39 |
#include "sad.h" |
#include "sad.h" |
40 |
#include "motion_inlines.h" |
#include "motion_inlines.h" |
41 |
|
|
|
|
|
42 |
static int32_t |
static int32_t |
43 |
ChromaSAD2(const int fx, const int fy, const int bx, const int by, |
ChromaSAD2(const int fx, const int fy, const int bx, const int by, |
44 |
const SearchData * const data) |
SearchData * const data) |
45 |
{ |
{ |
46 |
int sad; |
int sad; |
47 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
48 |
uint8_t *f_refu, *f_refv, *b_refu, *b_refv; |
uint8_t *f_refu, *f_refv, *b_refu, *b_refv; |
49 |
|
int offset, filter; |
50 |
|
|
51 |
const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = { |
const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = { |
52 |
NULL, |
NULL, |
55 |
interpolate8x8_halfpel_hv |
interpolate8x8_halfpel_hv |
56 |
}; |
}; |
57 |
|
|
58 |
int offset = (fx>>1) + (fy>>1)*stride; |
offset = (fx>>1) + (fy>>1)*stride; |
59 |
int filter = ((fx & 1) << 1) | (fy & 1); |
filter = ((fx & 1) << 1) | (fy & 1); |
60 |
|
|
61 |
if (filter != 0) { |
if (filter != 0) { |
62 |
f_refu = data->RefQ; |
f_refu = data->RefQ; |
88 |
} |
} |
89 |
|
|
90 |
static void |
static void |
91 |
CheckCandidateInt(const int xf, const int yf, const SearchData * const data, const unsigned int Direction) |
CheckCandidateInt(const int x, const int y, SearchData * const data, const unsigned int Direction) |
92 |
{ |
{ |
93 |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
int32_t sad, xf, yf, xb, yb, xcf, ycf, xcb, ycb; |
94 |
uint32_t t; |
uint32_t t; |
95 |
|
|
96 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
97 |
VECTOR *current; |
VECTOR *current; |
98 |
|
|
99 |
if ((xf > data->max_dx) || (xf < data->min_dx) || |
if ((x > data->max_dx) || (x < data->min_dx) || |
100 |
(yf > data->max_dy) || (yf < data->min_dy)) |
(y > data->max_dy) || (y < data->min_dy)) |
101 |
return; |
return; |
102 |
|
|
103 |
|
if (Direction == 1) { /* x and y mean forward vector */ |
104 |
|
VECTOR backward = data->qpel_precision ? data->currentQMV[1] : data->currentMV[1]; |
105 |
|
xb = backward.x; |
106 |
|
yb = backward.y; |
107 |
|
xf = x; yf = y; |
108 |
|
} else { /* x and y mean backward vector */ |
109 |
|
VECTOR forward = data->qpel_precision ? data->currentQMV[0] : data->currentMV[0]; |
110 |
|
xf = forward.x; |
111 |
|
yf = forward.y; |
112 |
|
xb = x; yb = y; |
113 |
|
} |
114 |
|
|
115 |
if (!data->qpel_precision) { |
if (!data->qpel_precision) { |
116 |
ReferenceF = GetReference(xf, yf, data); |
ReferenceF = GetReference(xf, yf, data); |
|
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
|
117 |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
118 |
current = data->currentMV; |
current = data->currentMV + Direction - 1; |
119 |
xcf = xf; ycf = yf; |
xcf = xf; ycf = yf; |
120 |
xcb = xb; ycb = yb; |
xcb = xb; ycb = yb; |
121 |
} else { |
} else { |
122 |
ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data); |
ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data); |
123 |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
current = data->currentQMV + Direction - 1; |
|
current = data->currentQMV; |
|
124 |
ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data); |
ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data); |
125 |
xcf = xf/2; ycf = yf/2; |
xcf = xf/2; ycf = yf/2; |
126 |
xcb = xb/2; ycb = yb/2; |
xcb = xb/2; ycb = yb/2; |
139 |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
140 |
|
|
141 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
142 |
*(data->iMinSAD) = sad; |
*data->iMinSAD = sad; |
143 |
current->x = xf; current->y = yf; |
current->x = x; current->y = y; |
144 |
*data->dir = Direction; |
data->dir = Direction; |
145 |
} |
} |
146 |
} |
} |
147 |
|
|
148 |
static void |
static void |
149 |
CheckCandidateDirect(const int x, const int y, const SearchData * const data, const unsigned int Direction) |
