37 |
#include "../prediction/mbprediction.h" |
#include "../prediction/mbprediction.h" |
38 |
#include "../global.h" |
#include "../global.h" |
39 |
#include "../utils/timer.h" |
#include "../utils/timer.h" |
40 |
|
#include "../image/interpolate8x8.h" |
41 |
#include "motion_est.h" |
#include "motion_est.h" |
42 |
#include "motion.h" |
#include "motion.h" |
43 |
#include "sad.h" |
#include "sad.h" |
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 |
|
|
55 |
|
#define GET_REFERENCE(X, Y, REF) { \ |
56 |
|
switch ( (((X)&1)<<1) + ((Y)&1) ) \ |
57 |
|
{ \ |
58 |
|
case 0 : REF = (uint8_t *)data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ |
59 |
|
case 1 : REF = (uint8_t *)data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ |
60 |
|
case 2 : REF = (uint8_t *)data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ |
61 |
|
default : REF = (uint8_t *)data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ |
62 |
|
} \ |
63 |
|
} |
64 |
|
|
65 |
#define iDiamondSize 2 |
#define iDiamondSize 2 |
66 |
|
|
67 |
static __inline int |
static __inline int |
95 |
static void |
static void |
96 |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
97 |
{ |
{ |
|
int32_t * const sad = data->temp; |
|
98 |
int t; |
int t; |
99 |
const uint8_t * Reference; |
const uint8_t * Reference; |
100 |
|
|
108 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
109 |
} |
} |
110 |
|
|
111 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
112 |
|
|
113 |
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
114 |
data->temp[0] += lambda_vec16[data->iQuant] * t; |
data->temp[0] += data->lambda16 * t; |
115 |
data->temp[1] += lambda_vec8[data->iQuant] * t; |
data->temp[1] += data->lambda8 * t; |
116 |
|
|
117 |
|
if (data->temp[0] < data->iMinSAD[0]) { |
118 |
|
data->iMinSAD[0] = data->temp[0]; |
119 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
120 |
|
*dir = Direction; } |
121 |
|
|
122 |
|
if (data->temp[1] < data->iMinSAD[1]) { |
123 |
|
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
124 |
|
if (data->temp[2] < data->iMinSAD[2]) { |
125 |
|
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
126 |
|
if (data->temp[3] < data->iMinSAD[3]) { |
127 |
|
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
128 |
|
if (data->temp[4] < data->iMinSAD[4]) { |
129 |
|
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
130 |
|
|
131 |
|
} |
132 |
|
|
133 |
|
static void |
134 |
|
CheckCandidate16Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
135 |
|
{ |
136 |
|
//this function expects X and Y in halfpel positions, but assumes that qpel positions are our goal |
137 |
|
int t; |
138 |
|
const uint8_t * Reference; |
139 |
|
|
140 |
|
if (( x > data->max_dx) || ( x < data->min_dx) |
141 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
142 |
|
|
143 |
|
switch ( ((x&1)<<1) + (y&1) ) { |
144 |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
145 |
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
146 |
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
147 |
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
148 |
|
} |
149 |
|
|
150 |
|
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
151 |
|
|
152 |
|
t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode); |
153 |
|
data->temp[0] += data->lambda16 * t; |
154 |
|
data->temp[1] += data->lambda8 * t; |
155 |
|
|
156 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (data->temp[0] < data->iMinSAD[0]) { |
157 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = data->temp[0]; |
186 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
187 |
} |
} |
188 |
|
|
189 |
sad = lambda_vec16[data->iQuant] * |
sad = data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
190 |
d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR); |
|
sad += sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
|
191 |
|
|
192 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
193 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
196 |
} |
} |
197 |
|
|
198 |
static void |
static void |
199 |
|
CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
200 |
|
|
201 |
|
// CheckCandidate16 variant which expects x and y in quarter pixel resolution |
202 |
|
// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) |
203 |
|
// around currentMV! |
204 |
|
{ |
205 |
|
int t; |
206 |
|
uint8_t * Reference = (uint8_t *)data->RefQ; |
207 |
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
208 |
|
VECTOR halfpelMV = *(data->currentMV); |
209 |
|
|
210 |
|
int32_t iEdgedWidth = data->iEdgedWidth; |
211 |
|
uint32_t rounding = data->rounding; |
212 |
|
|
213 |
|
if (( x > data->max_dx) || ( x < data->min_dx) |
214 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
215 |
|
|
216 |
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases |
217 |
|
switch( ((x&1)<<1) + (y&1) ) |
218 |
|
{ |
219 |
|
case 0: // pure halfpel position - shouldn't happen during a refinement step |
220 |
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); |
221 |
|
break; |
222 |
|
|
223 |
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
224 |
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
225 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
226 |
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); |
227 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); |
228 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); |
229 |
|
break; |
230 |
|
|
231 |
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
232 |
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); |
233 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
234 |
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); |
235 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); |
236 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); |
237 |
|
break; |
238 |
|
|
239 |
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
240 |
|
// bottom left/right) during qpel refinement |
241 |
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
242 |
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); |
243 |
|
GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); |
244 |
|
|
245 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
246 |
|
