31 |
#include <assert.h> |
#include <assert.h> |
32 |
#include <stdio.h> |
#include <stdio.h> |
33 |
#include <stdlib.h> |
#include <stdlib.h> |
34 |
|
#include <string.h> // memcpy |
35 |
|
|
36 |
#include "../encoder.h" |
#include "../encoder.h" |
37 |
#include "../utils/mbfunctions.h" |
#include "../utils/mbfunctions.h" |
48 |
#define FINAL_SKIP_THRESH (50) |
#define FINAL_SKIP_THRESH (50) |
49 |
#define MAX_SAD00_FOR_SKIP (20) |
#define MAX_SAD00_FOR_SKIP (20) |
50 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
|
#define SKIP_THRESH_B (15) |
|
51 |
|
|
52 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
53 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
54 |
|
|
|
#define iDiamondSize 2 |
|
|
|
|
55 |
static __inline int |
static __inline int |
56 |
d_mv_bits(int x, int y, const uint32_t iFcode) |
d_mv_bits(int x, int y, const uint32_t iFcode, const int qpel, const int rrv) |
57 |
{ |
{ |
58 |
int xb, yb; |
int xb, yb; |
59 |
|
if (qpel) { x *= 2; y *= 2;} |
60 |
|
else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
61 |
|
|
62 |
if (x == 0) xb = 1; |
if (x == 0) xb = 1; |
63 |
else { |
else { |
86 |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
dx = (dx >> 1) + roundtab_79[dx & 0x3]; |
87 |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
dy = (dy >> 1) + roundtab_79[dy & 0x3]; |
88 |
|
|
89 |
switch (((dx & 1) << 1) + (dy & 1)) { // ((dx%2)?2:0)+((dy%2)?1:0) |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently |
90 |
|
|
91 |
|
switch (((dx & 1) << 1) | (dy & 1)) { |
92 |
case 0: |
case 0: |
93 |
sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
sad = sad8(data->CurU, data->RefCU + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
94 |
sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
sad += sad8(data->CurV, data->RefCV + (dy/2) * (data->iEdgedWidth/2) + dx/2, data->iEdgedWidth/2); |
115 |
sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2); |
sad += sad8(data->CurV, data->RefQ, data->iEdgedWidth/2); |
116 |
break; |
break; |
117 |
} |
} |
118 |
|
data->temp[5] = dx; data->temp[6] = dy; data->temp[7] = sad; //backup |
119 |
return sad; |
return sad; |
120 |
} |
} |
121 |
|
|
132 |
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
133 |
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
134 |
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
|
|
135 |
} |
} |
136 |
} |
} |
137 |
|
|
239 |
|
|
240 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
241 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
242 |
xc = x/2; yc = y/2; //for chroma sad |
xc = x/2; yc = y/2; //for chroma sad |
243 |
current = data->currentQMV; |
current = data->currentQMV; |
244 |
} else { |
} else { |
245 |
switch ( ((x&1)<<1) + (y&1) ) { |
Reference = GetReference(x, y, 0, data); |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
|
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
246 |
current = data->currentMV; |
current = data->currentMV; |
247 |
xc = x; yc = y; |
xc = x; yc = y; |
248 |
} |
} |
249 |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0); |
250 |
|
|
251 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
252 |
|
|
272 |
} |
} |
273 |
|
|
274 |
static void |
static void |
275 |
|
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
276 |
|
{ |
277 |
|
int t; |
278 |
|
const uint8_t * Reference; |
279 |
|
|
280 |
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value |
281 |
|
( x > data->max_dx) || ( x < data->min_dx) |
282 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
283 |
|
|
284 |
|
Reference = GetReference(x, y, 0, data); |
285 |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, 0, 1); |
286 |
|
|
287 |
|
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
288 |
|
|
289 |
|
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
290 |
|
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
291 |
|
|
292 |
|
if (data->temp[0] < data->iMinSAD[0]) { |
293 |
|
data->iMinSAD[0] = data->temp[0]; |
294 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
295 |
|
*dir = Direction; } |
296 |
|
|
297 |
|
if (data->temp[1] < data->iMinSAD[1]) { |
298 |
|
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
299 |
|
if (data->temp[2] < data->iMinSAD[2]) { |
300 |
|
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
301 |
|
if (data->temp[3] < data->iMinSAD[3]) { |
302 |
|
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
303 |
|
if (data->temp[4] < data->iMinSAD[4]) { |
304 |
|
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
305 |
|
} |
306 |
|
|
307 |
|
static void |
308 |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
309 |
{ |
{ |
310 |
int32_t sad; |
int32_t sad; |
315 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
316 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
317 |
|
|
318 |
|
if (data->rrv) |
319 |
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) ) return; //non-zero integer value |
320 |
|
|
321 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
322 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
323 |
current = data->currentQMV; |
current = data->currentQMV; |
324 |
} else { |
} else { |
325 |
switch ( ((x&1)<<1) + (y&1) ) { |
Reference = GetReference(x, y, 0, data); |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
|
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
326 |
current = data->currentMV; |
current = data->currentMV; |
327 |
} |
} |
328 |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, |
329 |
|
data->qpel && !data->qpel_precision && !data->rrv, data->rrv); |
330 |
|
|
331 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
332 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)/1000; |
372 |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
373 |
current = data->currentQMV; |
current = data->currentQMV; |
374 |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
|
375 |
} else { |
} else { |
376 |
ReferenceF = Interpolate16x16qpel(2*xf, 2*yf, 0, data); |
ReferenceF = GetReference(xf, yf, 0, data); |
377 |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
378 |
ReferenceB = Interpolate16x16qpel(2*xb, 2*yb, 1, data); |
ReferenceB = GetReference(xb, yb, 1, data); |
379 |
current = data->currentMV; |
current = data->currentMV; |
|
if (data->qpel) |
|
|
