46 |
#include "../utils/emms.h" |
#include "../utils/emms.h" |
47 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
48 |
|
|
49 |
|
/***************************************************************************** |
50 |
|
* Modified rounding tables -- declared in motion.h |
51 |
|
* Original tables see ISO spec tables 7-6 -> 7-9 |
52 |
|
****************************************************************************/ |
53 |
|
|
54 |
|
const uint32_t roundtab[16] = |
55 |
|
{0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; |
56 |
|
|
57 |
|
/* K = 4 */ |
58 |
|
const uint32_t roundtab_76[16] = |
59 |
|
{ 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1 }; |
60 |
|
|
61 |
|
/* K = 2 */ |
62 |
|
const uint32_t roundtab_78[8] = |
63 |
|
{ 0, 0, 1, 1, 0, 0, 0, 1 }; |
64 |
|
|
65 |
|
/* K = 1 */ |
66 |
|
const uint32_t roundtab_79[4] = |
67 |
|
{ 0, 1, 0, 0 }; |
68 |
|
|
69 |
#define INITIAL_SKIP_THRESH (10) |
#define INITIAL_SKIP_THRESH (10) |
70 |
#define FINAL_SKIP_THRESH (50) |
#define FINAL_SKIP_THRESH (50) |
71 |
#define MAX_SAD00_FOR_SKIP (20) |
#define MAX_SAD00_FOR_SKIP (20) |
74 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
75 |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
76 |
|
|
77 |
|
/***************************************************************************** |
78 |
|
* Code |
79 |
|
****************************************************************************/ |
80 |
|
|
81 |
static __inline uint32_t |
static __inline uint32_t |
82 |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
83 |
{ |
{ |
249 |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
250 |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
251 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
252 |
case 0: // pure halfpel position |
case 3: // x and y in qpel resolution - the "corners" (top left/right and |
253 |
return (uint8_t *) ref1; |
// bottom left/right) during qpel refinement |
254 |
|
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
255 |
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
256 |
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
257 |
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
258 |
|
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
259 |
|
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
260 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
261 |
break; |
break; |
262 |
|
|
263 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
272 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
273 |
break; |
break; |
274 |
|
|
275 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
default: // pure halfpel position |
276 |
// bottom left/right) during qpel refinement |
return (uint8_t *) ref1; |
277 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
|
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
|
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
|
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
|
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
|
|
break; |
|
278 |
} |
} |
279 |
return Reference; |
return Reference; |
280 |
} |
} |
319 |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
320 |
break; |
break; |
321 |
|
|
322 |
case 0: // pure halfpel position |
default: // pure halfpel position |
323 |
return (uint8_t *) ref1; |
return (uint8_t *) ref1; |
324 |
} |
} |
325 |
return Reference; |
return Reference; |
405 |
uint32_t t; |
uint32_t t; |
406 |
const uint8_t * Reference; |
const uint8_t * Reference; |
407 |
|
|
408 |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero even value |
409 |
(x > data->max_dx) || (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
410 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
411 |
|
|
440 |
uint32_t t; |
uint32_t t; |
441 |
VECTOR * current; |
VECTOR * current; |
442 |
|
|
443 |
if ( (x > data->max_dx) | ( x < data->min_dx) |
if ( (x > data->max_dx) || ( x < data->min_dx) |
444 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