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
247 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
248 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
249 |
|
break; |
250 |
|
} |
251 |
|
|
252 |
|
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1); |
253 |
|
|
254 |
|
t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); |
255 |
|
data->temp[0] += data->lambda16 * t; |
256 |
|
data->temp[1] += data->lambda8 * t; |
257 |
|
|
258 |
|
if (data->temp[0] < data->iMinSAD[0]) { |
259 |
|
data->iMinSAD[0] = data->temp[0]; |
260 |
|
data->currentQMV[0].x = x; data->currentQMV[0].y = y; |
261 |
|
/* *dir = Direction;*/ } |
262 |
|
|
263 |
|
if (data->temp[1] < data->iMinSAD[1]) { |
264 |
|
data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; } |
265 |
|
if (data->temp[2] < data->iMinSAD[2]) { |
266 |
|
data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; } |
267 |
|
if (data->temp[3] < data->iMinSAD[3]) { |
268 |
|
data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; } |
269 |
|
if (data->temp[4] < data->iMinSAD[4]) { |
270 |
|
data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; } |
271 |
|
} |
272 |
|
|
273 |
|
static void |
274 |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
275 |
{ |
{ |
276 |
int32_t sad; |
int32_t sad; |
278 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
279 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
280 |
|
|
281 |
sad = lambda_vec16[data->iQuant] * |
sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
|
d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
|
|
|
|
sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
|
282 |
data->iEdgedWidth, 256*4096); |
data->iEdgedWidth, 256*4096); |
283 |
|
|
284 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
313 |
default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break; |
default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break; |
314 |
} |
} |
315 |
|
|
316 |
sad = lambda_vec16[data->iQuant] * |
sad = data->lambda16 * |
317 |
( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) + |
( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) + |
318 |
d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) ); |
d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) ); |
319 |
|
|
336 |
|
|
337 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
338 |
|
|
339 |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
sad = data->lambda16 * d_mv_bits(x, y, 1); |
340 |
|
|
341 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
342 |
mvs.x = data->directmvF[k].x + x; |
mvs.x = data->directmvF[k].x + x; |
391 |
|
|
392 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
393 |
|
|
394 |
sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1); |
sad = data->lambda16 * d_mv_bits(x, y, 1); |
395 |
|
|
396 |
mvs.x = data->directmvF[0].x + x; |
mvs.x = data->directmvF[0].x + x; |
397 |
b_mvs.x = ((x == 0) ? |
b_mvs.x = ((x == 0) ? |
448 |
} |
} |
449 |
|
|
450 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
451 |
sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
sad += data->lambda8 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
452 |
|
|
453 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
454 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
456 |
*dir = Direction; } |
*dir = Direction; } |
457 |
} |
} |
458 |
|
|
459 |
|
static void |
460 |
|
CheckCandidate8Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
461 |
|
{ |
462 |
|
int32_t sad; |
463 |
|
const uint8_t * Reference; |
464 |
|
|
465 |
|
if (( x > data->max_dx) || ( x < data->min_dx) |
466 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
467 |
|
|
468 |
|
switch ( ((x&1)<<1) + (y&1) ) |
469 |
|
{ |
470 |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
471 |
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
472 |
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
473 |
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
474 |
|
} |
475 |
|
|
476 |
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
477 |
|
sad += data->lambda8 * d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode); |
478 |
|
|
479 |
|
if (sad < *(data->iMinSAD)) { |
480 |
|
*(data->iMinSAD) = sad; |
481 |
|
data->currentMV->x = x; data->currentMV->y = y; |
482 |
|
*dir = Direction; } |
483 |
|
} |
484 |
|
|
485 |
|
static void |
486 |
|
CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
487 |
|
// CheckCandidate16no4v variant which expects x and y in quarter pixel resolution |
488 |
|
// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) |
489 |
|
// around currentMV! |
490 |
|
|
491 |
|
{ |
492 |
|
int32_t sad; |
493 |
|
uint8_t *Reference = (uint8_t *) data->RefQ; |
494 |
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
495 |
|
VECTOR halfpelMV = *(data->currentMV); |
496 |
|
|
497 |
|
int32_t iEdgedWidth = data->iEdgedWidth; |
498 |
|
uint32_t rounding = data->rounding; |
499 |
|
|
500 |
|
if (( x > data->max_dx) || ( x < data->min_dx) |
501 |
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
502 |
|
|
503 |
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); |
504 |
|
switch( ((x&1)<<1) + (y&1) ) |
505 |
|
{ |
506 |
|
case 0: // pure halfpel position - shouldn't happen during a refinement step |
507 |
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); |
508 |
|
break; |
509 |
|
|
510 |
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
511 |
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
512 |
|
|
513 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
514 |
|
break; |
515 |
|
|
516 |
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
517 |
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); |
518 |
|
|
519 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
520 |
|
break; |
521 |
|
|
522 |
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
523 |
|
// bottom left/right) during qpel refinement |
524 |
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
525 |
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); |
526 |
|
GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); |
527 |
|
|
528 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
529 |
|
break; |
530 |
|
} |
531 |
|
|
532 |
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
