39 |
#include "motion.h" |
#include "motion.h" |
40 |
#include "sad.h" |
#include "sad.h" |
41 |
#include "motion_inlines.h" |
#include "motion_inlines.h" |
42 |
|
#include "motion_smp.h" |
43 |
|
|
44 |
|
|
45 |
static const int xvid_me_lambda_vec8[32] = |
static const int xvid_me_lambda_vec8[32] = |
46 |
{ 0 ,(int)(1.00235 * NEIGH_TEND_8X8 + 0.5), |
{ 0 ,(int)(1.0 * NEIGH_TEND_8X8 + 0.5), |
47 |
(int)(1.15582 + NEIGH_TEND_8X8 + 0.5), (int)(1.31976*NEIGH_TEND_8X8 + 0.5), |
(int)(2.0*NEIGH_TEND_8X8 + 0.5), (int)(3.0*NEIGH_TEND_8X8 + 0.5), |
48 |
(int)(1.49591*NEIGH_TEND_8X8 + 0.5), (int)(1.68601*NEIGH_TEND_8X8 + 0.5), |
(int)(4.0*NEIGH_TEND_8X8 + 0.5), (int)(5.0*NEIGH_TEND_8X8 + 0.5), |
49 |
(int)(1.89187*NEIGH_TEND_8X8 + 0.5), (int)(2.11542*NEIGH_TEND_8X8 + 0.5), |
(int)(6.0*NEIGH_TEND_8X8 + 0.5), (int)(7.0*NEIGH_TEND_8X8 + 0.5), |
50 |
(int)(2.35878*NEIGH_TEND_8X8 + 0.5), (int)(2.62429*NEIGH_TEND_8X8 + 0.5), |
(int)(8.0*NEIGH_TEND_8X8 + 0.5), (int)(9.0*NEIGH_TEND_8X8 + 0.5), |
51 |
(int)(2.91455*NEIGH_TEND_8X8 + 0.5), (int)(3.23253*NEIGH_TEND_8X8 + 0.5), |
(int)(10.0*NEIGH_TEND_8X8 + 0.5), (int)(11.0*NEIGH_TEND_8X8 + 0.5), |
52 |
(int)(3.58158*NEIGH_TEND_8X8 + 0.5), (int)(3.96555*NEIGH_TEND_8X8 + 0.5), |
(int)(12.0*NEIGH_TEND_8X8 + 0.5), (int)(13.0*NEIGH_TEND_8X8 + 0.5), |
53 |
(int)(4.38887*NEIGH_TEND_8X8 + 0.5), (int)(4.85673*NEIGH_TEND_8X8 + 0.5), |
(int)(14.0*NEIGH_TEND_8X8 + 0.5), (int)(15.0*NEIGH_TEND_8X8 + 0.5), |
54 |
(int)(5.37519*NEIGH_TEND_8X8 + 0.5), (int)(5.95144*NEIGH_TEND_8X8 + 0.5), |
(int)(16.0*NEIGH_TEND_8X8 + 0.5), (int)(17.0*NEIGH_TEND_8X8 + 0.5), |
55 |
(int)(6.59408*NEIGH_TEND_8X8 + 0.5), (int)(7.31349*NEIGH_TEND_8X8 + 0.5), |
(int)(18.0*NEIGH_TEND_8X8 + 0.5), (int)(19.0*NEIGH_TEND_8X8 + 0.5), |
56 |
(int)(8.12242*NEIGH_TEND_8X8 + 0.5), (int)(9.03669*NEIGH_TEND_8X8 + 0.5), |
(int)(20.0*NEIGH_TEND_8X8 + 0.5), (int)(21.0*NEIGH_TEND_8X8 + 0.5), |
57 |
(int)(10.0763*NEIGH_TEND_8X8 + 0.5), (int)(11.2669*NEIGH_TEND_8X8 + 0.5), |
(int)(22.0*NEIGH_TEND_8X8 + 0.5), (int)(23.0*NEIGH_TEND_8X8 + 0.5), |
58 |
(int)(12.6426*NEIGH_TEND_8X8 + 0.5), (int)(14.2493*NEIGH_TEND_8X8 + 0.5), |
(int)(24.0*NEIGH_TEND_8X8 + 0.5), (int)(25.0*NEIGH_TEND_8X8 + 0.5), |
59 |
(int)(16.1512*NEIGH_TEND_8X8 + 0.5), (int)(18.442*NEIGH_TEND_8X8 + 0.5), |
(int)(26.0*NEIGH_TEND_8X8 + 0.5), (int)(27.0*NEIGH_TEND_8X8 + 0.5), |
60 |
(int)(21.2656*NEIGH_TEND_8X8 + 0.5), (int)(24.8580*NEIGH_TEND_8X8 + 0.5), |
(int)(28.0*NEIGH_TEND_8X8 + 0.5), (int)(29.0*NEIGH_TEND_8X8 + 0.5), |
61 |
(int)(29.6436*NEIGH_TEND_8X8 + 0.5), (int)(36.4949*NEIGH_TEND_8X8 + 0.5) |
(int)(30.0*NEIGH_TEND_8X8 + 0.5), (int)(31.0*NEIGH_TEND_8X8 + 0.5) |
62 |
}; |
}; |
63 |
|
|
64 |
static void |
static void |
82 |
} |
} |
83 |
|
|
84 |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp); |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp); |
85 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision); |
86 |
|
|
87 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t); |
88 |
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
data->temp[0] += (data->lambda8 * t); |
89 |
|
|
90 |
if (data->chroma) { |
if (data->chroma) { |
91 |
if (sad >= data->iMinSAD[0]) goto no16; |
if (sad >= data->iMinSAD[0]) goto no16; |
129 |
} |
} |
130 |
|
|
131 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
132 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision); |
133 |
|
|
134 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
sad += (data->lambda8 * t); |
135 |
|
|
136 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
137 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
140 |
} |
} |
141 |
} |
} |
142 |
|
|
|
static void |
|
|
CheckCandidate32(const int x, const int y, SearchData * const data, const unsigned int Direction) |
|
|
{ |
|
|
uint32_t t; |
|
|
const uint8_t * Reference; |
|
|
int sad; |
|
|
|
|
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
|
|
(x > data->max_dx) || (x < data->min_dx) |
|
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|
|
|
|
|
Reference = GetReference(x, y, data); |
|
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
|
|
|
|
|
sad = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp); |
|
|
|
|
|
sad += (data->lambda16 * t * sad) >> 10; |
|
|
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
|
|
|
|
|
if (sad < data->iMinSAD[0]) { |
|
|
data->iMinSAD[0] = sad; |
|
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
|
|
data->dir = Direction; |
|
|
} |
|
|
|
|
|
if (data->temp[0] < data->iMinSAD[1]) { |
|
|
data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
|
|
if (data->temp[1] < data->iMinSAD[2]) { |
|
|
data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
|
|
if (data->temp[2] < data->iMinSAD[3]) { |
|
|
data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
|
|
if (data->temp[3] < data->iMinSAD[4]) { |
|
|
data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
|
