2 |
* |
* |
3 |
* Modifications: |
* Modifications: |
4 |
* |
* |
5 |
|
* 25.04.2002 partial prevMB conversion |
6 |
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
7 |
* 14.04.2002 added MotionEstimationBVOP() |
* 14.04.2002 added MotionEstimationBVOP() |
8 |
* 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
* 02.04.2002 add EPZS(^2) as ME algorithm, use PMV_USESQUARES to choose between |
56 |
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
57 |
/* nb = vop pixels * 2^(bpp-8) */ |
/* nb = vop pixels * 2^(bpp-8) */ |
58 |
#define MV16_00_BIAS (128+1) |
#define MV16_00_BIAS (128+1) |
59 |
|
#define MV8_00_BIAS (0) |
60 |
|
|
61 |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
62 |
#define INTER_BIAS 512 |
#define INTER_BIAS 512 |
81 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
82 |
const int x, const int y, |
const int x, const int y, |
83 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
84 |
|
const uint32_t iQuant, |
85 |
|
const uint32_t iFcode, |
86 |
const MBParam * const pParam, |
const MBParam * const pParam, |
87 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
88 |
|
const MACROBLOCK * const prevMBs, |
89 |
VECTOR * const currMV, |
VECTOR * const currMV, |
90 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
91 |
|
|
97 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
98 |
const int x, const int y, |
const int x, const int y, |
99 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
100 |
|
const uint32_t iQuant, |
101 |
|
const uint32_t iFcode, |
102 |
const MBParam * const pParam, |
const MBParam * const pParam, |
103 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
104 |
|
const MACROBLOCK * const prevMBs, |
105 |
VECTOR * const currMV, |
VECTOR * const currMV, |
106 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
107 |
|
|
113 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
114 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
115 |
const int x, const int y, |
const int x, const int y, |
116 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
117 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
118 |
|
const uint32_t iQuant, |
119 |
|
const uint32_t iFcode, |
120 |
const MBParam * const pParam, |
const MBParam * const pParam, |
121 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
122 |
|
const MACROBLOCK * const prevMBs, |
123 |
VECTOR * const currMV, |
VECTOR * const currMV, |
124 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
125 |
|
|
130 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
131 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
132 |
const int x, const int y, |
const int x, const int y, |
133 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
134 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
135 |
|
const uint32_t iQuant, |
136 |
|
const uint32_t iFcode, |
137 |
const MBParam * const pParam, |
const MBParam * const pParam, |
138 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
139 |
|
const MACROBLOCK * const prevMBs, |
140 |
VECTOR * const currMV, |
VECTOR * const currMV, |
141 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
142 |
|
|
246 |
#endif |
#endif |
247 |
|
|
248 |
bool MotionEstimation( |
bool MotionEstimation( |
|
MACROBLOCK * const pMBs, |
|
249 |
MBParam * const pParam, |
MBParam * const pParam, |
250 |
const IMAGE * const pRef, |
FRAMEINFO * const current, |
251 |
|
FRAMEINFO * const reference, |
252 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
253 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
254 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
|
IMAGE * const pCurrent, |
|
255 |
const uint32_t iLimit) |
const uint32_t iLimit) |
256 |
|
|
257 |
{ |
{ |
258 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
259 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
260 |
|
MACROBLOCK * pMBs = current->mbs; |
261 |
|
IMAGE * pCurrent = ¤t->image; |
262 |
|
|
263 |
|
MACROBLOCK * prevMBs = reference->mbs; // previous frame |
264 |
|
IMAGE * pRef = &reference->image; |
265 |
|
|
266 |
|
|
267 |
uint32_t i, j, iIntra = 0; |
uint32_t i, j, iIntra = 0; |
268 |
|
|
276 |
if (sadInit) |
if (sadInit) |
277 |
(*sadInit)(); |
(*sadInit)(); |
278 |
|
|
279 |
|
|
280 |
|
/* eventhough we have a seperate prevMBs, |
281 |
|
pmvfast/epsz does something "funny" with the previous frames data */ |
282 |
|
|
283 |
|
/* for (i = 0; i < iHcount; i++) |
284 |
|
for (j = 0; j < iWcount; j++) |
285 |
|
{ |
286 |
|
pMBs[j + i * iWcount].mvs[0] = prevMBs[j + i * iWcount].mvs[0]; |
287 |
|