445 |
|
|
446 |
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; //non-zero even value |
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; //non-zero even value |
447 |
|
|
506 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
507 |
VECTOR *current; |
VECTOR *current; |
508 |
|
|
509 |
if ( (xf > data->max_dx) | (xf < data->min_dx) |
if ((xf > data->max_dx) || (xf < data->min_dx) || |
510 |
| (yf > data->max_dy) | (yf < data->min_dy) ) return; |
(yf > data->max_dy) || (yf < data->min_dy)) |
511 |
|
return; |
512 |
|
|
513 |
if (!data->qpel_precision) { |
if (!data->qpel_precision) { |
514 |
ReferenceF = GetReference(xf, yf, data); |
ReferenceF = GetReference(xf, yf, data); |
553 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
554 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
555 |
|
|
556 |
if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
557 |
|
|
558 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
559 |
mvs.x = data->directmvF[k].x + x; |
mvs.x = data->directmvF[k].x + x; |
566 |
data->directmvB[k].y |
data->directmvB[k].y |
567 |
: mvs.y - data->referencemv[k].y); |
: mvs.y - data->referencemv[k].y); |
568 |
|
|
569 |
if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx) |
if ((mvs.x > data->max_dx) || (mvs.x < data->min_dx) || |
570 |
| (mvs.y > data->max_dy) | (mvs.y < data->min_dy) |
(mvs.y > data->max_dy) || (mvs.y < data->min_dy) || |
571 |
| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx) |
(b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) || |
572 |
| (b_mvs.y > data->max_dy) | (b_mvs.y < data->min_dy) ) return; |
(b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) |
573 |
|
return; |
574 |
|
|
575 |
if (data->qpel) { |
if (data->qpel) { |
576 |
xcf += mvs.x/2; ycf += mvs.y/2; |
xcf += mvs.x/2; ycf += mvs.y/2; |
612 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
613 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
614 |
|
|
615 |
if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
616 |
|
|
617 |
mvs.x = data->directmvF[0].x + x; |
mvs.x = data->directmvF[0].x + x; |
618 |
b_mvs.x = ((x == 0) ? |
b_mvs.x = ((x == 0) ? |
624 |
data->directmvB[0].y |
data->directmvB[0].y |
625 |
: mvs.y - data->referencemv[0].y); |
: mvs.y - data->referencemv[0].y); |
626 |
|
|
627 |
if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx) |
if ( (mvs.x > data->max_dx) || (mvs.x < data->min_dx) |
628 |
| (mvs.y > data->max_dy) | (mvs.y < data->min_dy) |
|| (mvs.y > data->max_dy) || (mvs.y < data->min_dy) |
629 |
| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx) |
|| (b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) |
630 |
| (b_mvs.y > data->max_dy) | (b_mvs.y < data->min_dy) ) return; |
|| (b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) ) return; |
631 |
|
|
632 |
if (data->qpel) { |
if (data->qpel) { |
633 |
xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); |
xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); |
661 |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
662 |
{ |
{ |
663 |
|
|
664 |
static int16_t in[64], coeff[64]; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
665 |
int32_t bits = 0, sum; |
int32_t bits = 0, sum; |
666 |
VECTOR * current; |
VECTOR * current; |
667 |
const uint8_t * ptr; |
const uint8_t * ptr; |
723 |
} |
} |
724 |
} |
} |
725 |
|
|
726 |
bits += cbpy_tab[15-(cbp>>2)].len; |
bits += xvid_cbpy_tab[15-(cbp>>2)].len; |
727 |
bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
728 |
|
|
729 |
if (bits < data->iMinSAD[0]) { |
if (bits < data->iMinSAD[0]) { |
746 |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
747 |
{ |
{ |
748 |
|
|
749 |
static int16_t in[64], coeff[64]; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
750 |
int32_t sum, bits; |
int32_t sum, bits; |
751 |
VECTOR * current; |
VECTOR * current; |
752 |
const uint8_t * ptr; |
const uint8_t * ptr; |
952 |