|
} |
|
|
|
|
143 |
int |
int |
144 |
xvid_me_SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
xvid_me_SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
145 |
const int x, const int y, |
const int x, const int y, |
146 |
const uint32_t stride, const uint32_t iQuant, int rrv) |
const uint32_t stride, const uint32_t iQuant) |
147 |
{ |
{ |
148 |
int offset = (x + y*stride)*8; |
int offset = (x + y*stride)*8; |
|
if(!rrv) { |
|
149 |
uint32_t sadC = sad8(current->u + offset, |
uint32_t sadC = sad8(current->u + offset, |
150 |
reference->u + offset, stride); |
reference->u + offset, stride); |
151 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
153 |
reference->v + offset, stride); |
reference->v + offset, stride); |
154 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
155 |
return 1; |
return 1; |
|
|
|
|
} else { |
|
|
uint32_t sadC = sad16(current->u + 2*offset, |
|
|
reference->u + 2*offset, stride, 256*4096); |
|
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
|
|
sadC += sad16(current->v + 2*offset, |
|
|
reference->v + 2*offset, stride, 256*4096); |
|
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
|
|
return 1; |
|
|
} |
|
156 |
} |
} |
157 |
|
|
158 |
/* |
/* |
268 |
const IMAGE * const pCurrent, |
const IMAGE * const pCurrent, |
269 |
const IMAGE * const pRef, |
const IMAGE * const pRef, |
270 |
const IMAGE * const vGMC, |
const IMAGE * const vGMC, |
271 |
const int coding_type) |
const int coding_type, |
272 |
|
const int skip_sad) |
273 |
{ |
{ |
274 |
int mode = MODE_INTER; |
int mode = MODE_INTER; |
275 |
int mcsel = 0; |
int mcsel = 0; |
295 |
} |
} |
296 |
|
|
297 |
/* 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?" */ |
298 |
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
if (skip_possible && (skip_sad < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
299 |
if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
if ( (100*skip_sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
300 |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) { |
301 |
mode = MODE_NOT_CODED; |
mode = MODE_NOT_CODED; |
302 |
sad = 0; |
sad = 0; |
303 |
} |
} |
329 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
330 |
|
|
331 |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
|
if (Data->rrv) InterBias *= 4; |
|
332 |
|
|
333 |
if (InterBias < sad) { |
if (InterBias < sad) { |
334 |
int32_t deviation; |
int32_t deviation = dev16(Data->Cur, Data->iEdgedWidth); |
|
if (!Data->rrv) |
|
|
deviation = dev16(Data->Cur, Data->iEdgedWidth); |
|
|
else |
|
|
deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ |
|
|
dev16(Data->Cur+16, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
|
|
|
|
335 |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
336 |
} |
} |
337 |
|
|
338 |
pMB->cbp = 63; |
pMB->cbp = 63; |
339 |
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; |
340 |
|
|
|
if (Data->rrv) { |
|
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
|
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
|
|
} |
|
|
|
|
341 |
if (mode == MODE_INTER && mcsel == 0) { |
if (mode == MODE_INTER && mcsel == 0) { |
342 |
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]; |
343 |
|
|
371 |
|
|
372 |
static __inline void |
static __inline void |
373 |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
374 |
int iHcount, const MACROBLOCK * const prevMB, int rrv) |
int iHcount, const MACROBLOCK * const prevMB) |
375 |
{ |
{ |
|
/* this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself */ |
|
|
if (rrv) { iWcount /= 2; iHcount /= 2; } |
|
376 |
|
|
377 |
if ( (y != 0) && (x < (iWcount-1)) ) { /* [5] top-right neighbour */ |
if ( (y != 0) && (x < (iWcount-1)) ) { /* [5] top-right neighbour */ |
378 |
pmv[5].x = EVEN(pmv[3].x); |
pmv[5].x = EVEN(pmv[3].x); |
397 |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); /* [6] right-down neighbour in last frame */ |
pmv[6].x = EVEN((prevMB+1+iWcount)->mvs[0].x); /* [6] right-down neighbour in last frame */ |
398 |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
399 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
|
|
|
|
if (rrv) { |
|
|
int i; |
|
|
for (i = 0; i < 7; i++) { |
|
|
pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); |
|
|
pmv[i].y = RRV_MV_SCALEUP(pmv[i].y); |
|
|
} |
|
|
} |
|
400 |
} |
} |
401 |
|
|
402 |
static void |
static void |
411 |
{ |
{ |
412 |
int i = 0; |
int i = 0; |
413 |
VECTOR vbest_q; int32_t sbest_q; |
VECTOR vbest_q; int32_t sbest_q; |
|
CheckFunc * CheckCandidate; |
|
414 |
*Data->iMinSAD = *(OldData->iMinSAD + 1 + block); |
*Data->iMinSAD = *(OldData->iMinSAD + 1 + block); |
415 |
*Data->currentMV = *(OldData->currentMV + 1 + block); |
*Data->currentMV = *(OldData->currentMV + 1 + block); |
416 |
*Data->currentQMV = *(OldData->currentQMV + 1 + block); |
*Data->currentQMV = *(OldData->currentQMV + 1 + block); |
418 |
if(Data->qpel) { |
if(Data->qpel) { |
