2 |
* |
* |
3 |
* Modifications: |
* Modifications: |
4 |
* |
* |
5 |
|
* 01.05.2002 updated MotionEstimationBVOP |
6 |
|
* 25.04.2002 partial prevMB conversion |
7 |
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
* 22.04.2002 remove some compile warning by chenm001 <chenm001@163.com> |
8 |
* 14.04.2002 added MotionEstimationBVOP() |
* 14.04.2002 added MotionEstimationBVOP() |
9 |
* 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 |
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) |
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, |
87 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
88 |
const int x, const int y, |
const int x, const int y, |
89 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
90 |
|
const uint32_t iQuant, |
91 |
|
const uint32_t iFcode, |
92 |
const MBParam * const pParam, |
const MBParam * const pParam, |
93 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
94 |
|
const MACROBLOCK * const prevMBs, |
95 |
VECTOR * const currMV, |
VECTOR * const currMV, |
96 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
97 |
|
|
103 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
104 |
const int x, const int y, |
const int x, const int y, |
105 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
106 |
|
const uint32_t iQuant, |
107 |
|
const uint32_t iFcode, |
108 |
const MBParam * const pParam, |
const MBParam * const pParam, |
109 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
110 |
|
const MACROBLOCK * const prevMBs, |
111 |
VECTOR * const currMV, |
VECTOR * const currMV, |
112 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
113 |
|
|
119 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
120 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
121 |
const int x, const int y, |
const int x, const int y, |
122 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
123 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
124 |
|
const uint32_t iQuant, |
125 |
|
const uint32_t iFcode, |
126 |
const MBParam * const pParam, |
const MBParam * const pParam, |
127 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
128 |
|
const MACROBLOCK * const prevMBs, |
129 |
VECTOR * const currMV, |
VECTOR * const currMV, |
130 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
131 |
|
|
136 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
137 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
138 |
const int x, const int y, |
const int x, const int y, |
139 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
140 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
141 |
|
const uint32_t iQuant, |
142 |
|
const uint32_t iFcode, |
143 |
const MBParam * const pParam, |
const MBParam * const pParam, |
144 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
145 |
|
const MACROBLOCK * const prevMBs, |
146 |
VECTOR * const currMV, |
VECTOR * const currMV, |
147 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
148 |
|
|
190 |
|
|
191 |
typedef MainSearch8Func* MainSearch8FuncPtr; |
typedef MainSearch8Func* MainSearch8FuncPtr; |
192 |
|
|
193 |
|
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
194 |
|
{ 0 ,(int)(1.00235+0.5), (int)(1.15582+0.5), (int)(1.31976+0.5), (int)(1.49591+0.5), (int)(1.68601+0.5), |
195 |
|
(int)(1.89187+0.5), (int)(2.11542+0.5), (int)(2.35878+0.5), (int)(2.62429+0.5), (int)(2.91455+0.5), |
196 |
|
(int)(3.23253+0.5), (int)(3.58158+0.5), (int)(3.96555+0.5), (int)(4.38887+0.5), (int)(4.85673+0.5), |
197 |
|
(int)(5.37519+0.5), (int)(5.95144+0.5), (int)(6.59408+0.5), (int)(7.31349+0.5), (int)(8.12242+0.5), |
198 |
|
(int)(9.03669+0.5), (int)(10.0763+0.5), (int)(11.2669+0.5), (int)(12.6426+0.5), (int)(14.2493+0.5), |
199 |
|
(int)(16.1512+0.5), (int)(18.442+0.5), (int)(21.2656+0.5), (int)(24.8580+0.5), (int)(29.6436+0.5), |
200 |
|
(int)(36.4949+0.5) }; |
201 |
|
|
202 |
|
static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ |
203 |
|
|
204 |
|
|
205 |
|
|
206 |
// mv.length table |
// mv.length table |
207 |
static const uint32_t mvtab[33] = { |
static const uint32_t mvtab[33] = { |
208 |
1, 2, 3, 4, 6, 7, 7, 7, |
1, 2, 3, 4, 6, 7, 7, 7, |
238 |
} |
} |
239 |
|
|
240 |
|
|
241 |
static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode) |
static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
242 |
{ |
{ |
243 |
return NEIGH_TEND_16X16 * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
244 |
} |
} |
245 |
|
|
246 |
static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode) |
static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
247 |
|
|
248 |
{ |
{ |
249 |
return NEIGH_TEND_8X8 * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
250 |
} |
} |
251 |
|
|
252 |
|
|
265 |
#endif |
#endif |
266 |
|
|
267 |
bool MotionEstimation( |
bool MotionEstimation( |
|
MACROBLOCK * const pMBs, |
|
268 |
MBParam * const pParam, |
MBParam * const pParam, |
269 |
const IMAGE * const pRef, |
FRAMEINFO * const current, |
270 |
|
FRAMEINFO * const reference, |
271 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
272 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
273 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
|
IMAGE * const pCurrent, |
|
274 |
const uint32_t iLimit) |
const uint32_t iLimit) |
275 |
|
|
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 * const pMBs = current->mbs; |
280 |
|
MACROBLOCK * const prevMBs = reference->mbs; // previous frame |
281 |
|
|
282 |
uint32_t i, j, iIntra = 0; |
const IMAGE * const pCurrent = ¤t->image; |
283 |
|
const IMAGE * const pRef = &reference->image; |
284 |
|
|
285 |
VECTOR mv16; |
const VECTOR zeroMV = {0,0}; |
|
VECTOR pmv16; |
|
286 |
|
|
287 |
int32_t sad8 = 0; |
int32_t x, y; |
288 |
int32_t sad16; |
int32_t iIntra = 0; |
289 |
int32_t deviation; |
VECTOR pmv; |
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]; |
|
|
|
|
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
j, i, pParam->motion_flags, |
|
|
pParam, pMBs, &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[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 |
if (pParam->global_flags & XVID_INTER4V) |
// 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 |
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
// levels. |
|
2 * j, 2 * i, mv16.x, mv16.y, pParam->motion_flags, |
|
|
pParam, pMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
|
333 |
|
|
|
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j + 1, 2 * i, mv16.x, mv16.y, pParam->motion_flags, |
|
|
pParam, pMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
|
334 |
|
|
|
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j, 2 * i + 1, mv16.x, mv16.y, pParam->motion_flags, |
|
|
pParam, pMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
|
335 |
|
|
336 |
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
if (current->global_flags & XVID_INTER4V) |
337 |
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, pParam->motion_flags, |
for (y = 0; y < iHcount; y++) |
338 |
pParam, pMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
for (x = 0; x < iWcount; x++) |
339 |
|
{ |
340 |
|
MACROBLOCK* const pMB = &pMBs[x + y * iWcount]; |
