171 |
const IMAGE *const pCurrent = ¤t->image; |
const IMAGE *const pCurrent = ¤t->image; |
172 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
173 |
|
|
174 |
const VECTOR zeroMV = { 0, 0 }; |
static const VECTOR zeroMV = { 0, 0 }; |
175 |
|
VECTOR predMV; |
176 |
|
|
|
long long time; |
|
177 |
int32_t x, y; |
int32_t x, y; |
178 |
int32_t iIntra = 0; |
int32_t iIntra = 0; |
179 |
VECTOR pmv; |
VECTOR pmv; |
186 |
|
|
187 |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
188 |
|
|
189 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
190 |
|
|
191 |
pMB->sad16 = |
pMB->sad16 = |
192 |
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
193 |
y, current->motion_flags, current->quant, |
x, y, predMV.x, predMV.y, predMV.x, predMV.y, |
194 |
|
current->motion_flags, current->quant, |
195 |
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
196 |
&pMB->pmvs[0]); |
&pMB->pmvs[0]); |
197 |
|
|
223 |
pMB->dquant == NO_CHANGE)) { |
pMB->dquant == NO_CHANGE)) { |
224 |
int32_t sad8 = IMV16X16 * current->quant; |
int32_t sad8 = IMV16X16 * current->quant; |
225 |
|
|
226 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
|
|
|
227 |
sad8 += pMB->sad8[0] = |
sad8 += pMB->sad8[0] = |
228 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
229 |
pCurrent, 2 * x, 2 * y, pMB->mv16.x, |
pCurrent, 2 * x, 2 * y, |
230 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
231 |
|
current->motion_flags, |
232 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
233 |
pMBs, prevMBs, &pMB->mvs[0], |
pMBs, prevMBs, &pMB->mvs[0], |
234 |
&pMB->pmvs[0]); |
&pMB->pmvs[0]); |
235 |
|
} |
236 |
|
if (sad8 < pMB->sad16) { |
237 |
|
|
238 |
if (sad8 < pMB->sad16) |
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 1); |
239 |
sad8 += pMB->sad8[1] = |
sad8 += pMB->sad8[1] = |
240 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
241 |
pCurrent, 2 * x + 1, 2 * y, pMB->mv16.x, |
pCurrent, 2 * x + 1, 2 * y, |
242 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
243 |
|
current->motion_flags, |
244 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
245 |
pMBs, prevMBs, &pMB->mvs[1], |
pMBs, prevMBs, &pMB->mvs[1], |
246 |
&pMB->pmvs[1]); |
&pMB->pmvs[1]); |
247 |
|
} |
248 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
249 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 2); |
250 |
sad8 += pMB->sad8[2] = |
sad8 += pMB->sad8[2] = |
251 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
252 |
pCurrent, 2 * x, 2 * y + 1, pMB->mv16.x, |
pCurrent, 2 * x, 2 * y + 1, |
253 |
pMB->mv16.y, current->motion_flags, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
254 |
|
current->motion_flags, |
255 |
current->quant, current->fcode, pParam, |
current->quant, current->fcode, pParam, |
256 |
pMBs, prevMBs, &pMB->mvs[2], |
pMBs, prevMBs, &pMB->mvs[2], |
257 |
&pMB->pmvs[2]); |
&pMB->pmvs[2]); |
258 |
|
} |
259 |
if (sad8 < pMB->sad16) |
if (sad8 < pMB->sad16) { |
260 |
|
predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 3); |
261 |
sad8 += pMB->sad8[3] = |
sad8 += pMB->sad8[3] = |
262 |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
263 |
pCurrent, 2 * x + 1, 2 * y + 1, |
pCurrent, 2 * x + 1, 2 * y + 1, |
264 |
pMB->mv16.x, pMB->mv16.y, |
pMB->mv16.x, pMB->mv16.y, predMV.x, predMV.y, |
265 |
current->motion_flags, current->quant, |
current->motion_flags, |
266 |
current->fcode, pParam, pMBs, prevMBs, |
current->quant, current->fcode, pParam, |
267 |
&pMB->mvs[3], &pMB->pmvs[3]); |
pMBs, prevMBs, |
268 |
|
&pMB->mvs[3], |
269 |
|
&pMB->pmvs[3]); |
270 |
|
} |
271 |
|
|
272 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
273 |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
295 |
return 0; |
return 0; |
296 |
} |
} |
297 |
|
|
298 |
|
|
299 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
300 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
301 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
302 |
{ \ |
{ \ |
303 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, 0, 0 , iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); \ |
304 |
iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
305 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
306 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
307 |
} |
} |
308 |
|
|
309 |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
310 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
311 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
312 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
313 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
314 |
} |
} |
318 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
319 |
{ \ |
{ \ |
320 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
321 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
322 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
323 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
324 |
} |
} |
328 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
329 |
{ \ |
{ \ |
330 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
331 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
332 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
333 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
334 |
} |
} |
338 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
339 |
{ \ |
{ \ |
340 |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
iSAD = sad16( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 16, X, Y, iEdgedWidth),iEdgedWidth, iMinSAD); \ |
341 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_16((X) - center_x, (Y) - center_y, (uint8_t)iFcode, iQuant);\ |
342 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
343 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
344 |
} |
} |
346 |
|
|
347 |
#define CHECK_MV8_ZERO {\ |
#define CHECK_MV8_ZERO {\ |
348 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, 0, 0 , iEdgedWidth), iEdgedWidth); \ |
349 |
iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8(-center_x, -center_y, (uint8_t)iFcode, iQuant);\ |
350 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
351 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
352 |
} |
} |
354 |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
355 |
{ \ |
{ \ |
356 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
357 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
358 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
359 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
360 |
} |
} |
364 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
365 |
{ \ |
{ \ |
366 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
367 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
368 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
369 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
370 |
} |
} |
374 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
375 |
{ \ |
{ \ |
376 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
377 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
378 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
379 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
380 |
} |
} |
384 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
385 |
{ \ |
{ \ |
386 |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
iSAD = sad8( cur, get_ref(pRef, pRefH, pRefV, pRefHV, x, y, 8, (X), (Y), iEdgedWidth),iEdgedWidth); \ |
387 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
iSAD += calc_delta_8((X)-center_x, (Y)-center_y, (uint8_t)iFcode, iQuant);\ |
388 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
389 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
390 |
} |
} |
439 |
const uint8_t * const cur, |
const uint8_t * const cur, |
440 |
const int x, |
const int x, |
441 |
const int y, |
const int y, |
442 |
int32_t startx, |
const int start_x, |
443 |
int32_t starty, |
const int start_y, |
444 |
int32_t iMinSAD, |
int iMinSAD, |
445 |
VECTOR * const currMV, |
VECTOR * const currMV, |
446 |
const VECTOR * const pmv, |
const int center_x, |
447 |
|
const int center_y, |
448 |
const int32_t min_dx, |
const int32_t min_dx, |
449 |
const int32_t max_dx, |
const int32_t max_dx, |
450 |
const int32_t min_dy, |
const int32_t min_dy, |
458 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
459 |
|
|
460 |
int32_t iDirection = 0; |
int32_t iDirection = 0; |
461 |
|
int32_t iDirectionBackup; |
462 |
int32_t iSAD; |
int32_t iSAD; |
463 |
VECTOR backupMV; |
VECTOR backupMV; |
464 |
|
|
465 |
backupMV.x = startx; |
backupMV.x = start_x; |
466 |
backupMV.y = starty; |
backupMV.y = start_y; |
467 |
|
|
468 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
469 |
|
|
472 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
473 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
474 |
|
|
475 |
if (iDirection) |
if (iDirection) { |
476 |
while (!iFound) { |
while (!iFound) { |
477 |
iFound = 1; |
iFound = 1; |
478 |
backupMV = *currMV; |
backupMV = *currMV; |
479 |
|
iDirectionBackup = iDirection; |
480 |
|
|
481 |
if (iDirection != 2) |
if (iDirectionBackup != 2) |
482 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
483 |
backupMV.y, 1); |
backupMV.y, 1); |
484 |
if (iDirection != 1) |
if (iDirectionBackup != 1) |
485 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
486 |
backupMV.y, 2); |
backupMV.y, 2); |
487 |
if (iDirection != 4) |
if (iDirectionBackup != 4) |
488 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
489 |
backupMV.y - iDiamondSize, 3); |
backupMV.y - iDiamondSize, 3); |
490 |
if (iDirection != 3) |
if (iDirectionBackup != 3) |
491 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
492 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
493 |
|
} |
494 |
} else { |
} else { |
495 |
currMV->x = startx; |
currMV->x = start_x; |
496 |
currMV->y = starty; |
currMV->y = start_y; |
497 |
} |
} |
498 |
return iMinSAD; |
return iMinSAD; |
499 |
} |
} |
506 |
const uint8_t * const cur, |
const uint8_t * const cur, |
507 |
const int x, |
const int x, |
508 |
const int y, |
const int y, |
509 |
int32_t startx, |
const int start_x, |
510 |
int32_t starty, |
const int start_y, |
511 |
int32_t iMinSAD, |
int iMinSAD, |
512 |
VECTOR * const currMV, |
VECTOR * const currMV, |
513 |
const VECTOR * const pmv, |
const int center_x, |
514 |
|
const int center_y, |
515 |
const int32_t min_dx, |
const int32_t min_dx, |
516 |
const int32_t max_dx, |
const int32_t max_dx, |
517 |
const int32_t min_dy, |
const int32_t min_dy, |
528 |
int32_t iSAD; |
int32_t iSAD; |
529 |
VECTOR backupMV; |
VECTOR backupMV; |
530 |
|
|
531 |
backupMV.x = startx; |
backupMV.x = start_x; |
532 |
backupMV.y = starty; |
backupMV.y = start_y; |
533 |
|
|
534 |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
/* It's one search with full square pattern, and new parts for all following diamonds */ |
535 |
|
|
554 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
555 |
|
|
556 |
|
|
557 |
if (iDirection) |
if (iDirection) { |
558 |
while (!iFound) { |
while (!iFound) { |
559 |
iFound = 1; |
iFound = 1; |
560 |
backupMV = *currMV; |
backupMV = *currMV; |
563 |
case 1: |
case 1: |
564 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
565 |
backupMV.y, 1); |
backupMV.y, 1); |
566 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
567 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
568 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
569 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
570 |
break; |
break; |
571 |
case 2: |
case 2: |
572 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
573 |
2); |
2); |
574 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
575 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
576 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
577 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
578 |
break; |
break; |
579 |
|
|
580 |
case 3: |
case 3: |
581 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
582 |
4); |
4); |
583 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
584 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