419 |
Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block); |
Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block); |
420 |
if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y, |
if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y, |
421 |
Data->predMV, Data->iFcode, 0, 0); |
Data->predMV, Data->iFcode, 0); |
422 |
} else { |
} else { |
423 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2, y/2, block); |
424 |
if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y, |
if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y, |
425 |
Data->predMV, Data->iFcode, 0, Data->rrv); |
Data->predMV, Data->iFcode, 0); |
426 |
} |
} |
427 |
|
|
428 |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
*(Data->iMinSAD) += (Data->lambda8 * i); |
429 |
|
|
430 |
if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) { |
if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) { |
431 |
|
|
432 |
vbest_q = Data->currentQMV[0]; |
vbest_q = Data->currentQMV[0]; |
433 |
sbest_q = Data->iMinSAD[0]; |
sbest_q = Data->iMinSAD[0]; |
434 |
|
|
|
if (Data->rrv) i = 16; else i = 8; |
|
435 |
|
|
436 |
Data->RefP[0] = OldData->RefP[0] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[0] = OldData->RefP[0] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
437 |
Data->RefP[1] = OldData->RefP[1] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[1] = OldData->RefP[1] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
438 |
Data->RefP[2] = OldData->RefP[2] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[2] = OldData->RefP[2] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
439 |
Data->RefP[3] = OldData->RefP[3] + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->RefP[3] = OldData->RefP[3] + 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
440 |
|
|
441 |
Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->Cur = OldData->Cur + 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
442 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
443 |
|
|
444 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
445 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); |
|
|
|
|
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
|
|
else CheckCandidate = CheckCandidate16no4v; |
|
446 |
|
|
447 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) { |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) { |
448 |
|
|
451 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = xvid_me_AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = xvid_me_AdvDiamondSearch; |
452 |
else MainSearchPtr = xvid_me_DiamondSearch; |
else MainSearchPtr = xvid_me_DiamondSearch; |
453 |
|
|
454 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate8); |
455 |
} |
} |
456 |
|
|
457 |
if(!Data->qpel) { |
if(!Data->qpel) { |
458 |
/* halfpel mode */ |
/* halfpel mode */ |
459 |
if (MotionFlags & XVID_ME_HALFPELREFINE8) |
if (MotionFlags & XVID_ME_HALFPELREFINE8) |
460 |
xvid_me_SubpelRefine(Data, CheckCandidate, 0); /* perform halfpel refine of current best vector */ |
/* perform halfpel refine of current best vector */ |
461 |
|
xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate8, 0); |
462 |
} else { |
} else { |
463 |
/* qpel mode */ |
/* qpel mode */ |
464 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
467 |
if(MotionFlags & XVID_ME_FASTREFINE8) { |
if(MotionFlags & XVID_ME_FASTREFINE8) { |
468 |
/* fast */ |
/* fast */ |
469 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
470 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
471 |
FullRefine_Fast(Data, CheckCandidate8, 0); |
FullRefine_Fast(Data, CheckCandidate8, 0); |
472 |
} else if(MotionFlags & XVID_ME_QUARTERPELREFINE8) { |
} else if(MotionFlags & XVID_ME_QUARTERPELREFINE8) { |
473 |
/* full */ |
/* full */ |
474 |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
475 |
xvid_me_SubpelRefine(Data, CheckCandidate8, 0); /* hpel part */ |
xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate8, 0); /* hpel part */ |
476 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
477 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
478 |
} |
} |
479 |
|
|
480 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
481 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
482 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
483 |
|
|
484 |
xvid_me_SubpelRefine(Data, CheckCandidate8, 0); /* qpel part */ |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate8, 0); /* qpel part */ |
485 |
} |
} |
486 |
} |
} |
487 |
|
|
490 |
|
|
491 |
} |
} |
492 |
|
|
493 |
if (Data->rrv) { |
if(Data->qpel) { |
|
Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x); |
|
|
Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y); |
|
|
} else if(Data->qpel) { |
|
494 |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
495 |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
496 |
pMB->qmvs[block] = *Data->currentQMV; |