t = d_mv_bits(2*xf - data->predMV.x, 2*yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(2*xb - data->bpredMV.x, 2*yb - data->bpredMV.y, data->iFcode); |
|
|
else |
|
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
|
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
|
380 |
} |
} |
381 |
|
|
382 |
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0) |
383 |
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0); |
384 |
|
|
385 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
386 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)/1000; |
387 |
|
|
431 |
if (sad > *(data->iMinSAD)) return; |
if (sad > *(data->iMinSAD)) return; |
432 |
} |
} |
433 |
|
|
434 |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, 1, 0, 0) * sad)/1000; |
435 |
|
|
436 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
437 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
472 |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
473 |
|
|
474 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
475 |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, 1, 0, 0) * sad)/1000; |
476 |
|
|
477 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
478 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
490 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
491 |
|
|
492 |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
493 |
else Reference = Interpolate16x16qpel(2*x, 2*y, 0, data); |
else Reference = GetReference(x, y, 0, data); |
494 |
|
|
495 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
496 |
if (data->qpel) t = d_mv_bits(2 * x - data->predMV.x, 2 * y - data->predMV.y, data->iFcode); |
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode, data->qpel && !data->qpel_precision, 0); |
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
497 |
|
|
498 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
499 |
|
|
515 |
|
|
516 |
int iDirection; |
int iDirection; |
517 |
|
|
518 |
do { |
for(;;) { //forever |
519 |
iDirection = 0; |
iDirection = 0; |
520 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
521 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
533 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
534 |
} else { // what remains here is up or down |
} else { // what remains here is up or down |
535 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
536 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); } |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
537 |
|
} |
538 |
|
|
539 |
if (iDirection) { |
if (iDirection) { |
540 |
bDirection += iDirection; |
bDirection += iDirection; |
541 |
x = data->currentMV->x; y = data->currentMV->y; } |
x = data->currentMV->x; y = data->currentMV->y; |
542 |
|
} |
543 |
} else { //about to quit, eh? not so fast.... |
} else { //about to quit, eh? not so fast.... |
544 |
switch (bDirection) { |
switch (bDirection) { |
545 |
case 2: |
case 2: |
590 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
591 |
} |
} |
592 |
} |
} |
|
while (1); //forever |
|
593 |
} |
} |
594 |
|
|
595 |
static void |
static void |
639 |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
640 |
} else { // what remains here is up or down |
} else { // what remains here is up or down |
641 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
642 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); } |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
643 |
|
} |
644 |
bDirection += iDirection; |
bDirection += iDirection; |
645 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
646 |
} |
} |
678 |
static __inline int |
static __inline int |
679 |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
680 |
const int x, const int y, |
const int x, const int y, |
681 |
const uint32_t iEdgedWidth, const uint32_t iQuant) |
const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv) |
682 |
|
|
683 |
{ |
{ |
684 |
/* keep repeating checks for all b-frames before this P frame, |
/* keep repeating checks for all b-frames before this P frame, |
685 |
to make sure that SKIP is possible (todo) |
to make sure that SKIP is possible (todo) |
686 |
how: if skip is not possible set sad00 to a very high value */ |
how: if skip is not possible set sad00 to a very high value */ |
687 |
|
if(rrv) { |
688 |
|
uint32_t sadC = sad16(current->u + x*16 + y*(iEdgedWidth/2)*16, |
689 |
|
reference->u + x*16 + y*(iEdgedWidth/2)*16, iEdgedWidth/2, 256*4096); |
690 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
691 |
|
sadC += sad16(current->v + (x + y*(iEdgedWidth/2))*16, |
692 |
|
reference->v + (x + y*(iEdgedWidth/2))*16, iEdgedWidth/2, 256*4096); |
693 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
694 |
|
return 1; |
695 |
|
} else { |
696 |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
697 |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
698 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
699 |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
700 |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
701 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
|
|
|
702 |
return 1; |
return 1; |
703 |
} |
} |
704 |
|
} |
705 |
|
|
706 |
static __inline void |
static __inline void |
707 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
731 |
|
|
732 |
const VECTOR zeroMV = { 0, 0 }; |
const VECTOR zeroMV = { 0, 0 }; |
733 |
|
|
734 |
|
uint32_t mb_width = pParam->mb_width; |
735 |
|
uint32_t mb_height = pParam->mb_height; |
736 |
|
|
737 |
uint32_t x, y; |
uint32_t x, y; |
738 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
739 |
int32_t InterBias, quant = current->quant, sad00; |
int32_t InterBias, quant = current->quant, sad00; |
740 |
uint8_t *qimage; |
uint8_t *qimage; |
741 |
|
|
742 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
743 |
int32_t temp[5]; |
int32_t temp[8]; |
744 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
745 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
746 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
747 |
SearchData Data; |
SearchData Data; |
748 |
|
memset(&Data, 0, sizeof(SearchData)); |
749 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
750 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
751 |
Data.currentQMV = currentQMV; |
Data.currentQMV = currentQMV; |
755 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