585 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
586 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
587 |
break; |
break; |
588 |
|
|
589 |
case 4: |
case 4: |
590 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
591 |
3); |
3); |
592 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
593 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
594 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
595 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
596 |
break; |
break; |
597 |
|
|
598 |
case 5: |
case 5: |
599 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
600 |
1); |
1); |
601 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
602 |
3); |
3); |
603 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
604 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
605 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
606 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
607 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
608 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
609 |
break; |
break; |
610 |
|
|
611 |
case 6: |
case 6: |
612 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
613 |
2); |
2); |
614 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
615 |
3); |
3); |
616 |
|
|
617 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
618 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
619 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
620 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
621 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
622 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
623 |
|
|
624 |
break; |
break; |
626 |
case 7: |
case 7: |
627 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
628 |
backupMV.y, 1); |
backupMV.y, 1); |
629 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
630 |
4); |
4); |
631 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
632 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
633 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
634 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
635 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
636 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
637 |
break; |
break; |
638 |
|
|
639 |
case 8: |
case 8: |
640 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
641 |
2); |
2); |
642 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
643 |
4); |
4); |
644 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
645 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
646 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
647 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
648 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
649 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
650 |
break; |
break; |
651 |
default: |
default: |
652 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
653 |
1); |
1); |
654 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
655 |
2); |
2); |
656 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
657 |
3); |
3); |
658 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
659 |
4); |
4); |
660 |
|
|
661 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
662 |
backupMV.y - iDiamondSize, 5); |
backupMV.y - iDiamondSize, 5); |
663 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
664 |
backupMV.y + iDiamondSize, 6); |
backupMV.y + iDiamondSize, 6); |
665 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
666 |
backupMV.y - iDiamondSize, 7); |
backupMV.y - iDiamondSize, 7); |
667 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
668 |
backupMV.y + iDiamondSize, 8); |
backupMV.y + iDiamondSize, 8); |
669 |
break; |
break; |
670 |
} |
} |
671 |
|
} |
672 |
} else { |
} else { |
673 |
currMV->x = startx; |
currMV->x = start_x; |
674 |
currMV->y = starty; |
currMV->y = start_y; |
675 |
} |
} |
676 |
return iMinSAD; |
return iMinSAD; |
677 |
} |
} |
685 |
const uint8_t * const cur, |
const uint8_t * const cur, |
686 |
const int x, |
const int x, |
687 |
const int y, |
const int y, |
688 |
int32_t startx, |
const int start_x, |
689 |
int32_t starty, |
const int start_y, |
690 |
int32_t iMinSAD, |
int iMinSAD, |
691 |
VECTOR * const currMV, |
VECTOR * const currMV, |
692 |
const VECTOR * const pmv, |
const int center_x, |
693 |
|
const int center_y, |
694 |
const int32_t min_dx, |
const int32_t min_dx, |
695 |
const int32_t max_dx, |
const int32_t max_dx, |
696 |
const int32_t min_dy, |
const int32_t min_dy, |
705 |
int32_t dx, dy; |
int32_t dx, dy; |
706 |
VECTOR backupMV; |
VECTOR backupMV; |
707 |
|
|
708 |
backupMV.x = startx; |
backupMV.x = start_x; |
709 |
backupMV.y = starty; |
backupMV.y = start_y; |
710 |
|
|
711 |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
712 |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
723 |
const uint8_t * const cur, |
const uint8_t * const cur, |
724 |
const int x, |
const int x, |
725 |
const int y, |
const int y, |
726 |
int32_t startx, |
int start_x, |
727 |
int32_t starty, |
int start_y, |
728 |
int32_t iMinSAD, |
int iMinSAD, |
729 |
VECTOR * const currMV, |
VECTOR * const currMV, |
730 |
const VECTOR * const pmv, |
const int center_x, |
731 |
|
const int center_y, |
732 |
const int32_t min_dx, |
const int32_t min_dx, |
733 |
const int32_t max_dx, |
const int32_t max_dx, |
734 |
const int32_t min_dy, |
const int32_t min_dy, |
745 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
746 |
|
|
747 |
if (iDirection) { |
if (iDirection) { |
748 |
CHECK_MV16_CANDIDATE(startx - iDiamondSize, starty); |
CHECK_MV16_CANDIDATE(start_x - iDiamondSize, start_y); |
749 |
CHECK_MV16_CANDIDATE(startx + iDiamondSize, starty); |
CHECK_MV16_CANDIDATE(start_x + iDiamondSize, start_y); |
750 |
CHECK_MV16_CANDIDATE(startx, starty - iDiamondSize); |
CHECK_MV16_CANDIDATE(start_x, start_y - iDiamondSize); |
751 |
CHECK_MV16_CANDIDATE(startx, starty + iDiamondSize); |
CHECK_MV16_CANDIDATE(start_x, start_y + iDiamondSize); |
752 |
} else { |
} else { |
753 |
int bDirection = 1 + 2 + 4 + 8; |
int bDirection = 1 + 2 + 4 + 8; |
754 |
|
|
755 |
do { |
do { |
756 |
iDirection = 0; |
iDirection = 0; |
757 |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
758 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
759 |
|
|
760 |
if (bDirection & 2) |
if (bDirection & 2) |
761 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
762 |
|
|
763 |
if (bDirection & 4) |
if (bDirection & 4) |
764 |
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
765 |
|
|
766 |
if (bDirection & 8) |
if (bDirection & 8) |
767 |
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
768 |
|
|
769 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
770 |
|
|
772 |
{ |
{ |
773 |
bDirection = iDirection; |
bDirection = iDirection; |
774 |
iDirection = 0; |
iDirection = 0; |
775 |
startx = currMV->x; |
start_x = currMV->x; |
776 |
starty = currMV->y; |
start_y = currMV->y; |
777 |
if (bDirection & 3) //our candidate is left or right |
if (bDirection & 3) //our candidate is left or right |
778 |
{ |
{ |
779 |
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
780 |
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV16_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
781 |
} else // what remains here is up or down |
} else // what remains here is up or down |
782 |
{ |
{ |
783 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
784 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
785 |
} |
} |
786 |
|
|
787 |
if (iDirection) { |
if (iDirection) { |
788 |
bDirection += iDirection; |
bDirection += iDirection; |
789 |
startx = currMV->x; |
start_x = currMV->x; |
790 |
starty = currMV->y; |
start_y = currMV->y; |
791 |
} |
} |
792 |
} else //about to quit, eh? not so fast.... |
} else //about to quit, eh? not so fast.... |
793 |
{ |
{ |
794 |
switch (bDirection) { |
switch (bDirection) { |
795 |
case 2: |
case 2: |
796 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
797 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
798 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
799 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
800 |
break; |
break; |
801 |
case 1: |
case 1: |
802 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
|
803 |
starty - iDiamondSize, 1 + 4); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
804 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
start_y - iDiamondSize, 1 + 4); |
805 |
starty + iDiamondSize, 1 + 8); |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
806 |
|
start_y + iDiamondSize, 1 + 8); |
807 |
break; |
break; |
808 |
case 2 + 4: |
case 2 + 4: |
809 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
810 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
811 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
812 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
813 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
814 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
815 |
break; |
break; |
816 |
case 4: |
case 4: |
817 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
818 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
819 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
820 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
821 |
break; |
break; |
822 |
case 8: |
case 8: |
823 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
824 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
825 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
826 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
827 |
break; |
break; |
828 |
case 1 + 4: |
case 1 + 4: |
829 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
830 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
831 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
832 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
833 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
834 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
835 |
break; |
break; |
836 |
case 2 + 8: |
case 2 + 8: |
837 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
838 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
839 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
840 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
841 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
842 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
843 |
break; |
break; |
844 |
case 1 + 8: |
case 1 + 8: |
845 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
846 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
847 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
848 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
849 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
850 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
851 |
break; |
break; |
852 |
default: //1+2+4+8 == we didn't find anything at all |
default: //1+2+4+8 == we didn't find anything at all |
853 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
854 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
855 |
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x - iDiamondSize, |
856 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
857 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
858 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
859 |
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV16_CANDIDATE_DIR(start_x + iDiamondSize, |
860 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
861 |
break; |
break; |
862 |
} |
} |
863 |
if (!iDirection) |
if (!iDirection) |
864 |
break; //ok, the end. really |
break; //ok, the end. really |
865 |
else { |
else { |
866 |
bDirection = iDirection; |
bDirection = iDirection; |
867 |
startx = currMV->x; |
start_x = currMV->x; |
868 |
starty = currMV->y; |
start_y = currMV->y; |
869 |
} |
} |
870 |
} |
} |
871 |
} |
} |
874 |
return iMinSAD; |
return iMinSAD; |
875 |
} |
} |
876 |
|
|
877 |
|
#define CHECK_MV16_F_INTERPOL(X,Y) { \ |
878 |
|
if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ |
879 |
|
&& ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ |
880 |
|
{ \ |
881 |
|
iSAD = sad16bi( cur, \ |