const uint32_t stride, const uint32_t iQuant, int rrv) |
const uint32_t stride, const uint32_t iQuant, int rrv) |
953 |
|
|
954 |
{ |
{ |
955 |
|
int offset = (x + y*stride)*8; |
956 |
if(!rrv) { |
if(!rrv) { |
957 |
uint32_t sadC = sad8(current->u + x*8 + y*stride*8, |
uint32_t sadC = sad8(current->u + offset, |
958 |
reference->u + x*8 + y*stride*8, stride); |
reference->u + offset, stride); |
959 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
960 |
sadC += sad8(current->v + (x + y*stride)*8, |
sadC += sad8(current->v + offset, |
961 |
reference->v + (x + y*stride)*8, stride); |
reference->v + offset, stride); |
962 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
963 |
return 1; |
return 1; |
964 |
|
|
965 |
} else { |
} else { |
966 |
uint32_t sadC = sad16(current->u + x*16 + y*stride*16, |
uint32_t sadC = sad16(current->u + 2*offset, |
967 |
reference->u + x*16 + y*stride*16, stride, 256*4096); |
reference->u + 2*offset, stride, 256*4096); |
968 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
969 |
sadC += sad16(current->v + (x + y*stride)*16, |
sadC += sad16(current->v + 2*offset, |
970 |
reference->v + (x + y*stride)*16, stride, 256*4096); |
reference->v + 2*offset, stride, 256*4096); |
971 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
972 |
return 1; |
return 1; |
973 |
} |
} |
1003 |
uint32_t x, y; |
uint32_t x, y; |
1004 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
1005 |
int32_t quant = current->quant, sad00; |
int32_t quant = current->quant, sad00; |
1006 |
|
int skip_thresh = INITIAL_SKIP_THRESH * |
1007 |
|
(current->global_flags & XVID_REDUCED ? 4:1) * |
1008 |
|
(current->global_flags & XVID_MODEDECISION_BITS ? 2:1); |
1009 |
|
|
1010 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
1011 |
int32_t temp[8]; |
int32_t temp[8]; |
1012 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
1013 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
1014 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
1015 |
|
DECLARE_ALIGNED_MATRIX(dct_space, 2, 64, int16_t, CACHE_LINE); |
1016 |
SearchData Data; |
SearchData Data; |
1017 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
1018 |
Data.iEdgedWidth = iEdgedWidth; |
Data.iEdgedWidth = iEdgedWidth; |
1025 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
1026 |
Data.chroma = MotionFlags & PMV_CHROMA16; |
Data.chroma = MotionFlags & PMV_CHROMA16; |
1027 |
Data.rrv = current->global_flags & XVID_REDUCED; |
Data.rrv = current->global_flags & XVID_REDUCED; |
1028 |
|
Data.dctSpace = dct_space; |
1029 |
|
|
1030 |
if ((current->global_flags & XVID_REDUCED)) { |
if ((current->global_flags & XVID_REDUCED)) { |
1031 |
mb_width = (pParam->width + 31) / 32; |
mb_width = (pParam->width + 31) / 32; |
1074 |
//initial skip decision |
//initial skip decision |
1075 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
1076 |
if (!(current->global_flags & XVID_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
if (!(current->global_flags & XVID_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
1077 |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * skip_thresh) |
1078 |
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)) { |
1079 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
1080 |
continue; |
continue; |
1087 |
current->global_flags & XVID_INTER4V, pMB); |
current->global_flags & XVID_INTER4V, pMB); |
1088 |
|
|
1089 |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
1090 |
if (!(current->global_flags & XVID_GMC)) { |
if (!(current->global_flags & XVID_GMC || current->global_flags & XVID_MODEDECISION_BITS)) { |
1091 |
if ( pMB->dquant == NO_CHANGE && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) { |
if ( pMB->dquant == NO_CHANGE && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) { |
|
if (!(current->global_flags & XVID_MODEDECISION_BITS)) { |
|
1092 |
if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
1093 |
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)) |