533 |
|
sad += data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); |
534 |
|
|
535 |
|
if (sad < *(data->iMinSAD)) { |
536 |
|
*(data->iMinSAD) = sad; |
537 |
|
data->currentQMV->x = x; data->currentQMV->y = y; |
538 |
|
*dir = Direction; } |
539 |
|
} |
540 |
|
|
541 |
/* CHECK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
542 |
|
|
543 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
706 |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
707 |
} |
} |
708 |
|
|
709 |
|
|
710 |
|
static void |
711 |
|
QuarterpelRefine(const SearchData * const data) |
712 |
|
{ |
713 |
|
/* Perform quarter pixel refinement*/ |
714 |
|
|
715 |
|
VECTOR backupMV = *(data->currentQMV); |
716 |
|
int iDirection; //not needed |
717 |
|
|
718 |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
719 |
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
720 |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); |
721 |
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); |
722 |
|
|
723 |
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
724 |
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); |
725 |
|
|
726 |
|
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
727 |
|
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
728 |
|
|
729 |
|
} |
730 |
|
|
731 |
static __inline int |
static __inline int |
732 |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
733 |
const int x, const int y, |
const int x, const int y, |
754 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
755 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
756 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
757 |
|
pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0; |
758 |
|
pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0; |
759 |
|
|
760 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
761 |
} |
} |
762 |
|
|
778 |
uint32_t x, y; |
uint32_t x, y; |
779 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
780 |
int32_t InterBias, quant = current->quant; |
int32_t InterBias, quant = current->quant; |
781 |
|
uint8_t *qimage; |
782 |
|
|
783 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
784 |
int32_t temp[5]; |
int32_t temp[5]; |
785 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
786 |
|
VECTOR currentQMV[5]; |
787 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
788 |
SearchData Data; |
SearchData Data; |
789 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
790 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
791 |
|
Data.currentQMV = currentQMV; |
792 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
793 |
Data.temp = temp; |
Data.temp = temp; |
794 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
795 |
|
Data.rounding = pParam->m_rounding_type; |
796 |
|
|
797 |
|
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
798 |
|
return 1; // allocate some mem for qpel interpolated blocks |
799 |
|
// somehow this is dirty since I think we shouldn't use malloc outside |
800 |
|
// encoder_create() - so please fix me! |
801 |
|
|
802 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
803 |
|
|
812 |
|
|
813 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
814 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
815 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; |
816 |
else |
} else { |
817 |
if (pMB->dquant != NO_CHANGE) { |
if (pMB->dquant != NO_CHANGE) { |
818 |
quant += DQtab[pMB->dquant]; |
quant += DQtab[pMB->dquant]; |
819 |
if (quant > 31) quant = 31; |
if (quant > 31) quant = 31; |
820 |
else if (quant < 1) quant = 1; |
else if (quant < 1) quant = 1; |
821 |
|
} |
822 |
pMB->quant = quant; |
pMB->quant = quant; |
823 |
} |
} |
824 |
|
|
825 |
//initial skip decision |
/* initial skip decision */ |
826 |
|
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
827 |
|
|
828 |
if ((pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant) |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
829 |
&& (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) ) { |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH) |
830 |
if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) { |
if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
831 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
832 |
continue; |
continue; |
833 |
} |
} |
834 |
} else sad00 = 256*4096; // skip not allowed - for final skip decision |
} |
835 |
|
|
836 |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x, |
837 |
y, current->motion_flags, pMB->quant, |
y, current->motion_flags, pMB->quant, |
838 |
&Data, pParam, pMBs, reference->mbs, |
&Data, pParam, pMBs, reference->mbs, |
839 |
current->global_flags & XVID_INTER4V, pMB); |
current->global_flags & XVID_INTER4V, pMB); |
840 |
|
|
841 |
/* 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?" */ |
842 |
if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
if (current->coding_type == P_VOP) { |
843 |
if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
844 |
{ SkipMacroblockP(pMB, sad00); continue; } |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) ) |
845 |
|
if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) { |
846 |
|
SkipMacroblockP(pMB, sad00); |
847 |
|
continue; |
848 |
|
} |
849 |
|
} |
850 |
|
|
851 |
/* finally, intra decision */ |
/* finally, intra decision */ |
852 |
|
|
863 |
pParam->edged_width); |
pParam->edged_width); |
864 |
|
|
865 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
866 |
if (++iIntra >= iLimit) return 1; |
if (++iIntra >= iLimit) { free(qimage); return 1; } |
867 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
868 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
869 |
pMB->mvs[3] = zeroMV; |
pMB->mvs[3] = zeroMV; |
870 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = |
871 |
|
pMB->qmvs[3] = zeroMV; |
872 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
873 |
pMB->sad8[3] = 0; |
pMB->sad8[3] = 0; |
874 |
} |
} |
875 |
} |
} |
876 |
} |
} |
877 |
} |
} |
878 |
|
free(qimage); |
879 |
|
|
880 |
|
if (current->coding_type == S_VOP) /* first GMC step only for S(GMC)-VOPs */ |
881 |
|
current->GMC_MV = GlobalMotionEst( pMBs, pParam, current->fcode ); |
882 |
|
else |
883 |
|
current->GMC_MV = zeroMV; |
884 |
|
|
885 |