882 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
883 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ |
884 |
|
iEdgedWidth); \ |
885 |
|
iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
886 |
|
iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
887 |
|
if (iSAD < iMinSAD) \ |
888 |
|
{ iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); } } \ |
889 |
|
} |
890 |
|
|
891 |
|
#define CHECK_MV16_F_INTERPOL_FOUND(X,Y) { \ |
892 |
|
if ( ((X) <= f_max_dx) && ((X) >= f_min_dx) \ |
893 |
|
&& ((Y) <= f_max_dy) && ((Y) >= f_min_dy) ) \ |
894 |
|
{ \ |
895 |
|
iSAD = sad16bi( cur, \ |
896 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
897 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, b_currMV->x, b_currMV->y, iEdgedWidth), \ |
898 |
|
iEdgedWidth); \ |
899 |
|
iSAD += calc_delta_16((X) - f_center_x, (Y) - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
900 |
|
iSAD += calc_delta_16(b_currMV->x - b_center_x, b_currMV->y - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
901 |
|
if (iSAD < iMinSAD) \ |
902 |
|
{ iMinSAD=iSAD; f_currMV->x=(X); f_currMV->y=(Y); iFound=0;} } \ |
903 |
|
} |
904 |
|
|
905 |
|
#define CHECK_MV16_B_INTERPOL(X,Y) { \ |
906 |
|
if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ |
907 |
|
&& ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ |
908 |
|
{ \ |
909 |
|
iSAD = sad16bi( cur, \ |
910 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ |
911 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
912 |
|
iEdgedWidth); \ |
913 |
|
iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
914 |
|
iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
915 |
|
if (iSAD < iMinSAD) \ |
916 |
|
{ iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); } } \ |
917 |
|
} |
918 |
|
|
919 |
|
#define CHECK_MV16_B_INTERPOL_FOUND(X,Y) { \ |
920 |
|
if ( ((X) <= b_max_dx) && ((X) >= b_min_dx) \ |
921 |
|
&& ((Y) <= b_max_dy) && ((Y) >= b_min_dy) ) \ |
922 |
|
{ \ |
923 |
|
iSAD = sad16bi( cur, \ |
924 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, x, y, 16, f_currMV->x, f_currMV->y, iEdgedWidth), \ |
925 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, x, y, 16, X, Y, iEdgedWidth), \ |
926 |
|
iEdgedWidth); \ |
927 |
|
iSAD += calc_delta_16(f_currMV->x - f_center_x, f_currMV->y - f_center_y, (uint8_t)f_iFcode, iQuant);\ |
928 |
|
iSAD += calc_delta_16((X) - b_center_x, (Y) - b_center_y, (uint8_t)b_iFcode, iQuant);\ |
929 |
|
if (iSAD < iMinSAD) \ |
930 |
|
{ iMinSAD=iSAD; b_currMV->x=(X); b_currMV->y=(Y); iFound=0;} } \ |
931 |
|
} |
932 |
|
|
933 |
|
int32_t |
934 |
|
Diamond16_InterpolMainSearch( |
935 |
|
const uint8_t * const f_pRef, |
936 |
|
const uint8_t * const f_pRefH, |
937 |
|
const uint8_t * const f_pRefV, |
938 |
|
const uint8_t * const f_pRefHV, |
939 |
|
|
940 |
|
const uint8_t * const cur, |
941 |
|
|
942 |
|
const uint8_t * const b_pRef, |
943 |
|
const uint8_t * const b_pRefH, |
944 |
|
const uint8_t * const b_pRefV, |
945 |
|
const uint8_t * const b_pRefHV, |
946 |
|
|
947 |
|
const int x, |
948 |
|
const int y, |
949 |
|
|
950 |
|
const int f_start_x, |
951 |
|
const int f_start_y, |
952 |
|
const int b_start_x, |
953 |
|
const int b_start_y, |
954 |
|
|
955 |
|
int iMinSAD, |
956 |
|
VECTOR * const f_currMV, |
957 |
|
VECTOR * const b_currMV, |
958 |
|
|
959 |
|
const int f_center_x, |
960 |
|
const int f_center_y, |
961 |
|
const int b_center_x, |
962 |
|
const int b_center_y, |
963 |
|
|
964 |
|
const int32_t f_min_dx, |
965 |
|
const int32_t f_max_dx, |
966 |
|
const int32_t f_min_dy, |
967 |
|
const int32_t f_max_dy, |
968 |
|
|
969 |
|
const int32_t b_min_dx, |
970 |
|
const int32_t b_max_dx, |
971 |
|
const int32_t b_min_dy, |
972 |
|
const int32_t b_max_dy, |
973 |
|
|
974 |
|
const int32_t iEdgedWidth, |
975 |
|
const int32_t iDiamondSize, |
976 |
|
|
977 |
|
const int32_t f_iFcode, |
978 |
|
const int32_t b_iFcode, |
979 |
|
|
980 |
|
const int32_t iQuant, |
981 |
|
int iFound) |
982 |
|
{ |
983 |
|
/* Do a diamond search around given starting point, return SAD of best */ |
984 |
|
|
985 |
|
int32_t iSAD; |
986 |
|
|
987 |
|
VECTOR f_backupMV; |
988 |
|
VECTOR b_backupMV; |
989 |
|
|
990 |
|
f_currMV->x = f_start_x; |
991 |
|
f_currMV->y = f_start_y; |
992 |
|
b_currMV->x = b_start_x; |
993 |
|
b_currMV->y = b_start_y; |
994 |
|
|
995 |
|
do |
996 |
|
{ |
997 |
|
iFound = 1; |
998 |
|
|
999 |
|
f_backupMV = *f_currMV; |
1000 |
|
|
1001 |
|
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x - iDiamondSize, f_backupMV.y); |
1002 |
|
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x + iDiamondSize, f_backupMV.y); |
1003 |
|
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y - iDiamondSize); |
1004 |
|
CHECK_MV16_F_INTERPOL_FOUND(f_backupMV.x, f_backupMV.y + iDiamondSize); |
1005 |
|
|
1006 |
|
b_backupMV = *b_currMV; |
1007 |
|
|
1008 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x - iDiamondSize, b_backupMV.y); |
1009 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x + iDiamondSize, b_backupMV.y); |
1010 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y - iDiamondSize); |
1011 |
|
CHECK_MV16_B_INTERPOL_FOUND(b_backupMV.x, b_backupMV.y + iDiamondSize); |
1012 |
|
|
1013 |
|
} while (!iFound); |
1014 |
|
|
1015 |
|
return iMinSAD; |
1016 |
|
} |
1017 |
|
|
1018 |
|
/* Sorry, these MACROS really got too large... I'll turn them into function soon! */ |
1019 |
|
|
1020 |
|
#define CHECK_MV16_DIRECT_FOUND(X,Y) \ |
1021 |
|
if ( (X)>=(-32) && (X)<=(31) && ((Y)>=-32) && ((Y)<=31) ) \ |
1022 |
|
{ int k;\ |
1023 |
|
VECTOR mvs,b_mvs; \ |
1024 |
|
iSAD = 0;\ |
1025 |
|
for (k = 0; k < 4; k++) { \ |
1026 |
|
mvs.x = (int32_t) ((TRB * directmv[k].x) / TRD + (X)); \ |
1027 |
|
b_mvs.x = (int32_t) (((X) == 0) \ |
1028 |
|
? ((TRB - TRD) * directmv[k].x) / TRD \ |
1029 |
|
: mvs.x - directmv[k].x); \ |
1030 |
|
\ |
1031 |
|
mvs.y = (int32_t) ((TRB * directmv[k].y) / TRD + (Y)); \ |
1032 |
|
b_mvs.y = (int32_t) (((Y) == 0) \ |
1033 |
|
? ((TRB - TRD) * directmv[k].y) / TRD \ |
1034 |
|
: mvs.y - directmv[k].y); \ |
1035 |
|
\ |
1036 |
|
if ( (mvs.x <= max_dx) && (mvs.x >= min_dx) \ |
1037 |
|
&& (mvs.y <= max_dy) && (mvs.y >= min_dy) \ |
1038 |
|
&& (b_mvs.x <= max_dx) && (b_mvs.x >= min_dx) \ |
1039 |
|
&& (b_mvs.y <= max_dy) && (b_mvs.y >= min_dy) ) { \ |
1040 |
|
iSAD += sad8bi( cur + 8*(k&1) + 8*(k>>1)*iEdgedWidth, \ |
1041 |
|
get_ref(f_pRef, f_pRefH, f_pRefV, f_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ |
1042 |
|
mvs.x, mvs.y, iEdgedWidth), \ |
1043 |
|
get_ref(b_pRef, b_pRefH, b_pRefV, b_pRefHV, 2*x+(k&1), 2*y+(k>>1), 8, \ |
1044 |
|
b_mvs.x, b_mvs.y, iEdgedWidth), \ |
1045 |
|
iEdgedWidth); \ |
1046 |
|
} \ |
1047 |
|
else \ |
1048 |
|
iSAD = 65535; \ |
1049 |
|
} \ |
1050 |
|
iSAD += calc_delta_16((X),(Y), 1, iQuant);\ |
1051 |
|
if (iSAD < iMinSAD) \ |
1052 |
|
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iFound=0; } \ |
1053 |
|
} |
1054 |
|
|
1055 |
|
|
1056 |
|
|
1057 |
|
int32_t |
1058 |
|
Diamond16_DirectMainSearch( |
1059 |
|
const uint8_t * const f_pRef, |
1060 |
|
const uint8_t * const f_pRefH, |
1061 |
|
const uint8_t * const f_pRefV, |
1062 |
|
const uint8_t * const f_pRefHV, |
1063 |
|
|
1064 |
|
const uint8_t * const cur, |
1065 |
|
|
1066 |
|
const uint8_t * const b_pRef, |
1067 |
|
const uint8_t * const b_pRefH, |
1068 |
|
const uint8_t * const b_pRefV, |
1069 |
|
const uint8_t * const b_pRefHV, |
1070 |
|
|
1071 |
|
const int x, |
1072 |
|
const int y, |
1073 |
|
|
1074 |
|
const int TRB, |
1075 |
|
const int TRD, |
1076 |
|
|
1077 |
|
const int start_x, |
1078 |
|
const int start_y, |
1079 |
|
|
1080 |
|
int iMinSAD, |
1081 |
|
VECTOR * const currMV, |
1082 |
|
const VECTOR * const directmv, |
1083 |
|
|
1084 |
|
const int32_t min_dx, |
1085 |
|
const int32_t max_dx, |
1086 |
|
const int32_t min_dy, |
1087 |
|
const int32_t max_dy, |
1088 |
|
|
1089 |
|
const int32_t iEdgedWidth, |
1090 |
|
const int32_t iDiamondSize, |
1091 |
|
|
1092 |
|
const int32_t iQuant, |
1093 |
|
int iFound) |
1094 |
|
{ |
1095 |
|
/* Do a diamond search around given starting point, return SAD of best */ |
1096 |
|
|
1097 |
|
int32_t iSAD; |
1098 |
|
|
1099 |
|
VECTOR backupMV; |
1100 |
|
|
1101 |
|
currMV->x = start_x; |
1102 |
|
currMV->y = start_y; |
1103 |
|
|
1104 |
|
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1105 |
|
|
1106 |
|
do |
1107 |
|
{ |
1108 |
|
iFound = 1; |
1109 |
|
|
1110 |
|
backupMV = *currMV; |
1111 |
|
|
1112 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x - iDiamondSize, backupMV.y); |
1113 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x + iDiamondSize, backupMV.y); |
1114 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y - iDiamondSize); |
1115 |
|
CHECK_MV16_DIRECT_FOUND(backupMV.x, backupMV.y + iDiamondSize); |
1116 |
|
|
1117 |
|
} while (!iFound); |
1118 |
|
|
1119 |
|
return iMinSAD; |
1120 |
|
} |
1121 |
|
|
1122 |
|
|
1123 |
int32_t |
int32_t |
1124 |
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
1125 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1128 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1129 |
const int x, |
const int x, |
1130 |
const int y, |
const int y, |
1131 |
int32_t startx, |
int start_x, |
1132 |
int32_t starty, |
int start_y, |
1133 |
int32_t iMinSAD, |
int iMinSAD, |
1134 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1135 |
const VECTOR * const pmv, |
const int center_x, |
1136 |
|
const int center_y, |
1137 |
const int32_t min_dx, |
const int32_t min_dx, |
1138 |
const int32_t max_dx, |
const int32_t max_dx, |
1139 |
const int32_t min_dy, |
const int32_t min_dy, |
1150 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
1151 |
|
|
1152 |
if (iDirection) { |
if (iDirection) { |
1153 |
CHECK_MV8_CANDIDATE(startx - iDiamondSize, starty); |
CHECK_MV8_CANDIDATE(start_x - iDiamondSize, start_y); |
1154 |
CHECK_MV8_CANDIDATE(startx + iDiamondSize, starty); |
CHECK_MV8_CANDIDATE(start_x + iDiamondSize, start_y); |
1155 |
CHECK_MV8_CANDIDATE(startx, starty - iDiamondSize); |
CHECK_MV8_CANDIDATE(start_x, start_y - iDiamondSize); |
1156 |
CHECK_MV8_CANDIDATE(startx, starty + iDiamondSize); |
CHECK_MV8_CANDIDATE(start_x, start_y + iDiamondSize); |
1157 |
} else { |
} else { |
1158 |
int bDirection = 1 + 2 + 4 + 8; |
int bDirection = 1 + 2 + 4 + 8; |
1159 |
|
|
1160 |
do { |
do { |
1161 |
iDirection = 0; |
iDirection = 0; |
1162 |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
if (bDirection & 1) //we only want to check left if we came from the right (our last motion was to the left, up-left or down-left) |
1163 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
1164 |
|
|
1165 |
if (bDirection & 2) |
if (bDirection & 2) |
1166 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1167 |
|
|
1168 |
if (bDirection & 4) |
if (bDirection & 4) |
1169 |
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
1170 |
|
|
1171 |
if (bDirection & 8) |
if (bDirection & 8) |
1172 |
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1173 |
|
|
1174 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
1175 |
|
|
1177 |
{ |
{ |
1178 |
bDirection = iDirection; |
bDirection = iDirection; |
1179 |
iDirection = 0; |
iDirection = 0; |
1180 |
startx = currMV->x; |
start_x = currMV->x; |
1181 |
starty = currMV->y; |
start_y = currMV->y; |
1182 |
if (bDirection & 3) //our candidate is left or right |
if (bDirection & 3) //our candidate is left or right |
1183 |
{ |
{ |
1184 |
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
CHECK_MV8_CANDIDATE_DIR(start_x, start_y + iDiamondSize, 8); |
1185 |
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
CHECK_MV8_CANDIDATE_DIR(start_x, start_y - iDiamondSize, 4); |
1186 |
} else // what remains here is up or down |
} else // what remains here is up or down |
1187 |
{ |
{ |
1188 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, start_y, 2); |
1189 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, start_y, 1); |
1190 |
} |
} |
1191 |
|
|
1192 |
if (iDirection) { |
if (iDirection) { |
1193 |
bDirection += iDirection; |