CheckCandidateDirect(const int x, const int y, SearchData * const data, const unsigned int Direction) |
150 |
{ |
{ |
151 |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
152 |
uint32_t k; |
uint32_t k; |
200 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
201 |
|
|
202 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
203 |
*(data->iMinSAD) = sad; |
data->iMinSAD[0] = sad; |
204 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
205 |
*data->dir = Direction; |
data->dir = Direction; |
206 |
} |
} |
207 |
} |
} |
208 |
|
|
209 |
static void |
static void |
210 |
CheckCandidateDirectno4v(const int x, const int y, const SearchData * const data, const unsigned int Direction) |
CheckCandidateDirectno4v(const int x, const int y, SearchData * const data, const unsigned int Direction) |
211 |
{ |
{ |
212 |
int32_t sad, xcf, ycf, xcb, ycb; |
int32_t sad, xcf, ycf, xcb, ycb; |
213 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
255 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
256 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
257 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
258 |
*data->dir = Direction; |
data->dir = Direction; |
259 |
} |
} |
260 |
} |
} |
261 |
|
|
262 |
void |
void |
263 |
CheckCandidate16no4v(const int x, const int y, const SearchData * const data, const unsigned int Direction) |
CheckCandidate16no4v(const int x, const int y, SearchData * const data, const unsigned int Direction) |
264 |
{ |
{ |
265 |
int32_t sad, xc, yc; |
int32_t sad, xc, yc; |
266 |
const uint8_t * Reference; |
const uint8_t * Reference; |
294 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
295 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
296 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
297 |
*data->dir = Direction; |
data->dir = Direction; |
298 |
} |
} |
299 |
} |
} |
300 |
|
|
302 |
ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) |
ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) |
303 |
{ |
{ |
304 |
/* the stupidiest function ever */ |
/* the stupidiest function ever */ |
305 |
return (mode == MODE_FORWARD) ? pMB->mvs[0] : pMB->b_mvs[0]; |
return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]); |
306 |
} |
} |
307 |
|
|
308 |
static void __inline |
static void __inline |
361 |
|
|
362 |
int i; |
int i; |
363 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
364 |
MainSearchFunc *MainSearchPtr; |
int threshA = (MotionFlags & XVID_ME_FASTREFINE16) ? 150 : 300; |
365 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
366 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
367 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
368 |
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; /* reset chroma-sad cache */ |
Data->chromaX = Data->chromaY = Data->chromaSAD = 256*4096; /* reset chroma-sad cache */ |
369 |
|
|
370 |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
371 |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
391 |
if (!vector_repeats(pmv, i) ) |
if (!vector_repeats(pmv, i) ) |
392 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
393 |
|
|
394 |
|
if (*Data->iMinSAD > 512) { |
395 |
|
unsigned int mask = make_mask(pmv, 7, Data->dir); |
396 |
|
|
397 |
|
MainSearchFunc *MainSearchPtr; |
398 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
399 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
400 |
else MainSearchPtr = xvid_me_DiamondSearch; |
else MainSearchPtr = xvid_me_DiamondSearch; |
401 |
|
|
|
if (*Data->iMinSAD > 512) { |
|
|
unsigned int mask = make_mask(pmv, 7, *Data->dir); |
|
402 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
403 |
} |
} |
404 |
|
|
|
xvid_me_SubpelRefine(Data, CheckCandidate16no4v); |
|
405 |
|
|
406 |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
|
407 |
|
|