return 0; |
return 0; |
886 |
} |
} |
887 |
|
|
943 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
944 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
945 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
946 |
|
const uint8_t * const pRefQ, |
947 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
948 |
const int x, |
const int x, |
949 |
const int y, |
const int y, |
962 |
|
|
963 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
964 |
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, |
965 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
966 |
|
|
967 |
Data->predMV = pmv[0]; |
Data->predMV = pmv[0]; |
968 |
|
|
969 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
970 |
Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; |
Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16; |
971 |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16; |
972 |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16; |
973 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
974 |
|
Data->RefQ = pRefQ; |
975 |
|
|
976 |
Data->iQuant = iQuant; |
Data->lambda16 = lambda_vec16[iQuant]; |
977 |
|
Data->lambda8 = lambda_vec8[iQuant]; |
978 |
|
|
979 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
980 |
Data->min_dx = EVEN(Data->min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
984 |
|
|
985 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
986 |
|
|
987 |
if (inter4v) CheckCandidate = CheckCandidate16; |
for(i = 0; i < 5; i++) |
988 |
else CheckCandidate = CheckCandidate16no4v; |
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
989 |
|
|
990 |
for(i = 0; i < 5; i++) Data->currentMV[i].x = Data->currentMV[i].y = 0; |
if (pParam->m_quarterpel) { |
991 |
|
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
992 |
|
i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode); |
993 |
|
} else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
994 |
|
|
|
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
|
995 |
Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i; |
996 |
Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i; |
997 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
1007 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
1008 |
prevMBs + x + y * pParam->mb_width); |
prevMBs + x + y * pParam->mb_width); |
1009 |
|
|
1010 |
|
if (pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q; |
1011 |
|
else |
1012 |
if (inter4v) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
1013 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1014 |
|
|
|
|
|
1015 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
1016 |
|
|
1017 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) { |
1045 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1046 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1047 |
|
|
1048 |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
1049 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
1050 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1051 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1058 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1059 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1060 |
|
|
1061 |
CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data); |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
1062 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
1063 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1064 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1069 |
|
|
1070 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1071 |
|
|
1072 |
|
for(i = 0; i < 5; i++) { |
1073 |
|
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
1074 |
|
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1075 |
|
} |
1076 |
|
|
1077 |
|
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1078 |
|
|
1079 |
|
CheckCandidate = CheckCandidate16_qpel; |
1080 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1081 |
|
pParam->width, pParam->height, Data->iFcode, 0); |
1082 |
|
|
1083 |
|
QuarterpelRefine(Data); |
1084 |
|
} |
1085 |
|
|
1086 |
if (inter4v) { |
if (inter4v) { |
1087 |
SearchData Data8; |
SearchData Data8; |
1088 |
Data8.iFcode = Data->iFcode; |
Data8.iFcode = Data->iFcode; |
1089 |
Data8.iQuant = Data->iQuant; |
Data8.lambda8 = Data->lambda8; |
1090 |
Data8.iEdgedWidth = Data->iEdgedWidth; |
Data8.iEdgedWidth = Data->iEdgedWidth; |
1091 |
|
Data8.RefQ = Data->RefQ; |
1092 |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1093 |
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); |
1094 |
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); |
1103 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1104 |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1105 |
|
|
1106 |
|
pMB->qmvs[0] = pMB->qmvs[1] |
1107 |
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1108 |
|
|
1109 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1110 |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1111 |
|
|
1112 |
|
if(pParam->m_quarterpel) { |
1113 |
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x; |
1114 |
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y; |
1115 |
|
} |
1116 |
|
else { |
1117 |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1118 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1119 |
|
} |
1120 |
} else { |
} else { |
1121 |
// INTER4V MODE; all other things are already set in Search8 |
// INTER4V MODE; all other things are already set in Search8 |
1122 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1123 |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
1124 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
1125 |
} |
} |
|
|
|
1126 |
} |
} |
1127 |
|
|
1128 |
static void |
static void |
1135 |
const int block, |
const int block, |
1136 |
SearchData * const Data) |
SearchData * const Data) |
1137 |
{ |
{ |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
|
1138 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1139 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1140 |
|
Data->currentQMV = OldData->currentQMV + 1 + block; |
1141 |
|
|
1142 |
if (block != 0) |
if(pParam->m_quarterpel) { |
1143 |
*(Data->iMinSAD) += lambda_vec8[Data->iQuant] * |
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1144 |