341 |
|
|
342 |
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
if (pMB->mode == MODE_INTRA) |
343 |
} |
continue; |
344 |
|
|
345 |
|
|
346 |
|
if ( (!(current->global_flags & XVID_LUMIMASKING) || pMB->dquant == NO_CHANGE) ) |
347 |
|
{ |
348 |
|
int32_t neigh=0; |
349 |
|
|
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, |
376 |
|
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
377 |
|
|
378 |
|
if (sad8 < pMB->sad16) |
379 |
|
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, |
383 |
|
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
384 |
|
|
385 |
|
if (sad8 < pMB->sad16) |
386 |
|
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, |
390 |
|
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
391 |
|
|
392 |
|
if (sad8 < pMB->sad16) |
393 |
|
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, |
397 |
|
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
398 |
|
|
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 (pMB->dquant == NO_CHANGE) { |
if (sad8 < pMB->sad16) |
|
if (((pParam->global_flags & XVID_INTER4V)==0) || |
|
|
(sad16 < (sad8 + (int32_t)(IMV16X16 * pParam->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->pmvs[0].x = pmv16.x; |
|
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
|
else |
|
|
pMB->mode = MODE_INTER4V; |
|
|
} |
|
|
else |
|
404 |
{ |
{ |
405 |
sad8 = sad16; |
pMB->mode = MODE_INTER4V; |
406 |
pMB->mode = MODE_INTER; |
pMB->sad8[0] *= 4; |
407 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
pMB->sad8[1] *= 4; |
408 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
pMB->sad8[2] *= 4; |
409 |
pMB->pmvs[0].x = pmv16.x; |
pMB->sad8[3] *= 4; |
410 |
pMB->pmvs[0].y = pmv16.y; |
continue; |
411 |
} |
} |
412 |
|
|
413 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
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) \ |
431 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
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 |
} |
} |
438 |
|
|
439 |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
#define NOCHECK_MV16_CANDIDATE(X,Y) { \ |
440 |
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); \ |
441 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
442 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
443 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
444 |
} |
} |
448 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
449 |
{ \ |
{ \ |
450 |
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); \ |
451 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
452 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
453 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
454 |
} |
} |
458 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
459 |
{ \ |
{ \ |
460 |
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); \ |
461 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
462 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
463 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
464 |
} |
} |
468 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
469 |
{ \ |
{ \ |
470 |
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); \ |
471 |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_16((X) - pmv[0].x, (Y) - pmv[0].y, (uint8_t)iFcode, iQuant);\ |
472 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
473 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
474 |
} |
} |
476 |
|
|
477 |
#define CHECK_MV8_ZERO {\ |
#define CHECK_MV8_ZERO {\ |
478 |
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); \ |
479 |
iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_8(-pmv[0].x, -pmv[0].y, (uint8_t)iFcode, iQuant);\ |
480 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
481 |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
{ iMinSAD=iSAD; currMV->x=0; currMV->y=0; } \ |
482 |
} |
} |
484 |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
#define NOCHECK_MV8_CANDIDATE(X,Y) \ |
485 |
{ \ |
{ \ |
486 |
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); \ |
487 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
488 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
489 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } \ |
490 |
} |
} |
494 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
495 |
{ \ |
{ \ |
496 |
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); \ |
497 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
498 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
499 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); } } \ |
500 |
} |
} |
504 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
505 |
{ \ |
{ \ |
506 |
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); \ |
507 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
508 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
509 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); } } \ |
510 |
} |
} |
514 |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
&& ((Y) <= max_dy) && ((Y) >= min_dy) ) \ |
515 |
{ \ |
{ \ |
516 |
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); \ |
517 |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode) * iQuant;\ |
iSAD += calc_delta_8((X)-pmv[0].x, (Y)-pmv[0].y, (uint8_t)iFcode, iQuant);\ |
518 |
if (iSAD < iMinSAD) \ |
if (iSAD < iMinSAD) \ |
519 |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
{ iMinSAD=iSAD; currMV->x=(X); currMV->y=(Y); iDirection=(D); iFound=0; } } \ |
520 |
} |
} |
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; |
781 |
return iMinSAD; |
return iMinSAD; |
782 |
} |
} |
783 |
|
|
784 |
|
int32_t AdvDiamond16_MainSearch( |
785 |
|
const uint8_t * const pRef, |
786 |
|
const uint8_t * const pRefH, |
787 |
|
const uint8_t * const pRefV, |
788 |
|
const uint8_t * const pRefHV, |
789 |
|
const uint8_t * const cur, |
790 |
|
const int x, const int y, |
791 |
|
int32_t startx, int32_t starty, |
792 |
|
int32_t iMinSAD, |
793 |
|
VECTOR * const currMV, |
794 |
|
const VECTOR * const pmv, |
795 |
|
const int32_t min_dx, const int32_t max_dx, |
796 |
|
const int32_t min_dy, const int32_t max_dy, |
797 |
|
const int32_t iEdgedWidth, |
798 |
|
const int32_t iDiamondSize, |
799 |
|
const int32_t iFcode, |
800 |
|
const int32_t iQuant, |
801 |
|
int iDirection) |
802 |
|
{ |
803 |
|
|
804 |
|
int32_t iSAD; |
805 |
|
|
806 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
807 |
|
|
808 |
|
if (iDirection) |
809 |
|
{ |
810 |
|
CHECK_MV16_CANDIDATE(startx-iDiamondSize, starty); |
811 |
|
CHECK_MV16_CANDIDATE(startx+iDiamondSize, starty); |
812 |
|
CHECK_MV16_CANDIDATE(startx, starty-iDiamondSize); |
813 |
|
CHECK_MV16_CANDIDATE(startx, starty+iDiamondSize); |
814 |
|
} |
815 |
|
else |
816 |
|
{ |
817 |
|
int bDirection = 1+2+4+8; |
818 |
|
do |
819 |
|
{ |
820 |
|
iDirection = 0; |
821 |
|
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) |
822 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize,starty,1); |
823 |
|
|
824 |
|
if (bDirection&2) |
825 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize,starty,2); |
826 |
|
|
827 |
|
if (bDirection&4) |
828 |
|
CHECK_MV16_CANDIDATE_DIR(startx,starty-iDiamondSize,4); |