756 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
757 |
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
758 |
|
Data.rrv = current->global_flags & XVID_REDUCED; |
759 |
|
|
760 |
|
if ((current->global_flags & XVID_REDUCED)) { |
761 |
|
mb_width = (pParam->width + 31) / 32; |
762 |
|
mb_height = (pParam->height + 31) / 32; |
763 |
|
Data.qpel = Data.chroma = 0; |
764 |
|
} |
765 |
|
|
766 |
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
767 |
return 1; // allocate some mem for qpel interpolated blocks |
return 1; // allocate some mem for qpel interpolated blocks |
770 |
Data.RefQ = qimage; |
Data.RefQ = qimage; |
771 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
772 |
|
|
773 |
for (y = 0; y < pParam->mb_height; y++) { |
for (y = 0; y < mb_height; y++) { |
774 |
for (x = 0; x < pParam->mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
775 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
776 |
|
|
777 |
pMB->sad16 |
if (Data.rrv) pMB->sad16 = |
778 |
= sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, |
sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32, |
779 |
|
pRef->y + (x + y * pParam->edged_width) * 32, |
780 |
|
pParam->edged_width, pMB->sad8 ); |
781 |
|
|
782 |
|
else pMB->sad16 = |
783 |
|
sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, |
784 |
pRef->y + (x + y * pParam->edged_width) * 16, |
pRef->y + (x + y * pParam->edged_width) * 16, |
785 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
786 |
|
|
809 |
//initial skip decision |
//initial skip decision |
810 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
811 |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
812 |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH) |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
813 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
814 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
815 |
continue; |
continue; |
816 |
} |
} |
824 |
/* 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?" */ |
825 |
if (current->coding_type == P_VOP) { |
if (current->coding_type == P_VOP) { |
826 |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
827 |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) ) |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1)) ) |
828 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
829 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
830 |
continue; |
continue; |
831 |
} |
} |
841 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
842 |
|
|
843 |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
844 |
|
if (Data.rrv) InterBias *= 4; //?? |
845 |
|
|
846 |
if (InterBias < pMB->sad16) { |
if (InterBias < pMB->sad16) { |
847 |
const int32_t deviation = |
int32_t deviation; |
848 |
dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
if (Data.rrv) { |
849 |
|
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32, |
850 |
|
pParam->edged_width) |
851 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16, |
852 |
|
pParam->edged_width) |
853 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width, |
854 |
|
pParam->edged_width) |
855 |
|
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1), |
856 |
|
pParam->edged_width); |
857 |
|
} else |
858 |
|
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
859 |
pParam->edged_width); |
pParam->edged_width); |
860 |
|
|
861 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
903 |
} |
} |
904 |
|
|
905 |
static __inline void |
static __inline void |
906 |
PreparePredictionsP(VECTOR * const pmv, int x, int y, const int iWcount, |
PreparePredictionsP(VECTOR * const pmv, int x, int y, int iWcount, |
907 |
const int iHcount, const MACROBLOCK * const prevMB) |
int iHcount, const MACROBLOCK * const prevMB, int rrv) |
908 |
{ |
{ |
909 |
|
|
910 |
//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
//this function depends on get_pmvdata which means that it sucks. It should get the predictions by itself |
911 |
|
if (rrv) { iWcount /= 2; iHcount /= 2; } |
912 |
|
|
913 |
if ( (y != 0) && (x != (iWcount-1)) ) { // [5] top-right neighbour |
if ( (y != 0) && (x < (iWcount-1)) ) { // [5] top-right neighbour |
914 |
pmv[5].x = EVEN(pmv[3].x); |
pmv[5].x = EVEN(pmv[3].x); |
915 |
pmv[5].y = EVEN(pmv[3].y); |
pmv[5].y = EVEN(pmv[3].y); |
916 |
} else pmv[5].x = pmv[5].y = 0; |
} else pmv[5].x = pmv[5].y = 0; |
922 |
else pmv[4].x = pmv[4].y = 0; |
else pmv[4].x = pmv[4].y = 0; |
923 |
|
|
924 |
// [1] median prediction |
// [1] median prediction |
925 |
pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); |
if (rrv) { //median is in halfzero-precision |
926 |
|
pmv[1].x = RRV_MV_SCALEUP(pmv[0].x); |
927 |
|
pmv[1].y = RRV_MV_SCALEUP(pmv[0].y); |
928 |
|
} else { pmv[1].x = EVEN(pmv[0].x); pmv[1].y = EVEN(pmv[0].y); } |
929 |
|
|
930 |
pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask |
pmv[0].x = pmv[0].y = 0; // [0] is zero; not used in the loop (checked before) but needed here for make_mask |
931 |
|
|
932 |
pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
pmv[2].x = EVEN(prevMB->mvs[0].x); // [2] is last frame |
933 |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
pmv[2].y = EVEN(prevMB->mvs[0].y); |
934 |
|
|
935 |
if ((x != iWcount-1) && (y != iHcount-1)) { |
if ((x < iWcount-1) && (y < iHcount-1)) { |
936 |
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 |
937 |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
pmv[6].y = EVEN((prevMB+1+iWcount)->mvs[0].y); |
938 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
939 |
|
|
940 |
|
if (rrv) { |
941 |
|
int i; |
942 |
|
for (i = 0; i < 7; i++) { |
943 |
|
pmv[i].x = RRV_MV_SCALEDOWN(pmv[i].x); |
944 |
|
pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // a trick |
945 |
|
} |
946 |
|
} |
947 |
} |
} |
948 |
|
|
949 |
static void |
static void |
967 |
int i, iDirection = 255, mask, threshA; |
int i, iDirection = 255, mask, threshA; |
968 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
969 |
|
|
|
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
|
970 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
971 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
972 |
|
|
973 |
|
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
974 |
|
|
975 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache |
976 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
if (Data->rrv) i = 2; else i = 1; |
977 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
978 |
|
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
979 |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
980 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
|
981 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16*i; |
982 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16*i; |
983 |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16*i; |
984 |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16*i; |
985 |
|
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
986 |
|
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
987 |
|
|
988 |
Data->lambda16 = lambda_vec16[iQuant]; |
Data->lambda16 = lambda_vec16[iQuant]; |
989 |
Data->lambda8 = lambda_vec8[iQuant]; |
Data->lambda8 = lambda_vec8[iQuant]; |
990 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
991 |
|
|
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
|
|
Data->min_dx = EVEN(Data->min_dx); |
|
|
Data->max_dx = EVEN(Data->max_dx); |
|
|
Data->min_dy = EVEN(Data->min_dy); |
|
|
Data->max_dy = EVEN(Data->max_dy); } |
|
|
|
|
992 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
993 |
|
|
994 |
for(i = 0; i < 5; i++) |
for(i = 0; i < 5; i++) |
997 |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
998 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
999 |
|
|
1000 |
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode, 0, 0); |
1001 |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
1002 |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
1003 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
1011 |
if (threshA > 1024) threshA = 1024; } |
if (threshA > 1024) threshA = 1024; } |
1012 |
|
|
1013 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
1014 |
prevMBs + x + y * pParam->mb_width); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
1015 |
|
|
1016 |
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
if (Data->rrv) CheckCandidate = CheckCandidate32; |
1017 |
|
else if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
1018 |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
1019 |
|
|
1020 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
1045 |
if (MotionFlags & PMV_EXTSEARCH16) { |
if (MotionFlags & PMV_EXTSEARCH16) { |
1046 |
int32_t bSAD; |
int32_t bSAD; |
1047 |
VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; |
VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0]; |
1048 |
|
if (Data->rrv) { |
1049 |
|
startMV.x = RRV_MV_SCALEUP(startMV.x); |
1050 |
|
startMV.y = RRV_MV_SCALEUP(startMV.y); |
1051 |
|
} else |
1052 |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
1053 |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
1054 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1062 |
} |
} |
1063 |
|
|
1064 |
backupMV = Data->currentMV[0]; |
backupMV = Data->currentMV[0]; |
1065 |
if (MotionFlags & PMV_HALFPELREFINE16) startMV.x = startMV.y = 1; |
if (!MotionFlags & PMV_HALFPELREFINE16 || Data->rrv) startMV.x = startMV.y = 0; |
1066 |
else startMV.x = startMV.y = 0; |
else startMV.x = startMV.y = 1; |
1067 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1068 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1069 |
|
|
1083 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1084 |
} |
} |
1085 |
|
|
1086 |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
if((!Data->rrv) && (pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1087 |
|
|
1088 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1089 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1090 |
pParam->width, pParam->height, Data->iFcode, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1091 |
|
|
1092 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1093 |
} |
} |
1100 |
Data8.iEdgedWidth = Data->iEdgedWidth; |
Data8.iEdgedWidth = Data->iEdgedWidth; |
1101 |
Data8.RefQ = Data->RefQ; |
Data8.RefQ = Data->RefQ; |
1102 |
Data8.qpel = Data->qpel; |
Data8.qpel = Data->qpel; |
1103 |
|
Data8.rrv = Data->rrv; |
1104 |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1105 |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
1106 |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
1123 |
} |
} |
1124 |
} |
} |
1125 |
|
|
1126 |
|
if (Data->rrv) { |
1127 |
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
1128 |
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1129 |
|
} |
1130 |
if (!(inter4v) || |
if (!(inter4v) || |
1131 |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1132 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1165 |
const int block, |
const int block, |
1166 |
SearchData * const Data) |
SearchData * const Data) |
1167 |
{ |
{ |
1168 |
|
int i = 0; |
1169 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1170 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1171 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
1172 |
|
|
1173 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1174 |
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); |
1175 |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
if (block != 0) i = d_mv_bits( Data->currentQMV->x - Data->predMV.x, |
1176 |
d_mv_bits( Data->currentQMV->x - Data->predMV.x, |
Data->currentQMV->y - Data->predMV.y, Data->iFcode, 0, 0); |
1177 |
Data->currentQMV->y - Data->predMV.y, |
|
|
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
|
1178 |
} else { |
} else { |
1179 |
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); |
1180 |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
if (block != 0) { |
1181 |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
if (block != 0) i = d_mv_bits( Data->currentMV->x - Data->predMV.x, |
1182 |
Data->currentMV->y - Data->predMV.y, |
Data->currentMV->y - Data->predMV.y, Data->iFcode, 0, Data->rrv); |
|
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
|
1183 |
} |
} |
1184 |
|
} |
1185 |
|
|
1186 |
|
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
1187 |
|
|
1188 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
1189 |
|
if (Data->rrv) i = 2; else i = 1; |
1190 |
|
|
1191 |
Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Ref = OldData->Ref + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1192 |
Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefH = OldData->RefH + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1193 |
Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefV = OldData->RefV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1194 |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefHV = OldData->RefHV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1195 |
|
|
1196 |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
1197 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1198 |
|
|
1199 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1200 |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv); |
1201 |
CheckCandidate = CheckCandidate8; |
|
1202 |
|
if (Data->rrv) CheckCandidate = CheckCandidate16no4v; |
1203 |
|
else CheckCandidate = CheckCandidate8; |
1204 |
|
|
1205 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1206 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1229 |
} |
} |
1230 |
} |
} |
1231 |
|
|
1232 |
if(pParam->m_quarterpel) { |
if(!Data->rrv && Data->qpel) { |
1233 |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
1234 |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
1235 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1236 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1237 |
pParam->width, pParam->height, OldData->iFcode, 0); |
pParam->width, pParam->height, OldData->iFcode, 1, 0); |
1238 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1239 |
} |
} |
1240 |
} |
} |
1241 |
} |
} |
1242 |
|
|
1243 |
if(pParam->m_quarterpel) { |
if (Data->rrv) { |
1244 |
|
Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x); |
1245 |
|
Data->currentMV->y = RRV_MV_SCALEDOWN(Data->currentMV->y); |
1246 |
|
} |
1247 |
|
|
1248 |
|
if(Data->qpel) { |
1249 |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
1250 |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
1251 |
pMB->qmvs[block] = *(Data->currentQMV); |
pMB->qmvs[block] = *(Data->currentQMV); |
1252 |
} |
} else { |
|
else { |
|
1253 |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1254 |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1255 |
} |
} |
1342 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1343 |
|
|
1344 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1345 |
pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, iFcode - Data->qpel, 0, 0); |
1346 |
|
|
1347 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1348 |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
1372 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
1373 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1374 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1375 |
pParam->width, pParam->height, iFcode, 0); |
pParam->width, pParam->height, iFcode, 1, 0); |
1376 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1377 |
} |
} |
1378 |
|
|
1404 |
|
|
1405 |
} |
} |
1406 |
|
|
1407 |
static int32_t |
static void |
1408 |
|
SkipDecisionB(const IMAGE * const pCur, |
1409 |
|
const IMAGE * const f_Ref, |
1410 |
|
const IMAGE * const b_Ref, |
1411 |
|
MACROBLOCK * const pMB, |
1412 |
|
const uint32_t quant, |
1413 |
|
const uint32_t x, const uint32_t y, |
1414 |
|
const SearchData * const Data) |
1415 |
|
{ |
1416 |
|
int dx, dy, b_dx, b_dy; |
1417 |
|
uint32_t sum; |
1418 |
|
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1419 |
|
if (Data->qpel) { |
1420 |
|
dy = Data->directmvF[0].y/2 + Data->directmvF[1].y/2 + |
1421 |
|
Data->directmvF[2].y/2 + Data->directmvF[3].y/2; |
1422 |
|
|
1423 |
|
dx = Data->directmvF[0].x/2 + Data->directmvF[1].x/2 + |
1424 |
|
Data->directmvF[2].x/2 + Data->directmvF[3].x/2; |
1425 |
|
|
1426 |
|
b_dy = Data->directmvB[0].y/2 + Data->directmvB[1].y/2 + |
1427 |
|
Data->directmvB[2].y/2 + Data->directmvB[3].y/2; |
1428 |
|
|
1429 |
|
b_dx = Data->directmvB[0].x/2 + Data->directmvB[1].x/2 + |
1430 |
|
Data->directmvB[2].x/2 + Data->directmvB[3].x/2; |
1431 |
|
|
1432 |
|
} else { |
1433 |
|
dy = Data->directmvF[0].y + Data->directmvF[1].y + |
1434 |
|
Data->directmvF[2].y + Data->directmvF[3].y; |
1435 |
|
|
1436 |
|
dx = Data->directmvF[0].x + Data->directmvF[1].x + |
1437 |
|
Data->directmvF[2].x + Data->directmvF[3].x; |
1438 |
|
|
1439 |
|
b_dy = Data->directmvB[0].y + Data->directmvB[1].y + |
1440 |
|
Data->directmvB[2].y + Data->directmvB[3].y; |
1441 |
|
|
1442 |
|
b_dx = Data->directmvB[0].x + Data->directmvB[1].x + |
1443 |
|
Data->directmvB[2].x + Data->directmvB[3].x; |
1444 |
|
} |
1445 |
|
|
1446 |
|
|
1447 |
|
dy = (dy >> 3) + roundtab_76[dy & 0xf]; |
1448 |
|
dx = (dx >> 3) + roundtab_76[dx & 0xf]; |
1449 |
|
b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf]; |
1450 |
|
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
1451 |
|
|
1452 |
|
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1453 |
|
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1454 |
|
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1455 |
|
Data->iEdgedWidth/2); |
1456 |
|
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
1457 |
|
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1458 |
|
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1459 |
|
Data->iEdgedWidth/2); |
1460 |
|
|
1461 |
|
if (sum < 2*MAX_CHROMA_SAD_FOR_SKIP * quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
1462 |
|
} |
1463 |
|
|
1464 |
|
|
1465 |
|
|
1466 |
|
static __inline uint32_t |
1467 |
SearchDirect(const IMAGE * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
1468 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1469 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1524 |
*best_sad = 256*4096; // in that case, we won't use direct mode |
*best_sad = 256*4096; // in that case, we won't use direct mode |
1525 |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
1526 |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
pMB->b_mvs[0].x = pMB->b_mvs[0].y = 0; |
1527 |
return 0; |
return 256*4096; |
1528 |
} |
} |
1529 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1530 |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0]; |
1541 |
|
|
1542 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1543 |
|
|
1544 |
// skip decision |
// initial (fast) skip decision |
1545 |
if (*Data->iMinSAD < pMB->quant * SKIP_THRESH_B) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH*2) { |
1546 |