|
if (block != 0) *(Data->iMinSAD) += Data->lambda8 * |
1145 |
|
d_mv_bits( Data->currentQMV->x - Data->predQMV.x, |
1146 |
|
Data->currentQMV->y - Data->predQMV.y, |
1147 |
|
Data->iFcode); |
1148 |
|
CheckCandidate = CheckCandidate8Q; |
1149 |
|
} else { |
1150 |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1151 |
|
if (block != 0) *(Data->iMinSAD) += Data->lambda8 * |
1152 |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
1153 |
Data->currentMV->y - Data->predMV.y, |
Data->currentMV->y - Data->predMV.y, |
1154 |
Data->iFcode); |
Data->iFcode); |
1155 |
|
CheckCandidate = CheckCandidate8; |
1156 |
|
} |
1157 |
|
|
1158 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
1159 |
|
|
1165 |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1166 |
|
|
1167 |
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, |
1168 |
pParam->width, pParam->height, OldData->iFcode); |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
|
|
|
|
CheckCandidate = CheckCandidate8; |
|
1169 |
|
|
1170 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1171 |
|
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1172 |
|
|
1173 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1174 |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
1175 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1176 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1177 |
|
|
1178 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); } |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1179 |
|
|
1180 |
|
if(*(Data->iMinSAD) < temp_sad) { |
1181 |
|
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1182 |
|
Data->currentQMV->y = 2 * Data->currentMV->y; |
1183 |
|
} |
1184 |
|
} |
1185 |
|
|
1186 |
|
if (MotionFlags & PMV_HALFPELREFINE8) { |
1187 |
|
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1188 |
|
|
1189 |
|
HalfpelRefine(Data); // perform halfpel refine of current best vector |
1190 |
|
|
1191 |
|
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
1192 |
|
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1193 |
|
Data->currentQMV->y = 2 * Data->currentMV->y; |
1194 |
|
} |
1195 |
|
} |
1196 |
|
|
1197 |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(Data); |
if(pParam->m_quarterpel) { |
1198 |
|
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
1199 |
|
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
1200 |
|
CheckCandidate = CheckCandidate8_qpel; |
1201 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1202 |
|
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
1203 |
|
QuarterpelRefine(Data); |
1204 |
|
} |
1205 |
|
} |
1206 |
} |
} |
1207 |
|
|
1208 |
|
if(pParam->m_quarterpel) { |
1209 |
|
pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x; |
1210 |
|
pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y; |
1211 |
|
} |
1212 |
|
else { |
1213 |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1214 |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1215 |
|
} |
1216 |
|
|
1217 |
pMB->mvs[block] = *(Data->currentMV); |
pMB->mvs[block] = *(Data->currentMV); |
1218 |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
pMB->qmvs[block] = *(Data->currentQMV); |
1219 |
|
|
1220 |
|
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
1221 |
} |
} |
1222 |
|
|
1223 |
/* B-frames code starts here */ |
/* B-frames code starts here */ |
1303 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1304 |
|
|
1305 |
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, |
1306 |
pParam->width, pParam->height, iFcode); |
pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
1307 |
|
|
1308 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1309 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
1330 |
|
|
1331 |
// three bits are needed to code backward mode. four for forward |
// three bits are needed to code backward mode. four for forward |
1332 |
// we treat the bits just like they were vector's |
// we treat the bits just like they were vector's |
1333 |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * lambda_vec16[Data->iQuant]; |
if (mode_current == MODE_FORWARD) *Data->iMinSAD += 4 * Data->lambda16 * 2; |
1334 |
else *Data->iMinSAD += 3 * lambda_vec16[Data->iQuant]; |
else *Data->iMinSAD += 3 * Data->lambda16 * 2; |
|
|
|
1335 |
|
|
1336 |
if (*Data->iMinSAD < *best_sad) { |
if (*Data->iMinSAD < *best_sad) { |
1337 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1338 |
pMB->mode = mode_current; |
pMB->mode = mode_current; |
1339 |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1340 |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
1341 |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV; |
1342 |
else pMB->b_mvs[0] = *Data->currentMV; |
else pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
1343 |
} |
} |
1344 |
|
|
1345 |
} |
} |
1416 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1417 |
|
|
1418 |
// skip decision |
// skip decision |
1419 |
if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) { |
if (*Data->iMinSAD - 2 * Data->lambda16 < (uint32_t)pMB->quant * SKIP_THRESH_B) { |
1420 |
//checking chroma. everything copied from MC |
//possible skip - checking chroma. everything copied from MC |
1421 |
//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 |
1422 |
int sum, dx, dy, b_dx, b_dy; |
int sum, dx, dy, b_dx, b_dy; |
1423 |
|
|
1442 |
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1443 |
Data->iEdgedWidth/2); |
Data->iEdgedWidth/2); |
1444 |
|
|
1445 |
if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data->iQuant) { |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
1446 |
pMB->mode = MODE_DIRECT_NONE_MV; |
pMB->mode = MODE_DIRECT_NONE_MV; |
1447 |
return *Data->iMinSAD; |
return *Data->iMinSAD; |
1448 |
} |
} |
1461 |
|
|
1462 |
HalfpelRefine(Data); |
HalfpelRefine(Data); |
1463 |
|
|
1464 |
*Data->iMinSAD += 1 * lambda_vec16[Data->iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's |
*Data->iMinSAD += 1 * Data->lambda16 * 2; // one bit is needed to code direct mode |
1465 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1466 |
|
|
1467 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1517 |
int iDirection, i, j; |
int iDirection, i, j; |
1518 |