829 |
|
|
830 |
|
if (bDirection&8) |
831 |
|
CHECK_MV16_CANDIDATE_DIR(startx,starty+iDiamondSize,8); |
832 |
|
|
833 |
|
/* now we're doing diagonal checks near our candidate */ |
834 |
|
|
835 |
|
if (iDirection) //checking if anything found |
836 |
|
{ |
837 |
|
bDirection = iDirection; |
838 |
|
iDirection = 0; |
839 |
|
startx=currMV->x; starty=currMV->y; |
840 |
|
if (bDirection & 3) //our candidate is left or right |
841 |
|
{ |
842 |
|
CHECK_MV16_CANDIDATE_DIR(startx,starty+iDiamondSize, 8); |
843 |
|
CHECK_MV16_CANDIDATE_DIR(startx,starty-iDiamondSize, 4); |
844 |
|
} |
845 |
|
else // what remains here is up or down |
846 |
|
{ |
847 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty, 2); |
848 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty, 1); |
849 |
|
} |
850 |
|
|
851 |
|
if (iDirection) |
852 |
|
{ bDirection+=iDirection; |
853 |
|
startx=currMV->x; starty=currMV->y; |
854 |
|
} |
855 |
|
} |
856 |
|
else //about to quit, eh? not so fast.... |
857 |
|
{ |
858 |
|
switch (bDirection) |
859 |
|
{ |
860 |
|
case 2: |
861 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
862 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
863 |
|
break; |
864 |
|
case 1: |
865 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
866 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
867 |
|
break; |
868 |
|
case 2+4: |
869 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
870 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
871 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
872 |
|
break; |
873 |
|
case 4: |
874 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
875 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
876 |
|
break; |
877 |
|
case 8: |
878 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
879 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
880 |
|
break; |
881 |
|
case 1+4: |
882 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
883 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
884 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
885 |
|
break; |
886 |
|
case 2+8: |
887 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
888 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
889 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
890 |
|
break; |
891 |
|
case 1+8: |
892 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
893 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
894 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
895 |
|
break; |
896 |
|
default: //1+2+4+8 == we didn't find anything at all |
897 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
898 |
|
CHECK_MV16_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
899 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
900 |
|
CHECK_MV16_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
901 |
|
break; |
902 |
|
} |
903 |
|
if (!iDirection) break; //ok, the end. really |
904 |
|
else |
905 |
|
{ bDirection=iDirection; |
906 |
|
startx=currMV->x; starty=currMV->y; |
907 |
|
} |
908 |
|
} |
909 |
|
} |
910 |
|
while (1); //forever |
911 |
|
} |
912 |
|
return iMinSAD; |
913 |
|
} |
914 |
|
|
915 |
|
int32_t AdvDiamond8_MainSearch( |
916 |
|
const uint8_t * const pRef, |
917 |
|
const uint8_t * const pRefH, |
918 |
|
const uint8_t * const pRefV, |
919 |
|
const uint8_t * const pRefHV, |
920 |
|
const uint8_t * const cur, |
921 |
|
const int x, const int y, |
922 |
|
int32_t startx, int32_t starty, |
923 |
|
int32_t iMinSAD, |
924 |
|
VECTOR * const currMV, |
925 |
|
const VECTOR * const pmv, |
926 |
|
const int32_t min_dx, const int32_t max_dx, |
927 |
|
const int32_t min_dy, const int32_t max_dy, |
928 |
|
const int32_t iEdgedWidth, |
929 |
|
const int32_t iDiamondSize, |
930 |
|
const int32_t iFcode, |
931 |
|
const int32_t iQuant, |
932 |
|
int iDirection) |
933 |
|
{ |
934 |
|
|
935 |
|
int32_t iSAD; |
936 |
|
|
937 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
938 |
|
|
939 |
|
if (iDirection) |
940 |
|
{ |
941 |
|
CHECK_MV8_CANDIDATE(startx-iDiamondSize, starty); |
942 |
|
CHECK_MV8_CANDIDATE(startx+iDiamondSize, starty); |
943 |
|
CHECK_MV8_CANDIDATE(startx, starty-iDiamondSize); |
944 |
|
CHECK_MV8_CANDIDATE(startx, starty+iDiamondSize); |
945 |
|
} |
946 |
|
else |
947 |
|
{ |
948 |
|
int bDirection = 1+2+4+8; |
949 |
|
do |
950 |
|
{ |
951 |
|
iDirection = 0; |
952 |
|
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) |
953 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize,starty,1); |
954 |
|
|
955 |
|
if (bDirection&2) |
956 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize,starty,2); |
957 |
|
|
958 |
|
if (bDirection&4) |
959 |
|
CHECK_MV8_CANDIDATE_DIR(startx,starty-iDiamondSize,4); |
960 |
|
|
961 |
|
if (bDirection&8) |
962 |
|
CHECK_MV8_CANDIDATE_DIR(startx,starty+iDiamondSize,8); |
963 |
|
|
964 |
|
/* now we're doing diagonal checks near our candidate */ |
965 |
|
|
966 |
|
if (iDirection) //checking if anything found |
967 |
|
{ |
968 |
|
bDirection = iDirection; |
969 |
|
iDirection = 0; |
970 |
|
startx=currMV->x; starty=currMV->y; |
971 |
|
if (bDirection & 3) //our candidate is left or right |
972 |
|
{ |
973 |
|
CHECK_MV8_CANDIDATE_DIR(startx,starty+iDiamondSize, 8); |
974 |
|
CHECK_MV8_CANDIDATE_DIR(startx,starty-iDiamondSize, 4); |
975 |
|
} |
976 |
|
else // what remains here is up or down |
977 |
|
{ |
978 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty, 2); |
979 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty, 1); |
980 |
|
} |
981 |
|
|
982 |
|
if (iDirection) |
983 |
|
{ bDirection+=iDirection; |
984 |
|
startx=currMV->x; starty=currMV->y; |
985 |
|
} |
986 |
|
} |
987 |
|
else //about to quit, eh? not so fast.... |
988 |
|
{ |
989 |
|
switch (bDirection) |
990 |
|
{ |
991 |
|
case 2: |
992 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
993 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
994 |
|
break; |
995 |
|
case 1: |
996 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
997 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
998 |
|
break; |
999 |
|
case 2+4: |
1000 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
1001 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
1002 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
1003 |
|
break; |
1004 |
|
case 4: |
1005 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
1006 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
1007 |
|
break; |
1008 |
|
case 8: |
1009 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
1010 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
1011 |
|
break; |
1012 |
|
case 1+4: |
1013 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
1014 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
1015 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
1016 |
|
break; |
1017 |
|
case 2+8: |
1018 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
1019 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
1020 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
1021 |
|
break; |
1022 |
|
case 1+8: |
1023 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
1024 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