1094 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
|
} else { // BITS mode decision |
|
|
if (pMB->sad16 > 10) |
|
|
SkipMacroblockP(pMB, sad00); // more than 10 bits would be used for this MB - skip |
|
|
|
|
|
} |
|
1095 |
} |
} |
1096 |
} |
} |
1097 |
if (pMB->mode == MODE_INTRA) |
if (pMB->mode == MODE_INTRA) |
1180 |
int mode = MODE_INTER; |
int mode = MODE_INTER; |
1181 |
|
|
1182 |
if (!(GlobalFlags & XVID_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision |
if (!(GlobalFlags & XVID_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision |
|
int intra = 0; |
|
1183 |
int sad; |
int sad; |
1184 |
int InterBias = MV16_INTER_BIAS; |
int InterBias = MV16_INTER_BIAS; |
1185 |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1186 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
1187 |
mode = 0; //inter |
mode = MODE_INTER; |
1188 |
sad = Data->iMinSAD[0]; |
sad = Data->iMinSAD[0]; |
1189 |
} else { |
} else { |
1190 |
mode = MODE_INTER4V; |
mode = MODE_INTER4V; |
1212 |
dev16(Data->Cur + 8*Data->iEdgedWidth, Data->iEdgedWidth) + |
dev16(Data->Cur + 8*Data->iEdgedWidth, Data->iEdgedWidth) + |
1213 |
dev16(Data->Cur+8+8*Data->iEdgedWidth, Data->iEdgedWidth); |
dev16(Data->Cur+8+8*Data->iEdgedWidth, Data->iEdgedWidth); |
1214 |
|
|
1215 |
if (deviation < (sad - InterBias)) return MODE_INTRA;// intra |
if (deviation < (sad - InterBias)) return MODE_INTRA; |
1216 |
} |
} |
1217 |
return mode; |
return mode; |
1218 |
|
|
1233 |
if (bits == 0) return MODE_INTER; // quick stop |
if (bits == 0) return MODE_INTER; // quick stop |
1234 |
|
|
1235 |
if (inter4v) { |
if (inter4v) { |
1236 |
int inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
int bits_inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1237 |
if (inter4v < bits) { Data->iMinSAD[0] = bits = inter4v; mode = MODE_INTER4V; } |
if (bits_inter4v < bits) { Data->iMinSAD[0] = bits = bits_inter4v; mode = MODE_INTER4V; } |
1238 |
} |
} |
1239 |
|
|
1240 |
|
|
1292 |
|
|
1293 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1294 |
|
|
1295 |
for(i = 0; i < 5; i++) |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
|
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
|
1296 |
|
|
1297 |
if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1298 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
1384 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1385 |
} |
} |
1386 |
|
|
1387 |
if (MotionFlags & PMV_QUARTERPELREFINE16) |
if (MotionFlags & PMV_QUARTERPELREFINE16) { |
1388 |
if ((!(MotionFlags & QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) { |
|
|
Data->qpel_precision = 1; |
|
1389 |
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, |
1390 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1391 |
|
|
1392 |
|
if ((!(MotionFlags & QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) { |
1393 |
|
Data->qpel_precision = 1; |
1394 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1395 |
} |
} |
1396 |
|
} |
1397 |
|
|
1398 |
if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0; |
if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0; |
1399 |
|
|
1714 |
|
|
1715 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1716 |
dy += Data->directmvF[k].y / div; |
dy += Data->directmvF[k].y / div; |
1717 |
dx += Data->directmvF[0].x / div; |
dx += Data->directmvF[k].x / div; |
1718 |
b_dy += Data->directmvB[0].y / div; |
b_dy += Data->directmvB[k].y / div; |
1719 |
b_dx += Data->directmvB[0].x / div; |
b_dx += Data->directmvB[k].x / div; |
1720 |
} |
} |
1721 |
|
|
1722 |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
1736 |
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, |
1737 |
stride); |
stride); |
1738 |
|
|
1739 |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
1740 |
|
pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
1741 |
|
for (k = 0; k < 4; k++) { |
1742 |
|
pMB->qmvs[k] = pMB->mvs[k]; |
1743 |
|