//possible skip - checking chroma. everything copied from MC |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data->chroma, Data); //possible skip - checking chroma |
1547 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
|
int sum, dx, dy, b_dx, b_dy; |
|
|
|
|
|
if (Data->qpel) { |
|
|
sum = pMB->mvs[0].y/2 + pMB->mvs[1].y/2 + pMB->mvs[2].y/2 + pMB->mvs[3].y/2; |
|
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
sum = pMB->mvs[0].x/2 + pMB->mvs[1].x/2 + pMB->mvs[2].x/2 + pMB->mvs[3].x/2; |
|
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
sum = pMB->b_mvs[0].y/2 + pMB->b_mvs[1].y/2 + pMB->b_mvs[2].y/2 + pMB->b_mvs[3].y/2; |
|
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
sum = pMB->b_mvs[0].x/2 + pMB->b_mvs[1].x/2 + pMB->b_mvs[2].x/2 + pMB->b_mvs[3].x/2; |
|
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
|
|
|
|
} else { |
|
|
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
|
|
sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
|
|
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
|
|
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
} |
|
|
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
|
|
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
|
|
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
|
|
Data->iEdgedWidth/2); |
|
|
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
|
|
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
|
|
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
|
|
Data->iEdgedWidth/2); |
|
|
|
|
|
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
|
|
pMB->mode = MODE_DIRECT_NONE_MV; |
|
|
return *Data->iMinSAD; |
|
|
} |
|
1548 |
} |
} |
1549 |
|
|
1550 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
1560 |
|
|
1561 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1562 |
|
|
|
*Data->iMinSAD += 1 * Data->lambda16; // one bit is needed to code direct mode |
|
1563 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1564 |
|
|
1565 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1596 |
} |
} |
1597 |
|
|
1598 |
|
|
1599 |
static __inline void |
static void |
1600 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const uint8_t * const f_Ref, |
1601 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1602 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1619 |
|
|
1620 |
{ |
{ |
1621 |
|
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
1622 |
int iDirection, i, j; |
int iDirection, i, j; |
1623 |
SearchData bData; |
SearchData bData; |
1624 |
|
|
1625 |
*(bData.iMinSAD = fData->iMinSAD) = 4096*256; |
fData->qpel_precision = 0; |
1626 |
bData.Cur = fData->Cur; |
memcpy(&bData, fData, sizeof(SearchData)); //quick copy of common data |
1627 |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
*fData->iMinSAD = 4096*256; |
1628 |
bData.currentMV = fData->currentMV + 1; bData.currentQMV = fData->currentQMV + 1; |
bData.currentMV ++; bData.currentQMV ++; |
|
bData.lambda16 = fData->lambda16; |
|
1629 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1630 |
|
|
1631 |
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
i = (x + y * fData->iEdgedWidth) * 16; |
1632 |
bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16; |
bData.bRef = fData->Ref = f_Ref + i; |
1633 |
bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16; |
bData.bRefH = fData->RefH = f_RefH + i; |
1634 |
bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16; |
bData.bRefV = fData->RefV = f_RefV + i; |
1635 |
bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16; |
bData.bRefHV = fData->RefHV = f_RefHV + i; |
1636 |
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
bData.Ref = fData->bRef = b_Ref + i; |
1637 |
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
bData.RefH = fData->bRefH = b_RefH + i; |
1638 |
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
bData.RefV = fData->bRefV = b_RefV + i; |
1639 |
bData.RefQ = fData->RefQ; |
bData.RefHV = fData->bRefHV = b_RefHV + i; |
|
fData->qpel_precision = bData.qpel_precision = 0; |
|
|
bData.rounding = 0; |
|
1640 |
|
|
1641 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1642 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
|
|
|
1643 |
fData->currentMV[0] = fData->currentMV[2]; |
fData->currentMV[0] = fData->currentMV[2]; |
1644 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel); |
|
1645 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0); |
1646 |
|
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0); |
1647 |
|
|
1648 |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1649 |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
1650 |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy; |
1651 |
if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
1652 |
|
|
1653 |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1654 |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx; |
1655 |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
1656 |
if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
1657 |
|
|
1658 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1659 |
|
|
1679 |
|
|
1680 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1681 |
|
|
|
*fData->iMinSAD += 2 * fData->lambda16; // two bits are needed to code interpolate mode. |
|
|
|
|
1682 |
if (fData->qpel) { |
if (fData->qpel) { |
1683 |
|
CheckCandidate = CheckCandidateInt; |
1684 |
fData->qpel_precision = bData.qpel_precision = 1; |
fData->qpel_precision = bData.qpel_precision = 1; |
1685 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 0); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0); |
1686 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 0); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0); |
1687 |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
1688 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
1689 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
1693 |
SubpelRefine(&bData); |
SubpelRefine(&bData); |
1694 |
} |
} |
1695 |
|
|
1696 |
|
*fData->iMinSAD += (2+2) * fData->lambda16; // two bits are needed to code interpolate mode. |
1697 |
|
|
1698 |
if (*fData->iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1699 |
*best_sad = *fData->iMinSAD; |
*best_sad = *fData->iMinSAD; |
1700 |
pMB->mvs[0] = fData->currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1735 |
