21 |
* along with this program ; if not, write to the Free Software |
* along with this program ; if not, write to the Free Software |
22 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
* |
* |
24 |
* $Id: estimation_bvop.c,v 1.1.2.3 2003-11-09 20:47:14 edgomez Exp $ |
* $Id: estimation_bvop.c,v 1.1.2.11 2003-12-18 14:47:44 edgomez Exp $ |
25 |
* |
* |
26 |
****************************************************************************/ |
****************************************************************************/ |
27 |
|
|
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; |
87 |
} |
} |
88 |
|
|
89 |
static void |
static void |
90 |
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) |
91 |
{ |
{ |
92 |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
int32_t sad, xf, yf, xb, yb, xcf, ycf, xcb, ycb; |
93 |
uint32_t t; |
uint32_t t; |
94 |
|
|
95 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
96 |
VECTOR *current; |
VECTOR *current; |
97 |
|
|
98 |
if ((xf > data->max_dx) || (xf < data->min_dx) || |
if ((x > data->max_dx) || (x < data->min_dx) || |
99 |
(yf > data->max_dy) || (yf < data->min_dy)) |
(y > data->max_dy) || (y < data->min_dy)) |
100 |
return; |
return; |
101 |
|
|
102 |
|
if (Direction == 1) { /* x and y mean forward vector */ |
103 |
|
VECTOR backward = data->qpel_precision ? data->currentQMV[1] : data->currentMV[1]; |
104 |
|
xb = backward.x; |
105 |
|
yb = backward.y; |
106 |
|
xf = x; yf = y; |
107 |
|
} else { /* x and y mean backward vector */ |
108 |
|
VECTOR forward = data->qpel_precision ? data->currentQMV[0] : data->currentMV[0]; |
109 |
|
xf = forward.x; |
110 |
|
yf = forward.y; |
111 |
|
xb = x; yb = y; |
112 |
|
} |
113 |
|
|
114 |
if (!data->qpel_precision) { |
if (!data->qpel_precision) { |
115 |
ReferenceF = GetReference(xf, yf, data); |
ReferenceF = GetReference(xf, yf, data); |
|
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
|
116 |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
117 |
current = data->currentMV; |
current = data->currentMV + Direction - 1; |
118 |
xcf = xf; ycf = yf; |
xcf = xf; ycf = yf; |
119 |
xcb = xb; ycb = yb; |
xcb = xb; ycb = yb; |
120 |
} else { |
} else { |
121 |
ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data); |
ReferenceF = xvid_me_interpolate16x16qpel(xf, yf, 0, data); |
122 |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
current = data->currentQMV + Direction - 1; |
|
current = data->currentQMV; |
|
123 |
ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data); |
ReferenceB = xvid_me_interpolate16x16qpel(xb, yb, 1, data); |
124 |
xcf = xf/2; ycf = yf/2; |
xcf = xf/2; ycf = yf/2; |
125 |
xcb = xb/2; ycb = yb/2; |
xcb = xb/2; ycb = yb/2; |
138 |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
139 |
|
|
140 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
141 |
*(data->iMinSAD) = sad; |
*data->iMinSAD = sad; |
142 |
current->x = xf; current->y = yf; |
current->x = x; current->y = y; |
143 |
*data->dir = Direction; |
data->dir = Direction; |
144 |
} |
} |
145 |
} |
} |
146 |
|
|
147 |
static void |
static void |
148 |
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) |
149 |
{ |
{ |
150 |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
151 |
uint32_t k; |
uint32_t k; |
199 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
200 |
|
|
201 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
202 |
*(data->iMinSAD) = sad; |
data->iMinSAD[0] = sad; |
203 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
204 |
*data->dir = Direction; |
data->dir = Direction; |
205 |
} |
} |
206 |
} |
} |
207 |
|
|
208 |
static void |
static void |
209 |
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) |
210 |
{ |
{ |
211 |
int32_t sad, xcf, ycf, xcb, ycb; |
int32_t sad, xcf, ycf, xcb, ycb; |
212 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
254 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
255 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
256 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