SearchData bData; |
SearchData bData; |
1519 |
|
|
1520 |
bData.iMinSAD = fData->iMinSAD; |
*(bData.iMinSAD = fData->iMinSAD) = 4096*256; |
|
*bData.iMinSAD = 4096*256; |
|
1521 |
bData.Cur = fData->Cur; |
bData.Cur = fData->Cur; |
1522 |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
1523 |
bData.currentMV = fData->currentMV + 1; |
bData.currentMV = fData->currentMV + 1; |
1524 |
bData.iQuant = fData->iQuant; |
bData.lambda16 = fData->lambda16; |
1525 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1526 |
|
|
1527 |
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16; |
1536 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1537 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
1538 |
|
|
1539 |
fData->currentMV[0] = pMB->mvs[0]; |
fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed |
1540 |
fData->currentMV[1] = pMB->b_mvs[0]; |
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); |
1541 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
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(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode); |
|
1542 |
|
|
1543 |
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; |
1544 |
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_dy; |
1576 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1577 |
|
|
1578 |
// two bits are needed to code interpolate mode. we treat the bits just like they were vector's |
// two bits are needed to code interpolate mode. we treat the bits just like they were vector's |
1579 |
*fData->iMinSAD += 2 * lambda_vec16[fData->iQuant]; |
*fData->iMinSAD += 2 * fData->lambda16 * 2; |
1580 |
|
|
1581 |
if (*fData->iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1582 |
*best_sad = *fData->iMinSAD; |
*best_sad = *fData->iMinSAD; |
1583 |
pMB->mvs[0] = fData->currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1628 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1629 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
1630 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
1631 |
Data.iQuant = frame->quant; |
Data.lambda16 = lambda_vec16[frame->quant]; |
1632 |
|
|
1633 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
1634 |
|
|
1640 |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
MACROBLOCK * const pMB = frame->mbs + i + j * pParam->mb_width; |
1641 |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
1642 |
|
|
1643 |
/* special case, if collocated block is SKIPed: 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 */ |
1644 |
if (b_mb->mode == MODE_NOT_CODED) { |
if ( (b_mb->mode == MODE_NOT_CODED) ) { |
1645 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
1646 |
continue; |
continue; |
1647 |
} |
} |
1648 |
|
|
1649 |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
1650 |
|
pMB->quant = frame->quant; |
1651 |
/* direct search comes first, because it (1) checks for SKIP-mode |
/* direct search comes first, because it (1) checks for SKIP-mode |
1652 |
and (2) sets very good predictions for forward and backward search */ |
and (2) sets very good predictions for forward and backward search */ |
1653 |
|
|
1727 |
|
|
1728 |
/* Hinted ME starts here */ |
/* Hinted ME starts here */ |
1729 |
|
|
1730 |
static __inline void |
static void |
1731 |
Search8hinted( const SearchData * const OldData, |
Search8hinted( const SearchData * const OldData, |
1732 |
const int x, const int y, |
const int x, const int y, |
1733 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1734 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1735 |
MACROBLOCK * const pMB, |
MACROBLOCK * const pMB, |
1736 |
const MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1737 |
const int block) |
const int block, |
1738 |
|
SearchData * const Data) |
1739 |
{ |
{ |
1740 |
SearchData Data; |
int32_t temp_sad; |
1741 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1742 |
|
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1743 |
|
Data->currentMV = OldData->currentMV + 1 + block; |
1744 |
|
Data->currentQMV = OldData->currentQMV + 1 + block; |
1745 |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1746 |
|
|
1747 |
Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
if(pParam->m_quarterpel) { |
1748 |
Data.iMinSAD = OldData->iMinSAD + 1 + block; |
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1749 |
Data.currentMV = OldData->currentMV+1+block; |
if (block != 0) *(Data->iMinSAD) += Data->lambda8 * |
1750 |
Data.iFcode = OldData->iFcode; |
d_mv_bits( Data->currentQMV->x - Data->predQMV.x, |
1751 |
Data.iQuant = OldData->iQuant; |
Data->currentQMV->y - Data->predQMV.y, |
1752 |
|
Data->iFcode); |
1753 |
Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
CheckCandidate = CheckCandidate8Q; |
1754 |
Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
} else { |
1755 |
Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
if (block != 0) *(Data->iMinSAD) += Data->lambda8 * |
1756 |
Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
d_mv_bits( Data->currentMV->x - Data->predMV.x, |
1757 |
Data.iEdgedWidth = pParam->edged_width; |
Data->currentMV->y - Data->predMV.y, |
1758 |
Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->iFcode); |
|
|
|
1759 |
CheckCandidate = CheckCandidate8; |
CheckCandidate = CheckCandidate8; |
1760 |
|
} |
1761 |
|
|
1762 |
if (block != 0) |
Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1763 |
*(Data.iMinSAD) += lambda_vec8[Data.iQuant] * |
Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1764 |
d_mv_bits( Data.currentMV->x - Data.predMV.x, |
Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1765 |
Data.currentMV->y - Data.predMV.y, |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1766 |
Data.iFcode); |
|
1767 |
|
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1768 |
|
|
1769 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1770 |
|
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
1771 |
|
|
1772 |
get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8, |
temp_sad = *(Data->iMinSAD); // store current MinSAD |
|
pParam->width, pParam->height, OldData->iFcode); |
|
|
|
|
|
if (pMB->mode == MODE_INTER4V) { |
|
|
int dummy; |
|
|
CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); } |
|
1773 |
|
|