1025 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
1026 |
|
break; |
1027 |
|
default: //1+2+4+8 == we didn't find anything at all |
1028 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty-iDiamondSize, 1+4); |
1029 |
|
CHECK_MV8_CANDIDATE_DIR(startx-iDiamondSize, starty+iDiamondSize, 1+8); |
1030 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty-iDiamondSize, 2+4); |
1031 |
|
CHECK_MV8_CANDIDATE_DIR(startx+iDiamondSize, starty+iDiamondSize, 2+8); |
1032 |
|
break; |
1033 |
|
} |
1034 |
|
if (!(iDirection)) break; //ok, the end. really |
1035 |
|
else |
1036 |
|
{ bDirection=iDirection; |
1037 |
|
startx=currMV->x; starty=currMV->y; |
1038 |
|
} |
1039 |
|
} |
1040 |
|
} |
1041 |
|
while (1); //forever |
1042 |
|
} |
1043 |
|
return iMinSAD; |
1044 |
|
} |
1045 |
|
|
1046 |
|
|
1047 |
int32_t Full8_MainSearch( |
int32_t Full8_MainSearch( |
1048 |
const uint8_t * const pRef, |
const uint8_t * const pRef, |
1049 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1122 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1123 |
const int x, const int y, |
const int x, const int y, |
1124 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1125 |
|
const uint32_t iQuant, |
1126 |
|
const uint32_t iFcode, |
1127 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1128 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1129 |
|
const MACROBLOCK * const prevMBs, |
1130 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1131 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1132 |
{ |
{ |
1133 |
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; |
|
1134 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1135 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1136 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1152 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
1153 |
int32_t psad[4]; |
int32_t psad[4]; |
1154 |
|
|
1155 |
MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
MainSearch16FuncPtr MainSearchPtr; |
1156 |
|
|
1157 |
|
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1158 |
|
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1159 |
|
|
1160 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1161 |
int32_t bPredEq; |
int32_t bPredEq; |
1194 |
|
|
1195 |
iFound=0; |
iFound=0; |
1196 |
|
|
|
/* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
|
|
vector of the median. |
|
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
|
|
*/ |
|
|
|
|
|
if ((bPredEq) && (MVequal(pmv[0],pMB->mvs[0]) ) ) |
|
|
iFound=2; |
|
|
|
|
|
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
|
|
Otherwise select large Diamond Search. |
|
|
*/ |
|
|
|
|
|
if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536) || (bPredEq) ) |
|
|
iDiamondSize=1; // halfpel! |
|
|
else |
|
|
iDiamondSize=2; // halfpel! |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPELDIAMOND16) ) |
|
|
iDiamondSize*=2; |
|
|
|
|
1197 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1198 |
MinSAD=SAD |
MinSAD=SAD |
1199 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1201 |
If SAD<=256 goto Step 10. |
If SAD<=256 goto Step 10. |
1202 |
*/ |
*/ |
1203 |
|
|
|
|
|
|
// Prepare for main loop |
|
|
|
|
1204 |
*currMV=pmv[0]; /* current best := prediction */ |
*currMV=pmv[0]; /* current best := prediction */ |
1205 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1206 |
{ /* This should NOT be necessary! */ |
{ /* This should NOT be necessary! */ |
1228 |
iMinSAD = sad16( cur, |
iMinSAD = sad16( cur, |
1229 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
1230 |
iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth, MV_MAX_ERROR); |
1231 |
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); |
1232 |
|
|
1233 |
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) ) ) |
1234 |
|
{ |
1235 |
|
if (iMinSAD < 2*iQuant) // high chances for SKIP-mode |
1236 |
|
{ |
1237 |
|
if (!MVzero(*currMV)) |
1238 |
{ |
{ |
1239 |
|
iMinSAD += MV16_00_BIAS; |
1240 |
|
CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
1241 |
|
iMinSAD -= MV16_00_BIAS; |
1242 |
|
} |
1243 |
|
} |
1244 |
|
|
1245 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1246 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1248 |
goto PMVfast16_Terminate_with_Refine; |
goto PMVfast16_Terminate_with_Refine; |
1249 |
} |
} |
1250 |
|
|
1251 |
|
|
1252 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1253 |
|
vector of the median. |
1254 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1255 |
|
*/ |
1256 |
|
|
1257 |
|
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[0]) ) ) |
1258 |
|
iFound=2; |
1259 |
|
|
1260 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1261 |
|
Otherwise select large Diamond Search. |
1262 |
|
*/ |
1263 |
|
|
1264 |
|
if ( (!MVzero(pmv[0])) || (threshB<1536) || (bPredEq) ) |
1265 |
|
iDiamondSize=1; // halfpel! |
1266 |
|
else |
1267 |
|
iDiamondSize=2; // halfpel! |
1268 |
|
|
1269 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND16) ) |
1270 |
|
iDiamondSize*=2; |
1271 |
|
|
1272 |
/* |
/* |
1273 |
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. |
1274 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1275 |
Let MinSAD be the smallest SAD up to this point. |
Let MinSAD be the smallest SAD up to this point. |
1276 |
If MV is (0,0) subtract offset. ******** WHAT'S THIS 'OFFSET' ??? *********** |
If MV is (0,0) subtract offset. |
1277 |
*/ |
*/ |
1278 |
|
|
1279 |
// (0,0) is always possible |
// (0,0) is always possible |
1280 |
|
|
1281 |
|
if (!MVzero(pmv[0])) |
1282 |
CHECK_MV16_ZERO; |
CHECK_MV16_ZERO; |
1283 |
|
|
1284 |
// previous frame MV is always possible |
// previous frame MV is always possible |
1285 |
CHECK_MV16_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); |
|
1286 |
|
if (!MVzero(prevMB->mvs[0])) |
1287 |
|
if (!MVequal(prevMB->mvs[0],pmv[0])) |
1288 |
|
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1289 |
|
|
1290 |
// left neighbour, if allowed |
// left neighbour, if allowed |
1291 |
if (x != 0) |
|
1292 |
|
if (!MVzero(pmv[1])) |
1293 |
|
if (!MVequal(pmv[1],prevMB->mvs[0])) |
1294 |
|
if (!MVequal(pmv[1],pmv[0])) |
1295 |
{ |
{ |
1296 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1297 |
{ pmv[1].x = EVEN(pmv[1].x); |
{ pmv[1].x = EVEN(pmv[1].x); |
1298 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
1299 |
} |
} |
1300 |
|
|
1301 |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
1302 |
} |
} |
1303 |
|
|
1304 |
// top neighbour, if allowed |
// top neighbour, if allowed |
1305 |
if (y != 0) |
if (!MVzero(pmv[2])) |
1306 |
|
if (!MVequal(pmv[2],prevMB->mvs[0])) |
1307 |
|
if (!MVequal(pmv[2],pmv[0])) |
1308 |
|
if (!MVequal(pmv[2],pmv[1])) |
1309 |
{ |
{ |
1310 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1311 |
{ pmv[2].x = EVEN(pmv[2].x); |
{ pmv[2].x = EVEN(pmv[2].x); |
1314 |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
1315 |
|
|
1316 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
1317 |
if ((uint32_t)x != (iWcount-1)) |
if (!MVzero(pmv[3])) |
1318 |
|
if (!MVequal(pmv[3],prevMB->mvs[0])) |
1319 |
|