257 |
*data->dir = Direction; |
data->dir = Direction; |
258 |
} |
} |
259 |
} |
} |
260 |
|
|
261 |
void |
void |
262 |
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) |
263 |
{ |
{ |
264 |
int32_t sad, xc, yc; |
int32_t sad, xc, yc; |
265 |
const uint8_t * Reference; |
const uint8_t * Reference; |
293 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
294 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
295 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
296 |
*data->dir = Direction; |
data->dir = Direction; |
297 |
|
} |
298 |
|
} |
299 |
|
|
300 |
|
void |
301 |
|
CheckCandidate16no4v_qpel(const int x, const int y, SearchData * const data, const unsigned int Direction) |
302 |
|
{ |
303 |
|
int32_t sad, xc, yc; |
304 |
|
const uint8_t * Reference; |
305 |
|
uint32_t t; |
306 |
|
|
307 |
|
if ( (x > data->max_dx) || ( x < data->min_dx) |
308 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
309 |
|
|
310 |
|
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ |
311 |
|
|
312 |
|
Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); |
313 |
|
|
314 |
|
xc = x/2; yc = y/2; |
315 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
316 |
|
data->qpel^data->qpel_precision, data->rrv); |
317 |
|
|
318 |
|
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
319 |
|
sad += (data->lambda16 * t * sad)>>10; |
320 |
|
|
321 |
|
if (data->chroma && sad < *data->iMinSAD) |
322 |
|
sad += xvid_me_ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
323 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
324 |
|
|
325 |
|
if (sad < *(data->iMinSAD)) { |
326 |
|
data->iMinSAD2 = *(data->iMinSAD); |
327 |
|
data->currentQMV2.x = data->currentQMV->x; |
328 |
|
data->currentQMV2.y = data->currentQMV->y; |
329 |
|
|
330 |
|
data->iMinSAD[0] = sad; |
331 |
|
data->currentQMV[0].x = x; data->currentQMV[0].y = y; |
332 |
|
} else if (sad < data->iMinSAD2) { |
333 |
|
data->iMinSAD2 = sad; |
334 |
|
data->currentQMV2.x = x; data->currentQMV2.y = y; |
335 |
} |
} |
336 |
} |
} |
337 |
|
|
398 |
|
|
399 |
int i; |
int i; |
400 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
|
MainSearchFunc *MainSearchPtr; |
|
401 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
402 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
403 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
404 |
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 */ |
405 |
|
|
406 |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
407 |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16; |
427 |
if (!vector_repeats(pmv, i) ) |
if (!vector_repeats(pmv, i) ) |
428 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
429 |
|
|
430 |
|
if (*Data->iMinSAD > 512) { |
431 |
|
unsigned int mask = make_mask(pmv, 7, Data->dir); |
432 |
|
|
433 |
|
MainSearchFunc *MainSearchPtr; |
434 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
435 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
436 |
else MainSearchPtr = xvid_me_DiamondSearch; |
else MainSearchPtr = xvid_me_DiamondSearch; |
437 |
|
|
|
if (*Data->iMinSAD > 512) { |
|
|
unsigned int mask = make_mask(pmv, 7, *Data->dir); |
|
438 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
439 |
} |
} |
440 |
|
|
446 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
447 |
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, |
448 |
pParam->width, pParam->height, iFcode, 2, 0); |
pParam->width, pParam->height, iFcode, 2, 0); |
449 |
|
|
450 |
|
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) { |
451 |
|
if(MotionFlags & XVID_ME_FASTREFINE16) |
452 |
|
SubpelRefine_Fast(Data, CheckCandidate16no4v_qpel); |
453 |
|
else |
454 |
xvid_me_SubpelRefine(Data, CheckCandidate16no4v); |
xvid_me_SubpelRefine(Data, CheckCandidate16no4v); |
455 |
} |
} |
456 |
|
} |
457 |
|
|
458 |
/* three bits are needed to code backward mode. four for forward */ |
/* three bits are needed to code backward mode. four for forward */ |
459 |
|
|
490 |
const uint32_t x, const uint32_t y, |