pMBs[j + i * iWcount].mvs[1] = prevMBs[j + i * iWcount].mvs[1]; |
288 |
|
pMBs[j + i * iWcount].mvs[2] = prevMBs[j + i * iWcount].mvs[2]; |
289 |
|
pMBs[j + i * iWcount].mvs[3] = prevMBs[j + i * iWcount].mvs[3]; |
290 |
|
} |
291 |
|
*/ |
292 |
|
/*dprintf("*** BEFORE ***"); |
293 |
|
for (i = 0; i < iHcount; i++) |
294 |
|
for (j = 0; j < iWcount; j++) |
295 |
|
{ |
296 |
|
dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, |
297 |
|
pMBs[j + i * iWcount].mode, |
298 |
|
pMBs[j + i * iWcount].dquant, |
299 |
|
pMBs[j + i * iWcount].mvs[0], |
300 |
|
pMBs[j + i * iWcount].mvs[1], |
301 |
|
pMBs[j + i * iWcount].mvs[2], |
302 |
|
pMBs[j + i * iWcount].mvs[3], |
303 |
|
prevMBs[j + i * iWcount].sad8[0], |
304 |
|
prevMBs[j + i * iWcount].sad8[1], |
305 |
|
prevMBs[j + i * iWcount].sad8[2], |
306 |
|
prevMBs[j + i * iWcount].sad8[3], |
307 |
|
prevMBs[j + i * iWcount].sad16); |
308 |
|
} |
309 |
|
*/ |
310 |
|
|
311 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
312 |
for (i = 0; i < iHcount; i++) |
for (i = 0; i < iHcount; i++) |
313 |
for (j = 0; j < iWcount; j++) |
for (j = 0; j < iWcount; j++) |
314 |
{ |
{ |
315 |
MACROBLOCK *pMB = &pMBs[j + i * iWcount]; |
MACROBLOCK *pMB = &pMBs[j + i * iWcount]; |
316 |
|
MACROBLOCK *prevMB = &prevMBs[j + i * iWcount]; |
317 |
|
|
318 |
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
319 |
j, i, pParam->motion_flags, |
j, i, current->motion_flags, current->quant, current->fcode, |
320 |
pParam, pMBs, &mv16, &pmv16); |
pParam, pMBs, prevMBs, &mv16, &pmv16); |
321 |
pMB->sad16=sad16; |
pMB->sad16=sad16; |
322 |
|
|
323 |
|
|
333 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
334 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
335 |
|
|
336 |
|
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
337 |
|
|
338 |
iIntra++; |
iIntra++; |
339 |
if(iIntra >= iLimit) |
if(iIntra >= iLimit) |
340 |
return 1; |
return 1; |
342 |
continue; |
continue; |
343 |
} |
} |
344 |
|
|
345 |
if (pParam->global_flags & XVID_INTER4V) |
if (current->global_flags & XVID_INTER4V) |
346 |
{ |
{ |
347 |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
348 |
2 * j, 2 * i, mv16.x, mv16.y, pParam->motion_flags, |
2 * j, 2 * i, mv16.x, mv16.y, |
349 |
pParam, pMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
current->motion_flags, current->quant, current->fcode, |
350 |
|
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
351 |
|
|
352 |
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
353 |
2 * j + 1, 2 * i, mv16.x, mv16.y, pParam->motion_flags, |
2 * j + 1, 2 * i, mv16.x, mv16.y, |
354 |
pParam, pMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
current->motion_flags, current->quant, current->fcode, |
355 |
|
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
356 |
|
|
357 |
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
358 |
2 * j, 2 * i + 1, mv16.x, mv16.y, pParam->motion_flags, |
2 * j, 2 * i + 1, mv16.x, mv16.y, |
359 |
pParam, pMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
current->motion_flags, current->quant, current->fcode, |
360 |
|
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
361 |
|
|
362 |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
363 |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, pParam->motion_flags, |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
364 |
pParam, pMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
current->motion_flags, current->quant, current->fcode, |
365 |
|
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
366 |
|
|
367 |
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
368 |
} |
} |
372 |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
373 |
*/ |
*/ |
374 |
|
|
375 |
if (pMB->dquant == NO_CHANGE) { |
if (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) |
376 |
if (((pParam->global_flags & XVID_INTER4V)==0) || |
{ |
377 |
(sad16 < (sad8 + (int32_t)(IMV16X16 * pParam->quant)))) { |
if (((current->global_flags & XVID_INTER4V)==0) || |
378 |
|
(sad16 < (sad8 + (int32_t)(IMV16X16 * current->quant)))) |
379 |
|
{ |
380 |
|
|
381 |
sad8 = sad16; |
sad8 = sad16; |
382 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
399 |
} |
} |
400 |
} |
} |
401 |
|
|
402 |
|
/* dprintf("*** AFTER ***", pMBs[0].b_mvs[0].x); |
403 |
|
for (i = 0; i < iHcount; i++) |
404 |
|
for (j = 0; j < iWcount; j++) |
405 |
|
{ |
406 |
|
dprintf(" [%i,%i] mode=%i dquant=%i mvs=(%i %i %i %i) sad8=(%i %i %i %i) sad16=(%i)", j,i, |
407 |
|
pMBs[j + i * iWcount].mode, |
408 |
|
pMBs[j + i * iWcount].dquant, |
409 |
|