bDirection += iDirection; |
1194 |
startx = currMV->x; |
start_x = currMV->x; |
1195 |
starty = currMV->y; |
start_y = currMV->y; |
1196 |
} |
} |
1197 |
} else //about to quit, eh? not so fast.... |
} else //about to quit, eh? not so fast.... |
1198 |
{ |
{ |
1199 |
switch (bDirection) { |
switch (bDirection) { |
1200 |
case 2: |
case 2: |
1201 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1202 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1203 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1204 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1205 |
break; |
break; |
1206 |
case 1: |
case 1: |
1207 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1208 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1209 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1210 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1211 |
break; |
break; |
1212 |
case 2 + 4: |
case 2 + 4: |
1213 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1214 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1215 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1216 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1217 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1218 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1219 |
break; |
break; |
1220 |
case 4: |
case 4: |
1221 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1222 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1223 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1224 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1225 |
break; |
break; |
1226 |
case 8: |
case 8: |
1227 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1228 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1229 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1230 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1231 |
break; |
break; |
1232 |
case 1 + 4: |
case 1 + 4: |
1233 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1234 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1235 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1236 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1237 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1238 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1239 |
break; |
break; |
1240 |
case 2 + 8: |
case 2 + 8: |
1241 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1242 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1243 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1244 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1245 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1246 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1247 |
break; |
break; |
1248 |
case 1 + 8: |
case 1 + 8: |
1249 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1250 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1251 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1252 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1253 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1254 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1255 |
break; |
break; |
1256 |
default: //1+2+4+8 == we didn't find anything at all |
default: //1+2+4+8 == we didn't find anything at all |
1257 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1258 |
starty - iDiamondSize, 1 + 4); |
start_y - iDiamondSize, 1 + 4); |
1259 |
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x - iDiamondSize, |
1260 |
starty + iDiamondSize, 1 + 8); |
start_y + iDiamondSize, 1 + 8); |
1261 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1262 |
starty - iDiamondSize, 2 + 4); |
start_y - iDiamondSize, 2 + 4); |
1263 |
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
CHECK_MV8_CANDIDATE_DIR(start_x + iDiamondSize, |
1264 |
starty + iDiamondSize, 2 + 8); |
start_y + iDiamondSize, 2 + 8); |
1265 |
break; |
break; |
1266 |
} |
} |
1267 |
if (!(iDirection)) |
if (!(iDirection)) |
1268 |
break; //ok, the end. really |
break; //ok, the end. really |
1269 |
else { |
else { |
1270 |
bDirection = iDirection; |
bDirection = iDirection; |
1271 |
startx = currMV->x; |
start_x = currMV->x; |
1272 |
starty = currMV->y; |
start_y = currMV->y; |
1273 |
} |
} |
1274 |
} |
} |
1275 |
} |
} |
1287 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1288 |
const int x, |
const int x, |
1289 |
const int y, |
const int y, |
1290 |
int32_t startx, |
const int start_x, |
1291 |
int32_t starty, |
const int start_y, |
1292 |
int32_t iMinSAD, |
int iMinSAD, |
1293 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1294 |
const VECTOR * const pmv, |
const int center_x, |
1295 |
|
const int center_y, |
1296 |
const int32_t min_dx, |
const int32_t min_dx, |
1297 |
const int32_t max_dx, |
const int32_t max_dx, |
1298 |
const int32_t min_dy, |
const int32_t min_dy, |
1307 |
int32_t dx, dy; |
int32_t dx, dy; |
1308 |
VECTOR backupMV; |
VECTOR backupMV; |
1309 |
|
|
1310 |
backupMV.x = startx; |
backupMV.x = start_x; |
1311 |
backupMV.y = starty; |
backupMV.y = start_y; |
1312 |
|
|
1313 |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
for (dx = min_dx; dx <= max_dx; dx += iDiamondSize) |
1314 |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
for (dy = min_dy; dy <= max_dy; dy += iDiamondSize) |
1317 |
return iMinSAD; |
return iMinSAD; |
1318 |
} |
} |
1319 |
|
|
1320 |
|
Halfpel8_RefineFuncPtr Halfpel8_Refine; |
1321 |
|
|
1322 |
int32_t |
int32_t |
1323 |
Halfpel16_Refine(const uint8_t * const pRef, |
Halfpel16_Refine(const uint8_t * const pRef, |
1329 |
const int y, |
const int y, |
1330 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1331 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1332 |
const VECTOR * const pmv, |
const int center_x, |
1333 |
|
const int center_y, |
1334 |
const int32_t min_dx, |
const int32_t min_dx, |
1335 |
const int32_t max_dx, |
const int32_t max_dx, |
1336 |
const int32_t min_dy, |
const int32_t min_dy, |
1359 |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
1360 |
|
|
1361 |
|
|
1362 |
|
|
1363 |
int32_t |
int32_t |
1364 |
PMVfastSearch16(const uint8_t * const pRef, |
PMVfastSearch16(const uint8_t * const pRef, |
1365 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1368 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1369 |
const int x, |
const int x, |
1370 |
const int y, |
const int y, |
1371 |
|
const int start_x, /* start is searched first, so it should contain the most */ |
1372 |
|
const int start_y, /* likely motion vector for this block */ |
1373 |
|
const int center_x, /* center is from where length of MVs is measured */ |
1374 |
|
const int center_y, |
1375 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1376 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1377 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1425 |
} |
} |
1426 |
|
|
1427 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
1428 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
1429 |
// bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
1430 |
|
|
|
/* fprintf(stderr,"pmv: %d %d / %d --- %d %d %d %d %d %d - %d %d %d\n", |
|
|
pmv[0].x,pmv[0].y,psad[0], |
|
|
pmv[1].x,pmv[1].y,pmv[2].x,pmv[2].y,pmv[3].x,pmv[3].y, |
|
|
psad[1],psad[2],psad[3]); |
|
|
*/ |
|
1431 |
if ((x == 0) && (y == 0)) { |
if ((x == 0) && (y == 0)) { |
1432 |
threshA = 512; |
threshA = 512; |
1433 |
threshB = 1024; |
threshB = 1024; |
1451 |
If SAD<=256 goto Step 10. |
If SAD<=256 goto Step 10. |
1452 |
*/ |
*/ |
1453 |
|
|
1454 |
*currMV = pmv[0]; /* current best := prediction */ |
currMV->x = start_x; |
1455 |
|
currMV->y = start_y; |
1456 |
|
|
1457 |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
1458 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
1459 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
1477 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
1478 |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
1479 |
iMinSAD += |
iMinSAD += |
1480 |
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
1481 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
1482 |
|
|
1483 |
if ((iMinSAD < 256) || |
if ((iMinSAD < 256) || |
1484 |
((MVequal(*currMV, prevMB->mvs[0])) && |
((MVequal(*currMV, prevMB->mvs[0])) && |
1485 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
1486 |
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
1487 |
{ |
{ |
1488 |
if (!MVzero(*currMV)) { |
if (!MVzero(*currMV)) { |
1586 |
|
|
1587 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
1588 |
(MVequal(*currMV, prevMB->mvs[0]) && |
(MVequal(*currMV, prevMB->mvs[0]) && |
1589 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
1590 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1591 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1592 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1614 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1615 |
|
|
1616 |
|
|
|
// fprintf(stderr,"Entering Diamond %d %d (%d):\n",x,y,iMinSAD); |
|
|
|
|
1617 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
1618 |
iSAD = |
iSAD = |
1619 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1620 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->x, currMV->y, iMinSAD, &newMV, center_x, center_y, |
1621 |
|
min_dx, max_dx, |
1622 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1623 |
iQuant, iFound); |
iQuant, iFound); |
1624 |
|
|
1633 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
1634 |
iSAD = |
iSAD = |
1635 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1636 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
center_x, center_y, iMinSAD, &newMV, center_x, center_y, |
1637 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1638 |
iDiamondSize, iFcode, iQuant, iFound); |
iDiamondSize, iFcode, iQuant, iFound); |
1639 |
|
|
1646 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1647 |
iSAD = |
iSAD = |
1648 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1649 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, |
1650 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dx, max_dx, min_dy, max_dy, |
1651 |
|
iEdgedWidth, iDiamondSize, iFcode, |
1652 |
iQuant, iFound); |
iQuant, iFound); |
1653 |
|
|
1654 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
1666 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1667 |
iMinSAD = |
iMinSAD = |
1668 |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1669 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
1670 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1671 |
|
|
|
/*fprintf(stderr,"Chosen for %d %d: %d %d - %d %d\n",x,y,currMV->x,currMV->y,pmv[0].x,pmv[0].y); |
|
|
*/ |
|
1672 |
PMVfast16_Terminate_without_Refine: |
PMVfast16_Terminate_without_Refine: |
1673 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
1674 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
1675 |
return iMinSAD; |
return iMinSAD; |
1676 |
} |
} |
1677 |
|
|
1688 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1689 |
const int x, |
const int x, |
1690 |
const int y, |
const int y, |
1691 |
int32_t startx, |
int32_t start_x, |
1692 |
int32_t starty, |
int32_t start_y, |
1693 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1694 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1695 |
const VECTOR * const pmv, |
const int center_x, |
1696 |
|
const int center_y, |
1697 |
const int32_t min_dx, |
const int32_t min_dx, |
1698 |
const int32_t max_dx, |
const int32_t max_dx, |
1699 |
const int32_t min_dy, |
const int32_t min_dy, |
1707 |
/* Do a diamond search around given starting point, return SAD of best */ |
/* Do a diamond search around given starting point, return SAD of best */ |
1708 |
|
|
1709 |
int32_t iDirection = 0; |
int32_t iDirection = 0; |
1710 |
|
int32_t iDirectionBackup; |
1711 |
int32_t iSAD; |
int32_t iSAD; |
1712 |
VECTOR backupMV; |
VECTOR backupMV; |
1713 |
|
|
1714 |
backupMV.x = startx; |
backupMV.x = start_x; |
1715 |
backupMV.y = starty; |
backupMV.y = start_y; |
1716 |
|
|
1717 |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
/* It's one search with full Diamond pattern, and only 3 of 4 for all following diamonds */ |