pMB->b_qmvs[k] = pMB->b_mvs[k]; |
1744 |
|
} |
1745 |
|
} |
1746 |
} |
} |
1747 |
|
|
1748 |
static __inline uint32_t |
static __inline uint32_t |
1831 |
} |
} |
1832 |
} |
} |
1833 |
|
|
1834 |
|
*Data->iMinSAD += Data->lambda16; |
1835 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
1836 |
|
|
1837 |
// DIRECT MODE DELTA VECTOR SEARCH. |
// DIRECT MODE DELTA VECTOR SEARCH. |
2158 |
|
|
2159 |
int i, mask; |
int i, mask; |
2160 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
2161 |
MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
2162 |
|
|
2163 |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
2164 |
|
|
2172 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
2173 |
|
|
2174 |
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, |
2175 |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0); |
2176 |
|
|
2177 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2178 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
2184 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
2185 |
|
|
2186 |
CheckCandidate32I(0, 0, 255, &i, Data); |
CheckCandidate32I(0, 0, 255, &i, Data); |
2187 |
|
Data->iMinSAD[1] -= 50; |
2188 |
|
Data->iMinSAD[2] -= 50; |
2189 |
|
Data->iMinSAD[3] -= 50; |
2190 |
|
Data->iMinSAD[4] -= 50; |
2191 |
|
|
2192 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) { |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
2193 |
|
|
2194 |
if (!(mask = make_mask(pmv, 1))) |
if (!(mask = make_mask(pmv, 1))) |
2195 |
CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data); |
CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data); |
2196 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
2197 |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
2198 |
|
|
2199 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) // diamond only if needed |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) // diamond only if needed |
2200 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
2201 |
|
} |
2202 |
|
|
2203 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
2204 |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width]; |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width]; |
2207 |
MB->sad16 = Data->iMinSAD[i+1]; |
MB->sad16 = Data->iMinSAD[i+1]; |
2208 |
} |
} |
2209 |
} |
} |
|
} |
|
2210 |
|
|
2211 |
#define INTRA_BIAS 2500 |
#define INTRA_THRESH 2400 |
2212 |
#define INTRA_THRESH 1500 |
#define INTER_THRESH 1100 |
|
#define INTER_THRESH 1400 |
|
2213 |
|
|
2214 |
int |
int |
2215 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
2216 |
FRAMEINFO * const Current, |
const FRAMEINFO * const Current, |
2217 |
MBParam * const pParam, |
const MBParam * const pParam, |
2218 |
int maxIntra, //maximum number if non-I frames |
const int maxIntra, //maximum number if non-I frames |
2219 |
int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
const int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
2220 |
int bCount) // number of B frames in a row |
const int bCount, // number of B frames in a row |
2221 |
|
const int b_thresh) |
2222 |
{ |
{ |
2223 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
2224 |
int sSAD = 0; |
int sSAD = 0; |
2225 |
MACROBLOCK * const pMBs = Current->mbs; |
MACROBLOCK * const pMBs = Current->mbs; |
2226 |
const IMAGE * const pCurrent = &Current->image; |
const IMAGE * const pCurrent = &Current->image; |
2227 |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + 10*b_thresh; |
2228 |
|
int s = 0, blocks = 0; |
2229 |
|
|
2230 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
2231 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
2234 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
2235 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
2236 |
Data.iFcode = Current->fcode; |
Data.iFcode = Current->fcode; |
|
Data.rrv = Current->global_flags & XVID_REDUCED; |
|
2237 |
Data.temp = temp; |
Data.temp = temp; |
2238 |
CheckCandidate = CheckCandidate32I; |