pMB->qmvs[block] = *Data->currentQMV; |
532 |
CheckFunc * CheckCandidate; |
CheckFunc * CheckCandidate; |
533 |
|
|
534 |
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, |
535 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); |
536 |
|
|
537 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, pmv, Data->temp); |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, pmv, Data->temp); |
538 |
|
|
539 |
Data->chromaX = Data->chromaY = 0; /* chroma-sad cache */ |
Data->chromaX = Data->chromaY = 0; /* chroma-sad cache */ |
540 |
i = Data->rrv ? 2 : 1; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
541 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
542 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
543 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
|
544 |
|
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
545 |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
546 |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16; |
547 |
Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
548 |
Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
549 |
Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
|
550 |
|
|
551 |
Data->lambda16 = xvid_me_lambda_vec16[pMB->quant]; |
Data->lambda16 = xvid_me_lambda_vec16[pMB->quant]; |
552 |
Data->lambda8 = xvid_me_lambda_vec8[pMB->quant]; |
Data->lambda8 = xvid_me_lambda_vec8[pMB->quant]; |
558 |
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); |
559 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
560 |
|
|
561 |
i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0); |
i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0); |
562 |
Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10); |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i); |
563 |
Data->iMinSAD[1] = pMB->sad8[0] + ((Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS)) >> 10); |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i); |
564 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
565 |
Data->iMinSAD[3] = pMB->sad8[2]; |
Data->iMinSAD[3] = pMB->sad8[2]; |
566 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
573 |
threshA = 512; |
threshA = 512; |
574 |
|
|
575 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
576 |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
prevMBs + x + y * pParam->mb_width); |
577 |
|
|
|
if (!Data->rrv) { |
|
578 |
if (inter4v) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
579 |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
|
} else CheckCandidate = CheckCandidate32; |
|
580 |
|
|
581 |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
582 |
|
|
611 |
if (Data->qpel) { |
if (Data->qpel) { |
612 |
startMV.x /= 2; |
startMV.x /= 2; |
613 |
startMV.y /= 2; |
startMV.y /= 2; |
|
} else if (Data->rrv) { |
|
|
startMV.x = RRV_MV_SCALEUP(startMV.x); |
|
|
startMV.y = RRV_MV_SCALEUP(startMV.y); |
|
614 |
} |
} |
615 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
616 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
619 |
xvid_me_DiamondSearch(startMV.x, startMV.y, Data, 255, CheckCandidate); |
xvid_me_DiamondSearch(startMV.x, startMV.y, Data, 255, CheckCandidate); |
620 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
621 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
622 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; |
623 |
|
} |
624 |
} |
} |
625 |
|
|
626 |
backupMV = Data->currentMV[0]; |
backupMV = Data->currentMV[0]; |
642 |
if(!Data->qpel) { |
if(!Data->qpel) { |
643 |
/* halfpel mode */ |
/* halfpel mode */ |
644 |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
645 |
xvid_me_SubpelRefine(Data, CheckCandidate, 0); |
xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate, 0); |
646 |
} else { |
} else { |
647 |
/* qpel mode */ |
/* qpel mode */ |
648 |
|
|
653 |
if(MotionFlags & XVID_ME_FASTREFINE16 && MotionFlags & XVID_ME_QUARTERPELREFINE16) { |
if(MotionFlags & XVID_ME_FASTREFINE16 && MotionFlags & XVID_ME_QUARTERPELREFINE16) { |
654 |
/* fast */ |
/* fast */ |
655 |
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, |
656 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
657 |
FullRefine_Fast(Data, CheckCandidate, 0); |
FullRefine_Fast(Data, CheckCandidate, 0); |
658 |
} else { |
} else { |
659 |
if(MotionFlags & (XVID_ME_QUARTERPELREFINE16 | XVID_ME_QUARTERPELREFINE16_RD)) { |
if(MotionFlags & (XVID_ME_QUARTERPELREFINE16 | XVID_ME_QUARTERPELREFINE16_RD)) { |
660 |
/* full */ |
/* full */ |
661 |
if (MotionFlags & XVID_ME_HALFPELREFINE16) { |
if (MotionFlags & XVID_ME_HALFPELREFINE16) { |
662 |
xvid_me_SubpelRefine(Data, CheckCandidate, 0); /* hpel part */ |
xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidate, 0); /* hpel part */ |
663 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
664 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; |
665 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