const uint32_t x, const uint32_t y, |
491 |
const SearchData * const Data) |
const SearchData * const Data) |
492 |
{ |
{ |
493 |
|
int k; |
494 |
|
|
495 |
|
if (!Data->chroma) { |
496 |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
497 |
int32_t sum; |
int32_t sum; |
|
int k; |
|
498 |
const uint32_t stride = Data->iEdgedWidth/2; |
const uint32_t stride = Data->iEdgedWidth/2; |
499 |
/* 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 */ |
500 |
|
|
522 |
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, |
523 |
stride); |
stride); |
524 |
|
|
525 |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) { |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ |
526 |
|
} |
527 |
|
|
528 |
|
/* skip */ |
529 |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
530 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
531 |
pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; |
pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; |
532 |
pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; |
pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; |
533 |
} |
} |
534 |
} |
} |
|
} |
|
535 |
|
|
536 |
static uint32_t |
static uint32_t |
537 |
SearchDirect(const IMAGE * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
611 |
/* initial (fast) skip decision */ |
/* initial (fast) skip decision */ |
612 |
if (*Data->iMinSAD < (int)Data->iQuant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
if (*Data->iMinSAD < (int)Data->iQuant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
613 |
/* possible skip */ |
/* possible skip */ |
|
if (Data->chroma) { |
|
|
pMB->mode = MODE_DIRECT_NONE_MV; |
|
|
return *Data->iMinSAD; /* skip. */ |
|
|
} else { |
|
614 |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
615 |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skip. */ |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; /* skipped */ |
|
} |
|
616 |
} |
} |
617 |
|
|
618 |
*Data->iMinSAD += Data->lambda16; |
*Data->iMinSAD += Data->lambda16; |
619 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
620 |
|
|
621 |
/* |
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 |
|
|
*/ |
|
|
|
|
622 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = xvid_me_SquareSearch; |
623 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = xvid_me_AdvDiamondSearch; |
624 |
else MainSearchPtr = xvid_me_DiamondSearch; |
else MainSearchPtr = xvid_me_DiamondSearch; |
626 |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
627 |
|
|
628 |
xvid_me_SubpelRefine(Data, CheckCandidate); |
xvid_me_SubpelRefine(Data, CheckCandidate); |
629 |
|
} |
630 |
|
|
631 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
632 |
|
|
662 |
return skip_sad; |
return skip_sad; |
663 |
} |
} |
664 |
|
|
665 |
|
|
666 |
|
static void set_range(int * range, SearchData * Data) |
667 |
|
{ |
668 |
|
Data->min_dx = range[0]; |
669 |
|
Data->max_dx = range[1]; |
670 |
|
Data->min_dy = range[2]; |
671 |
|
Data->max_dy = range[3]; |
672 |
|
} |
673 |
|
|
674 |
|
static void |
675 |
|
SubpelRefine_dir(SearchData * const data, CheckFunc * const CheckCandidate, const int dir) |
676 |
|
{ |
677 |
|
/* Do a half-pel or q-pel refinement */ |
678 |
|
const VECTOR centerMV = data->qpel_precision ? |
679 |
|
data->currentQMV[dir-1] : data->currentMV[dir-1]; |
680 |
|
|
681 |
|
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, dir); |
682 |
|
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, dir); |
683 |
|
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, dir); |
684 |
|
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, dir); |
685 |
|
CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, dir); |
686 |
|
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, dir); |
687 |
|
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, dir); |
688 |
|
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, dir); |
689 |
|
} |
690 |
|
|
691 |
static void |
static void |
692 |
SearchInterpolate(const IMAGE * const f_Ref, |
