60 |
#define MV8_00_BIAS (0) |
#define MV8_00_BIAS (0) |
61 |
|
|
62 |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
63 |
#define INTER_BIAS 512 |
#define MV16_INTER_BIAS 512 |
64 |
|
|
65 |
/* Parameters which control inter/inter4v decision */ |
/* Parameters which control inter/inter4v decision */ |
66 |
#define IMV16X16 5 |
#define IMV16X16 5 |
69 |
#define NEIGH_TEND_16X16 2 |
#define NEIGH_TEND_16X16 2 |
70 |
#define NEIGH_TEND_8X8 2 |
#define NEIGH_TEND_8X8 2 |
71 |
|
|
|
|
|
72 |
// fast ((A)/2)*2 |
// fast ((A)/2)*2 |
73 |
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
74 |
|
|
75 |
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
76 |
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
77 |
|
|
78 |
int32_t PMVfastSearch16( |
int32_t PMVfastSearch16( |
79 |
const uint8_t * const pRef, |
const uint8_t * const pRef, |
272 |
{ |
{ |
273 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
274 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
275 |
MACROBLOCK * pMBs = current->mbs; |
MACROBLOCK * const pMBs = current->mbs; |
276 |
IMAGE * pCurrent = ¤t->image; |
MACROBLOCK * const prevMBs = reference->mbs; // previous frame |
|
|
|
|
MACROBLOCK * prevMBs = reference->mbs; // previous frame |
|
|
IMAGE * pRef = &reference->image; |
|
|
|
|
277 |
|
|
278 |
uint32_t i, j, iIntra = 0; |
const IMAGE * const pCurrent = ¤t->image; |
279 |
|
const IMAGE * const pRef = &reference->image; |
280 |
|
|
281 |
VECTOR mv16; |
const VECTOR zeroMV = {0,0}; |
|
VECTOR pmv16; |
|
282 |
|
|
283 |
int32_t sad8 = 0; |
int32_t x, y; |
284 |
int32_t sad16; |
int32_t iIntra = 0; |
285 |
int32_t deviation; |
VECTOR pmv; |
286 |
|
|
287 |
if (sadInit) |
if (sadInit) |
288 |
(*sadInit)(); |
(*sadInit)(); |
289 |
|
|
290 |
// note: i==horizontal, j==vertical |
for (y = 0; y < iHcount; y++) |
291 |
for (i = 0; i < iHcount; i++) |
for (x = 0; x < iWcount; x++) |
|
for (j = 0; j < iWcount; j++) |
|
292 |
{ |
{ |
293 |
MACROBLOCK *pMB = &pMBs[j + i * iWcount]; |
MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; |
|
MACROBLOCK *prevMB = &prevMBs[j + i * iWcount]; |
|
|
|
|
|
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
j, i, current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &mv16, &pmv16); |
|
|
pMB->sad16=sad16; |
|
294 |
|
|
295 |
|
pMB->sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
296 |
|
x, y, current->motion_flags, current->quant, current->fcode, |
297 |
|
pParam, pMBs, prevMBs, &pMB->mv16, &pMB->pmvs[0]); |
298 |
|
} |
299 |
|
|
300 |
/* decide: MODE_INTER or MODE_INTRA |
for (y = 0; y < iHcount; y++) |
301 |
if (dev_intra < sad_inter - 2 * nb) use_intra |
for (x = 0; x < iWcount; x++) |
302 |
*/ |
{ |
303 |
|
MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; |
304 |
|
|
305 |
deviation = dev16(pCurrent->y + j*16 + i*16*pParam->edged_width, pParam->edged_width); |
if (0 < (pMB->sad16 - MV16_INTER_BIAS)) |
306 |
|
{ |
307 |
|
int32_t deviation; |
308 |
|
deviation = dev16(pCurrent->y + x*16 + y*16*pParam->edged_width, |
309 |
|
pParam->edged_width); |
310 |
|
|
311 |
if (deviation < (sad16 - INTER_BIAS)) |
if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) |
312 |
{ |
{ |
313 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
314 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
pMB->mv16 = pMB->mvs[0] = pMB->mvs[1] |
315 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
= pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
316 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] |
317 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
= pMB->sad8[2] = pMB->sad8[3] = 0; |
318 |
|
|
319 |
iIntra++; |
iIntra++; |
320 |
if(iIntra >= iLimit) |
if(iIntra >= iLimit) |
322 |
|
|
323 |
continue; |
continue; |
324 |
} |
} |
325 |
|
} |
326 |
|
|
327 |
if (current->global_flags & XVID_INTER4V) |
if ( (current->global_flags & XVID_INTER4V) |
328 |
{ |
&& (!(current->global_flags & XVID_LUMIMASKING) |
329 |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|| pMB->dquant == NO_CHANGE) ) |
330 |
2 * j, 2 * i, mv16.x, mv16.y, |
{ |
331 |
|
int32_t sad8 = 129; //IMV16X16 * current->quant; |
332 |
|
|
333 |
|
if (sad8 < pMB->sad16) |
334 |
|
sad8 += pMB->sad8[0] |
335 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
336 |
|
2*x, 2*y, pMB->mv16.x, pMB->mv16.y, |
337 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
338 |