1774 |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch; |
1775 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1776 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1777 |
|
|
1778 |
(*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1779 |
|
|
1780 |
|
if(*(Data->iMinSAD) < temp_sad) { |
1781 |
|
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1782 |
|
Data->currentQMV->y = 2 * Data->currentMV->y; |
1783 |
|
} |
1784 |
|
|
1785 |
|
if (MotionFlags & PMV_HALFPELREFINE8) { |
1786 |
|
temp_sad = *(Data->iMinSAD); // store current MinSAD |
1787 |
|
|
1788 |
if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data); |
HalfpelRefine(Data); // perform halfpel refine of current best vector |
1789 |
|
|
1790 |
pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x; |
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
1791 |
pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y; |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1792 |
pMB->mvs[block] = *(Data.currentMV); |
Data->currentQMV->y = 2 * Data->currentMV->y; |
1793 |
pMB->sad8[block] = 4 * (*(Data.iMinSAD)); |
} |
1794 |
} |
} |
1795 |
|
|
1796 |
|
if(pParam->m_quarterpel) { |
1797 |
|
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
1798 |
|
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
1799 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1800 |
|
pParam->width, pParam->height, Data->iFcode, 0); |
1801 |
|
CheckCandidate = CheckCandidate8_qpel; |
1802 |
|
QuarterpelRefine(Data); |
1803 |
|
} |
1804 |
|
pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x; |
1805 |
|
pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y; |
1806 |
|
} else { |
1807 |
|
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1808 |
|
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1809 |
|
} |
1810 |
|
|
1811 |
|
pMB->mvs[block] = *(Data->currentMV); |
1812 |
|
pMB->qmvs[block] = *(Data->currentQMV); |
1813 |
|
|
1814 |
|
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
1815 |
|
} |
1816 |
|
|
1817 |
static void |
static void |
1818 |
SearchPhinted ( const uint8_t * const pRef, |
SearchPhinted ( const uint8_t * const pRef, |
1838 |
|
|
1839 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1840 |
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, |
1841 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
1842 |
|
|
1843 |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16; |
1844 |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
Data->Ref = pRef + (x + iEdgedWidth*y)*16; |
1845 |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
Data->RefH = pRefH + (x + iEdgedWidth*y) * 16; |
1846 |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
Data->RefV = pRefV + (x + iEdgedWidth*y) * 16; |
1847 |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16; |
1848 |
Data->iQuant = iQuant; |
Data->lambda16 = lambda_vec16[iQuant]; |
1849 |
|
Data->lambda8 = lambda_vec8[iQuant]; |
1850 |
|
|
1851 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1852 |
Data->min_dx = EVEN(Data->min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
1859 |
|
|
1860 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1861 |
|
|
1862 |
if (inter4v) |
if(pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q; |
1863 |
CheckCandidate = CheckCandidate16; |
else |
1864 |
|
if (inter4v) CheckCandidate = CheckCandidate16; |
1865 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1866 |
|
|
|
|
|
1867 |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
1868 |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
pMB->mvs[0].y = EVEN(pMB->mvs[0].y); |
1869 |
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->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed |
1891 |
|
|
1892 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
1893 |
|
|
1894 |
if (inter4v) |
for(i = 0; i < 5; i++) { |
1895 |
for(i = 0; i < 4; i++) |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
1896 |
Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i); |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1897 |
|
} |
1898 |
|
|
1899 |
|
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1900 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1901 |
|
pParam->width, pParam->height, Data->iFcode, 0); |
1902 |
|
CheckCandidate = CheckCandidate16_qpel; |
1903 |
|
QuarterpelRefine(Data); |
1904 |
|
} |
1905 |
|
|
1906 |
|
if (inter4v) { |
1907 |
|
SearchData Data8; |
1908 |
|
Data8.iFcode = Data->iFcode; |
1909 |
|
Data8.lambda8 = Data->lambda8; |
1910 |
|
Data8.iEdgedWidth = Data->iEdgedWidth; |
1911 |
|
Data8.RefQ = Data->RefQ; |
1912 |
|
Search8hinted(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1913 |
|
Search8hinted(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
1914 |
|
Search8hinted(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
1915 |
|
Search8hinted(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1916 |
|
} |
1917 |
|
|
1918 |
if (!(inter4v) || |
if (!(inter4v) || |
1919 |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + |
1920 |
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1921 |
// INTER MODE |
// INTER MODE |
|
|
|
1922 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
1923 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1924 |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1925 |
|
|
1926 |
|
pMB->qmvs[0] = pMB->qmvs[1] |
1927 |
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1928 |
|
|
1929 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1930 |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1931 |
|
|
1932 |
|
if(pParam->m_quarterpel) { |
1933 |
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x; |
1934 |
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y; |
1935 |
|
} |
1936 |
|
else { |
1937 |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1938 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1939 |
|
} |
1940 |
} else { |
} else { |
1941 |
// INTER4V MODE; all other things are already set in Search8hinted |
// INTER4V MODE; all other things are already set in Search8hinted |
1942 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1959 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
1960 |
|
|
1961 |
uint32_t x, y; |
uint32_t x, y; |