666 |
} |
} |
667 |
} |
} |
668 |
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, |
669 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
670 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
671 |
if(MotionFlags & XVID_ME_QUARTERPELREFINE16) |
if(MotionFlags & XVID_ME_QUARTERPELREFINE16) |
672 |
xvid_me_SubpelRefine(Data, CheckCandidate, 0); /* qpel part */ |
xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate, 0); /* qpel part */ |
673 |
} |
} |
674 |
} |
} |
675 |
} |
} |
707 |
} else Data->iMinSAD[1] = 4096*256; |
} else Data->iMinSAD[1] = 4096*256; |
708 |
} |
} |
709 |
|
|
710 |
|
static int |
711 |
|
InitialSkipDecisionP(int sad00, |
712 |
|
const MBParam * pParam, |
713 |
|
const FRAMEINFO * current, |
714 |
|
MACROBLOCK * pMB, |
715 |
|
const MACROBLOCK * prevMB, |
716 |
|
int x, int y, |
717 |
|
const SearchData * Data, |
718 |
|
const IMAGE * const pGMC, |
719 |
|
const IMAGE * const pCurrent, |
720 |
|
const IMAGE * const pRef, |
721 |
|
const uint32_t MotionFlags) |
722 |
|
{ |
723 |
|
const unsigned int iEdgedWidth = pParam->edged_width; |
724 |
|
|
725 |
|
int skip_thresh = INITIAL_SKIP_THRESH * \ |
726 |
|
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
727 |
|
int stat_thresh = 0; |
728 |
|
|
729 |
|
/* initial skip decision */ |
730 |
|
if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
731 |
|
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
732 |
|
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant)) { |
733 |
|
ZeroMacroblockP(pMB, sad00); |
734 |
|
pMB->mode = MODE_NOT_CODED; |
735 |
|
return 1; |
736 |
|
} |
737 |
|
} |
738 |
|
|
739 |
|
if(MotionFlags & XVID_ME_DETECT_STATIC_MOTION) { |
740 |
|
VECTOR *cmpMV; |
741 |
|
VECTOR staticMV = { 0, 0 }; |
742 |
|
const MACROBLOCK * pMBs = current->mbs; |
743 |
|
|
744 |
|
if (current->coding_type == S_VOP) |
745 |
|
cmpMV = &pMB->amv; |
746 |
|
else |
747 |
|
cmpMV = &staticMV; |
748 |
|
|
749 |
|
if(x > 0 && y > 0 && x < pParam->mb_width) { |
750 |
|
if(MVequal((&pMBs[(x-1) + y * pParam->mb_width])->mvs[0], *cmpMV) && |
751 |
|
MVequal((&pMBs[x + (y-1) * pParam->mb_width])->mvs[0], *cmpMV) && |
752 |
|
MVequal((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mvs[0], *cmpMV) && |
753 |
|
MVequal(prevMB->mvs[0], *cmpMV)) { |
754 |
|
stat_thresh = MAX((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
755 |
|
MAX((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
756 |
|
MAX((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
757 |
|
prevMB->sad16))); |
758 |
|
} else { |
759 |
|
stat_thresh = MIN((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
760 |
|
MIN((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
761 |
|
MIN((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
762 |
|
prevMB->sad16))); |
763 |
|
} |
764 |
|
} |
765 |
|
} |
766 |
|
|
767 |
|
/* favorize (0,0) or global vector for cartoons */ |
768 |
|
if (current->vop_flags & XVID_VOP_CARTOON) { |
769 |
|
if (current->coding_type == S_VOP) { |
770 |
|
int32_t iSAD = sad16(pCurrent->y + (x + y * iEdgedWidth) * 16, |
771 |
|
pGMC->y + 16*y*iEdgedWidth + 16*x, iEdgedWidth, 65536); |
772 |
|
|
773 |
|
if (Data->chroma) { |
774 |
|
iSAD += sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, pGMC->u + 8*y*(iEdgedWidth/2) + 8*x, iEdgedWidth/2); |
775 |
|
iSAD += sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, pGMC->v + 8*y*(iEdgedWidth/2) + 8*x, iEdgedWidth/2); |
776 |
|
} |
777 |
|
|
778 |
|
if (iSAD <= stat_thresh) { /* mode decision GMC */ |
779 |
|
pMB->mode = MODE_INTER; |
780 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = iSAD; |
781 |
|
pMB->mcsel = 1; |
782 |
|
if (Data->qpel) { |
783 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; |
784 |
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; |
785 |
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; |
786 |
|
} else |
787 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; |
788 |
|
|
789 |
|
return 1; |
790 |
|
} |
791 |
|
} |
792 |
|
else if (sad00 < stat_thresh) { |
793 |
|
VECTOR predMV; |
794 |
|
if (Data->qpel) |
795 |
|
predMV = get_qpmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
796 |
|
else |
797 |
|
predMV = get_pmv2(current->mbs, pParam->mb_width, 0, x, y, 0); |
798 |
|
|
799 |
|
ZeroMacroblockP(pMB, sad00); |
800 |
|
pMB->cbp = 0x3f; |
801 |
|
pMB->pmvs[0].x = - predMV.x; |
802 |
|
pMB->pmvs[0].y = - predMV.y; |
803 |
|
return 1; |
804 |
|
} |
805 |
|
} |
806 |
|
|
807 |
|
return 0; |
808 |
|
} |
809 |
|
|
810 |
static __inline uint32_t |
static __inline uint32_t |
811 |
MakeGoodMotionFlags(const uint32_t MotionFlags, const uint32_t VopFlags, const uint32_t VolFlags) |
MakeGoodMotionFlags(const uint32_t MotionFlags, const uint32_t VopFlags, const uint32_t VolFlags) |
812 |
{ |
{ |
841 |
if (!(VopFlags & XVID_VOP_HALFPEL)) |
if (!(VopFlags & XVID_VOP_HALFPEL)) |
842 |