if (!MVequal(pmv[3],pmv[0])) |
1320 |
|
if (!MVequal(pmv[3],pmv[1])) |
1321 |
|
if (!MVequal(pmv[3],pmv[2])) |
1322 |
{ |
{ |
1323 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1324 |
{ pmv[3].x = EVEN(pmv[3].x); |
{ pmv[3].x = EVEN(pmv[3].x); |
1328 |
} |
} |
1329 |
} |
} |
1330 |
|
|
1331 |
|
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96)*/ ) |
1332 |
|
iMinSAD -= MV16_00_BIAS; |
1333 |
|
|
1334 |
|
|
1335 |
/* Step 6: If MinSAD <= thresa goto Step 10. |
/* Step 6: If MinSAD <= thresa goto Step 10. |
1336 |
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. |
1337 |
*/ |
*/ |
1338 |
|
|
1339 |
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) ) ) |
1340 |
{ |
{ |
1341 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1342 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1355 |
Refine by using small diamond and goto step 10. |
Refine by using small diamond and goto step 10. |
1356 |
*/ |
*/ |
1357 |
|
|
1358 |
|
if (MotionFlags & PMV_USESQUARES16) |
1359 |
|
MainSearchPtr = Square16_MainSearch; |
1360 |
|
else |
1361 |
|
if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
1362 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
1363 |
|
else |
1364 |
|
MainSearchPtr = Diamond16_MainSearch; |
1365 |
|
|
1366 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1367 |
|
|
1368 |
/* 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 */ |
1369 |
iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1370 |
x, y, |
x, y, |
1371 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->x, currMV->y, iMinSAD, &newMV, |
1372 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1382 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1383 |
|
|
1384 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0],backupMV)) ) |
1385 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
{ iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1386 |
x, y, |
x, y, |
1387 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
1388 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1395 |
} |
} |
1396 |
|
|
1397 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
1398 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
{ iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1399 |
x, y, |
x, y, |
1400 |
0, 0, iMinSAD, &newMV, |
0, 0, iMinSAD, &newMV, |
1401 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1530 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1531 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1532 |
const int x, const int y, |
const int x, const int y, |
1533 |
const int start_x, int start_y, |
const int start_x, const int start_y, |
1534 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1535 |
|
const uint32_t iQuant, |
1536 |
|
const uint32_t iFcode, |
1537 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1538 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1539 |
|
const MACROBLOCK * const prevMBs, |
1540 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1541 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1542 |
{ |
{ |
1543 |
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; |
|
1544 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1545 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1546 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
1558 |
int32_t psad[4]; |
int32_t psad[4]; |
1559 |
VECTOR newMV; |
VECTOR newMV; |
1560 |
VECTOR backupMV; |
VECTOR backupMV; |
1561 |
|
VECTOR startMV; |
1562 |
|
|
1563 |
MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1564 |
|
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1565 |
|
|
1566 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1567 |
int32_t iFound,bPredEq; |
int32_t iFound,bPredEq; |
1568 |
int32_t iMinSAD,iSAD; |
int32_t iMinSAD,iSAD; |
1569 |
|
|
1570 |
int32_t iSubBlock = ((y&1)<<1) + (x&1); |
int32_t iSubBlock = (y&1)+(y&1) + (x&1); |
1571 |
|
|
1572 |
|
MainSearch8FuncPtr MainSearchPtr; |
1573 |
|
|
1574 |
|
/* Init variables */ |
1575 |
|
startMV.x = start_x; |
1576 |
|
startMV.y = start_y; |
1577 |
|
|
1578 |
/* Get maximum range */ |
/* Get maximum range */ |
1579 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1580 |
x, y, 8, iWidth, iHeight, iFcode); |
x, y, 8, iWidth, iHeight, iFcode); |
1581 |
|
|
|
/* we work with abs. MVs, not relative to prediction, so range is relative to 0,0 */ |
|
|
|
|
1582 |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
1583 |
{ min_dx = EVEN(min_dx); |
{ min_dx = EVEN(min_dx); |
1584 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1606 |
|
|
1607 |
iFound=0; |
iFound=0; |
1608 |
|
|
|
/* Step 2: Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
|
|
vector of the median. |
|
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
|
|
*/ |
|
|
|
|
|
if ((bPredEq) && (MVequal(pmv[0],pMB->mvs[iSubBlock]) ) ) |
|
|
iFound=2; |
|
|
|
|
|
/* Step 3: If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
|
|
Otherwise select large Diamond Search. |
|
|
*/ |
|
|
|
|
|
if ( (pmv[0].x != 0) || (pmv[0].y != 0) || (threshB<1536/4) || (bPredEq) ) |
|
|
iDiamondSize=1; // 1 halfpel! |
|
|
else |
|
|
iDiamondSize=2; // 2 halfpel = 1 full pixel! |
|
|
|
|
|
if (!(MotionFlags & PMV_HALFPELDIAMOND8) ) |
|
|
iDiamondSize*=2; |
|
|
|
|
1609 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1610 |
MinSAD=SAD |
MinSAD=SAD |
1611 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1616 |
|
|
1617 |
// Prepare for main loop |
// Prepare for main loop |
1618 |
|
|
1619 |
currMV->x=start_x; /* start with mv16 */ |
// if (MotionFlags & PMV_USESQUARES8) |
1620 |
currMV->y=start_y; |
// MainSearchPtr = Square8_MainSearch; |
1621 |
|
// else |
1622 |
|
|
1623 |
|
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
1624 |
|
MainSearchPtr = AdvDiamond8_MainSearch; |
1625 |
|
else |
1626 |
|
MainSearchPtr = Diamond8_MainSearch; |
1627 |
|
|
1628 |
|
|
1629 |
|
*currMV = startMV; |
1630 |
|
|
1631 |
iMinSAD = sad8( cur, |
iMinSAD = sad8( cur, |
1632 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
1633 |
iEdgedWidth); |
iEdgedWidth); |
1634 |
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); |
1635 |
|
|
1636 |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,pMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < pMB->sad8[iSubBlock]) ) ) |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) |
1637 |
|
&& ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1638 |
{ |
{ |
1639 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1640 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1642 |
goto PMVfast8_Terminate_with_Refine; |
goto PMVfast8_Terminate_with_Refine; |
1643 |
} |
} |
1644 |
|
|
1645 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1646 |
|
vector of the median. |
1647 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1648 |
|
*/ |
1649 |
|
|
1650 |
|
if ((bPredEq) && (MVequal(pmv[0],prevMB->mvs[iSubBlock]) ) ) |
1651 |
|
iFound=2; |
1652 |
|
|
1653 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1654 |
|
Otherwise select large Diamond Search. |
1655 |
|
*/ |
1656 |
|
|
1657 |
|
if ( (!MVzero(pmv[0])) || (threshB<1536/4) || (bPredEq) ) |
1658 |