1962 |
|
uint8_t * qimage; |
1963 |
int32_t temp[5], quant = current->quant; |
int32_t temp[5], quant = current->quant; |
1964 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
1965 |
VECTOR currentMV[5]; |
VECTOR currentMV[5], currentQMV[5]; |
1966 |
SearchData Data; |
SearchData Data; |
1967 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1968 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
1969 |
|
Data.currentQMV = currentQMV; |
1970 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
1971 |
Data.temp = temp; |
Data.temp = temp; |
1972 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
1973 |
|
Data.rounding = pParam->m_rounding_type; |
1974 |
|
|
1975 |
|
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
1976 |
|
return; // allocate some mem for qpel interpolated blocks |
1977 |
|
// somehow this is dirty since I think we shouldn't use malloc outside |
1978 |
|
// encoder_create() - so please fix me! |
1979 |
|
|
1980 |
|
Data.RefQ = qimage; |
1981 |
|
|
1982 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
1983 |
|
|
1993 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
1994 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
1995 |
pMB->quant = current->quant; } |
pMB->quant = current->quant; } |
1996 |
else |
else { |
1997 |
if (pMB->dquant != NO_CHANGE) { |
if (pMB->dquant != NO_CHANGE) { |
1998 |
quant += DQtab[pMB->dquant]; |
quant += DQtab[pMB->dquant]; |
1999 |
if (quant > 31) quant = 31; |
if (quant > 31) quant = 31; |
2000 |
else if (quant < 1) quant = 1; |
else if (quant < 1) quant = 1; |
2001 |
|
} |
2002 |
pMB->quant = quant; |
pMB->quant = quant; |
2003 |
} |
} |
2004 |
|
|
2009 |
|
|
2010 |
} |
} |
2011 |
} |
} |
2012 |
|
free(qimage); |
2013 |
} |
} |
2014 |
|
|
2015 |
static __inline int |
static __inline int |
2023 |
SearchData * const Data) |
SearchData * const Data) |
2024 |
{ |
{ |
2025 |
|
|
2026 |
int i, mask; |
int i = 255, mask; |
2027 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
2028 |
|
|
2029 |
*(Data->iMinSAD) = MV_MAX_ERROR; |
*(Data->iMinSAD) = MV_MAX_ERROR; |
2030 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
2031 |
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, |
2032 |
pParam->width, pParam->height, Data->iFcode); |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
2033 |
|
|
2034 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2035 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
2050 |
|
|
2051 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
2052 |
|
|
2053 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = *Data->currentMV; |
|
= pMB->mvs[2] = pMB->mvs[3] = *Data->currentMV; // all, for future get_pmv() |
|
2054 |
|
|
2055 |
return *(Data->iMinSAD); |
return *(Data->iMinSAD); |
2056 |
} |
} |
2075 |
Data.currentMV = ¤tMV; |
Data.currentMV = ¤tMV; |
2076 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2077 |
Data.iFcode = iFcode; |
Data.iFcode = iFcode; |
|
Data.iQuant = 2; |
|
2078 |
|
|
2079 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2080 |
|
|
2081 |
for (y = 0; y < pParam->mb_height-1; y++) { |
for (y = 1; y < pParam->mb_height-1; y++) { |
2082 |
for (x = 0; x < pParam->mb_width; x++) { |
for (x = 1; x < pParam->mb_width-1; x++) { |
2083 |
int sad, dev; |
int sad, dev; |
2084 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
2085 |
|
|
2086 |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
2087 |
pParam, pMBs, pMB, &Data); |
pParam, pMBs, pMB, &Data); |
2088 |
|
|
|
if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work |
|
2089 |
if (sad > INTRA_THRESH) { |
if (sad > INTRA_THRESH) { |
2090 |
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
2091 |
pParam->edged_width); |
pParam->edged_width); |
2094 |
} |
} |
2095 |
sSAD += sad; |
sSAD += sad; |
2096 |
} |
} |
|
|
|
|
} |
|
2097 |
} |
} |
2098 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
2099 |
if (sSAD > INTER_THRESH ) return 1; //P frame |
if (sSAD > INTER_THRESH ) return 1; //P frame |
2126 |
min = -min; |
min = -min; |
2127 |
max += 1; |
max += 1; |
2128 |
if (min > max) max = min; |
if (min > max) max = min; |
2129 |
|
if (pParam->m_quarterpel) max *= 2; |
2130 |
|
|
2131 |
for (i = 1; (max > 32 << (i - 1)); i++); |
for (i = 1; (max > 32 << (i - 1)); i++); |
2132 |
return i; |
return i; |
2133 |
} |
} |
2134 |
|
|
2135 |
|
|
2136 |
|
/* Global Motion Estimation, (C) 2002 by sysKin */ |
2137 |
|
/* calculate global translation from local (halfpel) motion field */ |
2138 |
|
|
2139 |
|
static VECTOR |
2140 |
|
GlobalMotionEst(const MACROBLOCK * const pMBs, const MBParam * const pParam, const uint32_t iFcode) |
2141 |
|
{ |
2142 |
|
|
2143 |
|
int count, bestcount = 0; |
2144 |
|
int x, y; |
2145 |
|
VECTOR gmc; |
2146 |
|
int step, min_x, max_x, min_y, max_y; |
2147 |
|
uint32_t mx, my; |
2148 |
|
|
2149 |
|
min_x = min_y = -32<<iFcode; |
2150 |
|
max_x = max_y = 32<<iFcode; |
2151 |
|
|
2152 |
|
for (step = 32; step >= 2; step /= 2) { |
2153 |
|
bestcount = 0; |
2154 |
|
for (y = min_y; y <= max_y; y += step) |
2155 |
|
for (x = min_x ; x <= max_x; x += step) { |
2156 |
|
count = 0; |
2157 |
|
//for all macroblocks |
2158 |
|
for (my = 1; my < pParam->mb_height-1; my++) |
2159 |
|
for (mx = 1; mx < pParam->mb_width-1; mx++) { |
2160 |
|
const MACROBLOCK *pMB = &pMBs[mx + my * pParam->mb_width]; |
2161 |
|
VECTOR mv; |
2162 |
|
|
2163 |
|
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) |
2164 |
|
continue; |
2165 |
|
|
2166 |
|
mv = pMB->mvs[0]; |
2167 |
|
if ( ABS(mv.x - x) <= step && ABS(mv.y - y) <= step ) /* GMC translation is always halfpel-res */ |
2168 |
|
count++; |
2169 |
|
} |
2170 |
|
if (count >= bestcount) { bestcount = count; gmc.x = x; gmc.y = y; } |
2171 |
|
} |
2172 |
|
min_x = gmc.x - step; |
2173 |
|
max_x = gmc.x + step; |
2174 |
|
min_y = gmc.y - step; |
2175 |
|
max_y = gmc.y + step; |
2176 |
|
|
2177 |
|
} |
2178 |
|
if (pParam->m_quarterpel) { |
2179 |
|
gmc.x *= 2; |
2180 |
|
gmc.y *= 2; /* we store the halfpel value as pseudo-qpel to make comparison easier */ |
2181 |
|
} |
2182 |
|
|
2183 |
|
if (bestcount < (pParam->mb_height-2)*(pParam->mb_width-2)/10) |
2184 |
|
gmc.x = gmc.y = 0; //no camara pan, no GMC |
2185 |
|
|
2186 |
|
return gmc; |
2187 |
|
} |