408 |
|
if(!Data->qpel) { |
409 |
|
/* halfpel mode */ |
410 |
|
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
411 |
|
xvid_me_SubpelRefine(Data, CheckCandidate16no4v, 0); |
412 |
|
} else { |
413 |
|
/* qpel mode */ |
414 |
|
if(MotionFlags & XVID_ME_FASTREFINE16) { |
415 |
|
/* fast */ |
416 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
417 |
|
pParam->width, pParam->height, Data->iFcode, 2, 0); |
418 |
|
if (*Data->iMinSAD < *best_sad + 2*threshA) |
419 |
|
FullRefine_Fast(Data, CheckCandidate16no4v, 0); |
420 |
|
} else { |
421 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
422 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
423 |
Data->qpel_precision = 1; |
if(MotionFlags & XVID_ME_QUARTERPELREFINE16) { |
424 |
|
/* full */ |
425 |
|
if (MotionFlags & XVID_ME_HALFPELREFINE16) { |
426 |
|
xvid_me_SubpelRefine(Data, CheckCandidate16no4v, 0); /* hpel part */ |
427 |
|
Data->currentQMV->x = 2*Data->currentMV->x; |
428 |
|
Data->currentQMV->y = 2*Data->currentMV->y; |
429 |
|
} |
430 |
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, |
431 |
pParam->width, pParam->height, iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
432 |
xvid_me_SubpelRefine(Data, CheckCandidate16no4v); |
Data->qpel_precision = 1; |
433 |
|
if (*Data->iMinSAD < *best_sad + threshA) |
434 |
|
xvid_me_SubpelRefine(Data, CheckCandidate16no4v, 0); /* qpel part */ |
435 |
|
} |
436 |
|
} |
437 |
} |
} |
438 |
|
|
439 |
/* three bits are needed to code backward mode. four for forward */ |
/* three bits are needed to code backward mode. four for forward */ |
471 |
const uint32_t x, const uint32_t y, |
const uint32_t x, const uint32_t y, |
472 |
const SearchData * const Data) |
const SearchData * const Data) |
473 |
{ |
{ |
474 |
|
int k; |
475 |
|
|
476 |
|
if (!Data->chroma) { |
477 |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
478 |
int32_t sum; |
int32_t sum; |
|
int k; |
|
479 |
const uint32_t stride = Data->iEdgedWidth/2; |
const uint32_t stride = Data->iEdgedWidth/2; |
480 |
/* 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 */ |
481 |
|
|
503 |
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, |
504 |
stride); |
stride); |
505 |
|
|
506 |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) { |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
507 |
|
} |
508 |
|
|
509 |
|
/* skip */ |
510 |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
511 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
512 |
pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; |
pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; |
513 |
pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; |
pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; |
514 |
} |
} |
515 |
} |
} |
|
} |
|
516 |
|
|
517 |
static uint32_t |
static uint32_t |
518 |
SearchDirect(const IMAGE * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
590 |
CheckCandidate(0, 0, Data, 255); |
CheckCandidate(0, 0, Data, 255); |
591 |
|
|
592 |
/* initial (fast) skip decision */ |
/* initial (fast) skip decision */ |
593 |
if (*Data->iMinSAD < (int)Data->iQuant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
if (*Data->iMinSAD < (int)Data->iQuant * INITIAL_SKIP_THRESH) { |
594 |
/* possible skip */ |
/* possible skip */ |
|
if (Data->chroma) { |
|
|
pMB->mode = MODE_DIRECT_NONE_MV; |
|
|
return *Data->iMinSAD; /* skip. */ |
|
|
} else { |
|
595 |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
596 |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skip. */ |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skipped */ |
|
} |
|
597 |
} |
} |
598 |
|
|
599 |
*Data->iMinSAD += Data->lambda16; |
*Data->iMinSAD += Data->lambda16; |
600 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
601 |
|
|
602 |
/* |
if (!(MotionFlags & XVID_ME_SKIP_DELTASEARCH)) { |
|