const IMAGE * const b_refHV) |
const IMAGE * const b_refHV) |
1736 |
{ |
{ |
1737 |
uint32_t i, j; |
uint32_t i, j; |
1738 |
int32_t best_sad, skip_sad; |
int32_t best_sad; |
1739 |
|
uint32_t skip_sad; |
1740 |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
1741 |
static const VECTOR zeroMV={0,0}; |
static const VECTOR zeroMV={0,0}; |
1742 |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
1753 |
int32_t iMinSAD; |
int32_t iMinSAD; |
1754 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
1755 |
VECTOR currentQMV[3]; |
VECTOR currentQMV[3]; |
1756 |
|
memset(&Data, 0, sizeof(SearchData)); |
1757 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1758 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
1759 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
1760 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant]; |
1761 |
|
Data.chroma = frame->quant; |
1762 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
1763 |
Data.rounding = 0; |
Data.rounding = 0; |
1764 |
|
|
1831 |
pMB, &best_sad, |
pMB, &best_sad, |
1832 |
&Data); |
&Data); |
1833 |
|
|
1834 |
|
// final skip decision |
1835 |
|
if ( (skip_sad < frame->quant * MAX_SAD00_FOR_SKIP*2) |
1836 |
|
&& ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) |
1837 |
|
SkipDecisionB(&frame->image, f_ref, b_ref, pMB,frame->quant, i, j, &Data); |
1838 |
|
|
1839 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1840 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1841 |
f_count++; |
f_count++; |
1842 |
if (pParam->m_quarterpel) f_predMV = pMB->qmvs[0]; |
if (Data.qpel) f_predMV = pMB->qmvs[0]; |
1843 |
else f_predMV = pMB->mvs[0]; |
else f_predMV = pMB->mvs[0]; |
1844 |
break; |
break; |
1845 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
1846 |
b_count++; |
b_count++; |
1847 |
if (pParam->m_quarterpel) b_predMV = pMB->b_qmvs[0]; |
if (Data.qpel) b_predMV = pMB->b_qmvs[0]; |
1848 |
else b_predMV = pMB->b_mvs[0]; |
else b_predMV = pMB->b_mvs[0]; |
1849 |
break; |
break; |
1850 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
1851 |
i_count++; |
i_count++; |
1852 |
if (pParam->m_quarterpel) { |
if (Data.qpel) { |
1853 |
f_predMV = pMB->qmvs[0]; |
f_predMV = pMB->qmvs[0]; |
1854 |
b_predMV = pMB->b_qmvs[0]; |
b_predMV = pMB->b_qmvs[0]; |
1855 |
} else { |
} else { |
1860 |
case MODE_DIRECT: |
case MODE_DIRECT: |
1861 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
1862 |
d_count++; |
d_count++; |
|
break; |
|
1863 |
default: |
default: |
1864 |
break; |
break; |
1865 |
} |
} |
1868 |
free(qimage); |
free(qimage); |
1869 |
} |
} |
1870 |
|
|
|
/* Hinted ME starts here */ |
|
|
|
|
|
static void |
|
|
SearchPhinted ( const IMAGE * const pRef, |
|
|
const uint8_t * const pRefH, |
|
|
const uint8_t * const pRefV, |
|
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
int inter4v, |
|
|
MACROBLOCK * const pMB, |
|
|
SearchData * const Data) |
|
|
{ |
|
|
|
|
|
int i, t; |
|
|
MainSearchFunc * MainSearchPtr; |
|
|
|
|
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
|
|
|
|
|
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
|
|
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
|
|
|
Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
|
|
Data->qpel_precision = 0; |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
|
|
Data->min_dx = EVEN(Data->min_dx); |
|
|
Data->max_dx = EVEN(Data->max_dx); |
|
|
Data->min_dy = EVEN(Data->min_dy); |
|
|
Data->max_dy = EVEN(Data->max_dy); |
|
|
} |
|
|
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
|
|
|
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
|
|
|
|
|
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
|
|
|
|
|
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
|
|
else CheckCandidate = CheckCandidate16no4v; |
|
|
|
|
|
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
|
|
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
|
|
if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed |
|
|
if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx; |
|
|
if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy; |
|
|
if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy; |
|
|
|
|
|
(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data); |
|
|
|
|
|
if (pMB->mode == MODE_INTER4V) |
|
|
for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search |
|
|
pMB->mvs[i].x = EVEN(pMB->mvs[i].x); |
|
|
pMB->mvs[i].y = EVEN(pMB->mvs[i].y); |
|
|
if (!(make_mask(pMB->mvs, i))) |
|
|
(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data); |
|
|
} |
|
|
|
|
|
if (MotionFlags & PMV_USESQUARES16) |
|
|
MainSearchPtr = SquareSearch; |
|
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
|
|
MainSearchPtr = AdvDiamondSearch; |
|
|
else MainSearchPtr = DiamondSearch; |
|
|
|
|
|
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
|
|
|
|
|
if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data); |
|
|
|
|
|
for(i = 0; i < 5; i++) { |
|
|
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
|
|
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
|
|
} |
|
|
|
|
|
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
|
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
|
|
pParam->width, pParam->height, Data->iFcode, 0); |
|
|
Data->qpel_precision = 1; |
|
|
SubpelRefine(Data); |
|
|
} |
|
|
|
|
|
if (inter4v) { |
|
|
SearchData Data8; |
|
|
Data8.iFcode = Data->iFcode; |
|
|
Data8.lambda8 = Data->lambda8; |
|
|
Data8.iEdgedWidth = Data->iEdgedWidth; |
|
|
Data8.RefQ = Data->RefQ; |
|
|
Data8.qpel = Data->qpel; |
|
|
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
|
|
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
|
|
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
|
|
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
|
|
|
|
|
if (Data->chroma) { |
|
|
int sumx, sumy, dx, dy; |
|
|
|
|
|
if(pParam->m_quarterpel) { |
|
|
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
|
|
sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
|
|
} else { |
|
|
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
} |
|
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
|
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
|
|
|
|
|
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
|
|
} |
|
|
} |
|
|
|
|
|
if (!(inter4v) || |
|
|