Flags &= ~(XVID_ME_EXTSEARCH16+XVID_ME_HALFPELREFINE16+XVID_ME_HALFPELREFINE8+XVID_ME_HALFPELREFINE16_RD+XVID_ME_HALFPELREFINE8_RD); |
Flags &= ~(XVID_ME_EXTSEARCH16+XVID_ME_HALFPELREFINE16+XVID_ME_HALFPELREFINE8+XVID_ME_HALFPELREFINE16_RD+XVID_ME_HALFPELREFINE8_RD); |
843 |
|
|
844 |
if ((VopFlags & XVID_VOP_GREYSCALE) || (VopFlags & XVID_VOP_REDUCED)) |
if (VopFlags & XVID_VOP_GREYSCALE) |
845 |
Flags &= ~(XVID_ME_CHROMA_PVOP + XVID_ME_CHROMA_BVOP); |
Flags &= ~(XVID_ME_CHROMA_PVOP + XVID_ME_CHROMA_BVOP); |
846 |
|
|
847 |
if (Flags & XVID_ME_FASTREFINE8) |
if (Flags & XVID_ME_FASTREFINE8) |
853 |
return Flags; |
return Flags; |
854 |
} |
} |
855 |
|
|
856 |
bool |
static __inline void |
857 |
|
motionStatsPVOP(int * const MVmax, int * const mvCount, int * const mvSum, |
858 |
|
const MACROBLOCK * const pMB, const int qpel) |
859 |
|
{ |
860 |
|
const VECTOR * const mv = qpel ? pMB->qmvs : pMB->mvs; |
861 |
|
int i; |
862 |
|
int max = *MVmax; |
863 |
|
|
864 |
|
switch (pMB->mode) { |
865 |
|
case MODE_INTER4V: |
866 |
|
*mvCount += 3; |
867 |
|
for(i = 3; i; i--) { |
868 |
|
if (mv[i].x > max) max = mv[i].x; |
869 |
|
else if (-mv[i].x - 1 > max) max = -mv[i].x - 1; |
870 |
|
*mvSum += mv[i].x * mv[i].x; |
871 |
|
if (mv[i].y > max) max = mv[i].y; |
872 |
|
else if (-mv[i].y - 1 > max) max = -mv[i].y - 1; |
873 |
|
*mvSum += mv[i].y * mv[i].y; |
874 |
|
} |
875 |
|
case MODE_INTER: |
876 |
|
(*mvCount)++; |
877 |
|
*mvSum += mv[0].x * mv[0].x; |
878 |
|
*mvSum += mv[0].y * mv[0].y; |
879 |
|
if (mv[0].x > max) max = mv[0].x; |
880 |
|
else if (-mv[0].x - 1 > max) max = -mv[0].x - 1; |
881 |
|
if (mv[0].y > max) max = mv[0].y; |
882 |
|
else if (-mv[0].y - 1 > max) max = -mv[0].y - 1; |
883 |
|
*MVmax = max; |
884 |
|
default: |
885 |
|
break; |
886 |
|
} |
887 |
|
} |
888 |
|
|
889 |
|
void |
890 |
MotionEstimation(MBParam * const pParam, |
MotionEstimation(MBParam * const pParam, |
891 |
FRAMEINFO * const current, |
FRAMEINFO * const current, |
892 |
FRAMEINFO * const reference, |
FRAMEINFO * const reference, |
900 |
const IMAGE *const pCurrent = ¤t->image; |
const IMAGE *const pCurrent = ¤t->image; |
901 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
902 |
|
|
903 |
uint32_t mb_width = pParam->mb_width; |
const uint32_t mb_width = pParam->mb_width; |
904 |
uint32_t mb_height = pParam->mb_height; |
const uint32_t mb_height = pParam->mb_height; |
905 |
const uint32_t iEdgedWidth = pParam->edged_width; |
const uint32_t iEdgedWidth = pParam->edged_width; |
906 |
const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags); |
const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags); |
907 |
int stat_thresh = 0; |
int stat_thresh = 0; |
908 |
|
int MVmax = 0, mvSum = 0, mvCount = 0; |
909 |
|
|
910 |
uint32_t x, y; |
uint32_t x, y; |
911 |
uint32_t iIntra = 0; |
int sad00; |
|
int32_t sad00; |
|
912 |
int skip_thresh = INITIAL_SKIP_THRESH * \ |
int skip_thresh = INITIAL_SKIP_THRESH * \ |
|
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
|
913 |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
914 |
|
int block = 0; |
915 |
|
|
916 |
/* some pre-initialized thingies for SearchP */ |
/* some pre-initialized thingies for SearchP */ |
917 |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
922 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
923 |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
924 |
Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP; |
Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP; |
|
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED) ? 1:0; |
|
925 |
Data.dctSpace = dct_space; |
Data.dctSpace = dct_space; |
926 |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
927 |
Data.mpeg_quant_matrices = pParam->mpeg_quant_matrices; |
Data.mpeg_quant_matrices = pParam->mpeg_quant_matrices; |
928 |
|
|
|
if ((current->vop_flags & XVID_VOP_REDUCED)) { |
|
|
mb_width = (pParam->width + 31) / 32; |
|
|
mb_height = (pParam->height + 31) / 32; |
|
|
Data.qpel = 0; |
|
|
} |
|
|
|
|
929 |
Data.RefQ = pRefV->u; /* a good place, also used in MC (for similar purpose) */ |
Data.RefQ = pRefV->u; /* a good place, also used in MC (for similar purpose) */ |
930 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
931 |
|
|
932 |
for (y = 0; y < mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
933 |
for (x = 0; x < mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
934 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[block]; |
935 |
MACROBLOCK *prevMB = &reference->mbs[x + y * pParam->mb_width]; |
MACROBLOCK *prevMB = &reference->mbs[block]; |
936 |
|
int skip; |
937 |
|
block++; |
938 |
|
|
939 |
if (!Data.rrv) pMB->sad16 = |
pMB->sad16 = |
940 |
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
941 |
pRef->y + (x + y * iEdgedWidth) * 16, |
pRef->y + (x + y * iEdgedWidth) * 16, |
942 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
943 |
|
|
944 |
else pMB->sad16 = |