SearchInterpolate(const IMAGE * const f_Ref, |
693 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
706 |
const VECTOR * const b_predMV, |
const VECTOR * const b_predMV, |
707 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
708 |
int32_t * const best_sad, |
int32_t * const best_sad, |
709 |
SearchData * const fData) |
SearchData * const Data) |
710 |
|
|
711 |
{ |
{ |
712 |
int i, j; |
int i, j; |
713 |
SearchData bData; |
int b_range[4], f_range[4]; |
714 |
|
int threshA = (MotionFlags & XVID_ME_FAST_MODEINTERPOLATE) ? 0 : 500; |
715 |
|
int threshB = (MotionFlags & XVID_ME_FAST_MODEINTERPOLATE) ? 0 : 300; |
716 |
|
|
717 |
|
Data->qpel_precision = 0; |
718 |
|
*Data->iMinSAD = 4096*256; |
719 |
|
Data->iFcode = fcode; Data->bFcode = bcode; |
720 |
|
|
721 |
|
i = (x + y * Data->iEdgedWidth) * 16; |
722 |
|
|
723 |
|
Data->RefP[0] = f_Ref->y + i; |
724 |
|
Data->RefP[2] = f_RefH + i; |
725 |
|
Data->RefP[1] = f_RefV + i; |
726 |
|
Data->RefP[3] = f_RefHV + i; |
727 |
|
Data->b_RefP[0] = b_Ref->y + i; |
728 |
|
Data->b_RefP[2] = b_RefH + i; |
729 |
|
Data->b_RefP[1] = b_RefV + i; |
730 |
|
Data->b_RefP[3] = b_RefHV + i; |
731 |
|
Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
732 |
|
Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
733 |
|
Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
734 |
|
Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
735 |
|
|
736 |
|
Data->predMV = *f_predMV; |
737 |
|
Data->bpredMV = *b_predMV; |
738 |
|
|
739 |
|
Data->currentMV[0] = Data->currentMV[2]; /* forward search left its vector here */ |
740 |
|
|
741 |
|
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); |
742 |
|
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); |
743 |
|
|
744 |
|
if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; |
745 |
|
if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; |
746 |
|
if (Data->currentMV[0].y > f_range[3]) Data->currentMV[0].y = f_range[3]; |
747 |
|
if (Data->currentMV[0].y < f_range[2]) Data->currentMV[0].y = f_range[2]; |
748 |
|
|
749 |
fData->qpel_precision = 0; |
if (Data->currentMV[1].x > b_range[1]) Data->currentMV[1].x = b_range[1]; |
750 |
memcpy(&bData, fData, sizeof(SearchData)); /* quick copy of common data */ |
if (Data->currentMV[1].x < b_range[0]) Data->currentMV[1].x = b_range[0]; |
751 |
*fData->iMinSAD = 4096*256; |
if (Data->currentMV[1].y > b_range[3]) Data->currentMV[1].y = b_range[3]; |
752 |
bData.currentMV++; bData.currentQMV++; |
if (Data->currentMV[1].y < b_range[2]) Data->currentMV[1].y = b_range[2]; |
|
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; |
|
753 |
|
|
754 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255); |
set_range(f_range, Data); |
755 |
|
|
756 |
|
CheckCandidateInt(Data->currentMV[0].x, Data->currentMV[0].y, Data, 1); |
757 |
|
|
758 |
/* diamond */ |
/* diamond */ |
759 |
do { |
do { |
760 |
*fData->dir = 255; |
Data->dir = 0; |
761 |
/* forward MV moves */ |
/* forward MV moves */ |
762 |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
i = Data->currentMV[0].x; j = Data->currentMV[0].y; |
763 |
|
|
764 |
CheckCandidateInt(i + 1, j, fData, 0); |
CheckCandidateInt(i + 1, j, Data, 1); |
765 |
CheckCandidateInt(i, j + 1, fData, 0); |
CheckCandidateInt(i, j + 1, Data, 1); |
766 |
CheckCandidateInt(i - 1, j, fData, 0); |
CheckCandidateInt(i - 1, j, Data, 1); |
767 |
CheckCandidateInt(i, j - 1, fData, 0); |
CheckCandidateInt(i, j - 1, Data, 1); |
768 |
|
|
769 |
/* backward MV moves */ |
/* backward MV moves */ |
770 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
set_range(b_range, Data); |
771 |
fData->currentMV[2] = fData->currentMV[0]; |
i = Data->currentMV[1].x; j = Data->currentMV[1].y; |
772 |
CheckCandidateInt(i + 1, j, &bData, 0); |
|
773 |
CheckCandidateInt(i, j + 1, &bData, 0); |
CheckCandidateInt(i + 1, j, Data, 2); |
774 |
CheckCandidateInt(i - 1, j, &bData, 0); |
CheckCandidateInt(i, j + 1, Data, 2); |
775 |
CheckCandidateInt(i, j - 1, &bData, 0); |
CheckCandidateInt(i - 1, j, Data, 2); |