1718 |
|
|
1721 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1722 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
1723 |
|
|
1724 |
if (iDirection) |
if (iDirection) { |
1725 |
while (!iFound) { |
while (!iFound) { |
1726 |
iFound = 1; |
iFound = 1; |
1727 |
backupMV = *currMV; // since iDirection!=0, this is well defined! |
backupMV = *currMV; // since iDirection!=0, this is well defined! |
1728 |
|
iDirectionBackup = iDirection; |
1729 |
|
|
1730 |
if (iDirection != 2) |
if (iDirectionBackup != 2) |
1731 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1732 |
backupMV.y, 1); |
backupMV.y, 1); |
1733 |
if (iDirection != 1) |
if (iDirectionBackup != 1) |
1734 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1735 |
backupMV.y, 2); |
backupMV.y, 2); |
1736 |
if (iDirection != 4) |
if (iDirectionBackup != 4) |
1737 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1738 |
backupMV.y - iDiamondSize, 3); |
backupMV.y - iDiamondSize, 3); |
1739 |
if (iDirection != 3) |
if (iDirectionBackup != 3) |
1740 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1741 |
backupMV.y + iDiamondSize, 4); |
backupMV.y + iDiamondSize, 4); |
1742 |
|
} |
1743 |
} else { |
} else { |
1744 |
currMV->x = startx; |
currMV->x = start_x; |
1745 |
currMV->y = starty; |
currMV->y = start_y; |
1746 |
} |
} |
1747 |
return iMinSAD; |
return iMinSAD; |
1748 |
} |
} |
1749 |
|
|
1750 |
|
|
1751 |
|
|
1752 |
|
|
1753 |
int32_t |
int32_t |
1754 |
Halfpel8_Refine(const uint8_t * const pRef, |
Square8_MainSearch(const uint8_t * const pRef, |
1755 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1756 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1757 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1758 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1759 |
const int x, |
const int x, |
1760 |
const int y, |
const int y, |
1761 |
VECTOR * const currMV, |
int32_t start_x, |
1762 |
|
int32_t start_y, |
1763 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1764 |
const VECTOR * const pmv, |
VECTOR * const currMV, |
1765 |
|
const int center_x, |
1766 |
|
const int center_y, |
1767 |
const int32_t min_dx, |
const int32_t min_dx, |
1768 |
const int32_t max_dx, |
const int32_t max_dx, |
1769 |
const int32_t min_dy, |
const int32_t min_dy, |
1770 |
const int32_t max_dy, |
const int32_t max_dy, |
1771 |
|
const int32_t iEdgedWidth, |
1772 |
|
const int32_t iDiamondSize, |
1773 |
const int32_t iFcode, |
const int32_t iFcode, |
1774 |
const int32_t iQuant, |
const int32_t iQuant, |
1775 |
const int32_t iEdgedWidth) |
int iFound) |
1776 |
{ |
{ |
1777 |
/* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ |
/* Do a square search around given starting point, return SAD of best */ |
1778 |
|
|
1779 |
|
int32_t iDirection = 0; |
1780 |
int32_t iSAD; |
int32_t iSAD; |
1781 |
VECTOR backupMV = *currMV; |
VECTOR backupMV; |
|
|
|
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y - 1); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y - 1); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y - 1); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y + 1); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y + 1); |
|
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y + 1); |
|
1782 |
|
|
1783 |
return iMinSAD; |
backupMV.x = start_x; |
1784 |
} |
backupMV.y = start_y; |
1785 |
|
|
1786 |
|
/* It's one search with full square pattern, and new parts for all following diamonds */ |
1787 |
|
|
1788 |
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
/* new direction are extra, so 1-4 is normal diamond |
1789 |
|
537 |
1790 |
|
1*2 |
1791 |
|
648 |
1792 |
|
*/ |
1793 |
|
|
1794 |
int32_t |
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, 1); |
1795 |
PMVfastSearch8(const uint8_t * const pRef, |
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, 2); |
1796 |
const uint8_t * const pRefH, |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, 3); |
1797 |
const uint8_t * const pRefV, |
CHECK_MV8_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, 4); |
|
const uint8_t * const pRefHV, |
|
|
const IMAGE * const pCur, |
|
|
const int x, |
|
|
const int y, |
|
|
const int start_x, |
|
|
const int start_y, |
|
|
const uint32_t MotionFlags, |
|
|
const uint32_t iQuant, |
|
|
const uint32_t iFcode, |
|
|
const MBParam * const pParam, |
|
|
const MACROBLOCK * const pMBs, |
|
|
const MACROBLOCK * const prevMBs, |
|
|
VECTOR * const currMV, |
|
|
VECTOR * const currPMV) |
|
|
{ |
|
|
const uint32_t iWcount = pParam->mb_width; |
|
|
const int32_t iWidth = pParam->width; |
|
|
const int32_t iHeight = pParam->height; |
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
1798 |
|
|
1799 |
const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; |
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1800 |
|
backupMV.y - iDiamondSize, 5); |
1801 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
1802 |
|
backupMV.y + iDiamondSize, 6); |
1803 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1804 |
|
backupMV.y - iDiamondSize, 7); |
1805 |
|
CHECK_MV8_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
1806 |
|
backupMV.y + iDiamondSize, 8); |
1807 |
|
|
1808 |
|
|
1809 |
|
if (iDirection) { |
1810 |
|
while (!iFound) { |
1811 |
|
iFound = 1; |
1812 |
|
backupMV = *currMV; |
1813 |
|
|
1814 |
|
switch (iDirection) { |
1815 |
|
case 1: |
1816 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1817 |
|
backupMV.y, 1); |
1818 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1819 |
|
backupMV.y - iDiamondSize, 5); |
1820 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1821 |
|
backupMV.y - iDiamondSize, 7); |
1822 |
|
break; |
1823 |
|
case 2: |
1824 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1825 |
|
2); |
1826 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1827 |
|
backupMV.y + iDiamondSize, 6); |
1828 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1829 |
|
backupMV.y + iDiamondSize, 8); |
1830 |
|
break; |
1831 |
|
|
1832 |
|
case 3: |
1833 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1834 |
|
4); |
1835 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1836 |
|
backupMV.y - iDiamondSize, 7); |
1837 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1838 |
|
backupMV.y + iDiamondSize, 8); |
1839 |
|
break; |
1840 |
|
|
1841 |
|
case 4: |
1842 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1843 |
|
3); |
1844 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1845 |
|
backupMV.y - iDiamondSize, 5); |
1846 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1847 |
|
backupMV.y + iDiamondSize, 6); |
1848 |
|
break; |
1849 |
|
|
1850 |
|
case 5: |
1851 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1852 |
|
1); |
1853 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1854 |
|
3); |
1855 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1856 |
|
backupMV.y - iDiamondSize, 5); |
1857 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1858 |
|
backupMV.y + iDiamondSize, 6); |
1859 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1860 |
|
backupMV.y - iDiamondSize, 7); |
1861 |
|
break; |
1862 |
|
|
1863 |
|
case 6: |
1864 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1865 |
|
2); |
1866 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1867 |
|
3); |
1868 |
|
|
1869 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1870 |
|
backupMV.y - iDiamondSize, 5); |
1871 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1872 |
|
backupMV.y + iDiamondSize, 6); |
1873 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1874 |
|
backupMV.y + iDiamondSize, 8); |
1875 |
|
|
1876 |
|
break; |
1877 |
|
|
1878 |
|
case 7: |
1879 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1880 |
|
backupMV.y, 1); |
1881 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1882 |
|
4); |
1883 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1884 |
|
backupMV.y - iDiamondSize, 5); |
1885 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1886 |
|
backupMV.y - iDiamondSize, 7); |
1887 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1888 |
|
backupMV.y + iDiamondSize, 8); |
1889 |
|
break; |
1890 |
|
|
1891 |
|
case 8: |
1892 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1893 |
|
2); |
1894 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1895 |
|
4); |
1896 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1897 |
|
backupMV.y + iDiamondSize, 6); |
1898 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1899 |
|
backupMV.y - iDiamondSize, 7); |
1900 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1901 |
|
backupMV.y + iDiamondSize, 8); |
1902 |
|
break; |
1903 |
|
default: |
1904 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, backupMV.y, |
1905 |
|
1); |
1906 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, backupMV.y, |
1907 |
|
2); |
1908 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y - iDiamondSize, |
1909 |
|
3); |
1910 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, backupMV.y + iDiamondSize, |
1911 |
|
4); |
1912 |
|
|
1913 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1914 |
|
backupMV.y - iDiamondSize, 5); |
1915 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1916 |
|
backupMV.y + iDiamondSize, 6); |
1917 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1918 |
|
backupMV.y - iDiamondSize, 7); |
1919 |
|
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1920 |
|
backupMV.y + iDiamondSize, 8); |
1921 |
|
break; |
1922 |
|
} |
1923 |
|
} |
1924 |
|
} else { |
1925 |
|
currMV->x = start_x; |
1926 |
|
currMV->y = start_y; |
1927 |
|
} |
1928 |
|
return iMinSAD; |
1929 |
|
} |
1930 |
|
|
1931 |
|
|
1932 |
|
|
1933 |
|
|
1934 |
|
|
1935 |
|
int32_t |
1936 |
|
Halfpel8_Refine_c(const uint8_t * const pRef, |
1937 |
|
const uint8_t * const pRefH, |
1938 |
|
const uint8_t * const pRefV, |
1939 |
|
const uint8_t * const pRefHV, |
1940 |
|
const uint8_t * const cur, |
1941 |
|
const int x, |
1942 |
|
const int y, |
1943 |
|
VECTOR * const currMV, |
1944 |
|
int32_t iMinSAD, |
1945 |
|
const int center_x, |
1946 |
|
const int center_y, |
1947 |
|
const int32_t min_dx, |
1948 |
|
const int32_t max_dx, |
1949 |
|
const int32_t min_dy, |
1950 |
|
const int32_t max_dy, |
1951 |
|
const int32_t iFcode, |
1952 |
|
const int32_t iQuant, |
1953 |
|
const int32_t iEdgedWidth) |
1954 |
|
{ |
1955 |
|
/* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ |
1956 |
|
|
1957 |
|
int32_t iSAD; |
1958 |
|
VECTOR backupMV = *currMV; |
1959 |
|
|
1960 |
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y - 1); |
1961 |
|
CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y - 1); |
1962 |
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y - 1); |
1963 |
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y); |
1964 |
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y); |
1965 |
|
CHECK_MV8_CANDIDATE(backupMV.x - 1, backupMV.y + 1); |
1966 |
|
CHECK_MV8_CANDIDATE(backupMV.x, backupMV.y + 1); |
1967 |
|
CHECK_MV8_CANDIDATE(backupMV.x + 1, backupMV.y + 1); |
1968 |
|
|
1969 |
|
return iMinSAD; |
1970 |
|
} |
1971 |
|
|
1972 |
|
|
1973 |
|
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
1974 |
|
|
1975 |
|
int32_t |
1976 |
|
PMVfastSearch8(const uint8_t * const pRef, |
1977 |
|
const uint8_t * const pRefH, |
1978 |
|
const uint8_t * const pRefV, |
1979 |
|
const uint8_t * const pRefHV, |
1980 |
|
const IMAGE * const pCur, |
1981 |
|
const int x, |
1982 |
|
const int y, |
1983 |
|
const int start_x, |
1984 |
|
const int start_y, |
1985 |
|
const int center_x, |
1986 |
|
const int center_y, |
1987 |
|
const uint32_t MotionFlags, |
1988 |
|
const uint32_t iQuant, |
1989 |
|
const uint32_t iFcode, |
1990 |
|
const MBParam * const pParam, |
1991 |
|
const MACROBLOCK * const pMBs, |
1992 |
|
const MACROBLOCK * const prevMBs, |
1993 |
|
VECTOR * const currMV, |
1994 |
|
VECTOR * const currPMV) |
1995 |
|
{ |
1996 |
|
const uint32_t iWcount = pParam->mb_width; |
1997 |
|
const int32_t iWidth = pParam->width; |
1998 |
|
const int32_t iHeight = pParam->height; |