CheckCandidate = CheckCandidate32I; |
2239 |
|
|
2240 |
if (intraCount != 0 && intraCount < 10) // we're right after an I frame |
if (intraCount != 0 && intraCount < 10) // we're right after an I frame |
2241 |
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
IntraThresh += 8 * (intraCount - 10) * (intraCount - 10); |
2242 |
else |
else |
2243 |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
2244 |
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
2245 |
|
|
2246 |
InterThresh += 400 * (1 - bCount); |
InterThresh -= (350 - 8*b_thresh) * bCount; |
2247 |
if (InterThresh < 300) InterThresh = 300; |
if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh; |
2248 |
|
|
2249 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2250 |
|
|
2251 |
for (y = 1; y < pParam->mb_height-1; y += 2) { |
for (y = 1; y < pParam->mb_height-1; y += 2) { |
2252 |
for (x = 1; x < pParam->mb_width-1; x += 2) { |
for (x = 1; x < pParam->mb_width-1; x += 2) { |
2253 |
int i; |
int i; |
2254 |
|
blocks += 4; |
2255 |
|
|
2256 |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
2257 |
|
else { //extrapolation of the vector found for last frame |
2258 |
|
pMBs[x + y * pParam->mb_width].mvs[0].x = |
2259 |
|
(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount; |
2260 |
|
pMBs[x + y * pParam->mb_width].mvs[0].y = |
2261 |
|
(pMBs[x + y * pParam->mb_width].mvs[0].y * (bCount+1) ) / bCount; |
2262 |
|
} |
2263 |
|
|
2264 |
MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data); |
MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data); |
2265 |
|
|
2271 |
pParam->edged_width); |
pParam->edged_width); |
2272 |
if (dev + IntraThresh < pMB->sad16) { |
if (dev + IntraThresh < pMB->sad16) { |
2273 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
2274 |
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; |
2275 |
} |
} |
2276 |
} |
} |
2277 |
|
if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) s++; |
2278 |
|
|
2279 |
sSAD += pMB->sad16; |
sSAD += pMB->sad16; |
2280 |
} |
} |
2281 |
} |
} |
2282 |
} |
} |
2283 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
|
2284 |
// if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP; |
sSAD /= blocks; |
2285 |
|
s = (10*s) / blocks; |
2286 |
|
|
2287 |
|
if (s > 5) sSAD += (s - 4) * (180 - 2*b_thresh); //static block - looks bad when in bframe... |
2288 |
|
|
2289 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2290 |
emms(); |
emms(); |
2291 |
return B_VOP; |
return B_VOP; |
|
|
|
2292 |
} |
} |
2293 |
|
|
2294 |
|
|
2320 |
double meanx,meany; |
double meanx,meany; |
2321 |
int num,oldnum; |
int num,oldnum; |
2322 |
|
|
2323 |
if (!MBmask) { fprintf(stderr,"Mem error\n"); return gmc;} |
if (!MBmask) { fprintf(stderr,"Mem error\n"); |
2324 |
|
gmc.duv[0].x= gmc.duv[0].y = |
2325 |
|
gmc.duv[1].x= gmc.duv[1].y = |
2326 |
|
gmc.duv[2].x= gmc.duv[2].y = 0; |
2327 |
|
return gmc; } |
2328 |
|
|
2329 |
// filter mask of all blocks |
// filter mask of all blocks |
2330 |
|
|
2331 |
for (my = 1; my < MBh-1; my++) |
for (my = 1; my < (uint32_t)MBh-1; my++) |
2332 |
for (mx = 1; mx < MBw-1; mx++) |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2333 |
{ |
{ |
2334 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2335 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
2345 |
MBmask[mbnum]=1; |
MBmask[mbnum]=1; |
2346 |
} |
} |
2347 |
|
|
2348 |
for (my = 1; my < MBh-1; my++) |
for (my = 1; my < (uint32_t)MBh-1; my++) |
2349 |
for (mx = 1; mx < MBw-1; mx++) |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2350 |
{ |
{ |
2351 |
const uint8_t *const pCur = current->image.y + 16*my*pParam->edged_width + 16*mx; |
const uint8_t *const pCur = current->image.y + 16*my*pParam->edged_width + 16*mx; |
2352 |
|
|
2354 |
if (!MBmask[mbnum]) |
if (!MBmask[mbnum]) |
2355 |
continue; |
continue; |
2356 |
|
|
2357 |
if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= grad ) |
if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= (uint32_t)grad ) |