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
|
|
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
|
|
// INTER MODE |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0] = pMB->mvs[1] |
|
|
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
|
|
|
|
|
pMB->qmvs[0] = pMB->qmvs[1] |
|
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
|
|
|
|
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
|
|
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
|
|
|
|
if(pParam->m_quarterpel) { |
|
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
|
|
} else { |
|
|
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
|
|
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
|
|
} |
|
|
} else { |
|
|
// INTER4V MODE; all other things are already set in Search8 |
|
|
pMB->mode = MODE_INTER4V; |
|
|
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] |
|
|
+ Data->iMinSAD[4] + IMV16X16 * iQuant; |
|
|
} |
|
|
|
|
|
} |
|
|
|
|
|
void |
|
|
MotionEstimationHinted( MBParam * const pParam, |
|
|
FRAMEINFO * const current, |
|
|
FRAMEINFO * const reference, |
|
|
const IMAGE * const pRefH, |
|
|
const IMAGE * const pRefV, |
|
|
const IMAGE * const pRefHV) |
|
|
{ |
|
|
MACROBLOCK *const pMBs = current->mbs; |
|
|
const IMAGE *const pCurrent = ¤t->image; |
|
|
const IMAGE *const pRef = &reference->image; |
|
|
|
|
|
uint32_t x, y; |
|
|
uint8_t * qimage; |
|
|
int32_t temp[5], quant = current->quant; |
|
|
int32_t iMinSAD[5]; |
|
|
VECTOR currentMV[5], currentQMV[5]; |
|
|
SearchData Data; |
|
|
Data.iEdgedWidth = pParam->edged_width; |
|
|
Data.currentMV = currentMV; |
|
|
Data.currentQMV = currentQMV; |
|
|
Data.iMinSAD = iMinSAD; |
|
|
Data.temp = temp; |
|
|
Data.iFcode = current->fcode; |
|
|
Data.rounding = pParam->m_rounding_type; |
|
|
Data.qpel = pParam->m_quarterpel; |
|
|
Data.chroma = current->global_flags & XVID_ME_COLOUR; |
|
|
|
|
|
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
|
|
return; // allocate some mem for qpel interpolated blocks |
|
|
// somehow this is dirty since I think we shouldn't use malloc outside |
|
|
// encoder_create() - so please fix me! |
|
|
|
|
|
Data.RefQ = qimage; |
|
|
|
|
|
if (sadInit) (*sadInit) (); |
|
|
|
|
|
for (y = 0; y < pParam->mb_height; y++) { |
|
|
for (x = 0; x < pParam->mb_width; x++) { |
|
|
|
|
|
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
|
|
|
|
|
//intra mode is copied from the first pass. At least for the time being |
|
|
if ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue; |
|
|
|
|
|
if (!(current->global_flags & XVID_LUMIMASKING)) { |
|
|
pMB->dquant = NO_CHANGE; |
|
|
pMB->quant = current->quant; } |
|
|
else { |
|
|
if (pMB->dquant != NO_CHANGE) { |
|
|
quant += DQtab[pMB->dquant]; |
|
|
if (quant > 31) quant = 31; |
|
|
else if (quant < 1) quant = 1; |
|
|
} |
|
|
pMB->quant = quant; |
|
|
} |
|
|
|
|
|
SearchPhinted(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
|
|
y, current->motion_flags, pMB->quant, |
|
|
pParam, pMBs, current->global_flags & XVID_INTER4V, pMB, |
|
|
&Data); |
|
|
|
|
|
} |
|
|
} |
|
|
free(qimage); |
|
|
} |
|
|
|
|
1871 |
static __inline int |
static __inline int |
1872 |
MEanalyzeMB ( const uint8_t * const pRef, |
MEanalyzeMB ( const uint8_t * const pRef, |
1873 |
const uint8_t * const pCur, |
const uint8_t * const pCur, |
1893 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
1894 |
|
|
1895 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1896 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, 0); |
1897 |
|
|
1898 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
1899 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
1904 |
pmv[2].y = EVEN(Data->predMV.y); |
pmv[2].y = EVEN(Data->predMV.y); |
1905 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
1906 |
|
|
1907 |
(*CheckCandidate)(0, 0, 255, &i, Data); |
CheckCandidate16no4vI(0, 0, 255, &i, Data); |
1908 |
|
|
1909 |
//early skip for 0,0 |
//early skip for 0,0 |
1910 |
if (*Data->iMinSAD < MAX_SAD00_FOR_SKIP * 4) { |
if (*Data->iMinSAD < MAX_SAD00_FOR_SKIP * 4) { |
1914 |
} |
} |
1915 |
|
|
1916 |
if (!(mask = make_mask(pmv, 1))) |
if (!(mask = make_mask(pmv, 1))) |
1917 |
(*CheckCandidate)(pmv[1].x, pmv[1].y, mask, &i, Data); |
CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data); |
1918 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
1919 |
(*CheckCandidate)(pmv[2].x, pmv[2].y, mask, &i, Data); |
CheckCandidate16no4vI(pmv[2].x, pmv[2].y, mask, &i, Data); |
1920 |
|
|
1921 |
if (*Data->iMinSAD > MAX_SAD00_FOR_SKIP * 4) // diamond only if needed |
if (*Data->iMinSAD > MAX_SAD00_FOR_SKIP * 6) // diamond only if needed |
1922 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1923 |
|
|
1924 |
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]; |
1978 |
pParam->edged_width); |
pParam->edged_width); |
1979 |
if (dev + IntraThresh < sad) { |
if (dev + IntraThresh < sad) { |
1980 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
1981 |
if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame |
if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP; |
1982 |
} |
} |
1983 |
} |
} |
1984 |
sSAD += sad; |
sSAD += sad; |
1985 |
} |
} |
1986 |
} |
} |
1987 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
1988 |
if (sSAD > InterThresh ) return 1; //P frame |
if (sSAD > InterThresh ) return P_VOP; |
1989 |
emms(); |
emms(); |
1990 |
return 0; // B frame |
return B_VOP; |
|
|
|
|
} |
|
|
|
|
|
int |
|
|
FindFcode( const MBParam * const pParam, |
|
|
const FRAMEINFO * const current) |
|
|
{ |
|
|
uint32_t x, y; |
|
|
int max = 0, min = 0, i; |
|
|
|
|
|
for (y = 0; y < pParam->mb_height; y++) { |
|
|
for (x = 0; x < pParam->mb_width; x++) { |
|
|
|
|
|
MACROBLOCK *pMB = ¤t->mbs[x + y * pParam->mb_width]; |
|
|
for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) { |
|
|
if (pMB->mvs[i].x > max) max = pMB->mvs[i].x; |
|
|
if (pMB->mvs[i].y > max) max = pMB->mvs[i].y; |
|
|
|
|
|
if (pMB->mvs[i].x < min) min = pMB->mvs[i].x; |
|
|
if (pMB->mvs[i].y < min) min = pMB->mvs[i].y; |
|
|
} |
|
|
} |
|
|
} |
|
|
|
|
|
min = -min; |
|
|
max += 1; |
|
|
if (min > max) max = min; |
|
|
if (pParam->m_quarterpel) max *= 2; |
|
1991 |
|
|
|
for (i = 1; (max > 32 << (i - 1)); i++); |
|
|
return i; |
|
1992 |
} |
} |
1993 |
|
|
1994 |
static void |
static void |