * iQuant; |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] |
1554 |
|
+ Data->iMinSAD[4] + IMV16X16 * iQuant; |
1555 |
} |
} |
1556 |
|
|
1557 |
} |
} |
1569 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
1570 |
|
|
1571 |
uint32_t x, y; |
uint32_t x, y; |
1572 |
|
int32_t temp[5]; |
1573 |
|
int32_t iMinSAD[5]; |
1574 |
|
VECTOR currentMV[5]; |
1575 |
|
SearchData Data; |
1576 |
|
Data.iEdgedWidth = pParam->edged_width; |
1577 |
|
Data.currentMV = currentMV; |
1578 |
|
Data.iMinSAD = iMinSAD; |
1579 |
|
Data.temp = temp; |
1580 |
|
Data.iFcode = current->fcode; |
1581 |
|
|
1582 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1583 |
|
|
1595 |
|
|
1596 |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1597 |
y, current->motion_flags, pMB->quant, |
y, current->motion_flags, pMB->quant, |
1598 |
current->fcode, pParam, pMBs, |
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
1599 |
current->global_flags & XVID_INTER4V, pMB); |
&Data); |
1600 |
|
|
1601 |
} |
} |
1602 |
} |
} |
1607 |
const uint8_t * const pCur, |
const uint8_t * const pCur, |
1608 |
const int x, |
const int x, |
1609 |
const int y, |
const int y, |
|
const uint32_t iFcode, |
|
1610 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1611 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1612 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB, |
1613 |
|
SearchData * const Data) |
1614 |
{ |
{ |
1615 |
|
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
1616 |
int i, mask; |
int i, mask; |
1617 |
VECTOR currentMV, pmv[3]; |
VECTOR pmv[3]; |
|
int32_t iMinSAD = MV_MAX_ERROR; |
|
|
SearchData Data; |
|
1618 |
|
|
1619 |
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
*(Data->iMinSAD) = MV_MAX_ERROR; |
1620 |
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1621 |
pParam->width, pParam->height, iFcode); |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1622 |
|
pParam->width, pParam->height, Data->iFcode); |
1623 |
|
|
1624 |
Data.Cur = pCur + (x + y * iEdgedWidth) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
1625 |
Data.iEdgedWidth = iEdgedWidth; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
|
Data.currentMV = ¤tMV; |
|
|
Data.iMinSAD = &iMinSAD; |
|
|
Data.Ref = pRef + (x + iEdgedWidth*y)*16; |
|
|
Data.iQuant = 2; |
|
|
Data.iFcode = iFcode; |
|
1626 |
|
|
1627 |
CheckCandidate = CheckCandidate16no4vI; |
CheckCandidate = CheckCandidate16no4vI; |
1628 |
|
|
1629 |
pmv[1].x = EVEN(pMB->mvs[0].x); |
pmv[1].x = EVEN(pMB->mvs[0].x); |
1630 |
pmv[1].y = EVEN(pMB->mvs[0].y); |
pmv[1].y = EVEN(pMB->mvs[0].y); |
1631 |
pmv[0].x = EVEN(Data.predMV.x); |
pmv[0].x = EVEN(Data->predMV.x); |
1632 |
pmv[0].y = EVEN(Data.predMV.y); |
pmv[0].y = EVEN(Data->predMV.y); |
1633 |
pmv[2].x = pmv[2].y = 0; |
pmv[2].x = pmv[2].y = 0; |
1634 |
|
|
1635 |
CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, &Data); |
CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data); |
1636 |
if (!(mask = make_mask(pmv, 1))) |
if (!(mask = make_mask(pmv, 1))) |
1637 |
CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, &Data); |
CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data); |
1638 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
1639 |
CheckCandidate16no4vI(0, 0, mask, &i, &Data); |
CheckCandidate16no4vI(0, 0, mask, &i, Data); |
1640 |
|
|
1641 |
DiamondSearch(currentMV.x, currentMV.y, &Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1642 |
|
|
1643 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1644 |
= pMB->mvs[2] = pMB->mvs[3] = currentMV; // all, for future get_pmv() |
= pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv() |
1645 |
|
|
1646 |
return iMinSAD; |
return *(Data->iMinSAD); |
1647 |
} |
} |
1648 |
|
|
1649 |
#define INTRA_THRESH 1350 |
#define INTRA_THRESH 1350 |
1659 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
1660 |
int sSAD = 0; |
int sSAD = 0; |
1661 |
|
|
1662 |
|
VECTOR currentMV; |
1663 |
|
int32_t iMinSAD; |
1664 |
|
SearchData Data; |
1665 |
|
Data.iEdgedWidth = pParam->edged_width; |
1666 |
|
Data.currentMV = ¤tMV; |
1667 |
|
Data.iMinSAD = &iMinSAD; |
1668 |
|
Data.iFcode = iFcode; |
1669 |
|
Data.iQuant = 2; |
1670 |
|
|
1671 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1672 |
|
|
1673 |
for (y = 0; y < pParam->mb_height-1; y++) { |
for (y = 0; y < pParam->mb_height-1; y++) { |
1674 |
for (x = 0; x < pParam->mb_width; x++) { |
for (x = 0; x < pParam->mb_width; x++) { |
1675 |
int sad, dev; |
int sad, dev; |
|
|
|
1676 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
1677 |
|
|
1678 |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
1679 |
iFcode, pParam, pMBs, pMB); |
pParam, pMBs, pMB, &Data); |
1680 |
|
|
1681 |
if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work |
if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work |
1682 |
if (sad > INTRA_THRESH) { |
if (sad > INTRA_THRESH) { |