* DIRECT MODE DELTA VECTOR SEARCH. |
|
|
* This has to be made more effective, but at the moment I'm happy it's running at all |
|
|
*/ |
|
|
|
|
603 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
604 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
605 |
else MainSearchPtr = xvid_me_DiamondSearch; |
else MainSearchPtr = xvid_me_DiamondSearch; |
606 |
|
|
607 |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
608 |
|
|
609 |
xvid_me_SubpelRefine(Data, CheckCandidate); |
xvid_me_SubpelRefine(Data, CheckCandidate, 0); |
610 |
|
} |
611 |
|
|
612 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
613 |
|
|
643 |
return skip_sad; |
return skip_sad; |
644 |
} |
} |
645 |
|
|
646 |
|
|
647 |
|
static void set_range(int * range, SearchData * Data) |
648 |
|
{ |
649 |
|
Data->min_dx = range[0]; |
650 |
|
Data->max_dx = range[1]; |
651 |
|
Data->min_dy = range[2]; |
652 |
|
Data->max_dy = range[3]; |
653 |
|
} |
654 |
|
|
655 |
static void |
static void |
656 |
SearchInterpolate(const IMAGE * const f_Ref, |
SearchInterpolate(const IMAGE * const f_Ref, |
657 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
670 |
const VECTOR * const b_predMV, |
const VECTOR * const b_predMV, |
671 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
672 |
int32_t * const best_sad, |
int32_t * const best_sad, |
673 |
SearchData * const fData) |
SearchData * const Data) |
674 |
|
|
675 |
{ |
{ |
676 |
int i, j; |
int i, j; |
677 |
SearchData bData; |
int b_range[4], f_range[4]; |
678 |
|
int threshA = (MotionFlags & XVID_ME_FAST_MODEINTERPOLATE) ? 250 : 500; |
679 |
|
int threshB = (MotionFlags & XVID_ME_FAST_MODEINTERPOLATE) ? 150 : 300; |
680 |
|
|
681 |
fData->qpel_precision = 0; |
Data->qpel_precision = 0; |
682 |
memcpy(&bData, fData, sizeof(SearchData)); /* quick copy of common data */ |
*Data->iMinSAD = 4096*256; |
683 |
*fData->iMinSAD = 4096*256; |
Data->iFcode = fcode; Data->bFcode = bcode; |
|
bData.currentMV++; bData.currentQMV++; |
|
|
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
|
|
|
|
|
i = (x + y * fData->iEdgedWidth) * 16; |
|
|
|
|
|
bData.b_RefP[0] = fData->RefP[0] = f_Ref->y + i; |
|
|
bData.b_RefP[2] = fData->RefP[2] = f_RefH + i; |
|
|
bData.b_RefP[1] = fData->RefP[1] = f_RefV + i; |
|
|
bData.b_RefP[3] = fData->RefP[3] = f_RefHV + i; |
|
|
bData.RefP[0] = fData->b_RefP[0] = b_Ref->y + i; |
|
|
bData.RefP[2] = fData->b_RefP[2] = b_RefH + i; |
|
|
bData.RefP[1] = fData->b_RefP[1] = b_RefV + i; |
|
|
bData.RefP[3] = fData->b_RefP[3] = b_RefHV + i; |
|
|
bData.b_RefP[4] = fData->RefP[4] = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
|
|
bData.dir = fData->dir; |
|
|
|
|
|
bData.bpredMV = fData->predMV = *f_predMV; |
|
|
fData->bpredMV = bData.predMV = *b_predMV; |
|
|
fData->currentMV[0] = fData->currentMV[2]; |
|
|
|
|
|
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode - fData->qpel, 1, 0); |
|
|
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode - fData->qpel, 1, 0); |
|
|
|
|
|
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
|
|
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
|
|
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy; |
|
|
if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
|
|
|
|
|
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
|
|
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx; |
|
|
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
|
|
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
|
684 |
|
|
685 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255); |
i = (x + y * Data->iEdgedWidth) * 16; |
686 |
|
|
687 |
|
Data->RefP[0] = f_Ref->y + i; |
688 |
|
Data->RefP[2] = f_RefH + i; |
689 |
|
Data->RefP[1] = f_RefV + i; |
690 |
|
Data->RefP[3] = f_RefHV + i; |
691 |
|
Data->b_RefP[0] = b_Ref->y + i; |
692 |
|
Data->b_RefP[2] = b_RefH + i; |
693 |
|
Data->b_RefP[1] = b_RefV + i; |