pMBs[j + i * iWcount].mvs[0], |
410 |
|
pMBs[j + i * iWcount].mvs[1], |
411 |
|
pMBs[j + i * iWcount].mvs[2], |
412 |
|
pMBs[j + i * iWcount].mvs[3], |
413 |
|
pMBs[j + i * iWcount].sad8[0], |
414 |
|
pMBs[j + i * iWcount].sad8[1], |
415 |
|
pMBs[j + i * iWcount].sad8[2], |
416 |
|
pMBs[j + i * iWcount].sad8[3], |
417 |
|
pMBs[j + i * iWcount].sad16); |
418 |
|
} |
419 |
|
*/ |
420 |
|
|
421 |
return 0; |
return 0; |
422 |
} |
} |
423 |
|
|
432 |
{ \ |
{ \ |
433 |
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); \ |
434 |
iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_16(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
|
if (iSAD <= iQuant * 96) \ |
|
|
iSAD -= MV16_00_BIAS; \ |
|
435 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
436 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } } \ |
437 |
} |
} |
529 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
530 |
const int x, const int y, |
const int x, const int y, |
531 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
532 |
|
const uint32_t iQuant, |
533 |
|
const uint32_t iFcode, |
534 |
MBParam * const pParam, |
MBParam * const pParam, |
535 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
536 |
|
const MACROBLOCK * const prevMBs, |
537 |
VECTOR * const currMV, |
VECTOR * const currMV, |
538 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
539 |
{ |
{ |
540 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
|
const int32_t iQuant = pParam->quant; |
|
541 |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
const uint8_t * cur = pCur->y + x*16 + y*16*iEdgedWidth; |
542 |
int32_t iSAD; |
int32_t iSAD; |
543 |
int32_t pred_x,pred_y; |
int32_t pred_x,pred_y; |
859 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
860 |
const int x, const int y, |
const int x, const int y, |
861 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
862 |
|
const uint32_t iQuant, |
863 |
|
const uint32_t iFcode, |
864 |
const MBParam * const pParam, |
const MBParam * const pParam, |
865 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
866 |
|
const MACROBLOCK * const prevMBs, |
867 |
VECTOR * const currMV, |
VECTOR * const currMV, |
868 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
869 |
{ |
{ |
870 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
|
const int32_t iFcode = pParam->fixed_code; |
|
|
const int32_t iQuant = pParam->quant; |
|
871 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
872 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
873 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
889 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
890 |
int32_t psad[4]; |
int32_t psad[4]; |
891 |
|
|
892 |
MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
893 |
|
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
894 |
|
|
895 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
896 |
int32_t bPredEq; |
int32_t bPredEq; |
934 |
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
935 |
*/ |
*/ |
936 |
|
|
937 |
if ((bPredEq) && (MVequal(pmv[0],pMB->mvs[0]) ) ) |
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[0]) ) ) |
938 |
iFound=2; |
iFound=2; |
939 |
|
|
940 |
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
988 |
iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth, MV_MAX_ERROR); |
989 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; |
990 |
|
|
991 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,pMB->mvs[0])) && ((uint32_t)iMinSAD < pMB->sad16) ) ) |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
992 |
{ |
{ |
993 |
|
|
994 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1001 |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1002 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1003 |
Let MinSAD be the smallest SAD up to this point. |
Let MinSAD be the smallest SAD up to this point. |
1004 |
If MV is (0,0) subtract offset. ******** WHAT'S THIS 'OFFSET' ??? *********** |
If MV is (0,0) subtract offset. |
1005 |
*/ |
*/ |
1006 |
|
|
1007 |
// (0,0) is always possible |
// (0,0) is always possible |
1009 |
CHECK_MV16_ZERO; |
CHECK_MV16_ZERO; |
1010 |
|
|
1011 |
// previous frame MV is always possible |
// previous frame MV is always possible |
1012 |
CHECK_MV16_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1013 |
|
|
1014 |
// left neighbour, if allowed |
// left neighbour, if allowed |
1015 |
if (x != 0) |
if (x != 0) |
1041 |
} |
} |
1042 |
} |
} |
1043 |
|
|
1044 |
|
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) && (iSAD <= iQuant * 96) ) |
1045 |
|