sad00 = 4*MAX(MAX(pMB->sad8[0], pMB->sad8[1]), MAX(pMB->sad8[2], pMB->sad8[3])); |
945 |
sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32, |
|
946 |
pRef->y + (x + y * iEdgedWidth) * 32, |
if (Data.chroma) { |
947 |
|
Data.chromaSAD = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, |
948 |
|
pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2) |
949 |
|
+ sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, |
950 |
|
pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
951 |
|
pMB->sad16 += Data.chromaSAD; |
952 |
|
sad00 += Data.chromaSAD; |
953 |
|
} |
954 |
|
|
955 |
|
skip = InitialSkipDecisionP(sad00, pParam, current, pMB, prevMB, x, y, &Data, pGMC, |
956 |
|
pCurrent, pRef, MotionFlags); |
957 |
|
if (skip) continue; |
958 |
|
|
959 |
|
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
960 |
|
y, MotionFlags, current->vop_flags, |
961 |
|
&Data, pParam, pMBs, reference->mbs, pMB); |
962 |
|
|
963 |
|
if (current->vop_flags & XVID_VOP_MODEDECISION_RD) |
964 |
|
xvid_me_ModeDecision_RD(&Data, pMB, pMBs, x, y, pParam, |
965 |
|
MotionFlags, current->vop_flags, current->vol_flags, |
966 |
|
pCurrent, pRef, pGMC, current->coding_type); |
967 |
|
|
968 |
|
else if (current->vop_flags & XVID_VOP_FAST_MODEDECISION_RD) |
969 |
|
xvid_me_ModeDecision_Fast(&Data, pMB, pMBs, x, y, pParam, |
970 |
|
MotionFlags, current->vop_flags, current->vol_flags, |
971 |
|
pCurrent, pRef, pGMC, current->coding_type); |
972 |
|
else |
973 |
|
ModeDecision_SAD(&Data, pMB, pMBs, x, y, pParam, |
974 |
|
MotionFlags, current->vop_flags, current->vol_flags, |
975 |
|
pCurrent, pRef, pGMC, current->coding_type, sad00); |
976 |
|
|
977 |
|
|
978 |
|
motionStatsPVOP(&MVmax, &mvCount, &mvSum, pMB, Data.qpel); |
979 |
|
} |
980 |
|
} |
981 |
|
|
982 |
|
current->fcode = getMinFcode(MVmax); |
983 |
|
current->sStat.iMvSum = mvSum; |
984 |
|
current->sStat.iMvCount = mvCount; |
985 |
|
} |
986 |
|
|
987 |
|
void |
988 |
|
MotionEstimateSMP(SMPmotionData * h) |
989 |
|
{ |
990 |
|
const MBParam * const pParam = h->pParam; |
991 |
|
const FRAMEINFO * const current = h->current; |
992 |
|
const FRAMEINFO * const reference = h->reference; |
993 |
|
const IMAGE * const pRefH = h->pRefH; |
994 |
|
const IMAGE * const pRefV = h->pRefV; |
995 |
|
const IMAGE * const pRefHV = h->pRefHV; |
996 |
|
const IMAGE * const pGMC = h->pGMC; |
997 |
|
uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, |
998 |
|
current->vop_flags, |
999 |
|
current->vol_flags); |
1000 |
|
|
1001 |
|
MACROBLOCK *const pMBs = current->mbs; |
1002 |
|
const IMAGE *const pCurrent = ¤t->image; |
1003 |
|
const IMAGE *const pRef = &reference->image; |
1004 |
|
|
1005 |
|
const uint32_t mb_width = pParam->mb_width; |
1006 |
|
const uint32_t mb_height = pParam->mb_height; |
1007 |
|
const uint32_t iEdgedWidth = pParam->edged_width; |
1008 |
|
int stat_thresh = 0; |
1009 |
|
int MVmax = 0, mvSum = 0, mvCount = 0; |
1010 |
|
int y_step = h->y_step; |
1011 |
|
int start_y = h->start_y; |
1012 |
|
|
1013 |
|
uint32_t x, y; |
1014 |
|
int sad00; |
1015 |
|
int skip_thresh = INITIAL_SKIP_THRESH * \ |
1016 |
|
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
1017 |
|
int block = start_y*mb_width; |
1018 |
|
int * complete_count_self = h->complete_count_self; |
1019 |
|
const int * complete_count_above = h->complete_count_above; |
1020 |
|
int max_mbs; |
1021 |
|
int current_mb = 0; |
1022 |
|
|
1023 |
|
/* some pre-initialized thingies for SearchP */ |
1024 |
|
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
1025 |
|
SearchData Data; |
1026 |
|
memset(&Data, 0, sizeof(SearchData)); |
1027 |
|
Data.iEdgedWidth = iEdgedWidth; |
1028 |
|
Data.iFcode = current->fcode; |
1029 |
|
Data.rounding = pParam->m_rounding_type; |
1030 |
|
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
1031 |
|
Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP; |
1032 |
|
Data.dctSpace = dct_space; |
1033 |
|
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
1034 |
|
Data.mpeg_quant_matrices = pParam->mpeg_quant_matrices; |
1035 |
|
|
1036 |
|
/* todo: sort out temp memory space */ |
1037 |
|
Data.RefQ = h->RefQ; |
1038 |
|
if (sadInit) (*sadInit) (); |
1039 |
|
|
1040 |
|
max_mbs = 0; |
1041 |
|
|
1042 |
|
for (y = start_y; y < mb_height; y += y_step) { |
1043 |
|
if (y == 0) max_mbs = mb_width; /* we can process all blocks of the first row */ |
1044 |
|
|
1045 |
|
for (x = 0; x < mb_width; x++) { |
1046 |
|
|
1047 |
|
MACROBLOCK *pMB, *prevMB; |
1048 |
|
int skip; |
1049 |
|
|
1050 |
|
pMB = &pMBs[block]; |
1051 |
|
prevMB = &reference->mbs[block]; |
1052 |
|
|
1053 |
|
pMB->sad16 = |
1054 |
|
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
1055 |
|
pRef->y + (x + y * iEdgedWidth) * 16, |
1056 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
1057 |
|
|
1058 |
|
sad00 = 4*MAX(MAX(pMB->sad8[0], pMB->sad8[1]), MAX(pMB->sad8[2], pMB->sad8[3])); |
1059 |
|
|