|
iDiamondSize=1; // 1 halfpel! |
1659 |
|
else |
1660 |
|
iDiamondSize=2; // 2 halfpel = 1 full pixel! |
1661 |
|
|
1662 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND8) ) |
1663 |
|
iDiamondSize*=2; |
1664 |
|
|
1665 |
|
|
1666 |
/* |
/* |
1667 |
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. |
1668 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1669 |
Let MinSAD be the smallest SAD up to this point. |
Let MinSAD be the smallest SAD up to this point. |
1670 |
If MV is (0,0) subtract offset. ******** WHAT'S THIS 'OFFSET' ??? *********** |
If MV is (0,0) subtract offset. |
1671 |
*/ |
*/ |
1672 |
|
|
1673 |
// the prediction might be even better than mv16 |
// the median prediction might be even better than mv16 |
1674 |
|
|
1675 |
|
if (!MVequal(pmv[0],startMV)) |
1676 |
CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); |
CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); |
1677 |
|
|
1678 |
// (0,0) is always possible |
// (0,0) if needed |
1679 |
|
if (!MVzero(pmv[0])) |
1680 |
|
if (!MVzero(startMV)) |
1681 |
CHECK_MV8_ZERO; |
CHECK_MV8_ZERO; |
1682 |
|
|
1683 |
// previous frame MV is always possible |
// previous frame MV if needed |
1684 |
CHECK_MV8_CANDIDATE(pMB->mvs[iSubBlock].x,pMB->mvs[iSubBlock].y); |
if (!MVzero(prevMB->mvs[iSubBlock])) |
1685 |
|
if (!MVequal(prevMB->mvs[iSubBlock],startMV)) |
1686 |
|
if (!MVequal(prevMB->mvs[iSubBlock],pmv[0])) |
1687 |
|
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
1688 |
|
|
1689 |
// left neighbour, if allowed |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
1690 |
if (psad[1] != MV_MAX_ERROR) |
{ |
1691 |
|
if (MotionFlags & PMV_QUICKSTOP16) |
1692 |
|
goto PMVfast8_Terminate_without_Refine; |
1693 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
1694 |
|
goto PMVfast8_Terminate_with_Refine; |
1695 |
|
} |
1696 |
|
|
1697 |
|
|
1698 |
|
// left neighbour, if allowed and needed |
1699 |
|
if (!MVzero(pmv[1])) |
1700 |
|
if (!MVequal(pmv[1],startMV)) |
1701 |
|
if (!MVequal(pmv[1],prevMB->mvs[iSubBlock])) |
1702 |
|
if (!MVequal(pmv[1],pmv[0])) |
1703 |
{ |
{ |
1704 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!(MotionFlags & PMV_HALFPEL8 )) |
1705 |
{ pmv[1].x = EVEN(pmv[1].x); |
{ pmv[1].x = EVEN(pmv[1].x); |
1708 |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
1709 |
} |
} |
1710 |
|
|
1711 |
// top neighbour, if allowed |
// top neighbour, if allowed and needed |
1712 |
if (psad[2] != MV_MAX_ERROR) |
if (!MVzero(pmv[2])) |
1713 |
|
if (!MVequal(pmv[2],startMV)) |
1714 |
|
if (!MVequal(pmv[2],prevMB->mvs[iSubBlock])) |
1715 |
|
if (!MVequal(pmv[2],pmv[0])) |
1716 |
|
if (!MVequal(pmv[2],pmv[1])) |
1717 |
{ |
{ |
1718 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!(MotionFlags & PMV_HALFPEL8 )) |
1719 |
{ pmv[2].x = EVEN(pmv[2].x); |
{ pmv[2].x = EVEN(pmv[2].x); |
1721 |
} |
} |
1722 |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
1723 |
|
|
1724 |
// top right neighbour, if allowed |
// top right neighbour, if allowed and needed |
1725 |
if (psad[3] != MV_MAX_ERROR) |
if (!MVzero(pmv[3])) |
1726 |
|
if (!MVequal(pmv[3],startMV)) |
1727 |
|
if (!MVequal(pmv[3],prevMB->mvs[iSubBlock])) |
1728 |
|
if (!MVequal(pmv[3],pmv[0])) |
1729 |
|
if (!MVequal(pmv[3],pmv[1])) |
1730 |
|
if (!MVequal(pmv[3],pmv[2])) |
1731 |
{ |
{ |
1732 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!(MotionFlags & PMV_HALFPEL8 )) |
1733 |
{ pmv[3].x = EVEN(pmv[3].x); |
{ pmv[3].x = EVEN(pmv[3].x); |
1737 |
} |
} |
1738 |
} |
} |
1739 |
|
|
1740 |
|
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1741 |
|
iMinSAD -= MV8_00_BIAS; |
1742 |
|
|
1743 |
|
|
1744 |
/* Step 6: If MinSAD <= thresa goto Step 10. |
/* Step 6: If MinSAD <= thresa goto Step 10. |
1745 |
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. |
1746 |
*/ |
*/ |
1747 |
|
|
1748 |
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]) ) ) |
1749 |
{ |
{ |
1750 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1751 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1766 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1767 |
|
|
1768 |
/* 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 */ |
1769 |
iSAD = Diamond8_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1770 |
x, y, |
x, y, |
1771 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->x, currMV->y, iMinSAD, &newMV, |
1772 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1782 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1783 |
|
|
1784 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0],backupMV)) ) |
1785 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
{ iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1786 |
x, y, |
x, y, |
1787 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
1788 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1795 |
} |
} |
1796 |
|
|
1797 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
1798 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
{ iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
1799 |
x, y, |
x, y, |
1800 |
0, 0, iMinSAD, &newMV, |
0, 0, iMinSAD, &newMV, |
1801 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
1835 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1836 |
const int x, const int y, |
const int x, const int y, |
1837 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1838 |
|
const uint32_t iQuant, |
1839 |
|
const uint32_t iFcode, |
1840 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1841 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
1842 |
|
const MACROBLOCK * const prevMBs, |
1843 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1844 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
1845 |
{ |
{ |
1846 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
1847 |
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; |
|
1848 |
|
|
1849 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
1850 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
1864 |
int32_t psad[8]; |
int32_t psad[8]; |
1865 |
|
|
1866 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
1867 |
MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1868 |
|
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1869 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
1870 |
|
|
1871 |
static int32_t thresh2; |
static int32_t thresh2; |
1872 |
int32_t bPredEq; |
int32_t bPredEq; |
1873 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
1874 |
|
|
1875 |
MainSearch16FuncPtr EPZSMainSearchPtr; |
MainSearch16FuncPtr MainSearchPtr; |
1876 |
|
|
1877 |
if (oldMBs == NULL) |
if (oldMBs == NULL) |
1878 |
{ oldMBs = (MACROBLOCK*) calloc(1,iWcount*iHcount*sizeof(MACROBLOCK)); |
{ oldMBs = (MACROBLOCK*) calloc(iWcount*iHcount,sizeof(MACROBLOCK)); |
1879 |
fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
1880 |
} |
} |
1881 |
oldMB = oldMBs + x + y * iWcount; |
oldMB = oldMBs + x + y * iWcount; |
1882 |
|
|
1884 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
1885 |
x, y, 16, iWidth, iHeight, iFcode); |
x, y, 16, iWidth, iHeight, iFcode); |
1886 |
|
|
|
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