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
339 |
|
|
340 |
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
341 |
2 * j + 1, 2 * i, mv16.x, mv16.y, |
sad8 += pMB->sad8[1] |
342 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
343 |
|
2*x+1, 2*y, pMB->mv16.x, pMB->mv16.y, |
344 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
345 |
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
346 |
|
|
347 |
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
348 |
2 * j, 2 * i + 1, mv16.x, mv16.y, |
sad8 += pMB->sad8[2] |
349 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
350 |
|
2*x, 2*y+1, pMB->mv16.x, pMB->mv16.y, |
351 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
352 |
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
353 |
|
|
354 |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
355 |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
sad8 += pMB->sad8[3] |
356 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
357 |
|
2*x+1, 2*y+1, pMB->mv16.x, pMB->mv16.y, |
358 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
359 |
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
360 |
|
|
|
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
|
|
} |
|
|
|
|
|
|
|
361 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
362 |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
363 |
*/ |
*/ |
364 |
|
|
365 |
if (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) |
if (sad8 < pMB->sad16) |
|
{ |
|
|
if (((current->global_flags & XVID_INTER4V)==0) || |
|
|
(sad16 < (sad8 + (int32_t)(IMV16X16 * current->quant)))) |
|
|
{ |
|
|
|
|
|
sad8 = sad16; |
|
|
pMB->mode = MODE_INTER; |
|
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
|
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
|
|
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad16; |
|
|
pMB->pmvs[0].x = pmv16.x; |
|
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
|
else |
|
366 |
{ |
{ |
367 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
368 |
pMB->sad8[0] *= 4; |
pMB->sad8[0] *= 4; |
369 |
pMB->sad8[1] *= 4; |
pMB->sad8[1] *= 4; |
370 |
pMB->sad8[2] *= 4; |
pMB->sad8[2] *= 4; |
371 |
pMB->sad8[3] *= 4; |
pMB->sad8[3] *= 4; |
372 |
|
continue; |
373 |
} |
} |
374 |
} |
} |
375 |
else |
|
|
{ |
|
|
sad8 = sad16; |
|
376 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
377 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
378 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = pMB->sad16; |
379 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad16; |
|
380 |
|
if (current->global_flags & XVID_INTER4V) |
381 |
|
{ pmv = get_pmv(pMBs, x, y, pParam->mb_width, 0); |
382 |
|
// get_pmv has to be called again. inter4v changes predictors |
383 |
|
|
384 |
pMB->pmvs[0].x = pmv16.x; |
pMB->pmvs[0].x = pMB->mv16.x - pmv.x; |
385 |
pMB->pmvs[0].y = pmv16.y; |
pMB->pmvs[0].y = pMB->mv16.y - pmv.y; |
386 |
} |
} |
387 |
} |
} |
|
|
|
388 |
return 0; |
return 0; |
389 |
} |
} |
390 |
|
|
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
|
|
|
|
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
|
|
|
|
|
|
|
391 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
392 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
393 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
851 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
852 |
int32_t psad[4]; |
int32_t psad[4]; |
853 |
|
|
854 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
855 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
856 |
|
|
857 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1249 |
VECTOR backupMV; |
VECTOR backupMV; |
1250 |
VECTOR startMV; |
VECTOR startMV; |
1251 |
|
|
1252 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1253 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1254 |
|
|
1255 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1541 |
int32_t psad[8]; |
int32_t psad[8]; |
1542 |
|
|
1543 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
1544 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1545 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1546 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
1547 |
|
|
1828 |
|
|
1829 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
1830 |
|
|
1831 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1832 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1833 |
|
|
1834 |
int32_t bPredEq; |
int32_t bPredEq; |