1999 |
|
const int32_t iEdgedWidth = pParam->edged_width; |
2000 |
|
|
2001 |
|
const uint8_t *cur = pCur->y + x * 8 + y * 8 * iEdgedWidth; |
2002 |
|
|
2003 |
int32_t iDiamondSize; |
int32_t iDiamondSize; |
2004 |
|
|
2040 |
} |
} |
2041 |
|
|
2042 |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
2043 |
bPredEq = get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
2044 |
// bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, (x >> 1), (y >> 1), iSubBlock, pmv, psad); |
2045 |
|
|
2046 |
if ((x == 0) && (y == 0)) { |
if ((x == 0) && (y == 0)) { |
2047 |
threshA = 512 / 4; |
threshA = 512 / 4; |
2048 |
threshB = 1024 / 4; |
threshB = 1024 / 4; |
2049 |
|
|
2050 |
} else { |
} else { |
2051 |
threshA = psad[0] / 4; /* good estimate */ |
threshA = psad[0] / 4; /* good estimate? */ |
2052 |
threshB = threshA + 256 / 4; |
threshB = threshA + 256 / 4; |
2053 |
if (threshA < 512 / 4) |
if (threshA < 512 / 4) |
2054 |
threshA = 512 / 4; |
threshA = 512 / 4; |
2070 |
|
|
2071 |
// Prepare for main loop |
// Prepare for main loop |
2072 |
|
|
2073 |
// if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES8) |
2074 |
// MainSearchPtr = Square8_MainSearch; |
MainSearchPtr = Square8_MainSearch; |
2075 |
// else |
else |
2076 |
|
|
2077 |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2078 |
MainSearchPtr = AdvDiamond8_MainSearch; |
MainSearchPtr = AdvDiamond8_MainSearch; |
2087 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
2088 |
iEdgedWidth), iEdgedWidth); |
iEdgedWidth), iEdgedWidth); |
2089 |
iMinSAD += |
iMinSAD += |
2090 |
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
2091 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
2092 |
|
|
2093 |
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
2094 |
&& ((uint32_t) iMinSAD < |
&& ((int32_t) iMinSAD < |
2095 |
prevMB->sad8[iSubBlock]))) { |
prevMB->sad8[iSubBlock]))) { |
2096 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2097 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
2130 |
// the median prediction might be even better than mv16 |
// the median prediction might be even better than mv16 |
2131 |
|
|
2132 |
if (!MVequal(pmv[0], startMV)) |
if (!MVequal(pmv[0], startMV)) |
2133 |
CHECK_MV8_CANDIDATE(pmv[0].x, pmv[0].y); |
CHECK_MV8_CANDIDATE(center_x, center_y); |
2134 |
|
|
2135 |
// (0,0) if needed |
// (0,0) if needed |
2136 |
if (!MVzero(pmv[0])) |
if (!MVzero(pmv[0])) |
2146 |
|
|
2147 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
2148 |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
2149 |
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
2150 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2151 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
2152 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2205 |
|
|
2206 |
if ((iMinSAD <= threshA) || |
if ((iMinSAD <= threshA) || |
2207 |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
2208 |
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
((int32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
2209 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2210 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
2211 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2227 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
2228 |
iSAD = |
iSAD = |
2229 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2230 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2231 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2232 |
iQuant, iFound); |
iQuant, iFound); |
2233 |
|
|
2242 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
2243 |
iSAD = |
iSAD = |
2244 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2245 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2246 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2247 |
iDiamondSize, iFcode, iQuant, iFound); |
iDiamondSize, iFcode, iQuant, iFound); |
2248 |
|
|
2255 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2256 |
iSAD = |
iSAD = |
2257 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2258 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2259 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2260 |
iQuant, iFound); |
iQuant, iFound); |
2261 |
|
|
2274 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
2275 |
iMinSAD = |
iMinSAD = |
2276 |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2277 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2278 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2279 |
|
|
2280 |
|
|
2281 |
PMVfast8_Terminate_without_Refine: |
PMVfast8_Terminate_without_Refine: |
2282 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2283 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2284 |
|
|
2285 |
return iMinSAD; |
return iMinSAD; |
2286 |
} |
} |
2293 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2294 |
const int x, |
const int x, |
2295 |
const int y, |
const int y, |
2296 |
|
const int start_x, |
2297 |
|
const int start_y, |
2298 |
|
const int center_x, |
2299 |
|
const int center_y, |
2300 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2301 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2302 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2355 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2356 |
} |
} |
2357 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2358 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
2359 |
// bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2360 |
|
|
2361 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2362 |
MinSAD=SAD |
MinSAD=SAD |
2367 |
|
|
2368 |
// Prepare for main loop |
// Prepare for main loop |
2369 |
|
|
2370 |
*currMV = pmv[0]; /* current best := median prediction */ |
currMV->x = start_x; |
2371 |
|
currMV->y = start_y; |
2372 |
|
|
2373 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2374 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2375 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2391 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
2392 |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2393 |
iMinSAD += |
iMinSAD += |
2394 |
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2395 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
2396 |
|
|
2397 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2398 |
if ((iMinSAD < 256) || |
if ((iMinSAD < 256) || |
2399 |
((MVequal(*currMV, prevMB->mvs[0])) && |
((MVequal(*currMV, prevMB->mvs[0])) && |
2400 |
((uint32_t) iMinSAD < prevMB->sad16))) { |
((int32_t) iMinSAD < prevMB->sad16))) { |
2401 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2402 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2403 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2457 |
|
|
2458 |
if ((iMinSAD <= thresh2) |
if ((iMinSAD <= thresh2) |
2459 |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
2460 |
((uint32_t) iMinSAD <= prevMB->sad16))) { |
((int32_t) iMinSAD <= prevMB->sad16))) { |
2461 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2462 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2463 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2515 |
|
|
2516 |
iSAD = |
iSAD = |
2517 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2518 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2519 |
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2520 |
|
|
2521 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
2530 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
2531 |
iSAD = |
iSAD = |
2532 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2533 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2534 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2535 |
2, iFcode, iQuant, 0); |
2, iFcode, iQuant, 0); |
2536 |
} |
} |
2543 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2544 |
iSAD = |
iSAD = |
2545 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2546 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2547 |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2548 |
|
|
2549 |
if (iSAD < iMinSAD) { |
if (iSAD < iMinSAD) { |
2559 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
2560 |
iMinSAD = |
iMinSAD = |
2561 |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2562 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2563 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2564 |
|
|
2565 |
EPZS16_Terminate_without_Refine: |
EPZS16_Terminate_without_Refine: |
2566 |
|
|
2567 |
*oldMB = *prevMB; |
*oldMB = *prevMB; |
2568 |
|
|
2569 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2570 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2571 |
return iMinSAD; |
return iMinSAD; |
2572 |
} |
} |
2573 |
|
|
2582 |
const int y, |
const int y, |
2583 |
const int start_x, |
const int start_x, |
2584 |
const int start_y, |
const int start_y, |
2585 |
|
const int center_x, |
2586 |
|
const int center_y, |
2587 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2588 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2589 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2638 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2639 |
} |
} |
2640 |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2641 |
bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv, psad); |
//bPredEq = get_pmvdata(pMBs, x >> 1, y >> 1, iWcount, iSubBlock, pmv[0].x, pmv[0].y, psad); |
2642 |
// bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); |
bPredEq = get_pmvdata2(pMBs, iWcount, 0, x >> 1, y >> 1, iSubBlock, pmv, psad); |
2643 |
|
|
2644 |
|
|
2645 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2674 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
2675 |
iEdgedWidth), iEdgedWidth); |
iEdgedWidth), iEdgedWidth); |
2676 |
iMinSAD += |
iMinSAD += |
2677 |
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
calc_delta_8(currMV->x - center_x, currMV->y - center_y, |
2678 |
(uint8_t) iFcode, iQuant); |
(uint8_t) iFcode, iQuant); |
2679 |
|
|
2680 |
|
|
2750 |
|
|
2751 |
// there is no EPZS^2 for inter4v at the moment |
// there is no EPZS^2 for inter4v at the moment |
2752 |
|
|
2753 |
// if (MotionFlags & PMV_USESQUARES8) |
if (MotionFlags & PMV_USESQUARES8) |
2754 |
// MainSearchPtr = Square8_MainSearch; |
MainSearchPtr = Square8_MainSearch; |
2755 |
// else |
else |
2756 |
|
|
2757 |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2758 |
MainSearchPtr = AdvDiamond8_MainSearch; |
MainSearchPtr = AdvDiamond8_MainSearch; |
2761 |
|
|
2762 |
iSAD = |
iSAD = |
2763 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2764 |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
2765 |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2766 |
iQuant, 0); |
iQuant, 0); |
2767 |
|
|
2777 |
if (!(MVequal(pmv[0], backupMV))) { |
if (!(MVequal(pmv[0], backupMV))) { |
2778 |
iSAD = |
iSAD = |
2779 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2780 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
2781 |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2782 |
iDiamondSize, iFcode, iQuant, 0); |
iDiamondSize, iFcode, iQuant, 0); |
2783 |
|
|
2790 |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2791 |
iSAD = |
iSAD = |
2792 |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2793 |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
2794 |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2795 |
iQuant, 0); |
iQuant, 0); |
2796 |
|
|
2807 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
2808 |
iMinSAD = |
iMinSAD = |
2809 |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2810 |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
2811 |
iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2812 |
|
|
2813 |
EPZS8_Terminate_without_Refine: |
EPZS8_Terminate_without_Refine: |
2814 |
|
|
2815 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - center_x; |
2816 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - center_y; |
2817 |
return iMinSAD; |
return iMinSAD; |
2818 |
} |
} |
2819 |
|
|
2820 |
|
|
2821 |
|
|
2822 |
|
int32_t |
2823 |
|
PMVfastIntSearch16(const uint8_t * const pRef, |
2824 |
|