694 |
|
Data->b_RefP[3] = b_RefHV + i; |
695 |
|
Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
696 |
|
Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
697 |
|
Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
698 |
|
Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
699 |
|
|
700 |
|
Data->predMV = *f_predMV; |
701 |
|
Data->bpredMV = *b_predMV; |
702 |
|
|
703 |
|
Data->currentMV[0] = Data->currentMV[2]; /* forward search left its vector here */ |
704 |
|
|
705 |
|
get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, fcode - Data->qpel, 1, 0); |
706 |
|
get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, bcode - Data->qpel, 1, 0); |
707 |
|
|
708 |
|
if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; |
709 |
|
if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; |
710 |
|
if (Data->currentMV[0].y > f_range[3]) Data->currentMV[0].y = f_range[3]; |
711 |
|
if (Data->currentMV[0].y < f_range[2]) Data->currentMV[0].y = f_range[2]; |
712 |
|
|
713 |
|
if (Data->currentMV[1].x > b_range[1]) Data->currentMV[1].x = b_range[1]; |
714 |
|
if (Data->currentMV[1].x < b_range[0]) Data->currentMV[1].x = b_range[0]; |
715 |
|
if (Data->currentMV[1].y > b_range[3]) Data->currentMV[1].y = b_range[3]; |
716 |
|
if (Data->currentMV[1].y < b_range[2]) Data->currentMV[1].y = b_range[2]; |
717 |
|
|
718 |
|
set_range(f_range, Data); |
719 |
|
|
720 |
|
CheckCandidateInt(Data->currentMV[0].x, Data->currentMV[0].y, Data, 1); |
721 |
|
|
722 |
/* diamond */ |
/* diamond */ |
723 |
do { |
do { |
724 |
*fData->dir = 255; |
Data->dir = 0; |
725 |
/* forward MV moves */ |
/* forward MV moves */ |
726 |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
i = Data->currentMV[0].x; j = Data->currentMV[0].y; |
727 |
|
|
728 |
CheckCandidateInt(i + 1, j, fData, 0); |
CheckCandidateInt(i + 1, j, Data, 1); |
729 |
CheckCandidateInt(i, j + 1, fData, 0); |
CheckCandidateInt(i, j + 1, Data, 1); |
730 |
CheckCandidateInt(i - 1, j, fData, 0); |
CheckCandidateInt(i - 1, j, Data, 1); |
731 |
CheckCandidateInt(i, j - 1, fData, 0); |
CheckCandidateInt(i, j - 1, Data, 1); |
732 |
|
|
733 |
/* backward MV moves */ |
/* backward MV moves */ |
734 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
set_range(b_range, Data); |
735 |
fData->currentMV[2] = fData->currentMV[0]; |
i = Data->currentMV[1].x; j = Data->currentMV[1].y; |
|
CheckCandidateInt(i + 1, j, &bData, 0); |
|
|
CheckCandidateInt(i, j + 1, &bData, 0); |
|
|
CheckCandidateInt(i - 1, j, &bData, 0); |
|
|
CheckCandidateInt(i, j - 1, &bData, 0); |
|
736 |
|
|
737 |
} while (!(*fData->dir)); |
CheckCandidateInt(i + 1, j, Data, 2); |
738 |
|
CheckCandidateInt(i, j + 1, Data, 2); |
739 |
|
CheckCandidateInt(i - 1, j, Data, 2); |
740 |
|
CheckCandidateInt(i, j - 1, Data, 2); |
741 |
|
|
742 |
|
set_range(f_range, Data); |
743 |
|
|
744 |
|
} while (Data->dir != 0); |
745 |
|
|
746 |
/* qpel refinement */ |
/* qpel refinement */ |
747 |
if (fData->qpel) { |
if (Data->qpel) { |
748 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*Data->iMinSAD > *best_sad + threshA) return; |
749 |
fData->qpel_precision = bData.qpel_precision = 1; |
Data->qpel_precision = 1; |
750 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); |
751 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
|
752 |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
Data->currentQMV[0].x = 2 * Data->currentMV[0].x; |
753 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
Data->currentQMV[0].y = 2 * Data->currentMV[0].y; |
754 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
Data->currentQMV[1].x = 2 * Data->currentMV[1].x; |
755 |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
Data->currentQMV[1].y = 2 * Data->currentMV[1].y; |
756 |
xvid_me_SubpelRefine(fData, CheckCandidateInt); |
|
757 |