2358 |
MBmask[mbnum] = 0; |
MBmask[mbnum] = 0; |
2359 |
if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= grad ) |
if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= (uint32_t)grad ) |
2360 |
MBmask[mbnum] = 0; |
MBmask[mbnum] = 0; |
2361 |
|
|
2362 |
} |
} |
2367 |
|
|
2368 |
a = b = c = n = 0; |
a = b = c = n = 0; |
2369 |
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
2370 |
for (my = 0; my < MBh; my++) |
for (my = 0; my < (uint32_t)MBh; my++) |
2371 |
for (mx = 0; mx < MBw; mx++) |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2372 |
{ |
{ |
2373 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2374 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
2405 |
|
|
2406 |
meanx = meany = 0.; |
meanx = meany = 0.; |
2407 |
oldnum = 0; |
oldnum = 0; |
2408 |
for (my = 0; my < MBh; my++) |
for (my = 0; my < (uint32_t)MBh; my++) |
2409 |
for (mx = 0; mx < MBw; mx++) |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2410 |
{ |
{ |
2411 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2412 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
2434 |
fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum); |
fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum); |
2435 |
*/ |
*/ |
2436 |
num = 0; |
num = 0; |
2437 |
for (my = 0; my < MBh; my++) |
for (my = 0; my < (uint32_t)MBh; my++) |
2438 |
for (mx = 0; mx < MBw; mx++) |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2439 |
{ |
{ |
2440 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2441 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
2556 |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
2557 |
SearchData Data2, *Data8 = &Data2; |
SearchData Data2, *Data8 = &Data2; |
2558 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
2559 |
int16_t in[64], coeff[64]; |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
2560 |
|
|
2561 |
memcpy(Data8, Data, sizeof(SearchData)); |
memcpy(Data8, Data, sizeof(SearchData)); |
2562 |
CheckCandidate = CheckCandidateBits8; |
CheckCandidate = CheckCandidateBits8; |
2683 |
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
2684 |
cbp |= 1 << (5 - 5); |
cbp |= 1 << (5 - 5); |
2685 |
} |
} |
2686 |
bits += cbpy_tab[15-(cbp>>2)].len; |
bits += xvid_cbpy_tab[15-(cbp>>2)].len; |
2687 |
bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
2688 |
} |
} |
2689 |
} |
} |
2698 |
int bits = 1; //this one is ac/dc prediction flag. always 1. |
int bits = 1; //this one is ac/dc prediction flag. always 1. |
2699 |
int cbp = 0, i, t, dc = 0, b_dc = 1024; |
int cbp = 0, i, t, dc = 0, b_dc = 1024; |
2700 |
const uint32_t iQuant = Data->lambda16; |
const uint32_t iQuant = Data->lambda16; |
2701 |
int16_t in[64], coeff[64]; |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
2702 |
|
|
2703 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
2704 |
uint32_t iDcScaler = get_dc_scaler(iQuant, 1); |
uint32_t iDcScaler = get_dc_scaler(iQuant, 1); |
2733 |
|
|
2734 |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
2735 |
if (t != 0) cbp |= 1 << (5 - 4); |
if (t != 0) cbp |= 1 << (5 - 4); |
|
Data->temp[4] = t; |
|
2736 |
|
|
2737 |
if (bits < Data->iMinSAD[0]) { |
if (bits < Data->iMinSAD[0]) { |
2738 |
|
iDcScaler = get_dc_scaler(iQuant, 1); |
2739 |
//chroma V |
//chroma V |
2740 |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
2741 |
fdct(in); |
fdct(in); |
2746 |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
2747 |
if (t != 0) cbp |= 1 << (5 - 5); |
if (t != 0) cbp |= 1 << (5 - 5); |
2748 |
|
|
2749 |
Data->temp[5] = t; |
bits += xvid_cbpy_tab[cbp>>2].len; |
2750 |
|
bits += mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
|
bits += t = cbpy_tab[cbp>>2].len; |
|
|
Data->temp[6] = t; |
|
|
|
|
|
bits += t = mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
|
|
Data->temp[7] = t; |
|
|
|
|
2751 |
} |
} |
2752 |
} |
} |
|
|
|
2753 |
return bits; |
return bits; |
2754 |
} |
} |