const uint8_t * const pRefH, |
2825 |
|
const uint8_t * const pRefV, |
2826 |
|
const uint8_t * const pRefHV, |
2827 |
|
const IMAGE * const pCur, |
2828 |
|
const int x, |
2829 |
|
const int y, |
2830 |
|
const int start_x, /* start should be most likely vector */ |
2831 |
|
const int start_y, |
2832 |
|
const int center_x, /* center is from where length of MVs is measured */ |
2833 |
|
const int center_y, |
2834 |
|
const uint32_t MotionFlags, |
2835 |
|
const uint32_t iQuant, |
2836 |
|
const uint32_t iFcode, |
2837 |
|
const MBParam * const pParam, |
2838 |
|
const MACROBLOCK * const pMBs, |
2839 |
|
const MACROBLOCK * const prevMBs, |
2840 |
|
VECTOR * const currMV, |
2841 |
|
VECTOR * const currPMV) |
2842 |
|
{ |
2843 |
|
const uint32_t iWcount = pParam->mb_width; |
2844 |
|
const int32_t iWidth = pParam->width; |
2845 |
|
const int32_t iHeight = pParam->height; |
2846 |
|
const int32_t iEdgedWidth = pParam->edged_width; |
2847 |
|
|
2848 |
|
const uint8_t *cur = pCur->y + x * 16 + y * 16 * iEdgedWidth; |
2849 |
|
const VECTOR zeroMV = { 0, 0 }; |
2850 |
|
|
2851 |
|
int32_t iDiamondSize; |
2852 |
|
|
2853 |
|
int32_t min_dx; |
2854 |
|
int32_t max_dx; |
2855 |
|
int32_t min_dy; |
2856 |
|
int32_t max_dy; |
2857 |
|
|
2858 |
|
int32_t iFound; |
2859 |
|
|
2860 |
|
VECTOR newMV; |
2861 |
|
VECTOR backupMV; |
2862 |
|
|
2863 |
|
VECTOR pmv[4]; |
2864 |
|
int32_t psad[4]; |
2865 |
|
|
2866 |
|
MainSearch16FuncPtr MainSearchPtr; |
2867 |
|
|
2868 |
|
const MACROBLOCK *const prevMB = prevMBs + x + y * iWcount; |
2869 |
|
MACROBLOCK *const pMB = pMBs + x + y * iWcount; |
2870 |
|
|
2871 |
|
int32_t threshA, threshB; |
2872 |
|
int32_t bPredEq; |
2873 |
|
int32_t iMinSAD, iSAD; |
2874 |
|
|
2875 |
|
|
2876 |
|
/* Get maximum range */ |
2877 |
|
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
2878 |
|
iFcode); |
2879 |
|
|
2880 |
|
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
2881 |
|
|
2882 |
|
if ((x == 0) && (y == 0)) { |
2883 |
|
threshA = 512; |
2884 |
|
threshB = 1024; |
2885 |
|
|
2886 |
|
bPredEq = 0; |
2887 |
|
psad[0] = psad[1] = psad[2] = psad[3] = 0; |
2888 |
|
*currMV = pmv[0] = pmv[1] = pmv[2] = pmv[3] = zeroMV; |
2889 |
|
|
2890 |
|
} else { |
2891 |
|
threshA = psad[0]; |
2892 |
|
threshB = threshA + 256; |
2893 |
|
if (threshA < 512) |
2894 |
|
threshA = 512; |
2895 |
|
if (threshA > 1024) |
2896 |
|
threshA = 1024; |
2897 |
|
if (threshB > 1792) |
2898 |
|
threshB = 1792; |
2899 |
|
|
2900 |
|
bPredEq = get_ipmvdata(pMBs, iWcount, 0, x, y, 0, pmv, psad); |
2901 |
|
*currMV = pmv[0]; /* current best := prediction */ |
2902 |
|
} |
2903 |
|
|
2904 |
|
iFound = 0; |
2905 |
|
|
2906 |
|
/* Step 4: Calculate SAD around the Median prediction. |
2907 |
|
MinSAD=SAD |
2908 |
|
If Motion Vector equal to Previous frame motion vector |
2909 |
|
and MinSAD<PrevFrmSAD goto Step 10. |
2910 |
|
If SAD<=256 goto Step 10. |
2911 |
|
*/ |
2912 |
|
|
2913 |
|
if (currMV->x > max_dx) { |
2914 |
|
currMV->x = EVEN(max_dx); |
2915 |
|
} |
2916 |
|
if (currMV->x < min_dx) { |
2917 |
|
currMV->x = EVEN(min_dx); |
2918 |
|
} |
2919 |
|
if (currMV->y > max_dy) { |
2920 |
|
currMV->y = EVEN(max_dy); |
2921 |
|
} |
2922 |
|
if (currMV->y < min_dy) { |
2923 |
|
currMV->y = EVEN(min_dy); |
2924 |
|
} |
2925 |
|
|
2926 |
|
iMinSAD = |
2927 |
|
sad16(cur, |
2928 |
|
get_iref_mv(pRef, x, y, 16, currMV, |
2929 |
|
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2930 |
|
iMinSAD += |
2931 |
|
calc_delta_16(currMV->x - center_x, currMV->y - center_y, |
2932 |
|
(uint8_t) iFcode, iQuant); |
2933 |
|
|
2934 |
|
if ((iMinSAD < 256) || |
2935 |
|
((MVequal(*currMV, prevMB->i_mvs[0])) && |
2936 |
|
((int32_t) iMinSAD < prevMB->i_sad16))) { |
2937 |
|
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
2938 |
|
{ |
2939 |
|
if (!MVzero(*currMV)) { |
2940 |
|
iMinSAD += MV16_00_BIAS; |
2941 |
|
CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
2942 |
|
iMinSAD -= MV16_00_BIAS; |
2943 |
|
} |
2944 |
|
} |
2945 |
|
|
2946 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
2947 |
|
goto PMVfastInt16_Terminate_with_Refine; |
2948 |
|
} |
2949 |
|
|
2950 |
|
|
2951 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
2952 |
|
vector of the median. |
2953 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
2954 |
|
*/ |
2955 |
|
|
2956 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->i_mvs[0]))) |
2957 |
|
iFound = 2; |
2958 |
|
|
2959 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
2960 |
|
Otherwise select large Diamond Search. |
2961 |
|
*/ |
2962 |
|
|
2963 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) |
2964 |
|
iDiamondSize = 2; // halfpel units! |
2965 |
|
else |
2966 |
|
iDiamondSize = 4; // halfpel units! |
2967 |
|
|
2968 |
|
/* |
2969 |
|
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
2970 |
|
Also calculate (0,0) but do not subtract offset. |
2971 |
|
Let MinSAD be the smallest SAD up to this point. |
2972 |
|
If MV is (0,0) subtract offset. |
2973 |
|
*/ |
2974 |
|
|
2975 |
|
// (0,0) is often a good choice |
2976 |
|
|
2977 |
|
if (!MVzero(pmv[0])) |
2978 |
|
CHECK_MV16_ZERO; |
2979 |
|
|
2980 |
|
// previous frame MV is always possible |
2981 |
|
|
2982 |
|
if (!MVzero(prevMB->i_mvs[0])) |
2983 |
|
if (!MVequal(prevMB->i_mvs[0], pmv[0])) |
2984 |
|
CHECK_MV16_CANDIDATE(prevMB->i_mvs[0].x, prevMB->i_mvs[0].y); |
2985 |
|
|
2986 |
|
// left neighbour, if allowed |
2987 |
|
|
2988 |
|
if (!MVzero(pmv[1])) |
2989 |
|
if (!MVequal(pmv[1], prevMB->i_mvs[0])) |
2990 |
|
if (!MVequal(pmv[1], pmv[0])) |
2991 |
|
CHECK_MV16_CANDIDATE(pmv[1].x, pmv[1].y); |
2992 |
|
|
2993 |
|
// top neighbour, if allowed |
2994 |
|
if (!MVzero(pmv[2])) |
2995 |
|
if (!MVequal(pmv[2], prevMB->i_mvs[0])) |
2996 |
|
if (!MVequal(pmv[2], pmv[0])) |
2997 |
|
if (!MVequal(pmv[2], pmv[1])) |
2998 |
|
CHECK_MV16_CANDIDATE(pmv[2].x, pmv[2].y); |
2999 |
|
|
3000 |
|
// top right neighbour, if allowed |
3001 |
|
if (!MVzero(pmv[3])) |
3002 |
|
if (!MVequal(pmv[3], prevMB->i_mvs[0])) |
3003 |
|
if (!MVequal(pmv[3], pmv[0])) |
3004 |
|
if (!MVequal(pmv[3], pmv[1])) |
3005 |
|
if (!MVequal(pmv[3], pmv[2])) |
3006 |
|
CHECK_MV16_CANDIDATE(pmv[3].x, |
3007 |
|
pmv[3].y); |
3008 |
|
|
3009 |
|
if ((MVzero(*currMV)) && |
3010 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
3011 |
|
iMinSAD -= MV16_00_BIAS; |
3012 |
|
|
3013 |
|
|
3014 |
|
/* Step 6: If MinSAD <= thresa goto Step 10. |
3015 |
|
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
3016 |
|
*/ |
3017 |
|
|
3018 |
|
if ((iMinSAD <= threshA) || |
3019 |
|
(MVequal(*currMV, prevMB->i_mvs[0]) && |
3020 |
|
((int32_t) iMinSAD < prevMB->i_sad16))) { |
3021 |
|
|
3022 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
3023 |
|
goto PMVfastInt16_Terminate_with_Refine; |
3024 |
|
} |
3025 |
|
|
3026 |
|
|
3027 |
|
/************ (Diamond Search) **************/ |
3028 |
|
/* |
3029 |
|
Step 7: Perform Diamond search, with either the small or large diamond. |
3030 |
|
If Found=2 only examine one Diamond pattern, and afterwards goto step 10 |
3031 |
|
Step 8: If small diamond, iterate small diamond search pattern until motion vector lies in the center of the diamond. |
3032 |
|
If center then goto step 10. |
3033 |
|
Step 9: If large diamond, iterate large diamond search pattern until motion vector lies in the center. |
3034 |
|
Refine by using small diamond and goto step 10. |
3035 |
|
*/ |
3036 |
|
|
3037 |
|
if (MotionFlags & PMV_USESQUARES16) |
3038 |
|
MainSearchPtr = Square16_MainSearch; |
3039 |
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
3040 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
3041 |
|
else |
3042 |
|
MainSearchPtr = Diamond16_MainSearch; |
3043 |
|
|
3044 |
|
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
3045 |
|
|
3046 |
|
|
3047 |
|
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
3048 |
|
iSAD = |
3049 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
3050 |
|
currMV->y, iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, |
3051 |
|
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
3052 |
|
iQuant, iFound); |
3053 |
|
|
3054 |
|
if (iSAD < iMinSAD) { |
3055 |
|
*currMV = newMV; |
3056 |
|
iMinSAD = iSAD; |
3057 |
|
} |
3058 |
|
|
3059 |
|
if (MotionFlags & PMV_EXTSEARCH16) { |
3060 |
|
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
3061 |
|
|
3062 |
|
if (!(MVequal(pmv[0], backupMV))) { |
3063 |
|
iSAD = |
3064 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
3065 |
|
pmv[0].x, pmv[0].y, iMinSAD, &newMV, center_x, center_y, |
3066 |
|
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
3067 |
|
iDiamondSize, iFcode, iQuant, iFound); |
3068 |
|
|
3069 |
|
if (iSAD < iMinSAD) { |
3070 |
|
*currMV = newMV; |
3071 |
|
iMinSAD = iSAD; |
3072 |
|
} |
3073 |
|
} |
3074 |
|
|
3075 |
|
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
3076 |
|
iSAD = |
3077 |
|
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
3078 |
|
iMinSAD, &newMV, center_x, center_y, min_dx, max_dx, min_dy, |
3079 |
|
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
3080 |
|
iQuant, iFound); |
3081 |
|
|
3082 |
|
if (iSAD < iMinSAD) { |
3083 |
|
*currMV = newMV; |
3084 |
|
iMinSAD = iSAD; |
3085 |
|
} |
3086 |
|
} |
3087 |
|
} |
3088 |
|
|
3089 |
|
/* |
3090 |
|
Step 10: The motion vector is chosen according to the block corresponding to MinSAD. |
3091 |
|
*/ |
3092 |
|
|
3093 |
|
PMVfastInt16_Terminate_with_Refine: |
3094 |
|
|
3095 |
|
pMB->i_mvs[0] = pMB->i_mvs[1] = pMB->i_mvs[2] = pMB->i_mvs[3] = pMB->i_mv16 = *currMV; |
3096 |
|
pMB->i_sad8[0] = pMB->i_sad8[1] = pMB->i_sad8[2] = pMB->i_sad8[3] = pMB->i_sad16 = iMinSAD; |
3097 |
|
|
3098 |
|
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
3099 |
|
iMinSAD = |
3100 |
|
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
3101 |
|
iMinSAD, center_x, center_y, min_dx, max_dx, min_dy, max_dy, |
3102 |
|
iFcode, iQuant, iEdgedWidth); |
3103 |
|
|
3104 |
|
pmv[0] = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); // get _REAL_ prediction (halfpel possible) |
3105 |
|
|
3106 |
|
PMVfastInt16_Terminate_without_Refine: |
3107 |
|
currPMV->x = currMV->x - center_x; |
3108 |
|
currPMV->y = currMV->y - center_y; |
3109 |
|
return iMinSAD; |
3110 |
|
} |
3111 |
|
|
3112 |
|
|
3113 |
|
|
3114 |
/* *********************************************************** |
/* *********************************************************** |
3115 |
bvop motion estimation |
bvop motion estimation |
|
// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) |
|
3116 |
***************************************************************/ |
***************************************************************/ |
3117 |
|
|
|
|
|
3118 |
void |
void |
3119 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
3120 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
3121 |
|
const int32_t time_bp, |
3122 |
|
const int32_t time_pp, |
3123 |
// forward (past) reference |
// forward (past) reference |
3124 |
const MACROBLOCK * const f_mbs, |
const MACROBLOCK * const f_mbs, |
3125 |
const IMAGE * const f_ref, |
const IMAGE * const f_ref, |
3133 |
const IMAGE * const b_refV, |
const IMAGE * const b_refV, |
3134 |
const IMAGE * const b_refHV) |
const IMAGE * const b_refHV) |
3135 |
{ |
{ |
3136 |
const uint32_t mb_width = pParam->mb_width; |
const int mb_width = pParam->mb_width; |
3137 |
const uint32_t mb_height = pParam->mb_height; |
const int mb_height = pParam->mb_height; |
3138 |
const int32_t edged_width = pParam->edged_width; |
const int edged_width = pParam->edged_width; |
3139 |
|
|
3140 |
uint32_t i, j; |
const int32_t iWidth = pParam->width; |
3141 |
|
const int32_t iHeight = pParam->height; |
|
int32_t f_sad16; |
|
|
int32_t b_sad16; |
|
|
int32_t i_sad16; |
|
|
int32_t d_sad16; |
|
|
int32_t best_sad; |
|
3142 |
|
|
3143 |
|
int i, j, k; |
3144 |
|
|
3145 |
|
static const VECTOR zeroMV={0,0}; |
3146 |
|
|
3147 |
|
int f_sad16; /* forward (as usual) search */ |
3148 |
|
int b_sad16; /* backward (only in b-frames) search */ |
3149 |
|
int i_sad16; /* interpolated (both direction, b-frames only) */ |