if (*fData->iMinSAD > *best_sad + 300) return; |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
758 |
fData->currentQMV[2] = fData->currentQMV[0]; |
xvid_me_SubpelRefine(Data, CheckCandidateInt, 1); |
759 |
xvid_me_SubpelRefine(&bData, CheckCandidateInt); |
|
760 |
} |
if (*Data->iMinSAD > *best_sad + threshB) return; |
761 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
762 |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
|
763 |
|
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
764 |
if (*fData->iMinSAD < *best_sad) { |
xvid_me_SubpelRefine(Data, CheckCandidateInt, 2); |
765 |
*best_sad = *fData->iMinSAD; |
} |
766 |
pMB->mvs[0] = fData->currentMV[0]; |
|
767 |
pMB->b_mvs[0] = fData->currentMV[1]; |
*Data->iMinSAD += 2 * Data->lambda16; /* two bits are needed to code interpolate mode. */ |
768 |
|
|
769 |
|
if (*Data->iMinSAD < *best_sad) { |
770 |
|
*best_sad = *Data->iMinSAD; |
771 |
|
pMB->mvs[0] = Data->currentMV[0]; |
772 |
|
pMB->b_mvs[0] = Data->currentMV[1]; |
773 |
pMB->mode = MODE_INTERPOLATE; |
pMB->mode = MODE_INTERPOLATE; |
774 |
if (fData->qpel) { |
if (Data->qpel) { |
775 |
pMB->qmvs[0] = fData->currentQMV[0]; |
pMB->qmvs[0] = Data->currentQMV[0]; |
776 |
pMB->b_qmvs[0] = fData->currentQMV[1]; |
pMB->b_qmvs[0] = Data->currentQMV[1]; |
777 |
pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
778 |
pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; |
pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; |
779 |
pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; |
pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; |
810 |
uint32_t skip_sad; |
uint32_t skip_sad; |
811 |
|
|
812 |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
813 |
|
MACROBLOCK *const pMBs = frame->mbs; |
814 |
|
|
815 |
VECTOR f_predMV, b_predMV; |
VECTOR f_predMV, b_predMV; |
816 |
|
|
820 |
/* some pre-inintialized data for the rest of the search */ |
/* some pre-inintialized data for the rest of the search */ |
821 |
|
|
822 |
SearchData Data; |
SearchData Data; |
|
int32_t iMinSAD; |
|
|
uint32_t dir; |
|
|
VECTOR currentMV[3]; |
|
|
VECTOR currentQMV[3]; |
|
|
int32_t temp[8]; |
|
823 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
824 |
|
|
825 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
|
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
|
|
Data.iMinSAD = &iMinSAD; |
|
|
Data.lambda16 = xvid_me_lambda_vec16[MAX(frame->quant-2, 2)]; |
|
826 |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
827 |
Data.rounding = 0; |
Data.rounding = 0; |
828 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
|
Data.temp = temp; |
|
|
Data.dir = &dir; |
|
829 |
Data.iQuant = frame->quant; |
Data.iQuant = frame->quant; |
830 |
|
|
831 |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */ |
838 |
for (i = 0; i < pParam->mb_width; i++) { |
for (i = 0; i < pParam->mb_width; i++) { |
839 |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
840 |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
841 |
|
int interpol_search = 0; |
842 |
|
int bf_search = 0; |
843 |
|
int bf_thresh = 0; |
844 |
|
|
845 |
/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */ |
/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */ |
846 |
if (b_reference->coding_type != S_VOP) |
if (b_reference->coding_type != S_VOP) |
847 |
if (b_mb->mode == MODE_NOT_CODED) { |
if (b_mb->mode == MODE_NOT_CODED) { |
848 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
849 |
pMB->mvs[0] = zeroMV; |
pMB->mvs[0] = pMB->b_mvs[0] = zeroMV; |
850 |
pMB->b_mvs[0] = zeroMV; |
pMB->sad16 = 0; |
851 |
continue; |
continue; |
852 |
} |
} |
853 |
|
|
854 |
|
Data.lambda16 = xvid_me_lambda_vec16[b_mb->quant]; |
855 |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
856 |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
857 |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