1060 |
if (Data.chroma) { |
if (Data.chroma) { |
1061 |
Data.chromaSAD = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, |
Data.chromaSAD = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, |
1062 |
pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2) |
pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2) |
1063 |
+ sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, |
+ sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, |
1064 |
pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
1065 |
pMB->sad16 += Data.chromaSAD; |
pMB->sad16 += Data.chromaSAD; |
1066 |
|
sad00 += Data.chromaSAD; |
1067 |
} |
} |
1068 |
|
|
1069 |
sad00 = pMB->sad16; |
if (current_mb >= max_mbs) { |
1070 |
|
/* we ME-ed all macroblocks we safely could. grab next portion */ |
1071 |
|
int above_count = *complete_count_above; /* sync point */ |
1072 |
|
if (above_count == mb_width) { |
1073 |
|
/* full line above is ready */ |
1074 |
|
above_count = mb_width+1; |
1075 |
|
if (y < mb_height-y_step) { |
1076 |
|
/* this is not last line, grab a portion of MBs from the next line too */ |
1077 |
|
above_count += MAX(0, complete_count_above[1] - 1); |
1078 |
|
} |
1079 |
|
} |
1080 |
|
|
1081 |
/* initial skip decision */ |
max_mbs = current_mb + above_count - x - 1; |
|
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
|
|
if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
|
|
if (pMB->dquant == 0 |
|
|
&& pMB->sad8[0] < pMB->quant * skip_thresh |
|
|
&& pMB->sad8[1] < pMB->quant * skip_thresh |
|
|
&& pMB->sad8[2] < pMB->quant * skip_thresh |
|
|
&& pMB->sad8[3] < pMB->quant * skip_thresh) |
|
1082 |
|
|
1083 |
if (Data.chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
if (current_mb >= max_mbs) { |
1084 |
ZeroMacroblockP(pMB, sad00); |
/* current workload is zero */ |
1085 |
pMB->mode = MODE_NOT_CODED; |
x--; |
1086 |
|
sched_yield(); |
1087 |
continue; |
continue; |
1088 |
} |
} |
1089 |
} |
} |
1090 |
|
|
1091 |
if(MotionFlags & XVID_ME_DETECT_STATIC_MOTION) { |
skip = InitialSkipDecisionP(sad00, pParam, current, pMB, prevMB, x, y, &Data, pGMC, |
1092 |
if(x > 0 && y > 0 && x < pParam->mb_width) { |
pCurrent, pRef, MotionFlags); |
1093 |
if(MVequal((&pMBs[(x-1) + y * pParam->mb_width])->mvs[0], zeroMV) && |
if (current_mb >= max_mbs) { |
1094 |
MVequal((&pMBs[x + (y-1) * pParam->mb_width])->mvs[0], zeroMV) && |
/* we ME-ed all macroblocks we safely could. grab next portion */ |
1095 |
MVequal((&pMBs[(x+1) + (y-1) * pParam->mb_width])->mvs[0], zeroMV) && |
int above_count = *complete_count_above; /* sync point */ |
1096 |
MVequal(prevMB->mvs[0], zeroMV)) { |
if (above_count == mb_width) { |
1097 |
stat_thresh = MAX((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
/* full line above is ready */ |
1098 |
MAX((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
above_count = mb_width+1; |
1099 |
MAX((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
if (y < mb_height-y_step) { |
1100 |
prevMB->sad16))); |
/* this is not last line, grab a portion of MBs from the next line too */ |
1101 |
} else { |
above_count += MAX(0, complete_count_above[1] - 1); |
|
stat_thresh = MIN((&pMBs[(x-1) + y * pParam->mb_width])->sad16, |
|
|
MIN((&pMBs[x + (y-1) * pParam->mb_width])->sad16, |
|
|
MIN((&pMBs[(x+1) + (y-1) * pParam->mb_width])->sad16, |
|
|
prevMB->sad16))); |
|
1102 |
} |
} |
1103 |
} |
} |
1104 |
|
|
1105 |
|
max_mbs = current_mb + above_count - x - 1; |
1106 |
|
|
1107 |
|
if (current_mb >= max_mbs) { |
1108 |
|
/* current workload is zero */ |
1109 |
|
x--; |
1110 |
|
sched_yield(); |
1111 |
|
continue; |
1112 |
|
} |
1113 |
} |
} |
1114 |
|
|
1115 |
/* favorize (0,0) vector for cartoons */ |
if (skip) { |
1116 |
if ((current->vop_flags & XVID_VOP_CARTOON) && |
current_mb++; |
1117 |
((sad00 < pMB->quant * 4 * skip_thresh) || (sad00 < stat_thresh))) { |
block++; |
1118 |
ZeroMacroblockP(pMB, sad00); |
*complete_count_self = x+1; |
1119 |
continue; |
continue; |
1120 |
} |
} |
1121 |
|
|
1135 |
else |
else |
1136 |
ModeDecision_SAD(&Data, pMB, pMBs, x, y, pParam, |
ModeDecision_SAD(&Data, pMB, pMBs, x, y, pParam, |
1137 |
MotionFlags, current->vop_flags, current->vol_flags, |
MotionFlags, current->vop_flags, current->vol_flags, |
1138 |
pCurrent, pRef, pGMC, current->coding_type); |
pCurrent, pRef, pGMC, current->coding_type, sad00); |
1139 |
|
|
1140 |
|
*complete_count_self = x+1; |
1141 |
|
|
1142 |
|
current_mb++; |
1143 |
|
block++; |
1144 |
|
|
1145 |
|
motionStatsPVOP(&MVmax, &mvCount, &mvSum, pMB, Data.qpel); |
1146 |
|
|
|
if (pMB->mode == MODE_INTRA) |
|
|
if (++iIntra > iLimit) return 1; |
|
1147 |
} |
} |
1148 |
|
block += (y_step-1)*pParam->mb_width; |
1149 |
|
complete_count_self++; |
1150 |
|
complete_count_above++; |
1151 |
} |
} |
1152 |
return 0; |
|
1153 |
|
h->MVmax = MVmax; |
1154 |
|
h->mvSum = mvSum; |
1155 |
|
h->mvCount = mvCount; |
1156 |
} |
} |
1157 |
|
|
1158 |
|
|