776 |
|
CheckCandidateInt(i, j - 1, Data, 2); |
777 |
|
|
778 |
} while (!(*fData->dir)); |
set_range(f_range, Data); |
779 |
|
|
780 |
|
} while (Data->dir != 0); |
781 |
|
|
782 |
/* qpel refinement */ |
/* qpel refinement */ |
783 |
if (fData->qpel) { |
if (Data->qpel) { |
784 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*Data->iMinSAD > *best_sad + threshA) return; |
785 |
fData->qpel_precision = bData.qpel_precision = 1; |
Data->qpel_precision = 1; |
786 |
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); |
787 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
|
788 |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
Data->currentQMV[0].x = 2 * Data->currentMV[0].x; |
789 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
Data->currentQMV[0].y = 2 * Data->currentMV[0].y; |
790 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
Data->currentQMV[1].x = 2 * Data->currentMV[1].x; |
791 |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
Data->currentQMV[1].y = 2 * Data->currentMV[1].y; |
792 |
xvid_me_SubpelRefine(fData, CheckCandidateInt); |
SubpelRefine_dir(Data, CheckCandidateInt, 1); |
793 |
if (*fData->iMinSAD > *best_sad + 300) return; |
if (*Data->iMinSAD > *best_sad + threshB) return; |
794 |
fData->currentQMV[2] = fData->currentQMV[0]; |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
795 |
xvid_me_SubpelRefine(&bData, CheckCandidateInt); |
SubpelRefine_dir(Data, CheckCandidateInt, 2); |
796 |
} |
} |
797 |
|
|
798 |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
*Data->iMinSAD += 2 * Data->lambda16; /* two bits are needed to code interpolate mode. */ |
799 |
|
|
800 |
if (*fData->iMinSAD < *best_sad) { |
if (*Data->iMinSAD < *best_sad) { |
801 |
*best_sad = *fData->iMinSAD; |
*best_sad = *Data->iMinSAD; |
802 |
pMB->mvs[0] = fData->currentMV[0]; |
pMB->mvs[0] = Data->currentMV[0]; |
803 |
pMB->b_mvs[0] = fData->currentMV[1]; |
pMB->b_mvs[0] = Data->currentMV[1]; |
804 |
pMB->mode = MODE_INTERPOLATE; |
pMB->mode = MODE_INTERPOLATE; |
805 |
if (fData->qpel) { |
if (Data->qpel) { |
806 |
pMB->qmvs[0] = fData->currentQMV[0]; |
pMB->qmvs[0] = Data->currentQMV[0]; |
807 |
pMB->b_qmvs[0] = fData->currentQMV[1]; |
pMB->b_qmvs[0] = Data->currentQMV[1]; |
808 |
pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
809 |
pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; |
pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; |
810 |
pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; |
pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; |
841 |
uint32_t skip_sad; |
uint32_t skip_sad; |
842 |
|
|
843 |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
844 |
|
MACROBLOCK *const pMBs = frame->mbs; |
845 |
|
|
846 |
VECTOR f_predMV, b_predMV; |
VECTOR f_predMV, b_predMV; |
847 |
|
|
851 |
/* some pre-inintialized data for the rest of the search */ |
/* some pre-inintialized data for the rest of the search */ |
852 |
|
|
853 |
SearchData Data; |
SearchData Data; |
|
int32_t iMinSAD; |
|
|
uint32_t dir; |
|
|
VECTOR currentMV[3]; |
|
|
VECTOR currentQMV[3]; |
|
|
int32_t temp[8]; |
|
854 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
855 |
|
|
856 |
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)]; |
|
857 |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
858 |
Data.rounding = 0; |
Data.rounding = 0; |
859 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
|
Data.temp = temp; |
|
|
Data.dir = &dir; |
|
860 |
Data.iQuant = frame->quant; |
Data.iQuant = frame->quant; |
861 |
|
|
862 |
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) */ |
869 |
for (i = 0; i < pParam->mb_width; i++) { |
for (i = 0; i < pParam->mb_width; i++) { |
870 |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
871 |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
872 |
|
int interpol_search; |
873 |
|
|
874 |
/* 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 */ |
875 |
if (b_reference->coding_type != S_VOP) |
if (b_reference->coding_type != S_VOP) |
876 |