3150 |
|
int d_sad16; /* direct mode (assume almost linear motion) */ |
3151 |
|
|
3152 |
|
int best_sad; |
3153 |
|
|
3154 |
|
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
3155 |
|
VECTOR f_interpolMV, b_interpolMV; |
3156 |
VECTOR pmv_dontcare; |
VECTOR pmv_dontcare; |
3157 |
|
|
3158 |
|
int min_dx, max_dx, min_dy, max_dy; |
3159 |
|
int f_min_dx, f_max_dx, f_min_dy, f_max_dy; |
3160 |
|
int b_min_dx, b_max_dx, b_min_dy, b_max_dy; |
3161 |
|
|
3162 |
|
int f_count=0; |
3163 |
|
int b_count=0; |
3164 |
|
int i_count=0; |
3165 |
|
int d_count=0; |
3166 |
|
|
3167 |
|
const int64_t TRB = (int32_t)time_pp - (int32_t)time_bp; |
3168 |
|
const int64_t TRD = (int32_t)time_pp; |
3169 |
|
|
3170 |
|
// fprintf(stderr,"TRB = %lld TRD = %lld time_bp =%d time_pp =%d\n\n",TRB,TRD,time_bp,time_pp); |
3171 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
3172 |
for (j = 0; j < mb_height; j++) { |
for (j = 0; j < mb_height; j++) { |
3173 |
|
|
3174 |
|
f_predMV = zeroMV; /* prediction is reset at left boundary */ |
3175 |
|
b_predMV = zeroMV; |
3176 |
|
|
3177 |
for (i = 0; i < mb_width; i++) { |
for (i = 0; i < mb_width; i++) { |
3178 |
MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; |
MACROBLOCK *mb = &frame->mbs[i + j * mb_width]; |
3179 |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
const MACROBLOCK *f_mb = &f_mbs[i + j * mb_width]; |
3180 |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
const MACROBLOCK *b_mb = &b_mbs[i + j * mb_width]; |
3181 |
|
|
3182 |
|
mb->deltamv=zeroMV; |
3183 |
|
|
3184 |
|
/* special case, if collocated block is SKIPed: encoding is forward (0,0), cpb=0 without further ado */ |
3185 |
|
|
3186 |
|
#ifndef _DISABLE_SKIP |
3187 |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
3188 |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
3189 |
mb->mode = MODE_NOT_CODED; |
mb->mode = MODE_NOT_CODED; |
3193 |
mb->b_mvs[0].y = 0; |
mb->b_mvs[0].y = 0; |
3194 |
continue; |
continue; |
3195 |
} |
} |
3196 |
/* force F_SAD16 |
#endif |
|
f_sad16 = 100; |
|
|
b_sad16 = 65535; |
|
3197 |
|
|
3198 |
mb->mode = MODE_FORWARD; |
if (b_mb->mode == MODE_INTER4V) |
3199 |
mb->mvs[0].x = 1; |
{ |
3200 |
mb->mvs[0].y = 1; |
d_sad16 = 0; |
3201 |
mb->b_mvs[0].x = 1; |
/* same method of scaling as in decoder.c, so we copy from there */ |
3202 |
mb->b_mvs[0].y = 1; |
for (k = 0; k < 4; k++) { |
3203 |
continue; |
|
3204 |
^^ force F_SAD16 */ |
mb->directmv[k] = b_mb->mvs[k]; |
3205 |
|
|
3206 |
|
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3207 |
|
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
3208 |
|
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
3209 |
|
: mb->mvs[k].x - mb->directmv[k].x); |
3210 |
|
|
3211 |
|
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3212 |
|
mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) |
3213 |
|
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
3214 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
3215 |
|
|
3216 |
|
d_sad16 += |
3217 |
|
sad8bi(frame->image.y + 2*(i+(k&1))*8 + 2*(j+(k>>1))*8*edged_width, |
3218 |
|
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3219 |
|
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->mvs[k], edged_width), |
3220 |
|
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3221 |
|
2*(i+(k&1)), 2*(j+(k>>1)), 8, &mb->b_mvs[k], edged_width), |
3222 |
|
edged_width); |
3223 |
|
} |
3224 |
|
} |
3225 |
|
else |
3226 |
|
{ |
3227 |
|
mb->directmv[3] = mb->directmv[2] = mb->directmv[1] = |
3228 |
|
mb->directmv[0] = b_mb->mvs[0]; |
3229 |
|
|
3230 |
|
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3231 |
|
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
3232 |
|
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
3233 |
|
: mb->mvs[0].x - mb->directmv[0].x); |
3234 |
|
|
3235 |
|
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
3236 |
|
mb->b_mvs[0].y = (int32_t) ((mb->directmv[0].y == 0) |
3237 |
|
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
3238 |
|
: mb->mvs[0].y - mb->directmv[0].y); |
3239 |
|
|
3240 |
|
d_sad16 = sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
3241 |
|
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3242 |
|
i, j, 16, &mb->mvs[0], edged_width), |
3243 |
|
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3244 |
|
i, j, 16, &mb->b_mvs[0], edged_width), |
3245 |
|
edged_width); |
3246 |
|
|
3247 |
|
} |
3248 |
|
d_sad16 += calc_delta_16(mb->deltamv.x, mb->deltamv.y, 1, frame->quant); |
3249 |
|
|
3250 |
// forward search |
// forward search |
3251 |
f_sad16 = |
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3252 |
SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
&frame->image, i, j, |
3253 |
&frame->image, i, j, frame->motion_flags, |
mb->mvs[0].x, mb->mvs[0].y, /* start point f_directMV */ |
3254 |
|
f_predMV.x, f_predMV.y, /* center is f-prediction */ |
3255 |
|
frame->motion_flags, |
3256 |
frame->quant, frame->fcode, pParam, |
frame->quant, frame->fcode, pParam, |
3257 |
f_mbs, f_mbs, /* todo */ |
f_mbs, f_mbs, |
3258 |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
&mb->mvs[0], &pmv_dontcare); |
3259 |
|
|
3260 |
|
|
3261 |
// backward search |
// backward search |
3262 |
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3263 |
&frame->image, i, j, frame->motion_flags, |
&frame->image, i, j, |
3264 |
|
mb->b_mvs[0].x, mb->b_mvs[0].y, /* start point b_directMV */ |
3265 |
|
b_predMV.x, b_predMV.y, /* center is b-prediction */ |
3266 |
|
frame->motion_flags, |
3267 |
frame->quant, frame->bcode, pParam, |
frame->quant, frame->bcode, pParam, |
3268 |
b_mbs, b_mbs, /* todo */ |
b_mbs, b_mbs, |
3269 |
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
&mb->b_mvs[0], &pmv_dontcare); |
3270 |
|
|
|
// interpolate search (simple, but effective) |
|
|
i_sad16 = 65535; |
|
|
|
|
|
/* |
|
|
x/y range somewhat buggy |
|
3271 |
i_sad16 = |
i_sad16 = |
3272 |
sad16bi_c(frame->image.y + i * 16 + j * 16 * edged_width, |
sad16bi(frame->image.y + i * 16 + j * 16 * edged_width, |
3273 |
get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
get_ref_mv(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3274 |
i, j, 16, mb->mvs[0].x, mb->mvs[0].y, |
i, j, 16, &mb->mvs[0], edged_width), |
3275 |
edged_width), get_ref(b_ref->y, b_refH->y, |
get_ref_mv(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3276 |
b_refV->y, b_refHV->y, |
i, j, 16, &mb->b_mvs[0], edged_width), |
|
i, j, 16, |
|
|
mb->b_mvs[0].x, |
|
|
mb->b_mvs[0].x, |
|
|
edged_width), |
|
3277 |
edged_width); |
edged_width); |
3278 |
*/ |
i_sad16 += calc_delta_16(mb->mvs[0].x-f_predMV.x, mb->mvs[0].y-f_predMV.y, |
3279 |
|
frame->fcode, frame->quant); |
3280 |
// TODO: direct search |
i_sad16 += calc_delta_16(mb->b_mvs[0].x-b_predMV.x, mb->b_mvs[0].y-b_predMV.y, |
3281 |
// predictor + range of [-32,32] |
frame->bcode, frame->quant); |
3282 |
d_sad16 = 65535; |
|
3283 |
|
get_range(&f_min_dx, &f_max_dx, &f_min_dy, &f_max_dy, i, j, 16, iWidth, iHeight, |
3284 |
|
frame->fcode); |
3285 |
|
get_range(&b_min_dx, &b_max_dx, &b_min_dy, &b_max_dy, i, j, 16, iWidth, iHeight, |
3286 |
|
frame->bcode); |
3287 |
|
|
3288 |
|
/* Interpolated MC motion vector search, this is tedious and more complicated because there are |
3289 |
|
two values for everything, always one for backward and one for forward ME. Still, we don't gain |
3290 |
|
much from this search, maybe it should simply be skipped and simply current i_sad16 value used |
3291 |
|
as "optimal". */ |
3292 |
|
|
3293 |
|
i_sad16 = Diamond16_InterpolMainSearch( |
3294 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3295 |
|
frame->image.y + i * 16 + j * 16 * edged_width, |
3296 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3297 |
|
i, j, |
3298 |
|
mb->mvs[0].x, mb->mvs[0].y, |
3299 |
|
mb->b_mvs[0].x, mb->b_mvs[0].y, |
3300 |
|
i_sad16, |
3301 |
|
&f_interpolMV, &b_interpolMV, |
3302 |
|
f_predMV.x, f_predMV.y, b_predMV.x, b_predMV.y, |
3303 |
|
f_min_dx, f_max_dx, f_min_dy, f_max_dy, |
3304 |
|
b_min_dx, b_max_dx, b_min_dy, b_max_dy, |
3305 |
|
edged_width, 1, |
3306 |
|
frame->fcode, frame->bcode,frame->quant,0); |
3307 |
|
|
3308 |
|
|
3309 |
|
/* DIRECT MODE DELTA VECTOR SEARCH. |
3310 |
|
This has to be made more effective, but at the moment I'm happy it's running at all */ |
3311 |
|
|
3312 |
|
/* range is taken without fcode restriction, just a hack instead of writing down the dimensions, of course */ |
3313 |
|
|
3314 |
|
get_range(&min_dx, &max_dx, &min_dy, &max_dy, i, j, 16, iWidth, iHeight, 19); |
3315 |
|
|
3316 |
|
d_sad16 = Diamond16_DirectMainSearch( |
3317 |
|
f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
3318 |
|
frame->image.y + i*16 + j*16*edged_width, |
3319 |
|
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
3320 |
|
i, j, |
3321 |
|
TRB,TRD, |
3322 |
|
0,0, |
3323 |
|
d_sad16, |
3324 |
|
&mb->deltamv, |
3325 |
|
mb->directmv, // this has to be pre-initialized with b_mb->mvs[} |
3326 |
|
min_dx, max_dx, min_dy, max_dy, |
3327 |
|
edged_width, 1, frame->quant, 0); |
3328 |
|
|
3329 |
|
|
3330 |
|
// i_sad16 = 65535; /* remove the comment to disable any of the MODEs */ |
3331 |
|
// f_sad16 = 65535; |
3332 |
|
// b_sad16 = 65535; |
3333 |
|
// d_sad16 = 65535; |
3334 |
|
|
3335 |
if (f_sad16 < b_sad16) { |
if (f_sad16 < b_sad16) { |
3336 |
best_sad = f_sad16; |
best_sad = f_sad16; |
3346 |
} |
} |
3347 |
|
|
3348 |
if (d_sad16 < best_sad) { |
if (d_sad16 < best_sad) { |
3349 |
|
|
3350 |
|
if (b_mb->mode == MODE_INTER4V) |
3351 |
|
{ |
3352 |
|
|
3353 |
|
/* how to calc vectors is defined in standard. mvs[] and b_mvs[] are only for motion compensation */ |
3354 |
|
/* for the bitstream, the value mb->deltamv is read directly */ |
3355 |
|
|
3356 |
|
for (k = 0; k < 4; k++) { |
3357 |
|
|
3358 |
|
mb->mvs[k].x = (int32_t) ((TRB * mb->directmv[k].x) / TRD + mb->deltamv.x); |
3359 |
|
mb->b_mvs[k].x = (int32_t) ((mb->deltamv.x == 0) |
3360 |
|
? ((TRB - TRD) * mb->directmv[k].x) / TRD |
3361 |
|
: mb->mvs[k].x - mb->directmv[k].x); |
3362 |
|
|
3363 |
|
mb->mvs[k].y = (int32_t) ((TRB * mb->directmv[k].y) / TRD + mb->deltamv.y); |
3364 |
|
mb->b_mvs[k].y = (int32_t) ((mb->deltamv.y == 0) |
3365 |
|
? ((TRB - TRD) * mb->directmv[k].y) / TRD |
3366 |
|
: mb->mvs[k].y - mb->directmv[k].y); |
3367 |
|
} |
3368 |
|
} |
3369 |
|
else |
3370 |
|
{ |
3371 |
|
mb->mvs[0].x = (int32_t) ((TRB * mb->directmv[0].x) / TRD + mb->deltamv.x); |
3372 |
|
|
3373 |
|
mb->b_mvs[0].x = (int32_t) ((mb->deltamv.x == 0) |
3374 |
|
? ((TRB - TRD) * mb->directmv[0].x) / TRD |
3375 |
|
: mb->mvs[0].x - mb->directmv[0].x); |
3376 |
|
|
3377 |
|
mb->mvs[0].y = (int32_t) ((TRB * mb->directmv[0].y) / TRD + mb->deltamv.y); |
3378 |
|
|
3379 |
|
mb->b_mvs[0].y = (int32_t) ((mb->deltamv.y == 0) |
3380 |
|
? ((TRB - TRD) * mb->directmv[0].y) / TRD |
3381 |
|
: mb->mvs[0].y - mb->directmv[0].y); |
3382 |
|
|
3383 |
|
mb->mvs[3] = mb->mvs[2] = mb->mvs[1] = mb->mvs[0]; |
3384 |
|
mb->b_mvs[3] = mb->b_mvs[2] = mb->b_mvs[1] = mb->b_mvs[0]; |
3385 |
|
} |
3386 |
|
|
3387 |
best_sad = d_sad16; |
best_sad = d_sad16; |
3388 |
mb->mode = MODE_DIRECT; |
mb->mode = MODE_DIRECT; |
3389 |
} |
} |
3390 |
|
|
3391 |
|
switch (mb->mode) |
3392 |
|
{ |
3393 |
|
case MODE_FORWARD: |
3394 |
|
f_count++; |
3395 |
|
f_predMV = mb->mvs[0]; |
3396 |
|
break; |
3397 |
|
case MODE_BACKWARD: |
3398 |
|
b_count++; |
3399 |
|
b_predMV = mb->b_mvs[0]; |
3400 |
|
|
3401 |
|
break; |
3402 |
|
case MODE_INTERPOLATE: |
3403 |
|
i_count++; |
3404 |
|
mb->mvs[0] = f_interpolMV; |
3405 |
|
mb->b_mvs[0] = b_interpolMV; |
3406 |
|
f_predMV = mb->mvs[0]; |
3407 |
|
b_predMV = mb->b_mvs[0]; |
3408 |
|
break; |
3409 |
|
case MODE_DIRECT: |
3410 |
|
d_count++; |
3411 |
|
break; |
3412 |
|
default: |
3413 |
|
break; |
3414 |
} |
} |
3415 |
|
|
3416 |
} |
} |
3417 |
} |
} |
3418 |
|
|
3419 |
|
#ifdef _DEBUG_BFRAME_STAT |
3420 |
|
fprintf(stderr,"B-Stat: F: %04d B: %04d I: %04d D: %04d\n", |
3421 |
|
f_count,b_count,i_count,d_count); |
3422 |
|
#endif |
3423 |
|
|
3424 |
|
} |