869 |
&best_sad, |
&best_sad, |
870 |
&Data); |
&Data); |
871 |
|
|
872 |
if (pMB->mode == MODE_DIRECT_NONE_MV) continue; |
if (pMB->mode == MODE_DIRECT_NONE_MV) { |
873 |
|
pMB->sad16 = best_sad; |
874 |
|
continue; |
875 |
|
} |
876 |
|
|
877 |
|
if (frame->motion_flags & XVID_ME_BFRAME_EARLYSTOP) { |
878 |
|
if(i > 0 && j > 0 && i < pParam->mb_width) { |
879 |
|
bf_thresh = MIN((&pMBs[(i-1) + j * pParam->mb_width])->sad16, |
880 |
|
MIN((&pMBs[i + (j-1) * pParam->mb_width])->sad16, |
881 |
|
(&pMBs[(i+1) + (j-1) * pParam->mb_width])->sad16)); |
882 |
|
|
883 |
|
if (((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_FORWARD) && |
884 |
|
((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_BACKWARD) && |
885 |
|
((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
886 |
|
bf_search++; |
887 |
|
|
888 |
|
if (((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_FORWARD) && |
889 |
|
((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_BACKWARD) && |
890 |
|
((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
891 |
|
bf_search++; |
892 |
|
|
893 |
|
if (((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_FORWARD) && |
894 |
|
((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_BACKWARD) && |
895 |
|
((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
896 |
|
bf_search++; |
897 |
|
} |
898 |
|
|
899 |
|
if ((best_sad < bf_thresh) && (bf_search == 3)) |
900 |
|
continue; |
901 |
|
} |
902 |
|
|
903 |
/* forward search */ |
/* forward search */ |
904 |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
917 |
MODE_BACKWARD, &Data); |
MODE_BACKWARD, &Data); |
918 |
|
|
919 |
/* 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 */ |
920 |
|
if (frame->motion_flags & XVID_ME_FAST_MODEINTERPOLATE) { |
921 |
|
|
922 |
|
if(i > 0 && j > 0 && i < pParam->mb_width) { |
923 |
|
if ((&pMBs[(i-1) + j * pParam->mb_width])->mode == MODE_INTERPOLATE) |
924 |
|
interpol_search++; |
925 |
|
if ((&pMBs[i + (j - 1) * pParam->mb_width])->mode == MODE_INTERPOLATE) |
926 |
|
interpol_search++; |
927 |
|
if ((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode == MODE_INTERPOLATE) |
928 |
|
interpol_search++; |
929 |
|
} |
930 |
|
else |
931 |
|
interpol_search = 1; |
932 |
|
|
933 |
|
interpol_search |= !(best_sad < 3 * Data.iQuant * MAX_SAD00_FOR_SKIP * (Data.chroma ? 3:2)); |
934 |
|
} |
935 |
|
else |
936 |
|
interpol_search = 1; |
937 |
|
|
938 |
|
if (interpol_search) { |
939 |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
940 |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
941 |
i, j, |
i, j, |
945 |
&f_predMV, &b_predMV, |
&f_predMV, &b_predMV, |
946 |
pMB, &best_sad, |
pMB, &best_sad, |
947 |
&Data); |
&Data); |
948 |
|
} |
949 |
|
|
950 |
/* final skip decision */ |
/* final skip decision */ |
951 |
if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * 2) |
if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * (Data.chroma ? 3:2) ) |
952 |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
953 |
|
|
954 |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
955 |
|
|
956 |
switch (pMB->mode) { |
switch (pMB->mode) { |
957 |
case MODE_FORWARD: |
case MODE_FORWARD: |
958 |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
959 |
|
pMB->sad16 = best_sad; |
960 |
break; |
break; |
961 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
962 |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
963 |
|
pMB->sad16 = best_sad; |
964 |
break; |
break; |
965 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
966 |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
967 |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
968 |
|
pMB->sad16 = best_sad; |
969 |
break; |
break; |
970 |
default: |
default: |
971 |
|
pMB->sad16 = best_sad; |
972 |
break; |
break; |
973 |
} |
} |
974 |
} |
} |
975 |
} |
} |
976 |
} |
} |
|
|
|