if (b_mb->mode == MODE_NOT_CODED) { |
if (b_mb->mode == MODE_NOT_CODED) { |
877 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
878 |
|
pMB->mvs[0] = pMB->b_mvs[0] = zeroMV; |
879 |
|
pMB->sad16 = 0; |
880 |
continue; |
continue; |
881 |
} |
} |
882 |
|
|
883 |
|
Data.lambda16 = xvid_me_lambda_vec16[b_mb->quant]; |
884 |
|
|
885 |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
886 |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
887 |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
899 |
&best_sad, |
&best_sad, |
900 |
&Data); |
&Data); |
901 |
|
|
902 |
if (pMB->mode == MODE_DIRECT_NONE_MV) continue; |
if (pMB->mode == MODE_DIRECT_NONE_MV) { |
903 |
|
pMB->sad16 = best_sad; |
904 |
|
continue; |
905 |
|
} |
906 |
|
|
907 |
|
if (frame->motion_flags & XVID_ME_BFRAME_EARLYSTOP) { |
908 |
|
int bf_search = 0; |
909 |
|
int bf_thresh = 0; |
910 |
|
|
911 |
|
if(i > 0 && j > 0 && i < pParam->mb_width) { |
912 |
|
bf_thresh = ((&pMBs[(i-1) + j * pParam->mb_width])->sad16 + |
913 |
|
(&pMBs[i + (j-1) * pParam->mb_width])->sad16 + |
914 |
|
(&pMBs[(i+1) + (j-1) * pParam->mb_width])->sad16) / 3; |
915 |
|
|
916 |
|
if (((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_FORWARD) && |
917 |
|
((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_BACKWARD) && |
918 |
|
((&pMBs[(i-1) + j * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
919 |
|
bf_search++; |
920 |
|
|
921 |
|
if (((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_FORWARD) && |
922 |
|
((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_BACKWARD) && |
923 |
|
((&pMBs[i + (j - 1) * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
924 |
|
bf_search++; |
925 |
|
|
926 |
|
if (((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_FORWARD) && |
927 |
|
((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_BACKWARD) && |
928 |
|
((&pMBs[(i + 1) + (j - 1) * pParam->mb_width])->mode != MODE_INTERPOLATE)) |
929 |
|
bf_search++; |
930 |
|
} |
931 |
|
|
932 |
|
if ((best_sad < bf_thresh) && (bf_search == 3)) |
933 |
|
continue; |
934 |
|
} |
935 |
|
|
936 |
/* forward search */ |
/* forward search */ |
937 |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
950 |
MODE_BACKWARD, &Data); |
MODE_BACKWARD, &Data); |
951 |
|
|
952 |
/* 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 */ |
953 |
|
if (frame->motion_flags & XVID_ME_FAST_MODEINTERPOLATE) |
954 |
|
interpol_search = (best_sad > Data.iQuant * 3 * MAX_SAD00_FOR_SKIP * (Data.chroma ? 3:2)); |
955 |
|
else |
956 |
|
interpol_search = 1; |
957 |
|
|
958 |
|
if (interpol_search) { |
959 |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
960 |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
961 |
i, j, |
i, j, |
965 |
&f_predMV, &b_predMV, |
&f_predMV, &b_predMV, |
966 |
pMB, &best_sad, |
pMB, &best_sad, |
967 |
&Data); |
&Data); |
968 |
|
} |
969 |
|
|
970 |
/* final skip decision */ |
/* final skip decision */ |
971 |
if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * 2) |
if ( (skip_sad < Data.iQuant * MAX_SAD00_FOR_SKIP * (Data.chroma ? 3:2) ) |
972 |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
973 |
|
|
974 |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
SkipDecisionB(&frame->image, f_ref, b_ref, pMB, i, j, &Data); |
975 |
|
|
976 |
switch (pMB->mode) { |
switch (pMB->mode) { |
977 |
case MODE_FORWARD: |
case MODE_FORWARD: |
978 |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
979 |
|
pMB->sad16 = best_sad; |
980 |
break; |
break; |
981 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
982 |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
983 |
|
pMB->sad16 = best_sad; |
984 |
break; |
break; |
985 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
986 |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
987 |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
988 |
|
pMB->sad16 = best_sad; |
989 |
break; |
break; |
990 |
default: |
default: |
991 |
|
pMB->sad16 = best_sad; |
992 |
break; |
break; |
993 |
} |
} |
994 |
} |
} |