iMinSAD -= MV16_00_BIAS; |
1046 |
|
|
1047 |
|
|
1048 |
/* Step 6: If MinSAD <= thresa goto Step 10. |
/* Step 6: If MinSAD <= thresa goto Step 10. |
1049 |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
1050 |
*/ |
*/ |
1051 |
|
|
1052 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,pMB->mvs[0]) && ((uint32_t)iMinSAD < pMB->sad16) ) ) |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1053 |
{ |
{ |
1054 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1055 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1235 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1236 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1237 |
const int x, const int y, |
const int x, const int y, |
1238 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
1239 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1240 |
|
const uint32_t iQuant, |
1241 |
|
const uint32_t iFcode, |
1242 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1243 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1244 |
|
const MACROBLOCK * const prevMBs, |
1245 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1246 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1247 |
{ |
{ |
1248 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
|
|
|
|
const int32_t iFcode = pParam->fixed_code; |
|
|
const int32_t iQuant = pParam->quant; |
|
1249 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1250 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1251 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1264 |
VECTOR newMV; |
VECTOR newMV; |
1265 |
VECTOR backupMV; |
VECTOR backupMV; |
1266 |
|
|
1267 |
MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1268 |
|
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1269 |
|
|
1270 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1271 |
int32_t iFound,bPredEq; |
int32_t iFound,bPredEq; |
1277 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1278 |
x, y, 8, iWidth, iHeight, iFcode); |
x, y, 8, iWidth, iHeight, iFcode); |
1279 |
|
|
|
/* we work with abs. MVs, not relative to prediction, so range is relative to 0,0 */ |
|
|
|
|
1280 |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
1281 |
{ min_dx = EVEN(min_dx); |
{ min_dx = EVEN(min_dx); |
1282 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1309 |
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1310 |
*/ |
*/ |
1311 |
|
|
1312 |
if ((bPredEq) && (MVequal(pmv[0],pMB->mvs[iSubBlock]) ) ) |
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[iSubBlock]) ) ) |
1313 |
iFound=2; |
iFound=2; |
1314 |
|
|
1315 |
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1342 |
iEdgedWidth); |
iEdgedWidth); |
1343 |
iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode) * iQuant; |
iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode) * iQuant; |
1344 |
|
|
1345 |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,pMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < pMB->sad8[iSubBlock]) ) ) |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1346 |
{ |
{ |
1347 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1348 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1355 |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
Step 5: Calculate SAD for motion vectors taken from left block, top, top-right, and Previous frame block. |
1356 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1357 |
Let MinSAD be the smallest SAD up to this point. |
Let MinSAD be the smallest SAD up to this point. |
1358 |
If MV is (0,0) subtract offset. ******** WHAT'S THIS 'OFFSET' ??? *********** |
If MV is (0,0) subtract offset. |
1359 |
*/ |
*/ |
1360 |
|
|
1361 |
// the prediction might be even better than mv16 |
// the prediction might be even better than mv16 |
1365 |
CHECK_MV8_ZERO; |
CHECK_MV8_ZERO; |
1366 |
|
|
1367 |
// previous frame MV is always possible |
// previous frame MV is always possible |
1368 |
CHECK_MV8_CANDIDATE(pMB->mvs[iSubBlock].x,pMB->mvs[iSubBlock].y); |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
1369 |
|
|
1370 |
// left neighbour, if allowed |
// left neighbour, if allowed |
1371 |
if (psad[1] != MV_MAX_ERROR) |
if (psad[1] != MV_MAX_ERROR) |
1397 |
} |
} |
1398 |
} |
} |
1399 |
|
|
1400 |
|
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) && (iSAD <= iQuant * 96) ) |
1401 |
|
iMinSAD -= MV8_00_BIAS; |
1402 |
|
|
1403 |
|
|
1404 |
/* Step 6: If MinSAD <= thresa goto Step 10. |
/* Step 6: If MinSAD <= thresa goto Step 10. |
1405 |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