|
|
|
|
1887 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16 )) |
1888 |
{ min_dx = EVEN(min_dx); |
{ min_dx = EVEN(min_dx); |
1889 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1923 |
iMinSAD = sad16( cur, |
iMinSAD = sad16( cur, |
1924 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
1925 |
iEdgedWidth, MV_MAX_ERROR); |
iEdgedWidth, MV_MAX_ERROR); |
1926 |
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); |
1927 |
|
|
1928 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
1929 |
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) ) ) |
1930 |
{ |
{ |
1931 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1932 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
1937 |
/************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
/************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
1938 |
|
|
1939 |
// previous frame MV |
// previous frame MV |
1940 |
CHECK_MV16_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1941 |
|
|
1942 |
// set threshhold based on Min of Prediction and SAD of collocated block |
// set threshhold based on Min of Prediction and SAD of collocated block |
1943 |
// CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want |
// CHECK_MV16 always uses iSAD for the SAD of last vector to check, so now iSAD is what we want |
1993 |
*/ |
*/ |
1994 |
|
|
1995 |
if ( (iMinSAD <= thresh2) |
if ( (iMinSAD <= thresh2) |
1996 |
|| ( MVequal(*currMV,pMB->mvs[0]) && ((uint32_t)iMinSAD <= pMB->sad16) ) ) |
|| ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
1997 |
{ |
{ |
1998 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1999 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2003 |
|
|
2004 |
/***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ |
/***** predictor SET C: acceleration MV (new!), neighbours in prev. frame(new!) ****/ |
2005 |
|
|
2006 |
backupMV = pMB->mvs[0]; // last MV |
backupMV = prevMB->mvs[0]; // collocated MV |
2007 |
backupMV.x += (pMB->mvs[0].x - oldMB->mvs[0].x ); // acceleration X |
backupMV.x += (prevMB->mvs[0].x - oldMB->mvs[0].x ); // acceleration X |
2008 |
backupMV.y += (pMB->mvs[0].y - oldMB->mvs[0].y ); // acceleration Y |
backupMV.y += (prevMB->mvs[0].y - oldMB->mvs[0].y ); // acceleration Y |
2009 |
|
|
2010 |
CHECK_MV16_CANDIDATE(backupMV.x,backupMV.y); |
CHECK_MV16_CANDIDATE(backupMV.x,backupMV.y); |
2011 |
|
|
2012 |
// left neighbour |
// left neighbour |
2013 |
if (x != 0) |
if (x != 0) |
2014 |
CHECK_MV16_CANDIDATE((oldMB-1)->mvs[0].x,oldMB->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB-1)->mvs[0].x,(prevMB-1)->mvs[0].y); |
2015 |
|
|
2016 |
// top neighbour |
// top neighbour |
2017 |
if (y != 0) |
if (y != 0) |
2018 |
CHECK_MV16_CANDIDATE((oldMB-iWcount)->mvs[0].x,oldMB->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB-iWcount)->mvs[0].x,(prevMB-iWcount)->mvs[0].y); |
2019 |
|
|
2020 |
// right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs |
// right neighbour, if allowed (this value is not written yet, so take it from pMB->mvs |
2021 |
|
|
2022 |
if ((uint32_t)x != iWcount-1) |
if ((uint32_t)x != iWcount-1) |
2023 |
CHECK_MV16_CANDIDATE((pMB+1)->mvs[0].x,oldMB->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB+1)->mvs[0].x,(prevMB+1)->mvs[0].y); |
2024 |
|
|
2025 |
// bottom neighbour, dito |
// bottom neighbour, dito |
2026 |
if ((uint32_t)y != iHcount-1) |
if ((uint32_t)y != iHcount-1) |
2027 |
CHECK_MV16_CANDIDATE((pMB+iWcount)->mvs[0].x,oldMB->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB+iWcount)->mvs[0].x,(prevMB+iWcount)->mvs[0].y); |
2028 |
|
|
2029 |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
2030 |
if (iMinSAD <= thresh2) |
if (iMinSAD <= thresh2) |
2041 |
|
|
2042 |
/* 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 */ |
2043 |
|
|
2044 |
if (MotionFlags & PMV_USESQUARES16) |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
|
EPZSMainSearchPtr = Square16_MainSearch; |
|
|
else |
|
|
EPZSMainSearchPtr = Diamond16_MainSearch; |
|
|
|
|
|
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
|
2045 |
x, y, |
x, y, |
2046 |
currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2047 |
2, iFcode, iQuant, 0); |
2, iFcode, iQuant, 0); |
2059 |
|
|
2060 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0],backupMV)) ) |
2061 |
{ |
{ |
2062 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
2063 |
x, y, |
x, y, |
2064 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
2065 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2073 |
|
|
2074 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
2075 |
{ |
{ |
2076 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
2077 |
x, y, |
x, y, |
2078 |
0, 0, iMinSAD, &newMV, |
0, 0, iMinSAD, &newMV, |
2079 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, /*iDiamondSize*/ 2, iFcode, iQuant, 0); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2080 |
|
|
2081 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) |
2082 |
{ |
{ |
2097 |
|
|
2098 |
EPZS16_Terminate_without_Refine: |
EPZS16_Terminate_without_Refine: |
2099 |
|
|
2100 |
*oldMB = *pMB; |
*oldMB = *prevMB; |
2101 |
|
|
2102 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - pmv[0].x; |
2103 |
currPMV->y = currMV->y - pmv[0].y; |
currPMV->y = currMV->y - pmv[0].y; |
2114 |
const int x, const int y, |
const int x, const int y, |
2115 |
const int start_x, const int start_y, |
const int start_x, const int start_y, |
2116 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2117 |
|
const uint32_t iQuant, |
2118 |
|
const uint32_t iFcode, |
2119 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2120 |
MACROBLOCK * const pMBs, |
const MACROBLOCK * const pMBs, |
2121 |
|
const MACROBLOCK * const prevMBs, |
2122 |
VECTOR * const currMV, |
VECTOR * const currMV, |
2123 |
VECTOR * const currPMV) |
VECTOR * const currPMV) |
2124 |
{ |
{ |
2125 |
const uint32_t iWcount = pParam->mb_width; |
/* Please not that EPZS might not be a good choice for 8x8-block motion search ! */ |
|
const int32_t iFcode = pParam->fixed_code; |
|
|
const int32_t iQuant = pParam->quant; |
|
2126 |
|
|
2127 |
|
const uint32_t iWcount = pParam->mb_width; |
2128 |
const int32_t iWidth = pParam->width; |
const int32_t iWidth = pParam->width; |
2129 |
const int32_t iHeight = pParam->height; |
const int32_t iHeight = pParam->height; |
2130 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
2146 |
|
|
2147 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
2148 |
|
|
2149 |
MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
2150 |
|
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
2151 |
|
|
2152 |
int32_t bPredEq; |
int32_t bPredEq; |
2153 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
2154 |
|
|
2155 |
MainSearch8FuncPtr EPZSMainSearchPtr; |
MainSearch8FuncPtr MainSearchPtr; |
2156 |
|
|
2157 |
/* Get maximum range */ |
/* Get maximum range */ |
2158 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
2201 |
iMinSAD = sad8( cur, |
iMinSAD = sad8( cur, |
2202 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
2203 |
iEdgedWidth); |
iEdgedWidth); |
2204 |
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); |
2205 |
|
|
2206 |
|
|
2207 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2215 |
|
|
2216 |
/************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
/************** This is predictor SET B: (0,0), prev.frame MV, neighbours **************/ |
2217 |
|
|
|
// previous frame MV |
|
|
CHECK_MV8_CANDIDATE(pMB->mvs[0].x,pMB->mvs[0].y); |
|
2218 |
|
|
2219 |
// MV=(0,0) is often a good choice |
// MV=(0,0) is often a good choice |
|
|
|
2220 |
CHECK_MV8_ZERO; |
CHECK_MV8_ZERO; |
2221 |
|
|
2222 |
|
// previous frame MV |
2223 |
|
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
2224 |
|
|
2225 |
|
// left neighbour, if allowed |
2226 |
|
if (psad[1] != MV_MAX_ERROR) |
2227 |
|
{ |
2228 |
|
if (!(MotionFlags & PMV_HALFPEL8 )) |
2229 |
|
{ pmv[1].x = EVEN(pmv[1].x); |
2230 |
|
pmv[1].y = EVEN(pmv[1].y); |
2231 |
|
} |
2232 |
|
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
2233 |
|
} |
2234 |
|
|
2235 |
|
// top neighbour, if allowed |
2236 |
|
if (psad[2] != MV_MAX_ERROR) |
2237 |
|
{ |
2238 |
|
if (!(MotionFlags & PMV_HALFPEL8 )) |
2239 |
|
{ pmv[2].x = EVEN(pmv[2].x); |
2240 |
|
pmv[2].y = EVEN(pmv[2].y); |
2241 |
|
} |
2242 |
|
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
2243 |
|
|
2244 |
|
// top right neighbour, if allowed |
2245 |
|
if (psad[3] != MV_MAX_ERROR) |
2246 |
|
{ |
2247 |
|
if (!(MotionFlags & PMV_HALFPEL8 )) |
2248 |
|
{ pmv[3].x = EVEN(pmv[3].x); |
2249 |
|
pmv[3].y = EVEN(pmv[3].y); |
2250 |
|
} |
2251 |
|
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
2252 |
|
} |
2253 |
|
} |
2254 |
|
|
2255 |
|
/* // this bias is zero anyway, at the moment! |
2256 |
|
|
2257 |
|
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) ) // && (iMinSAD <= iQuant * 96) |
2258 |
|
iMinSAD -= MV8_00_BIAS; |
2259 |
|
|
2260 |
|
*/ |
2261 |
|
|
2262 |
/* Terminate if MinSAD <= T_2 |
/* Terminate if MinSAD <= T_2 |
2263 |
Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] |
Terminate if MV[t] == MV[t-1] and MinSAD[t] <= MinSAD[t-1] |
2264 |
*/ |
*/ |
2271 |
goto EPZS8_Terminate_with_Refine; |
goto EPZS8_Terminate_with_Refine; |
2272 |
} |
} |
2273 |
|
|
2274 |
/************ (if Diamond Search) **************/ |
/************ (Diamond Search) **************/ |
2275 |
|
|
2276 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2277 |
|
|
2278 |
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) |
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) |
2279 |
iDiamondSize *= 2; |
iDiamondSize *= 2; |
2280 |
|
|
2281 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
/* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
2282 |
|
|
2283 |
|
/* // there is no EPZS^2 for inter4v at the moment |
2284 |
|
|
2285 |
|
if (MotionFlags & PMV_USESQUARES8) |
2286 |
|
MainSearchPtr = Square8_MainSearch; |
2287 |
|
else |
2288 |
|
*/ |
2289 |
|
|
2290 |
// if (MotionFlags & PMV_USESQUARES8) |
// if (MotionFlags & PMV_USESQUARES8) |
2291 |
// EPZSMainSearchPtr = Square8_MainSearch; |
// MainSearchPtr = Square8_MainSearch; |
2292 |
// else |
// else |
|
EPZSMainSearchPtr = Diamond8_MainSearch; |
|
2293 |
|
|
2294 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2295 |
|
MainSearchPtr = AdvDiamond8_MainSearch; |
2296 |
|
else |
2297 |
|
MainSearchPtr = Diamond8_MainSearch; |
2298 |
|
|
2299 |
|
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
2300 |
x, y, |
x, y, |
2301 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->x, currMV->y, iMinSAD, &newMV, |
2302 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2303 |
iDiamondSize, iFcode, iQuant, 00); |
iDiamondSize, iFcode, iQuant, 0); |
2304 |
|
|
2305 |
|
|
2306 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) |
2315 |
|
|
2316 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0],backupMV)) ) |
2317 |
{ |
{ |
2318 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
2319 |
x, y, |
x, y, |
2320 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
2321 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
2329 |
|
|
2330 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
2331 |
{ |
{ |
2332 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = (*MainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
2333 |
x, y, |
x, y, |
2334 |
0, 0, iMinSAD, &newMV, |
0, 0, iMinSAD, &newMV, |
2335 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
2367 |
// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) |
// TODO: need to incorporate prediction here (eg. sad += calc_delta_16) |
2368 |
***************************************************************/ |
***************************************************************/ |
2369 |
|
|
2370 |
/* |
|
2371 |
void MotionEstimationBVOP( |
void MotionEstimationBVOP( |
2372 |
MBParam * const pParam, |
MBParam * const pParam, |
2373 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
2389 |
const uint32_t mb_height = pParam->mb_height; |
const uint32_t mb_height = pParam->mb_height; |
2390 |
const int32_t edged_width = pParam->edged_width; |
const int32_t edged_width = pParam->edged_width; |
2391 |
|
|
2392 |
int32_t i,j; |
uint32_t i,j; |
2393 |
|
|
2394 |
int32_t f_sad16; |
int32_t f_sad16; |
2395 |
int32_t b_sad16; |
int32_t b_sad16; |
2413 |
&& b_mb->mvs[0].x == 0 |
&& b_mb->mvs[0].x == 0 |
2414 |
&& b_mb->mvs[0].y == 0) |
&& b_mb->mvs[0].y == 0) |
2415 |
{ |
{ |
2416 |
mb->mode = MB_IGNORE; |
mb->mode = MODE_NOT_CODED; |
2417 |
mb->mvs[0].x = 0; |
mb->mvs[0].x = 0; |
2418 |
mb->mvs[0].y = 0; |
mb->mvs[0].y = 0; |
2419 |
mb->b_mvs[0].x = 0; |
mb->b_mvs[0].x = 0; |
2428 |
i, j, |
i, j, |
2429 |
frame->motion_flags, frame->quant, frame->fcode, |
frame->motion_flags, frame->quant, frame->fcode, |
2430 |
pParam, |
pParam, |
2431 |
f_mbs, |
f_mbs, f_mbs /* todo */, |
2432 |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
2433 |
|
|
2434 |
// backward search |
// backward search |
2437 |
i, j, |
i, j, |
2438 |
frame->motion_flags, frame->quant, frame->bcode, |
frame->motion_flags, frame->quant, frame->bcode, |
2439 |
pParam, |
pParam, |
2440 |
b_mbs, |
b_mbs, b_mbs, /* todo */ |
2441 |
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
2442 |
|
|
2443 |
// interpolate search (simple, but effective) |
// interpolate search (simple, but effective) |
2457 |
if (f_sad16 < b_sad16) |
if (f_sad16 < b_sad16) |
2458 |
{ |
{ |
2459 |
best_sad = f_sad16; |
best_sad = f_sad16; |
2460 |
mb->mode = MB_FORWARD; |
mb->mode = MODE_FORWARD; |
2461 |
} |
} |
2462 |
else |
else |
2463 |
{ |
{ |
2464 |
best_sad = b_sad16; |
best_sad = b_sad16; |
2465 |
mb->mode = MB_BACKWARD; |
mb->mode = MODE_BACKWARD; |
2466 |
} |
} |
2467 |
|
|
2468 |
if (i_sad16 < best_sad) |
if (i_sad16 < best_sad) |
2469 |
{ |
{ |
2470 |
best_sad = i_sad16; |
best_sad = i_sad16; |
2471 |
mb->mode = MB_INTERPOLATE; |
mb->mode = MODE_INTERPOLATE; |
2472 |
} |
} |
2473 |
|
|
2474 |
if (d_sad16 < best_sad) |
if (d_sad16 < best_sad) |
2475 |
{ |
{ |
2476 |
best_sad = d_sad16; |
best_sad = d_sad16; |
2477 |
mb->mode = MB_DIRECT; |
mb->mode = MODE_DIRECT; |
2478 |
} |
} |
2479 |
|
|
2480 |
} |
} |
2481 |
} |
} |
2482 |
} |
} |
|
|
|
|
*/ |
|