54 |
#define MV16_THRESHOLD 192 |
#define MV16_THRESHOLD 192 |
55 |
#define MV8_THRESHOLD 56 |
#define MV8_THRESHOLD 56 |
56 |
|
|
57 |
|
#define NEIGH_MOVE_THRESH 0 |
58 |
|
// how much a block's MV must differ from his neighbour |
59 |
|
// to be search for INTER4V. The more, the faster... |
60 |
|
|
61 |
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
/* sad16(0,0) bias; mpeg4 spec suggests nb/2+1 */ |
62 |
/* nb = vop pixels * 2^(bpp-8) */ |
/* nb = vop pixels * 2^(bpp-8) */ |
63 |
#define MV16_00_BIAS (128+1) |
#define MV16_00_BIAS (128+1) |
64 |
#define MV8_00_BIAS (0) |
#define MV8_00_BIAS (0) |
65 |
|
|
66 |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
67 |
#define INTER_BIAS 512 |
#define MV16_INTER_BIAS 512 |
68 |
|
|
69 |
/* Parameters which control inter/inter4v decision */ |
/* Parameters which control inter/inter4v decision */ |
70 |
#define IMV16X16 5 |
#define IMV16X16 5 |
73 |
#define NEIGH_TEND_16X16 2 |
#define NEIGH_TEND_16X16 2 |
74 |
#define NEIGH_TEND_8X8 2 |
#define NEIGH_TEND_8X8 2 |
75 |
|
|
|
|
|
76 |
// fast ((A)/2)*2 |
// fast ((A)/2)*2 |
77 |
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
#define EVEN(A) (((A)<0?(A)+1:(A)) & ~1) |
78 |
|
|
79 |
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
80 |
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
81 |
|
|
82 |
int32_t PMVfastSearch16( |
int32_t PMVfastSearch16( |
83 |
const uint8_t * const pRef, |
const uint8_t * const pRef, |
276 |
{ |
{ |
277 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
278 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
279 |
MACROBLOCK * pMBs = current->mbs; |
MACROBLOCK * const pMBs = current->mbs; |
280 |
IMAGE * pCurrent = ¤t->image; |
MACROBLOCK * const prevMBs = reference->mbs; // previous frame |
281 |
|
|
282 |
MACROBLOCK * prevMBs = reference->mbs; // previous frame |
const IMAGE * const pCurrent = ¤t->image; |
283 |
IMAGE * pRef = &reference->image; |
const IMAGE * const pRef = &reference->image; |
284 |
|
|
285 |
|
const VECTOR zeroMV = {0,0}; |
286 |
|
|
287 |
uint32_t i, j, iIntra = 0; |
int32_t x, y; |
288 |
|
int32_t iIntra = 0; |
289 |
VECTOR mv16; |
VECTOR pmv; |
|
VECTOR pmv16; |
|
|
|
|
|
int32_t sad8 = 0; |
|
|
int32_t sad16; |
|
|
int32_t deviation; |
|
290 |
|
|
291 |
if (sadInit) |
if (sadInit) |
292 |
(*sadInit)(); |
(*sadInit)(); |
293 |
|
|
294 |
// note: i==horizontal, j==vertical |
for (y = 0; y < iHcount; y++) |
295 |
for (i = 0; i < iHcount; i++) |
for (x = 0; x < iWcount; x++) |
|
for (j = 0; j < iWcount; j++) |
|
296 |
{ |
{ |
297 |
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; |
|
298 |
|
|
299 |
|
pMB->sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
300 |
|
x, y, current->motion_flags, current->quant, current->fcode, |
301 |
|
pParam, pMBs, prevMBs, &pMB->mv16, &pMB->pmvs[0]); |
302 |
|
|
|
/* decide: MODE_INTER or MODE_INTRA |
|
|
if (dev_intra < sad_inter - 2 * nb) use_intra |
|
|
*/ |
|
303 |
|
|
304 |
deviation = dev16(pCurrent->y + j*16 + i*16*pParam->edged_width, pParam->edged_width); |
if (0 < (pMB->sad16 - MV16_INTER_BIAS)) |
305 |
|
{ |
306 |
|
int32_t deviation; |
307 |
|
deviation = dev16(pCurrent->y + x*16 + y*16*pParam->edged_width, |
308 |
|
pParam->edged_width); |
309 |
|
|
310 |
if (deviation < (sad16 - INTER_BIAS)) |
if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) |
311 |
{ |
{ |
312 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
313 |
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] |
314 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
= pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
315 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] |
316 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
= pMB->sad8[2] = pMB->sad8[3] = 0; |
317 |
|
|
318 |
iIntra++; |
iIntra++; |
319 |
if(iIntra >= iLimit) |
if(iIntra >= iLimit) |
321 |
|
|
322 |
continue; |
continue; |
323 |
} |
} |
324 |
|
} |
325 |
|
pMB->mode = MODE_INTER; |
326 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
327 |
|
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = pMB->sad16; |
328 |
|
} |
329 |
|
|
330 |
|
// we try to do as few INTER4V-searches as possible. So we split ME in two parts, normal |
331 |
|
// SEARCH16 and only for special blocks SEARCH8. May this should be modified for quality |
332 |
|
// levels. |
333 |
|
|
334 |
|
|
335 |
|
|
336 |
if (current->global_flags & XVID_INTER4V) |
if (current->global_flags & XVID_INTER4V) |
337 |
|
for (y = 0; y < iHcount; y++) |
338 |
|
for (x = 0; x < iWcount; x++) |
339 |
|
{ |
340 |
|
MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; |
341 |
|
|
342 |
|
if (pMB->mode == MODE_INTRA) |
343 |
|
continue; |
344 |
|
|
345 |
|
|
346 |
|
if ( (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) ) |
347 |
{ |
{ |
348 |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
int32_t neigh=0; |
349 |
2 * j, 2 * i, mv16.x, mv16.y, |
|
350 |
|
if (x>0) |
351 |
|
{ neigh += abs((pMB->mv16.x)-((pMB-1)->mv16.x)); |
352 |
|
neigh += abs((pMB->mv16.y)-((pMB-1)->mv16.y)); |
353 |
|
} |
354 |
|
if (y>0) |
355 |
|
{ neigh += abs((pMB->mv16.x)-((pMB-iWcount)->mv16.x)); |
356 |
|
neigh += abs((pMB->mv16.y)-((pMB-iWcount)->mv16.y)); |
357 |
|
} |
358 |
|
if (x<(iWcount-1)) |
359 |
|
{ neigh += abs((pMB->mv16.x)-((pMB+1)->mv16.x)); |
360 |
|
neigh += abs((pMB->mv16.y)-((pMB+1)->mv16.y)); |
361 |
|
} |
362 |
|
if (y<(iHcount-1)) |
363 |
|
{ neigh += abs((pMB->mv16.x)-((pMB+iHcount)->mv16.x)); |
364 |
|
neigh += abs((pMB->mv16.y)-((pMB+iHcount)->mv16.y)); |
365 |
|
} |
366 |
|
|
367 |
|
if (neigh > NEIGH_MOVE_THRESH) |
368 |
|
{ |
369 |
|
int32_t sad8 = IMV16X16 * current->quant; |
370 |
|
|
371 |
|
if (sad8 < pMB->sad16) |
372 |
|
sad8 += pMB->sad8[0] |
373 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
374 |
|
2*x, 2*y, pMB->mv16.x, pMB->mv16.y, |
375 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
376 |
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
377 |
|
|
378 |
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
379 |
2 * j + 1, 2 * i, mv16.x, mv16.y, |
sad8 += pMB->sad8[1] |
380 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
381 |
|
2*x+1, 2*y, pMB->mv16.x, pMB->mv16.y, |
382 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
383 |
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
384 |
|
|
385 |
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
386 |
2 * j, 2 * i + 1, mv16.x, mv16.y, |
sad8 += pMB->sad8[2] |
387 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
388 |
|
2*x, 2*y+1, pMB->mv16.x, pMB->mv16.y, |
389 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
390 |
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
391 |
|
|
392 |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (sad8 < pMB->sad16) |
393 |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
sad8 += pMB->sad8[3] |
394 |
|
= SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
395 |
|
2*x+1, 2*y+1, pMB->mv16.x, pMB->mv16.y, |
396 |
current->motion_flags, current->quant, current->fcode, |
current->motion_flags, current->quant, current->fcode, |
397 |
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
398 |
|
|
|
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
|
|
} |
|
|
|
|
|
|
|
399 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
400 |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
401 |
*/ |
*/ |
402 |
|
|
403 |
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 |
|
404 |
{ |
{ |
405 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
406 |
pMB->sad8[0] *= 4; |
pMB->sad8[0] *= 4; |
407 |
pMB->sad8[1] *= 4; |
pMB->sad8[1] *= 4; |
408 |
pMB->sad8[2] *= 4; |
pMB->sad8[2] *= 4; |
409 |
pMB->sad8[3] *= 4; |
pMB->sad8[3] *= 4; |
410 |
|
continue; |
411 |
} |
} |
|
} |
|
|
else |
|
|
{ |
|
|
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; |
|
412 |
|
|
413 |
pMB->pmvs[0].x = pmv16.x; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
414 |
} |
} |
415 |
|
|
|
return 0; |
|
416 |
} |
} |
417 |
|
|
418 |
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
// get_pmv has to be called again, because inter4v changes predictors |
419 |
|
|
420 |
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
pmv = get_pmv(pMBs, x, y, pParam->mb_width, 0); |
421 |
|
pMB->pmvs[0].x = pMB->mv16.x - pmv.x; /* the other pmvs are only needed in INTER4V-mode */ |
422 |
|
pMB->pmvs[0].y = pMB->mv16.y - pmv.y; |
423 |
|
|
424 |
|
} |
425 |
|
|
426 |
|
return 0; |
427 |
|
} |
428 |
|
|
429 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
430 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
889 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
890 |
int32_t psad[4]; |
int32_t psad[4]; |
891 |
|
|
892 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
893 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
894 |
|
|
895 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1285 |
int32_t psad[4]; |
int32_t psad[4]; |
1286 |
VECTOR newMV; |
VECTOR newMV; |
1287 |
VECTOR backupMV; |
VECTOR backupMV; |
1288 |
VECTOR startMV = { start_x, start_y }; |
VECTOR startMV; |
1289 |
|
|
1290 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1291 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1292 |
|
|
1293 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1296 |
|
|
1297 |
int32_t iSubBlock = (y&1)+(y&1) + (x&1); |
int32_t iSubBlock = (y&1)+(y&1) + (x&1); |
1298 |
|
|
1299 |
|
/* Init variables */ |
1300 |
|
startMV.x = start_x; |
1301 |
|
startMV.y = start_y; |
1302 |
|
|
1303 |
/* Get maximum range */ |
/* Get maximum range */ |
1304 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1305 |
x, y, 8, iWidth, iHeight, iFcode); |
x, y, 8, iWidth, iHeight, iFcode); |
1579 |
int32_t psad[8]; |
int32_t psad[8]; |
1580 |
|
|
1581 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
1582 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1583 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1584 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
1585 |
|
|
1866 |
|
|
1867 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
1868 |
|
|
1869 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1870 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1871 |
|
|
1872 |
int32_t bPredEq; |
int32_t bPredEq; |