If Motion Vector equal to Previous frame motion vector and MinSAD<PrevFrmSAD goto Step 10. |
1406 |
*/ |
*/ |
1407 |
|
|
1408 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,pMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < pMB->sad8[iSubBlock]) ) ) |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1409 |
{ |
{ |
1410 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1411 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1495 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1496 |
const int x, const int y, |
const int x, const int y, |
1497 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1498 |
|
const uint32_t iQuant, |
1499 |
|
const uint32_t iFcode, |
1500 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1501 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1502 |
|
const MACROBLOCK * const prevMBs, |
1503 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1504 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1505 |
{ |
{ |
1506 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
1507 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
|
const int32_t iFcode = pParam->fixed_code; |
|
|
const int32_t iQuant = pParam->quant; |
|
1508 |
|
|
1509 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1510 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1524 |
int32_t psad[8]; |
int32_t psad[8]; |
1525 |
|
|
1526 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
1527 |
MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1528 |
|
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1529 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
1530 |
|
|
1531 |
static int32_t thresh2; |
static int32_t thresh2; |
1536 |
|
|
1537 |
if (oldMBs == NULL) |
if (oldMBs == NULL) |
1538 |
{ oldMBs = (MACROBLOCK*) calloc(1,iWcount*iHcount*sizeof(MACROBLOCK)); |
{ oldMBs = (MACROBLOCK*) calloc(1,iWcount*iHcount*sizeof(MACROBLOCK)); |
1539 |
fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
1540 |
} |
} |
1541 |
oldMB = oldMBs + x + y * iWcount; |
oldMB = oldMBs + x + y * iWcount; |
1542 |
|
|
1544 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1545 |
x, y, 16, iWidth, iHeight, iFcode); |
x, y, 16, iWidth, iHeight, iFcode); |
1546 |
|
|
|
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
|
|
|
|
1547 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1548 |
{ min_dx = EVEN(min_dx); |
{ min_dx = EVEN(min_dx); |
1549 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1586 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode) * iQuant; |
1587 |
|
|
1588 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
1589 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,pMB->mvs[0])) && ((uint32_t)iMinSAD < pMB->sad16) ) ) |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,pMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
1590 |
{ |
{ |
1591 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1592 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
1653 |
*/ |
*/ |
1654 |
|
|
1655 |
if ( (iMinSAD <= thresh2) |
if ( (iMinSAD <= thresh2) |
1656 |
|| ( MVequal(*currMV,pMB->mvs[0]) && ((uint32_t)iMinSAD <= pMB->sad16) ) ) |
|| ( MVequal(*currMV,pMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
1657 |
{ |
{ |
1658 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1659 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
1779 |
const int x, const int y, |
const int x, const int y, |
1780 |
const int start_x, const int start_y, |
const int start_x, const int start_y, |
1781 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1782 |
|
const uint32_t iQuant, |
1783 |
|
const uint32_t iFcode, |
1784 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1785 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1786 |
|
const MACROBLOCK * const prevMBs, |
1787 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1788 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1789 |
{ |
{ |
1790 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
|
const int32_t iFcode = pParam->fixed_code; |
|
|
const int32_t iQuant = pParam->quant; |
|
|
|
|
1791 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1792 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1793 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1809 |
|
|
1810 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
1811 |
|
|
1812 |
MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1813 |
|
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1814 |
|
|
1815 |
int32_t bPredEq; |
int32_t bPredEq; |
1816 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |