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(const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
83 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
84 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
85 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
86 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
87 |
const int x, const int y, |
const int x, |
88 |
|
const int y, |
89 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
90 |
const uint32_t iQuant, |
const uint32_t iQuant, |
91 |
const uint32_t iFcode, |
const uint32_t iFcode, |
95 |
VECTOR * const currMV, |
VECTOR * const currMV, |
96 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
97 |
|
|
98 |
int32_t EPZSSearch16( |
int32_t EPZSSearch16(const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
99 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
100 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
101 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
102 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
103 |
const int x, const int y, |
const int x, |
104 |
|
const int y, |
105 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
106 |
const uint32_t iQuant, |
const uint32_t iQuant, |
107 |
const uint32_t iFcode, |
const uint32_t iFcode, |
112 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
113 |
|
|
114 |
|
|
115 |
int32_t PMVfastSearch8( |
int32_t PMVfastSearch8(const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
116 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
117 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
118 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
119 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
120 |
const int x, const int y, |
const int x, |
121 |
const int start_x, const int start_y, |
const int y, |
122 |
|
const int start_x, |
123 |
|
const int start_y, |
124 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
125 |
const uint32_t iQuant, |
const uint32_t iQuant, |
126 |
const uint32_t iFcode, |
const uint32_t iFcode, |
130 |
VECTOR * const currMV, |
VECTOR * const currMV, |
131 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
132 |
|
|
133 |
int32_t EPZSSearch8( |
int32_t EPZSSearch8(const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
134 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
135 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
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, |
139 |
const int start_x, const int start_y, |
const int y, |
140 |
|
const int start_x, |
141 |
|
const int start_y, |
142 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
143 |
const uint32_t iQuant, |
const uint32_t iQuant, |
144 |
const uint32_t iFcode, |
const uint32_t iFcode, |
149 |
VECTOR * const currPMV); |
VECTOR * const currPMV); |
150 |
|
|
151 |
|
|
152 |
typedef int32_t (MainSearch16Func)( |
typedef int32_t(MainSearch16Func) (const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
153 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
154 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
155 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
156 |
const uint8_t * const cur, |
const uint8_t * const cur, |
157 |
const int x, const int y, |
const int x, |
158 |
int32_t startx, int32_t starty, |
const int y, |
159 |
|
int32_t startx, |
160 |
|
int32_t starty, |
161 |
int32_t iMinSAD, |
int32_t iMinSAD, |
162 |
VECTOR * const currMV, |
VECTOR * const currMV, |
163 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
164 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
165 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
166 |
|
const int32_t min_dy, |
167 |
|
const int32_t max_dy, |
168 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
169 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
170 |
const int32_t iFcode, |
const int32_t iFcode, |
174 |
typedef MainSearch16Func* MainSearch16FuncPtr; |
typedef MainSearch16Func* MainSearch16FuncPtr; |
175 |
|
|
176 |
|
|
177 |
typedef int32_t (MainSearch8Func)( |
typedef int32_t(MainSearch8Func) (const uint8_t * const pRef, |
|
const uint8_t * const pRef, |
|
178 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
179 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
180 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
181 |
const uint8_t * const cur, |
const uint8_t * const cur, |
182 |
const int x, const int y, |
const int x, |
183 |
int32_t startx, int32_t starty, |
const int y, |
184 |
|
int32_t startx, |
185 |
|
int32_t starty, |
186 |
int32_t iMinSAD, |
int32_t iMinSAD, |
187 |
VECTOR * const currMV, |
VECTOR * const currMV, |
188 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
189 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
190 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
191 |
|
const int32_t min_dy, |
192 |
|
const int32_t max_dy, |
193 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
194 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
195 |
const int32_t iFcode, |
const int32_t iFcode, |
199 |
typedef MainSearch8Func* MainSearch8FuncPtr; |
typedef MainSearch8Func* MainSearch8FuncPtr; |
200 |
|
|
201 |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
static int32_t lambda_vec16[32] = /* rounded values for lambda param for weight of motion bits as in modified H.26L */ |
202 |
{ 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), |
{ 0, (int) (1.00235 + 0.5), (int) (1.15582 + 0.5), (int) (1.31976 + 0.5), |
203 |
(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), |
(int) (1.49591 + 0.5), (int) (1.68601 + 0.5), |
204 |
(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), |
(int) (1.89187 + 0.5), (int) (2.11542 + 0.5), (int) (2.35878 + 0.5), |
205 |
(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), |
(int) (2.62429 + 0.5), (int) (2.91455 + 0.5), |
206 |
(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), |
(int) (3.23253 + 0.5), (int) (3.58158 + 0.5), (int) (3.96555 + 0.5), |
207 |
(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), |
(int) (4.38887 + 0.5), (int) (4.85673 + 0.5), |
208 |
(int)(36.4949+0.5) }; |
(int) (5.37519 + 0.5), (int) (5.95144 + 0.5), (int) (6.59408 + 0.5), |
209 |
|
(int) (7.31349 + 0.5), (int) (8.12242 + 0.5), |
210 |
|
(int) (9.03669 + 0.5), (int) (10.0763 + 0.5), (int) (11.2669 + 0.5), |
211 |
|
(int) (12.6426 + 0.5), (int) (14.2493 + 0.5), |
212 |
|
(int) (16.1512 + 0.5), (int) (18.442 + 0.5), (int) (21.2656 + 0.5), |
213 |
|
(int) (24.8580 + 0.5), (int) (29.6436 + 0.5), |
214 |
|
(int) (36.4949 + 0.5) |
215 |
|
}; |
216 |
|
|
217 |
static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ |
static int32_t *lambda_vec8 = lambda_vec16; /* same table for INTER and INTER4V for now*/ |
218 |
|
|
227 |
}; |
}; |
228 |
|
|
229 |
|
|
230 |
static __inline uint32_t mv_bits(int32_t component, const uint32_t iFcode) |
static __inline uint32_t |
231 |
|
mv_bits(int32_t component, |
232 |
|
const uint32_t iFcode) |
233 |
{ |
{ |
234 |
if (component == 0) |
if (component == 0) |
235 |
return 1; |
return 1; |
237 |
if (component < 0) |
if (component < 0) |
238 |
component = -component; |
component = -component; |
239 |
|
|
240 |
if (iFcode == 1) |
if (iFcode == 1) { |
|
{ |
|
241 |
if (component > 32) |
if (component > 32) |
242 |
component = 32; |
component = 32; |
243 |
|
|
254 |
} |
} |
255 |
|
|
256 |
|
|
257 |
static __inline uint32_t calc_delta_16(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
static __inline uint32_t |
258 |
|
calc_delta_16(const int32_t dx, |
259 |
|
const int32_t dy, |
260 |
|
const uint32_t iFcode, |
261 |
|
const uint32_t iQuant) |
262 |
{ |
{ |
263 |
return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_16X16 * lambda_vec16[iQuant] * (mv_bits(dx, iFcode) + |
264 |
|
mv_bits(dy, iFcode)); |
265 |
} |
} |
266 |
|
|
267 |
static __inline uint32_t calc_delta_8(const int32_t dx, const int32_t dy, const uint32_t iFcode, const uint32_t iQuant) |
static __inline uint32_t |
268 |
|
calc_delta_8(const int32_t dx, |
269 |
|
const int32_t dy, |
270 |
|
const uint32_t iFcode, |
271 |
|
const uint32_t iQuant) |
272 |
{ |
{ |
273 |
return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + mv_bits(dy, iFcode)); |
return NEIGH_TEND_8X8 * lambda_vec8[iQuant] * (mv_bits(dx, iFcode) + |
274 |
|
mv_bits(dy, iFcode)); |
275 |
} |
} |
276 |
|
|
277 |
|
|
289 |
//#define SEARCH8 EPZSSearch8 |
//#define SEARCH8 EPZSSearch8 |
290 |
#endif |
#endif |
291 |
|
|
292 |
bool MotionEstimation( |
bool |
293 |
MBParam * const pParam, |
MotionEstimation(MBParam * const pParam, |
294 |
FRAMEINFO * const current, |
FRAMEINFO * const current, |
295 |
FRAMEINFO * const reference, |
FRAMEINFO * const reference, |
296 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
297 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
298 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
299 |
const uint32_t iLimit) |
const uint32_t iLimit) |
|
|
|
300 |
{ |
{ |
301 |
const uint32_t iWcount = pParam->mb_width; |
const uint32_t iWcount = pParam->mb_width; |
302 |
const uint32_t iHcount = pParam->mb_height; |
const uint32_t iHcount = pParam->mb_height; |
303 |
MACROBLOCK * pMBs = current->mbs; |
MACROBLOCK *const pMBs = current->mbs; |
304 |
IMAGE * pCurrent = ¤t->image; |
MACROBLOCK *const prevMBs = reference->mbs; |
305 |
|
const IMAGE *const pCurrent = ¤t->image; |
306 |
MACROBLOCK * prevMBs = reference->mbs; // previous frame |
const IMAGE *const pRef = &reference->image; |
307 |
IMAGE * pRef = &reference->image; |
|
308 |
|
const VECTOR zeroMV = { 0, 0 }; |
309 |
|
|
310 |
uint32_t i, j, iIntra = 0; |
int32_t x, y; |
311 |
|
int32_t iIntra = 0; |
312 |
VECTOR mv16; |
VECTOR pmv; |
|
VECTOR pmv16; |
|
|
|
|
|
int32_t sad8 = 0; |
|
|
int32_t sad16; |
|
|
int32_t deviation; |
|
313 |
|
|
314 |
if (sadInit) |
if (sadInit) |
315 |
(*sadInit)(); |
(*sadInit)(); |
316 |
|
|
317 |
// note: i==horizontal, j==vertical |
for (y = 0; y < iHcount; y++) |
318 |
for (i = 0; i < iHcount; i++) |
for (x = 0; x < iWcount; x++) { |
319 |
for (j = 0; j < iWcount; j++) |
MACROBLOCK *const pMB = &pMBs[x + y * iWcount]; |
320 |
{ |
|
321 |
MACROBLOCK *pMB = &pMBs[j + i * iWcount]; |
pMB->sad16 = |
322 |
MACROBLOCK *prevMB = &prevMBs[j + i * iWcount]; |
SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
323 |
|
y, current->motion_flags, current->quant, |
324 |
sad16 = SEARCH16(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
current->fcode, pParam, pMBs, prevMBs, &pMB->mv16, |
325 |
j, i, current->motion_flags, current->quant, current->fcode, |
&pMB->pmvs[0]); |
|
pParam, pMBs, prevMBs, &mv16, &pmv16); |
|
|
pMB->sad16=sad16; |
|
326 |
|
|
327 |
|
if (0 < (pMB->sad16 - MV16_INTER_BIAS)) { |
328 |
|
int32_t deviation; |
329 |
|
|
330 |
/* decide: MODE_INTER or MODE_INTRA |
deviation = |
331 |
if (dev_intra < sad_inter - 2 * nb) use_intra |
dev16(pCurrent->y + x * 16 + y * 16 * pParam->edged_width, |
332 |
*/ |
pParam->edged_width); |
|
|
|
|
deviation = dev16(pCurrent->y + j*16 + i*16*pParam->edged_width, pParam->edged_width); |
|
333 |
|
|
334 |
if (deviation < (sad16 - INTER_BIAS)) |
if (deviation < (pMB->sad16 - MV16_INTER_BIAS)) { |
|
{ |
|
335 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
336 |
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] = pMB->mvs[2] = |
337 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
pMB->mvs[3] = zeroMV; |
338 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
339 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
pMB->sad8[3] = 0; |
340 |
|
|
341 |
iIntra++; |
iIntra++; |
342 |
if(iIntra >= iLimit) |
if(iIntra >= iLimit) |
344 |
|
|
345 |
continue; |
continue; |
346 |
} |
} |
|
|
|
|
if (current->global_flags & XVID_INTER4V) |
|
|
{ |
|
|
pMB->sad8[0] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j, 2 * i, mv16.x, mv16.y, |
|
|
current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &pMB->mvs[0], &pMB->pmvs[0]); |
|
|
|
|
|
pMB->sad8[1] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j + 1, 2 * i, mv16.x, mv16.y, |
|
|
current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &pMB->mvs[1], &pMB->pmvs[1]); |
|
|
|
|
|
pMB->sad8[2] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j, 2 * i + 1, mv16.x, mv16.y, |
|
|
current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &pMB->mvs[2], &pMB->pmvs[2]); |
|
|
|
|
|
pMB->sad8[3] = SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, |
|
|
2 * j + 1, 2 * i + 1, mv16.x, mv16.y, |
|
|
current->motion_flags, current->quant, current->fcode, |
|
|
pParam, pMBs, prevMBs, &pMB->mvs[3], &pMB->pmvs[3]); |
|
|
|
|
|
sad8 = pMB->sad8[0] + pMB->sad8[1] + pMB->sad8[2] + pMB->sad8[3]; |
|
347 |
} |
} |
348 |
|
|
349 |
|
pmv = pMB->pmvs[0]; |
350 |
|
if (current->global_flags & XVID_INTER4V) |
351 |
|
if ((!(current->global_flags & XVID_LUMIMASKING) || |
352 |
|
pMB->dquant == NO_CHANGE)) { |
353 |
|
int32_t sad8 = IMV16X16 * current->quant; |
354 |
|
|
355 |
|
if (sad8 < pMB->sad16) |
356 |
|
|
357 |
|
sad8 += pMB->sad8[0] = |
358 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
359 |
|
pCurrent, 2 * x, 2 * y, pMB->mv16.x, |
360 |
|
pMB->mv16.y, current->motion_flags, |
361 |
|
current->quant, current->fcode, pParam, |
362 |
|
pMBs, prevMBs, &pMB->mvs[0], |
363 |
|
&pMB->pmvs[0]); |
364 |
|
|
365 |
|
if (sad8 < pMB->sad16) |
366 |
|
sad8 += pMB->sad8[1] = |
367 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
368 |
|
pCurrent, 2 * x + 1, 2 * y, pMB->mv16.x, |
369 |
|
pMB->mv16.y, current->motion_flags, |
370 |
|
current->quant, current->fcode, pParam, |
371 |
|
pMBs, prevMBs, &pMB->mvs[1], |
372 |
|
&pMB->pmvs[1]); |
373 |
|
|
374 |
|
if (sad8 < pMB->sad16) |
375 |
|
sad8 += pMB->sad8[2] = |
376 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
377 |
|
pCurrent, 2 * x, 2 * y + 1, pMB->mv16.x, |
378 |
|
pMB->mv16.y, current->motion_flags, |
379 |
|
current->quant, current->fcode, pParam, |
380 |
|
pMBs, prevMBs, &pMB->mvs[2], |
381 |
|
&pMB->pmvs[2]); |
382 |
|
|
383 |
|
if (sad8 < pMB->sad16) |
384 |
|
sad8 += pMB->sad8[3] = |
385 |
|
SEARCH8(pRef->y, pRefH->y, pRefV->y, pRefHV->y, |
386 |
|
pCurrent, 2 * x + 1, 2 * y + 1, |
387 |
|
pMB->mv16.x, pMB->mv16.y, |
388 |
|
current->motion_flags, current->quant, |
389 |
|
current->fcode, pParam, pMBs, prevMBs, |
390 |
|
&pMB->mvs[3], &pMB->pmvs[3]); |
391 |
|
|
392 |
/* decide: MODE_INTER or MODE_INTER4V |
/* decide: MODE_INTER or MODE_INTER4V |
393 |
mpeg4: if (sad8 < sad16 - nb/2+1) use_inter4v |
mpeg4: if (sad8 < pMB->sad16 - nb/2+1) use_inter4v |
394 |
*/ |
*/ |
395 |
|
|
396 |
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 |
|
|
{ |
|
397 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
398 |
pMB->sad8[0] *= 4; |
pMB->sad8[0] *= 4; |
399 |
pMB->sad8[1] *= 4; |
pMB->sad8[1] *= 4; |
400 |
pMB->sad8[2] *= 4; |
pMB->sad8[2] *= 4; |
401 |
pMB->sad8[3] *= 4; |
pMB->sad8[3] *= 4; |
402 |
|
continue; |
403 |
} |
} |
404 |
|
|
405 |
} |
} |
406 |
else |
|
|
{ |
|
|
sad8 = sad16; |
|
407 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
408 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = mv16.x; |
pMB->pmvs[0] = pmv; /* pMB->pmvs[1] = pMB->pmvs[2] = pMB->pmvs[3] are not needed for INTER */ |
409 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = mv16.y; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mv16; |
410 |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad16; |
pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = |
411 |
|
pMB->sad16; |
412 |
|
|
|
pMB->pmvs[0].x = pmv16.x; |
|
|
pMB->pmvs[0].y = pmv16.y; |
|
|
} |
|
413 |
} |
} |
|
|
|
414 |
return 0; |
return 0; |
415 |
} |
} |
416 |
|
|
|
#define MVzero(A) ( ((A).x)==(0) && ((A).y)==(0) ) |
|
|
|
|
|
#define MVequal(A,B) ( ((A).x)==((B).x) && ((A).y)==((B).y) ) |
|
|
|
|
|
|
|
417 |
#define CHECK_MV16_ZERO {\ |
#define CHECK_MV16_ZERO {\ |
418 |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
if ( (0 <= max_dx) && (0 >= min_dx) \ |
419 |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
&& (0 <= max_dy) && (0 >= min_dy) ) \ |
548 |
} |
} |
549 |
*/ |
*/ |
550 |
|
|
551 |
int32_t Diamond16_MainSearch( |
int32_t |
552 |
const uint8_t * const pRef, |
Diamond16_MainSearch(const uint8_t * const pRef, |
553 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
554 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
555 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
556 |
const uint8_t * const cur, |
const uint8_t * const cur, |
557 |
const int x, const int y, |
const int x, |
558 |
int32_t startx, int32_t starty, |
const int y, |
559 |
|
int32_t startx, |
560 |
|
int32_t starty, |
561 |
int32_t iMinSAD, |
int32_t iMinSAD, |
562 |
VECTOR * const currMV, |
VECTOR * const currMV, |
563 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
564 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
565 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
566 |
|
const int32_t min_dy, |
567 |
|
const int32_t max_dy, |
568 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
569 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
570 |
const int32_t iFcode, |
const int32_t iFcode, |
576 |
int32_t iDirection=0; |
int32_t iDirection=0; |
577 |
int32_t iSAD; |
int32_t iSAD; |
578 |
VECTOR backupMV; |
VECTOR backupMV; |
579 |
|
|
580 |
backupMV.x = startx; |
backupMV.x = startx; |
581 |
backupMV.y = starty; |
backupMV.y = starty; |
582 |
|
|
588 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
589 |
|
|
590 |
if (iDirection) |
if (iDirection) |
591 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
592 |
iFound = 1; |
iFound = 1; |
593 |
backupMV=*currMV; |
backupMV=*currMV; |
594 |
|
|
595 |
if ( iDirection != 2) |
if ( iDirection != 2) |
596 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
597 |
|
backupMV.y, 1); |
598 |
if ( iDirection != 1) |
if ( iDirection != 1) |
599 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
600 |
|
backupMV.y, 2); |
601 |
if ( iDirection != 4) |
if ( iDirection != 4) |
602 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
603 |
|
backupMV.y - iDiamondSize, 3); |
604 |
if ( iDirection != 3) |
if ( iDirection != 3) |
605 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x, |
606 |
} |
backupMV.y + iDiamondSize, 4); |
607 |
else |
} else { |
|
{ |
|
608 |
currMV->x = startx; |
currMV->x = startx; |
609 |
currMV->y = starty; |
currMV->y = starty; |
610 |
} |
} |
611 |
return iMinSAD; |
return iMinSAD; |
612 |
} |
} |
613 |
|
|
614 |
int32_t Square16_MainSearch( |
int32_t |
615 |
const uint8_t * const pRef, |
Square16_MainSearch(const uint8_t * const pRef, |
616 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
617 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
618 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
619 |
const uint8_t * const cur, |
const uint8_t * const cur, |
620 |
const int x, const int y, |
const int x, |
621 |
int32_t startx, int32_t starty, |
const int y, |
622 |
|
int32_t startx, |
623 |
|
int32_t starty, |
624 |
int32_t iMinSAD, |
int32_t iMinSAD, |
625 |
VECTOR * const currMV, |
VECTOR * const currMV, |
626 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
627 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
628 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
629 |
|
const int32_t min_dy, |
630 |
|
const int32_t max_dy, |
631 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
632 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
633 |
const int32_t iFcode, |
const int32_t iFcode, |
639 |
int32_t iDirection=0; |
int32_t iDirection=0; |
640 |
int32_t iSAD; |
int32_t iSAD; |
641 |
VECTOR backupMV; |
VECTOR backupMV; |
642 |
|
|
643 |
backupMV.x = startx; |
backupMV.x = startx; |
644 |
backupMV.y = starty; |
backupMV.y = starty; |
645 |
|
|
656 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
657 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
658 |
|
|
659 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
660 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
backupMV.y - iDiamondSize, 5); |
661 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
662 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
backupMV.y + iDiamondSize, 6); |
663 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
664 |
|
backupMV.y - iDiamondSize, 7); |
665 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
666 |
|
backupMV.y + iDiamondSize, 8); |
667 |
|
|
668 |
|
|
669 |
if (iDirection) |
if (iDirection) |
670 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
671 |
iFound = 1; |
iFound = 1; |
672 |
backupMV=*currMV; |
backupMV=*currMV; |
673 |
|
|
674 |
switch (iDirection) |
switch (iDirection) { |
|
{ |
|
675 |
case 1: |
case 1: |
676 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
677 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
backupMV.y, 1); |
678 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
679 |
|
backupMV.y - iDiamondSize, 5); |
680 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
681 |
|
backupMV.y - iDiamondSize, 7); |
682 |
break; |
break; |
683 |
case 2: |
case 2: |
684 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
685 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
2); |
686 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
687 |
|
backupMV.y + iDiamondSize, 6); |
688 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
689 |
|
backupMV.y + iDiamondSize, 8); |
690 |
break; |
break; |
691 |
|
|
692 |
case 3: |
case 3: |
693 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
694 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
695 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
696 |
|
backupMV.y - iDiamondSize, 7); |
697 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
698 |
|
backupMV.y + iDiamondSize, 8); |
699 |
break; |
break; |
700 |
|
|
701 |
case 4: |
case 4: |
702 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
703 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
704 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
705 |
|
backupMV.y - iDiamondSize, 5); |
706 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
707 |
|
backupMV.y + iDiamondSize, 6); |
708 |
break; |
break; |
709 |
|
|
710 |
case 5: |
case 5: |
711 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
712 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
1); |
713 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
714 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
3); |
715 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
716 |
|
backupMV.y - iDiamondSize, 5); |
717 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
718 |
|
backupMV.y + iDiamondSize, 6); |
719 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
720 |
|
backupMV.y - iDiamondSize, 7); |
721 |
break; |
break; |
722 |
|
|
723 |
case 6: |
case 6: |
724 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
725 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
2); |
726 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
727 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
728 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
|
729 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
730 |
|
backupMV.y - iDiamondSize, 5); |
731 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
732 |
|
backupMV.y + iDiamondSize, 6); |
733 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
734 |
|
backupMV.y + iDiamondSize, 8); |
735 |
|
|
736 |
break; |
break; |
737 |
|
|
738 |
case 7: |
case 7: |
739 |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
740 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
backupMV.y, 1); |
741 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
742 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
743 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
744 |
|
backupMV.y - iDiamondSize, 5); |
745 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
746 |
|
backupMV.y - iDiamondSize, 7); |
747 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
748 |
|
backupMV.y + iDiamondSize, 8); |
749 |
break; |
break; |
750 |
|
|
751 |
case 8: |
case 8: |
752 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
753 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
2); |
754 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
755 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
756 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
757 |
|
backupMV.y + iDiamondSize, 6); |
758 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
759 |
|
backupMV.y - iDiamondSize, 7); |
760 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
761 |
|
backupMV.y + iDiamondSize, 8); |
762 |
break; |
break; |
763 |
default: |
default: |
764 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, backupMV.y, |
765 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y,2); |
1); |
766 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, backupMV.y, |
767 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
2); |
768 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y - iDiamondSize, |
769 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y-iDiamondSize,5); |
3); |
770 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x-iDiamondSize,backupMV.y+iDiamondSize,6); |
CHECK_MV16_CANDIDATE_DIR(backupMV.x, backupMV.y + iDiamondSize, |
771 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y-iDiamondSize,7); |
4); |
772 |
CHECK_MV16_CANDIDATE_DIR(backupMV.x+iDiamondSize,backupMV.y+iDiamondSize,8); |
|
773 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
774 |
|
backupMV.y - iDiamondSize, 5); |
775 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x - iDiamondSize, |
776 |
|
backupMV.y + iDiamondSize, 6); |
777 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
778 |
|
backupMV.y - iDiamondSize, 7); |
779 |
|
CHECK_MV16_CANDIDATE_DIR(backupMV.x + iDiamondSize, |
780 |
|
backupMV.y + iDiamondSize, 8); |
781 |
break; |
break; |
782 |
} |
} |
783 |
} |
} else { |
|
else |
|
|
{ |
|
784 |
currMV->x = startx; |
currMV->x = startx; |
785 |
currMV->y = starty; |
currMV->y = starty; |
786 |
} |
} |
788 |
} |
} |
789 |
|
|
790 |
|
|
791 |
int32_t Full16_MainSearch( |
int32_t |
792 |
const uint8_t * const pRef, |
Full16_MainSearch(const uint8_t * const pRef, |
793 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
794 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
795 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
796 |
const uint8_t * const cur, |
const uint8_t * const cur, |
797 |
const int x, const int y, |
const int x, |
798 |
int32_t startx, int32_t starty, |
const int y, |
799 |
|
int32_t startx, |
800 |
|
int32_t starty, |
801 |
int32_t iMinSAD, |
int32_t iMinSAD, |
802 |
VECTOR * const currMV, |
VECTOR * const currMV, |
803 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
804 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
805 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
806 |
|
const int32_t min_dy, |
807 |
|
const int32_t max_dy, |
808 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
809 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
810 |
const int32_t iFcode, |
const int32_t iFcode, |
814 |
int32_t iSAD; |
int32_t iSAD; |
815 |
int32_t dx,dy; |
int32_t dx,dy; |
816 |
VECTOR backupMV; |
VECTOR backupMV; |
817 |
|
|
818 |
backupMV.x = startx; |
backupMV.x = startx; |
819 |
backupMV.y = starty; |
backupMV.y = starty; |
820 |
|
|
825 |
return iMinSAD; |
return iMinSAD; |
826 |
} |
} |
827 |
|
|
828 |
int32_t Full8_MainSearch( |
int32_t |
829 |
const uint8_t * const pRef, |
AdvDiamond16_MainSearch(const uint8_t * const pRef, |
830 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
831 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
832 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
833 |
const uint8_t * const cur, |
const uint8_t * const cur, |
834 |
const int x, const int y, |
const int x, |
835 |
int32_t startx, int32_t starty, |
const int y, |
836 |
|
int32_t startx, |
837 |
|
int32_t starty, |
838 |
int32_t iMinSAD, |
int32_t iMinSAD, |
839 |
VECTOR * const currMV, |
VECTOR * const currMV, |
840 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
841 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
842 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
843 |
|
const int32_t min_dy, |
844 |
|
const int32_t max_dy, |
845 |
|
const int32_t iEdgedWidth, |
846 |
|
const int32_t iDiamondSize, |
847 |
|
const int32_t iFcode, |
848 |
|
const int32_t iQuant, |
849 |
|
int iDirection) |
850 |
|
{ |
851 |
|
|
852 |
|
int32_t iSAD; |
853 |
|
|
854 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
855 |
|
|
856 |
|
if (iDirection) { |
857 |
|
CHECK_MV16_CANDIDATE(startx - iDiamondSize, starty); |
858 |
|
CHECK_MV16_CANDIDATE(startx + iDiamondSize, starty); |
859 |
|
CHECK_MV16_CANDIDATE(startx, starty - iDiamondSize); |
860 |
|
CHECK_MV16_CANDIDATE(startx, starty + iDiamondSize); |
861 |
|
} else { |
862 |
|
int bDirection = 1 + 2 + 4 + 8; |
863 |
|
|
864 |
|
do { |
865 |
|
iDirection = 0; |
866 |
|
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) |
867 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
868 |
|
|
869 |
|
if (bDirection & 2) |
870 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
871 |
|
|
872 |
|
if (bDirection & 4) |
873 |
|
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
874 |
|
|
875 |
|
if (bDirection & 8) |
876 |
|
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
877 |
|
|
878 |
|
/* now we're doing diagonal checks near our candidate */ |
879 |
|
|
880 |
|
if (iDirection) //checking if anything found |
881 |
|
{ |
882 |
|
bDirection = iDirection; |
883 |
|
iDirection = 0; |
884 |
|
startx = currMV->x; |
885 |
|
starty = currMV->y; |
886 |
|
if (bDirection & 3) //our candidate is left or right |
887 |
|
{ |
888 |
|
CHECK_MV16_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
889 |
|
CHECK_MV16_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
890 |
|
} else // what remains here is up or down |
891 |
|
{ |
892 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
893 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
894 |
|
} |
895 |
|
|
896 |
|
if (iDirection) { |
897 |
|
bDirection += iDirection; |
898 |
|
startx = currMV->x; |
899 |
|
starty = currMV->y; |
900 |
|
} |
901 |
|
} else //about to quit, eh? not so fast.... |
902 |
|
{ |
903 |
|
switch (bDirection) { |
904 |
|
case 2: |
905 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
906 |
|
starty - iDiamondSize, 2 + 4); |
907 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
908 |
|
starty + iDiamondSize, 2 + 8); |
909 |
|
break; |
910 |
|
case 1: |
911 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
912 |
|
starty - iDiamondSize, 1 + 4); |
913 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
914 |
|
starty + iDiamondSize, 1 + 8); |
915 |
|
break; |
916 |
|
case 2 + 4: |
917 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
918 |
|
starty - iDiamondSize, 1 + 4); |
919 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
920 |
|
starty - iDiamondSize, 2 + 4); |
921 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
922 |
|
starty + iDiamondSize, 2 + 8); |
923 |
|
break; |
924 |
|
case 4: |
925 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
926 |
|
starty - iDiamondSize, 2 + 4); |
927 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
928 |
|
starty - iDiamondSize, 1 + 4); |
929 |
|
break; |
930 |
|
case 8: |
931 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
932 |
|
starty + iDiamondSize, 2 + 8); |
933 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
934 |
|
starty + iDiamondSize, 1 + 8); |
935 |
|
break; |
936 |
|
case 1 + 4: |
937 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
938 |
|
starty + iDiamondSize, 1 + 8); |
939 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
940 |
|
starty - iDiamondSize, 1 + 4); |
941 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
942 |
|
starty - iDiamondSize, 2 + 4); |
943 |
|
break; |
944 |
|
case 2 + 8: |
945 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
946 |
|
starty - iDiamondSize, 1 + 4); |
947 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
948 |
|
starty + iDiamondSize, 1 + 8); |
949 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
950 |
|
starty + iDiamondSize, 2 + 8); |
951 |
|
break; |
952 |
|
case 1 + 8: |
953 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
954 |
|
starty - iDiamondSize, 2 + 4); |
955 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
956 |
|
starty + iDiamondSize, 2 + 8); |
957 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
958 |
|
starty + iDiamondSize, 1 + 8); |
959 |
|
break; |
960 |
|
default: //1+2+4+8 == we didn't find anything at all |
961 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
962 |
|
starty - iDiamondSize, 1 + 4); |
963 |
|
CHECK_MV16_CANDIDATE_DIR(startx - iDiamondSize, |
964 |
|
starty + iDiamondSize, 1 + 8); |
965 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
966 |
|
starty - iDiamondSize, 2 + 4); |
967 |
|
CHECK_MV16_CANDIDATE_DIR(startx + iDiamondSize, |
968 |
|
starty + iDiamondSize, 2 + 8); |
969 |
|
break; |
970 |
|
} |
971 |
|
if (!iDirection) |
972 |
|
break; //ok, the end. really |
973 |
|
else { |
974 |
|
bDirection = iDirection; |
975 |
|
startx = currMV->x; |
976 |
|
starty = currMV->y; |
977 |
|
} |
978 |
|
} |
979 |
|
} |
980 |
|
while (1); //forever |
981 |
|
} |
982 |
|
return iMinSAD; |
983 |
|
} |
984 |
|
|
985 |
|
int32_t |
986 |
|
AdvDiamond8_MainSearch(const uint8_t * const pRef, |
987 |
|
const uint8_t * const pRefH, |
988 |
|
const uint8_t * const pRefV, |
989 |
|
const uint8_t * const pRefHV, |
990 |
|
const uint8_t * const cur, |
991 |
|
const int x, |
992 |
|
const int y, |
993 |
|
int32_t startx, |
994 |
|
int32_t starty, |
995 |
|
int32_t iMinSAD, |
996 |
|
VECTOR * const currMV, |
997 |
|
const VECTOR * const pmv, |
998 |
|
const int32_t min_dx, |
999 |
|
const int32_t max_dx, |
1000 |
|
const int32_t min_dy, |
1001 |
|
const int32_t max_dy, |
1002 |
|
const int32_t iEdgedWidth, |
1003 |
|
const int32_t iDiamondSize, |
1004 |
|
const int32_t iFcode, |
1005 |
|
const int32_t iQuant, |
1006 |
|
int iDirection) |
1007 |
|
{ |
1008 |
|
|
1009 |
|
int32_t iSAD; |
1010 |
|
|
1011 |
|
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
1012 |
|
|
1013 |
|
if (iDirection) { |
1014 |
|
CHECK_MV8_CANDIDATE(startx - iDiamondSize, starty); |
1015 |
|
CHECK_MV8_CANDIDATE(startx + iDiamondSize, starty); |
1016 |
|
CHECK_MV8_CANDIDATE(startx, starty - iDiamondSize); |
1017 |
|
CHECK_MV8_CANDIDATE(startx, starty + iDiamondSize); |
1018 |
|
} else { |
1019 |
|
int bDirection = 1 + 2 + 4 + 8; |
1020 |
|
|
1021 |
|
do { |
1022 |
|
iDirection = 0; |
1023 |
|
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) |
1024 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
1025 |
|
|
1026 |
|
if (bDirection & 2) |
1027 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
1028 |
|
|
1029 |
|
if (bDirection & 4) |
1030 |
|
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
1031 |
|
|
1032 |
|
if (bDirection & 8) |
1033 |
|
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
1034 |
|
|
1035 |
|
/* now we're doing diagonal checks near our candidate */ |
1036 |
|
|
1037 |
|
if (iDirection) //checking if anything found |
1038 |
|
{ |
1039 |
|
bDirection = iDirection; |
1040 |
|
iDirection = 0; |
1041 |
|
startx = currMV->x; |
1042 |
|
starty = currMV->y; |
1043 |
|
if (bDirection & 3) //our candidate is left or right |
1044 |
|
{ |
1045 |
|
CHECK_MV8_CANDIDATE_DIR(startx, starty + iDiamondSize, 8); |
1046 |
|
CHECK_MV8_CANDIDATE_DIR(startx, starty - iDiamondSize, 4); |
1047 |
|
} else // what remains here is up or down |
1048 |
|
{ |
1049 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, starty, 2); |
1050 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, starty, 1); |
1051 |
|
} |
1052 |
|
|
1053 |
|
if (iDirection) { |
1054 |
|
bDirection += iDirection; |
1055 |
|
startx = currMV->x; |
1056 |
|
starty = currMV->y; |
1057 |
|
} |
1058 |
|
} else //about to quit, eh? not so fast.... |
1059 |
|
{ |
1060 |
|
switch (bDirection) { |
1061 |
|
case 2: |
1062 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1063 |
|
starty - iDiamondSize, 2 + 4); |
1064 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1065 |
|
starty + iDiamondSize, 2 + 8); |
1066 |
|
break; |
1067 |
|
case 1: |
1068 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1069 |
|
starty - iDiamondSize, 1 + 4); |
1070 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1071 |
|
starty + iDiamondSize, 1 + 8); |
1072 |
|
break; |
1073 |
|
case 2 + 4: |
1074 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1075 |
|
starty - iDiamondSize, 1 + 4); |
1076 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1077 |
|
starty - iDiamondSize, 2 + 4); |
1078 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1079 |
|
starty + iDiamondSize, 2 + 8); |
1080 |
|
break; |
1081 |
|
case 4: |
1082 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1083 |
|
starty - iDiamondSize, 2 + 4); |
1084 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1085 |
|
starty - iDiamondSize, 1 + 4); |
1086 |
|
break; |
1087 |
|
case 8: |
1088 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1089 |
|
starty + iDiamondSize, 2 + 8); |
1090 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1091 |
|
starty + iDiamondSize, 1 + 8); |
1092 |
|
break; |
1093 |
|
case 1 + 4: |
1094 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1095 |
|
starty + iDiamondSize, 1 + 8); |
1096 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1097 |
|
starty - iDiamondSize, 1 + 4); |
1098 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1099 |
|
starty - iDiamondSize, 2 + 4); |
1100 |
|
break; |
1101 |
|
case 2 + 8: |
1102 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1103 |
|
starty - iDiamondSize, 1 + 4); |
1104 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1105 |
|
starty + iDiamondSize, 1 + 8); |
1106 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1107 |
|
starty + iDiamondSize, 2 + 8); |
1108 |
|
break; |
1109 |
|
case 1 + 8: |
1110 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1111 |
|
starty - iDiamondSize, 2 + 4); |
1112 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1113 |
|
starty + iDiamondSize, 2 + 8); |
1114 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1115 |
|
starty + iDiamondSize, 1 + 8); |
1116 |
|
break; |
1117 |
|
default: //1+2+4+8 == we didn't find anything at all |
1118 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1119 |
|
starty - iDiamondSize, 1 + 4); |
1120 |
|
CHECK_MV8_CANDIDATE_DIR(startx - iDiamondSize, |
1121 |
|
starty + iDiamondSize, 1 + 8); |
1122 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1123 |
|
starty - iDiamondSize, 2 + 4); |
1124 |
|
CHECK_MV8_CANDIDATE_DIR(startx + iDiamondSize, |
1125 |
|
starty + iDiamondSize, 2 + 8); |
1126 |
|
break; |
1127 |
|
} |
1128 |
|
if (!(iDirection)) |
1129 |
|
break; //ok, the end. really |
1130 |
|
else { |
1131 |
|
bDirection = iDirection; |
1132 |
|
startx = currMV->x; |
1133 |
|
starty = currMV->y; |
1134 |
|
} |
1135 |
|
} |
1136 |
|
} |
1137 |
|
while (1); //forever |
1138 |
|
} |
1139 |
|
return iMinSAD; |
1140 |
|
} |
1141 |
|
|
1142 |
|
|
1143 |
|
int32_t |
1144 |
|
Full8_MainSearch(const uint8_t * const pRef, |
1145 |
|
const uint8_t * const pRefH, |
1146 |
|
const uint8_t * const pRefV, |
1147 |
|
const uint8_t * const pRefHV, |
1148 |
|
const uint8_t * const cur, |
1149 |
|
const int x, |
1150 |
|
const int y, |
1151 |
|
int32_t startx, |
1152 |
|
int32_t starty, |
1153 |
|
int32_t iMinSAD, |
1154 |
|
VECTOR * const currMV, |
1155 |
|
const VECTOR * const pmv, |
1156 |
|
const int32_t min_dx, |
1157 |
|
const int32_t max_dx, |
1158 |
|
const int32_t min_dy, |
1159 |
|
const int32_t max_dy, |
1160 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
1161 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
1162 |
const int32_t iFcode, |
const int32_t iFcode, |
1166 |
int32_t iSAD; |
int32_t iSAD; |
1167 |
int32_t dx,dy; |
int32_t dx,dy; |
1168 |
VECTOR backupMV; |
VECTOR backupMV; |
1169 |
|
|
1170 |
backupMV.x = startx; |
backupMV.x = startx; |
1171 |
backupMV.y = starty; |
backupMV.y = starty; |
1172 |
|
|
1179 |
|
|
1180 |
|
|
1181 |
|
|
1182 |
int32_t Halfpel16_Refine( |
int32_t |
1183 |
const uint8_t * const pRef, |
Halfpel16_Refine(const uint8_t * const pRef, |
1184 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1185 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1186 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1187 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1188 |
const int x, const int y, |
const int x, |
1189 |
|
const int y, |
1190 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1191 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1192 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
1193 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
1194 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
1195 |
|
const int32_t min_dy, |
1196 |
|
const int32_t max_dy, |
1197 |
const int32_t iFcode, |
const int32_t iFcode, |
1198 |
const int32_t iQuant, |
const int32_t iQuant, |
1199 |
const int32_t iEdgedWidth) |
const int32_t iEdgedWidth) |
1218 |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
1219 |
|
|
1220 |
|
|
1221 |
int32_t PMVfastSearch16( |
int32_t |
1222 |
const uint8_t * const pRef, |
PMVfastSearch16(const uint8_t * const pRef, |
1223 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1224 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1225 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1226 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1227 |
const int x, const int y, |
const int x, |
1228 |
|
const int y, |
1229 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1230 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1231 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1257 |
VECTOR pmv[4]; |
VECTOR pmv[4]; |
1258 |
int32_t psad[4]; |
int32_t psad[4]; |
1259 |
|
|
1260 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
MainSearch16FuncPtr MainSearchPtr; |
1261 |
|
|
1262 |
|
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1263 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1264 |
|
|
1265 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1267 |
int32_t iMinSAD,iSAD; |
int32_t iMinSAD,iSAD; |
1268 |
|
|
1269 |
/* Get maximum range */ |
/* Get maximum range */ |
1270 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
1271 |
x, y, 16, iWidth, iHeight, iFcode); |
iFcode); |
1272 |
|
|
1273 |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
1274 |
|
|
1275 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1276 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
1277 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1278 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
1279 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
1280 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
|
|
|
1281 |
|
|
1282 |
|
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
1283 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
1284 |
|
|
1285 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
1286 |
threshA = 512; |
threshA = 512; |
1287 |
threshB = 1024; |
threshB = 1024; |
1288 |
|
|
1289 |
} |
} else { |
|
else |
|
|
{ |
|
1290 |
threshA = psad[0]; |
threshA = psad[0]; |
1291 |
threshB = threshA+256; |
threshB = threshA+256; |
1292 |
if (threshA< 512) threshA = 512; |
if (threshA < 512) |
1293 |
if (threshA>1024) threshA = 1024; |
threshA = 512; |
1294 |
if (threshB>1792) threshB = 1792; |
if (threshA > 1024) |
1295 |
|
threshA = 1024; |
1296 |
|
if (threshB > 1792) |
1297 |
|
threshB = 1792; |
1298 |
} |
} |
1299 |
|
|
1300 |
iFound=0; |
iFound=0; |
1301 |
|
|
|
/* 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],prevMB->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; |
|
|
|
|
1302 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1303 |
MinSAD=SAD |
MinSAD=SAD |
1304 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1306 |
If SAD<=256 goto Step 10. |
If SAD<=256 goto Step 10. |
1307 |
*/ |
*/ |
1308 |
|
|
|
|
|
|
// Prepare for main loop |
|
|
|
|
1309 |
*currMV=pmv[0]; /* current best := prediction */ |
*currMV=pmv[0]; /* current best := prediction */ |
1310 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16)) { /* This should NOT be necessary! */ |
|
{ /* This should NOT be necessary! */ |
|
1311 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
1312 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
1313 |
} |
} |
1314 |
|
|
1315 |
if (currMV->x > max_dx) |
if (currMV->x > max_dx) { |
|
{ |
|
1316 |
currMV->x=max_dx; |
currMV->x=max_dx; |
1317 |
} |
} |
1318 |
if (currMV->x < min_dx) |
if (currMV->x < min_dx) { |
|
{ |
|
1319 |
currMV->x=min_dx; |
currMV->x=min_dx; |
1320 |
} |
} |
1321 |
if (currMV->y > max_dy) |
if (currMV->y > max_dy) { |
|
{ |
|
1322 |
currMV->y=max_dy; |
currMV->y=max_dy; |
1323 |
} |
} |
1324 |
if (currMV->y < min_dy) |
if (currMV->y < min_dy) { |
|
{ |
|
1325 |
currMV->y=min_dy; |
currMV->y=min_dy; |
1326 |
} |
} |
1327 |
|
|
1328 |
iMinSAD = sad16( cur, |
iMinSAD = |
1329 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
sad16(cur, |
1330 |
iEdgedWidth, MV_MAX_ERROR); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
1331 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
1332 |
|
iMinSAD += |
1333 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV,prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
1334 |
{ |
(uint8_t) iFcode, iQuant); |
1335 |
|
|
1336 |
|
if ((iMinSAD < 256) || |
1337 |
|
((MVequal(*currMV, prevMB->mvs[0])) && |
1338 |
|
((uint32_t) iMinSAD < prevMB->sad16))) { |
1339 |
|
if (iMinSAD < 2 * iQuant) // high chances for SKIP-mode |
1340 |
|
{ |
1341 |
|
if (!MVzero(*currMV)) { |
1342 |
|
iMinSAD += MV16_00_BIAS; |
1343 |
|
CHECK_MV16_ZERO; // (0,0) saves space for letterboxed pictures |
1344 |
|
iMinSAD -= MV16_00_BIAS; |
1345 |
|
} |
1346 |
|
} |
1347 |
|
|
1348 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1349 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1351 |
goto PMVfast16_Terminate_with_Refine; |
goto PMVfast16_Terminate_with_Refine; |
1352 |
} |
} |
1353 |
|
|
1354 |
|
|
1355 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1356 |
|
vector of the median. |
1357 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1358 |
|
*/ |
1359 |
|
|
1360 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[0]))) |
1361 |
|
iFound = 2; |
1362 |
|
|
1363 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1364 |
|
Otherwise select large Diamond Search. |
1365 |
|
*/ |
1366 |
|
|
1367 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536) || (bPredEq)) |
1368 |
|
iDiamondSize = 1; // halfpel! |
1369 |
|
else |
1370 |
|
iDiamondSize = 2; // halfpel! |
1371 |
|
|
1372 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND16)) |
1373 |
|
iDiamondSize *= 2; |
1374 |
|
|
1375 |
/* |
/* |
1376 |
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. |
1377 |
Also calculate (0,0) but do not subtract offset. |
Also calculate (0,0) but do not subtract offset. |
1381 |
|
|
1382 |
// (0,0) is always possible |
// (0,0) is always possible |
1383 |
|
|
1384 |
|
if (!MVzero(pmv[0])) |
1385 |
CHECK_MV16_ZERO; |
CHECK_MV16_ZERO; |
1386 |
|
|
1387 |
// previous frame MV is always possible |
// previous frame MV is always possible |
1388 |
|
|
1389 |
|
if (!MVzero(prevMB->mvs[0])) |
1390 |
|
if (!MVequal(prevMB->mvs[0], pmv[0])) |
1391 |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
CHECK_MV16_CANDIDATE(prevMB->mvs[0].x,prevMB->mvs[0].y); |
1392 |
|
|
1393 |
// left neighbour, if allowed |
// left neighbour, if allowed |
1394 |
if (x != 0) |
|
1395 |
{ |
if (!MVzero(pmv[1])) |
1396 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[1], prevMB->mvs[0])) |
1397 |
{ pmv[1].x = EVEN(pmv[1].x); |
if (!MVequal(pmv[1], pmv[0])) { |
1398 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
1399 |
|
pmv[1].x = EVEN(pmv[1].x); |
1400 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
1401 |
} |
} |
1402 |
|
|
1403 |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
1404 |
} |
} |
|
|
|
1405 |
// top neighbour, if allowed |
// top neighbour, if allowed |
1406 |
if (y != 0) |
if (!MVzero(pmv[2])) |
1407 |
{ |
if (!MVequal(pmv[2], prevMB->mvs[0])) |
1408 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[2], pmv[0])) |
1409 |
{ pmv[2].x = EVEN(pmv[2].x); |
if (!MVequal(pmv[2], pmv[1])) { |
1410 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
1411 |
|
pmv[2].x = EVEN(pmv[2].x); |
1412 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
1413 |
} |
} |
1414 |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
1415 |
|
|
1416 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
1417 |
if ((uint32_t)x != (iWcount-1)) |
if (!MVzero(pmv[3])) |
1418 |
{ |
if (!MVequal(pmv[3], prevMB->mvs[0])) |
1419 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!MVequal(pmv[3], pmv[0])) |
1420 |
{ pmv[3].x = EVEN(pmv[3].x); |
if (!MVequal(pmv[3], pmv[1])) |
1421 |
|
if (!MVequal(pmv[3], pmv[2])) { |
1422 |
|
if (!(MotionFlags & PMV_HALFPEL16)) { |
1423 |
|
pmv[3].x = EVEN(pmv[3].x); |
1424 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
1425 |
} |
} |
1426 |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV16_CANDIDATE(pmv[3].x, |
1427 |
|
pmv[3].y); |
1428 |
} |
} |
1429 |
} |
} |
1430 |
|
|
1431 |
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96)*/ ) |
if ((MVzero(*currMV)) && |
1432 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1433 |
iMinSAD -= MV16_00_BIAS; |
iMinSAD -= MV16_00_BIAS; |
1434 |
|
|
1435 |
|
|
1437 |
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. |
1438 |
*/ |
*/ |
1439 |
|
|
1440 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
if ((iMinSAD <= threshA) || |
1441 |
{ |
(MVequal(*currMV, prevMB->mvs[0]) && |
1442 |
|
((uint32_t) iMinSAD < prevMB->sad16))) { |
1443 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1444 |
goto PMVfast16_Terminate_without_Refine; |
goto PMVfast16_Terminate_without_Refine; |
1445 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1457 |
Refine by using small diamond and goto step 10. |
Refine by using small diamond and goto step 10. |
1458 |
*/ |
*/ |
1459 |
|
|
1460 |
|
if (MotionFlags & PMV_USESQUARES16) |
1461 |
|
MainSearchPtr = Square16_MainSearch; |
1462 |
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
1463 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
1464 |
|
else |
1465 |
|
MainSearchPtr = Diamond16_MainSearch; |
1466 |
|
|
1467 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1468 |
|
|
1469 |
/* 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 */ |
1470 |
iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1471 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
1472 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
1473 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1474 |
|
iQuant, iFound); |
1475 |
|
|
1476 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1477 |
*currMV = newMV; |
*currMV = newMV; |
1478 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1479 |
} |
} |
1480 |
|
|
1481 |
if (MotionFlags & PMV_EXTSEARCH16) |
if (MotionFlags & PMV_EXTSEARCH16) { |
|
{ |
|
1482 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1483 |
|
|
1484 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
1485 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1486 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1487 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
1488 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1489 |
|
iDiamondSize, iFcode, iQuant, iFound); |
1490 |
|
|
1491 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1492 |
*currMV = newMV; |
*currMV = newMV; |
1493 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1494 |
} |
} |
1495 |
} |
} |
1496 |
|
|
1497 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1498 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1499 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1500 |
0, 0, iMinSAD, &newMV, |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
1501 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1502 |
|
iQuant, iFound); |
1503 |
|
|
1504 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1505 |
*currMV = newMV; |
*currMV = newMV; |
1506 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1507 |
} |
} |
1514 |
|
|
1515 |
PMVfast16_Terminate_with_Refine: |
PMVfast16_Terminate_with_Refine: |
1516 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
1517 |
iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
1518 |
x, y, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1519 |
currMV, iMinSAD, |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
1520 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1521 |
|
|
1522 |
PMVfast16_Terminate_without_Refine: |
PMVfast16_Terminate_without_Refine: |
1523 |
currPMV->x = currMV->x - pmv[0].x; |
currPMV->x = currMV->x - pmv[0].x; |
1530 |
|
|
1531 |
|
|
1532 |
|
|
1533 |
int32_t Diamond8_MainSearch( |
int32_t |
1534 |
const uint8_t * const pRef, |
Diamond8_MainSearch(const uint8_t * const pRef, |
1535 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1536 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1537 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1538 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1539 |
const int x, const int y, |
const int x, |
1540 |
int32_t startx, int32_t starty, |
const int y, |
1541 |
|
int32_t startx, |
1542 |
|
int32_t starty, |
1543 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1544 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1545 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
1546 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
1547 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
1548 |
|
const int32_t min_dy, |
1549 |
|
const int32_t max_dy, |
1550 |
const int32_t iEdgedWidth, |
const int32_t iEdgedWidth, |
1551 |
const int32_t iDiamondSize, |
const int32_t iDiamondSize, |
1552 |
const int32_t iFcode, |
const int32_t iFcode, |
1558 |
int32_t iDirection=0; |
int32_t iDirection=0; |
1559 |
int32_t iSAD; |
int32_t iSAD; |
1560 |
VECTOR backupMV; |
VECTOR backupMV; |
1561 |
|
|
1562 |
backupMV.x = startx; |
backupMV.x = startx; |
1563 |
backupMV.y = starty; |
backupMV.y = starty; |
1564 |
|
|
1570 |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV8_CANDIDATE_DIR(backupMV.x,backupMV.y+iDiamondSize,4); |
1571 |
|
|
1572 |
if (iDirection) |
if (iDirection) |
1573 |
while (!iFound) |
while (!iFound) { |
|
{ |
|
1574 |
iFound = 1; |
iFound = 1; |
1575 |
backupMV=*currMV; // since iDirection!=0, this is well defined! |
backupMV=*currMV; // since iDirection!=0, this is well defined! |
1576 |
|
|
1577 |
if ( iDirection != 2) |
if ( iDirection != 2) |
1578 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x-iDiamondSize,backupMV.y,1); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x - iDiamondSize, |
1579 |
|
backupMV.y, 1); |
1580 |
if ( iDirection != 1) |
if ( iDirection != 1) |
1581 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x+iDiamondSize,backupMV.y,2); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x + iDiamondSize, |
1582 |
|
backupMV.y, 2); |
1583 |
if ( iDirection != 4) |
if ( iDirection != 4) |
1584 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y-iDiamondSize,3); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1585 |
|
backupMV.y - iDiamondSize, 3); |
1586 |
if ( iDirection != 3) |
if ( iDirection != 3) |
1587 |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x,backupMV.y+iDiamondSize,4); |
CHECK_MV8_CANDIDATE_FOUND(backupMV.x, |
1588 |
} |
backupMV.y + iDiamondSize, 4); |
1589 |
else |
} else { |
|
{ |
|
1590 |
currMV->x = startx; |
currMV->x = startx; |
1591 |
currMV->y = starty; |
currMV->y = starty; |
1592 |
} |
} |
1593 |
return iMinSAD; |
return iMinSAD; |
1594 |
} |
} |
1595 |
|
|
1596 |
int32_t Halfpel8_Refine( |
int32_t |
1597 |
const uint8_t * const pRef, |
Halfpel8_Refine(const uint8_t * const pRef, |
1598 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1599 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1600 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1601 |
const uint8_t * const cur, |
const uint8_t * const cur, |
1602 |
const int x, const int y, |
const int x, |
1603 |
|
const int y, |
1604 |
VECTOR * const currMV, |
VECTOR * const currMV, |
1605 |
int32_t iMinSAD, |
int32_t iMinSAD, |
1606 |
const VECTOR * const pmv, |
const VECTOR * const pmv, |
1607 |
const int32_t min_dx, const int32_t max_dx, |
const int32_t min_dx, |
1608 |
const int32_t min_dy, const int32_t max_dy, |
const int32_t max_dx, |
1609 |
|
const int32_t min_dy, |
1610 |
|
const int32_t max_dy, |
1611 |
const int32_t iFcode, |
const int32_t iFcode, |
1612 |
const int32_t iQuant, |
const int32_t iQuant, |
1613 |
const int32_t iEdgedWidth) |
const int32_t iEdgedWidth) |
1632 |
|
|
1633 |
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
#define PMV_HALFPEL8 (PMV_HALFPELDIAMOND8|PMV_HALFPELREFINE8) |
1634 |
|
|
1635 |
int32_t PMVfastSearch8( |
int32_t |
1636 |
const uint8_t * const pRef, |
PMVfastSearch8(const uint8_t * const pRef, |
1637 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1638 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1639 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1640 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1641 |
const int x, const int y, |
const int x, |
1642 |
const int start_x, const int start_y, |
const int y, |
1643 |
|
const int start_x, |
1644 |
|
const int start_y, |
1645 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1646 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1647 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1669 |
int32_t psad[4]; |
int32_t psad[4]; |
1670 |
VECTOR newMV; |
VECTOR newMV; |
1671 |
VECTOR backupMV; |
VECTOR backupMV; |
1672 |
|
VECTOR startMV; |
1673 |
|
|
1674 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
1675 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
1676 |
|
|
1677 |
static int32_t threshA,threshB; |
static int32_t threshA,threshB; |
1678 |
int32_t iFound,bPredEq; |
int32_t iFound,bPredEq; |
1679 |
int32_t iMinSAD,iSAD; |
int32_t iMinSAD,iSAD; |
1680 |
|
|
1681 |
int32_t iSubBlock = ((y&1)<<1) + (x&1); |
int32_t iSubBlock = (y & 1) + (y & 1) + (x & 1); |
1682 |
|
|
1683 |
|
MainSearch8FuncPtr MainSearchPtr; |
1684 |
|
|
1685 |
|
/* Init variables */ |
1686 |
|
startMV.x = start_x; |
1687 |
|
startMV.y = start_y; |
1688 |
|
|
1689 |
/* Get maximum range */ |
/* Get maximum range */ |
1690 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
1691 |
x, y, 8, iWidth, iHeight, iFcode); |
iFcode); |
1692 |
|
|
1693 |
if (!(MotionFlags & PMV_HALFPELDIAMOND8 )) |
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) { |
1694 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
1695 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
1696 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
1697 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
1698 |
} /* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
} |
|
|
|
1699 |
|
|
1700 |
bPredEq = get_pmvdata(pMBs, (x>>1), (y>>1), iWcount, iSubBlock, pmv, psad); |
/* because we might use IF (dx>max_dx) THEN dx=max_dx; */ |
1701 |
|
bPredEq = |
1702 |
|
get_pmvdata(pMBs, (x >> 1), (y >> 1), iWcount, iSubBlock, pmv, psad); |
1703 |
|
|
1704 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
1705 |
threshA = 512/4; |
threshA = 512/4; |
1706 |
threshB = 1024/4; |
threshB = 1024/4; |
1707 |
|
|
1708 |
} |
} else { |
|
else |
|
|
{ |
|
1709 |
threshA = psad[0]/4; /* good estimate */ |
threshA = psad[0]/4; /* good estimate */ |
1710 |
threshB = threshA+256/4; |
threshB = threshA+256/4; |
1711 |
if (threshA< 512/4) threshA = 512/4; |
if (threshA < 512 / 4) |
1712 |
if (threshA>1024/4) threshA = 1024/4; |
threshA = 512 / 4; |
1713 |
if (threshB>1792/4) threshB = 1792/4; |
if (threshA > 1024 / 4) |
1714 |
|
threshA = 1024 / 4; |
1715 |
|
if (threshB > 1792 / 4) |
1716 |
|
threshB = 1792 / 4; |
1717 |
} |
} |
1718 |
|
|
1719 |
iFound=0; |
iFound=0; |
1720 |
|
|
|
/* 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],prevMB->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; |
|
|
|
|
1721 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
1722 |
MinSAD=SAD |
MinSAD=SAD |
1723 |
If Motion Vector equal to Previous frame motion vector |
If Motion Vector equal to Previous frame motion vector |
1728 |
|
|
1729 |
// Prepare for main loop |
// Prepare for main loop |
1730 |
|
|
1731 |
currMV->x=start_x; /* start with mv16 */ |
// if (MotionFlags & PMV_USESQUARES8) |
1732 |
currMV->y=start_y; |
// MainSearchPtr = Square8_MainSearch; |
1733 |
|
// else |
1734 |
|
|
1735 |
iMinSAD = sad8( cur, |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
1736 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
MainSearchPtr = AdvDiamond8_MainSearch; |
1737 |
iEdgedWidth); |
else |
1738 |
iMinSAD += calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, (uint8_t)iFcode, iQuant); |
MainSearchPtr = Diamond8_MainSearch; |
1739 |
|
|
1740 |
if ( (iMinSAD < 256/4 ) || ( (MVequal(*currMV,prevMB->mvs[iSubBlock])) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
|
1741 |
{ |
*currMV = startMV; |
1742 |
|
|
1743 |
|
iMinSAD = |
1744 |
|
sad8(cur, |
1745 |
|
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
1746 |
|
iEdgedWidth), iEdgedWidth); |
1747 |
|
iMinSAD += |
1748 |
|
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
1749 |
|
(uint8_t) iFcode, iQuant); |
1750 |
|
|
1751 |
|
if ((iMinSAD < 256 / 4) || ((MVequal(*currMV, prevMB->mvs[iSubBlock])) |
1752 |
|
&& ((uint32_t) iMinSAD < |
1753 |
|
prevMB->sad8[iSubBlock]))) { |
1754 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1755 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1756 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1757 |
goto PMVfast8_Terminate_with_Refine; |
goto PMVfast8_Terminate_with_Refine; |
1758 |
} |
} |
1759 |
|
|
1760 |
|
/* Step 2 (lazy eval): Calculate Distance= |MedianMVX| + |MedianMVY| where MedianMV is the motion |
1761 |
|
vector of the median. |
1762 |
|
If PredEq=1 and MVpredicted = Previous Frame MV, set Found=2 |
1763 |
|
*/ |
1764 |
|
|
1765 |
|
if ((bPredEq) && (MVequal(pmv[0], prevMB->mvs[iSubBlock]))) |
1766 |
|
iFound = 2; |
1767 |
|
|
1768 |
|
/* Step 3 (lazy eval): If Distance>0 or thresb<1536 or PredEq=1 Select small Diamond Search. |
1769 |
|
Otherwise select large Diamond Search. |
1770 |
|
*/ |
1771 |
|
|
1772 |
|
if ((!MVzero(pmv[0])) || (threshB < 1536 / 4) || (bPredEq)) |
1773 |
|
iDiamondSize = 1; // 1 halfpel! |
1774 |
|
else |
1775 |
|
iDiamondSize = 2; // 2 halfpel = 1 full pixel! |
1776 |
|
|
1777 |
|
if (!(MotionFlags & PMV_HALFPELDIAMOND8)) |
1778 |
|
iDiamondSize *= 2; |
1779 |
|
|
1780 |
|
|
1781 |
/* |
/* |
1782 |
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. |
1785 |
If MV is (0,0) subtract offset. |
If MV is (0,0) subtract offset. |
1786 |
*/ |
*/ |
1787 |
|
|
1788 |
// the prediction might be even better than mv16 |
// the median prediction might be even better than mv16 |
1789 |
|
|
1790 |
|
if (!MVequal(pmv[0], startMV)) |
1791 |
CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); |
CHECK_MV8_CANDIDATE(pmv[0].x,pmv[0].y); |
1792 |
|
|
1793 |
// (0,0) is always possible |
// (0,0) if needed |
1794 |
|
if (!MVzero(pmv[0])) |
1795 |
|
if (!MVzero(startMV)) |
1796 |
CHECK_MV8_ZERO; |
CHECK_MV8_ZERO; |
1797 |
|
|
1798 |
// previous frame MV is always possible |
// previous frame MV if needed |
1799 |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
if (!MVzero(prevMB->mvs[iSubBlock])) |
1800 |
|
if (!MVequal(prevMB->mvs[iSubBlock], startMV)) |
1801 |
|
if (!MVequal(prevMB->mvs[iSubBlock], pmv[0])) |
1802 |
|
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x, |
1803 |
|
prevMB->mvs[iSubBlock].y); |
1804 |
|
|
1805 |
|
if ((iMinSAD <= threshA) || |
1806 |
|
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1807 |
|
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1808 |
|
if (MotionFlags & PMV_QUICKSTOP16) |
1809 |
|
goto PMVfast8_Terminate_without_Refine; |
1810 |
|
if (MotionFlags & PMV_EARLYSTOP16) |
1811 |
|
goto PMVfast8_Terminate_with_Refine; |
1812 |
|
} |
1813 |
|
|
1814 |
// left neighbour, if allowed |
// left neighbour, if allowed and needed |
1815 |
if (psad[1] != MV_MAX_ERROR) |
if (!MVzero(pmv[1])) |
1816 |
{ |
if (!MVequal(pmv[1], startMV)) |
1817 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[1], prevMB->mvs[iSubBlock])) |
1818 |
{ pmv[1].x = EVEN(pmv[1].x); |
if (!MVequal(pmv[1], pmv[0])) { |
1819 |
|
if (!(MotionFlags & PMV_HALFPEL8)) { |
1820 |
|
pmv[1].x = EVEN(pmv[1].x); |
1821 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
1822 |
} |
} |
1823 |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
1824 |
} |
} |
1825 |
|
// top neighbour, if allowed and needed |
1826 |
// top neighbour, if allowed |
if (!MVzero(pmv[2])) |
1827 |
if (psad[2] != MV_MAX_ERROR) |
if (!MVequal(pmv[2], startMV)) |
1828 |
{ |
if (!MVequal(pmv[2], prevMB->mvs[iSubBlock])) |
1829 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[2], pmv[0])) |
1830 |
{ pmv[2].x = EVEN(pmv[2].x); |
if (!MVequal(pmv[2], pmv[1])) { |
1831 |
|
if (!(MotionFlags & PMV_HALFPEL8)) { |
1832 |
|
pmv[2].x = EVEN(pmv[2].x); |
1833 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
1834 |
} |
} |
1835 |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
1836 |
|
|
1837 |
// top right neighbour, if allowed |
// top right neighbour, if allowed and needed |
1838 |
if (psad[3] != MV_MAX_ERROR) |
if (!MVzero(pmv[3])) |
1839 |
{ |
if (!MVequal(pmv[3], startMV)) |
1840 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!MVequal(pmv[3], prevMB->mvs[iSubBlock])) |
1841 |
{ pmv[3].x = EVEN(pmv[3].x); |
if (!MVequal(pmv[3], pmv[0])) |
1842 |
|
if (!MVequal(pmv[3], pmv[1])) |
1843 |
|
if (!MVequal(pmv[3], pmv[2])) { |
1844 |
|
if (! |
1845 |
|
(MotionFlags & |
1846 |
|
PMV_HALFPEL8)) { |
1847 |
|
pmv[3].x = EVEN(pmv[3].x); |
1848 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
1849 |
} |
} |
1850 |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV8_CANDIDATE(pmv[3].x, |
1851 |
|
pmv[3].y); |
1852 |
} |
} |
1853 |
} |
} |
1854 |
|
|
1855 |
if ( (MVzero(*currMV)) && (!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
if ((MVzero(*currMV)) && |
1856 |
|
(!MVzero(pmv[0])) /* && (iMinSAD <= iQuant * 96) */ ) |
1857 |
iMinSAD -= MV8_00_BIAS; |
iMinSAD -= MV8_00_BIAS; |
1858 |
|
|
1859 |
|
|
1861 |
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. |
1862 |
*/ |
*/ |
1863 |
|
|
1864 |
if ( (iMinSAD <= threshA) || ( MVequal(*currMV,prevMB->mvs[iSubBlock]) && ((uint32_t)iMinSAD < prevMB->sad8[iSubBlock]) ) ) |
if ((iMinSAD <= threshA) || |
1865 |
{ |
(MVequal(*currMV, prevMB->mvs[iSubBlock]) && |
1866 |
|
((uint32_t) iMinSAD < prevMB->sad8[iSubBlock]))) { |
1867 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
1868 |
goto PMVfast8_Terminate_without_Refine; |
goto PMVfast8_Terminate_without_Refine; |
1869 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
1883 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
1884 |
|
|
1885 |
/* 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 */ |
1886 |
iSAD = Diamond8_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1887 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
1888 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
1889 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1890 |
|
iQuant, iFound); |
1891 |
|
|
1892 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1893 |
*currMV = newMV; |
*currMV = newMV; |
1894 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1895 |
} |
} |
1896 |
|
|
1897 |
if (MotionFlags & PMV_EXTSEARCH8) |
if (MotionFlags & PMV_EXTSEARCH8) { |
|
{ |
|
1898 |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended: search (up to) two more times: orignal prediction and (0,0) */ |
1899 |
|
|
1900 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
1901 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1902 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
1903 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
1904 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
1905 |
|
iDiamondSize, iFcode, iQuant, iFound); |
1906 |
|
|
1907 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1908 |
*currMV = newMV; |
*currMV = newMV; |
1909 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1910 |
} |
} |
1911 |
} |
} |
1912 |
|
|
1913 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
1914 |
{ iSAD = Diamond16_MainSearch(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
1915 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
1916 |
0, 0, iMinSAD, &newMV, |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
1917 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, iFound); |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
1918 |
|
iQuant, iFound); |
1919 |
|
|
1920 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
1921 |
*currMV = newMV; |
*currMV = newMV; |
1922 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
1923 |
} |
} |
1930 |
|
|
1931 |
PMVfast8_Terminate_with_Refine: |
PMVfast8_Terminate_with_Refine: |
1932 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
1933 |
iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
1934 |
x, y, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
1935 |
currMV, iMinSAD, |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
1936 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
1937 |
|
|
1938 |
|
|
1939 |
PMVfast8_Terminate_without_Refine: |
PMVfast8_Terminate_without_Refine: |
1943 |
return iMinSAD; |
return iMinSAD; |
1944 |
} |
} |
1945 |
|
|
1946 |
int32_t EPZSSearch16( |
int32_t |
1947 |
const uint8_t * const pRef, |
EPZSSearch16(const uint8_t * const pRef, |
1948 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1949 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1950 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1951 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
1952 |
const int x, const int y, |
const int x, |
1953 |
|
const int y, |
1954 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1955 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1956 |
const uint32_t iFcode, |
const uint32_t iFcode, |
1981 |
int32_t psad[8]; |
int32_t psad[8]; |
1982 |
|
|
1983 |
static MACROBLOCK * oldMBs = NULL; |
static MACROBLOCK * oldMBs = NULL; |
1984 |
const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
|
1985 |
|
// const MACROBLOCK * const pMB = pMBs + x + y * iWcount; |
1986 |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + x + y * iWcount; |
1987 |
MACROBLOCK * oldMB = NULL; |
MACROBLOCK * oldMB = NULL; |
1988 |
|
|
1990 |
int32_t bPredEq; |
int32_t bPredEq; |
1991 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
1992 |
|
|
1993 |
MainSearch16FuncPtr EPZSMainSearchPtr; |
MainSearch16FuncPtr MainSearchPtr; |
1994 |
|
|
1995 |
if (oldMBs == NULL) |
if (oldMBs == NULL) { |
1996 |
{ oldMBs = (MACROBLOCK*) calloc(iWcount*iHcount,sizeof(MACROBLOCK)); |
oldMBs = (MACROBLOCK *) calloc(iWcount * iHcount, sizeof(MACROBLOCK)); |
1997 |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
// fprintf(stderr,"allocated %d bytes for oldMBs\n",iWcount*iHcount*sizeof(MACROBLOCK)); |
1998 |
} |
} |
1999 |
oldMB = oldMBs + x + y * iWcount; |
oldMB = oldMBs + x + y * iWcount; |
2000 |
|
|
2001 |
/* Get maximum range */ |
/* Get maximum range */ |
2002 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 16, iWidth, iHeight, |
2003 |
x, y, 16, iWidth, iHeight, iFcode); |
iFcode); |
2004 |
|
|
2005 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2006 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
2007 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
2008 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
2009 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2010 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
2011 |
|
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2012 |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
bPredEq = get_pmvdata(pMBs, x, y, iWcount, 0, pmv, psad); |
2013 |
|
|
2014 |
/* Step 4: Calculate SAD around the Median prediction. |
/* Step 4: Calculate SAD around the Median prediction. |
2021 |
// Prepare for main loop |
// Prepare for main loop |
2022 |
|
|
2023 |
*currMV=pmv[0]; /* current best := median prediction */ |
*currMV=pmv[0]; /* current best := median prediction */ |
2024 |
if (!(MotionFlags & PMV_HALFPEL16)) |
if (!(MotionFlags & PMV_HALFPEL16)) { |
|
{ |
|
2025 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2026 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2027 |
} |
} |
2037 |
|
|
2038 |
/***************** This is predictor SET A: only median prediction ******************/ |
/***************** This is predictor SET A: only median prediction ******************/ |
2039 |
|
|
2040 |
iMinSAD = sad16( cur, |
iMinSAD = |
2041 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, iEdgedWidth), |
sad16(cur, |
2042 |
iEdgedWidth, MV_MAX_ERROR); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 16, currMV, |
2043 |
iMinSAD += calc_delta_16(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth, MV_MAX_ERROR); |
2044 |
|
iMinSAD += |
2045 |
|
calc_delta_16(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
2046 |
|
(uint8_t) iFcode, iQuant); |
2047 |
|
|
2048 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2049 |
if ( (iMinSAD < 256 ) || ( (MVequal(*currMV, prevMB->mvs[0])) && ((uint32_t)iMinSAD < prevMB->sad16) ) ) |
if ((iMinSAD < 256) || |
2050 |
{ |
((MVequal(*currMV, prevMB->mvs[0])) && |
2051 |
|
((uint32_t) iMinSAD < prevMB->sad16))) { |
2052 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2053 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2054 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2063 |
// set threshhold based on Min of Prediction and SAD of collocated block |
// set threshhold based on Min of Prediction and SAD of collocated block |
2064 |
// 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 |
2065 |
|
|
2066 |
if ((x==0) && (y==0) ) |
if ((x == 0) && (y == 0)) { |
|
{ |
|
2067 |
thresh2 = 512; |
thresh2 = 512; |
2068 |
} |
} else { |
|
else |
|
|
{ |
|
2069 |
/* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ |
/* T_k = 1.2 * MIN(SAD_top,SAD_left,SAD_topleft,SAD_coll) +128; [Tourapis, 2002] */ |
2070 |
|
|
2071 |
thresh2 = MIN(psad[0],iSAD)*6/5 + 128; |
thresh2 = MIN(psad[0],iSAD)*6/5 + 128; |
2077 |
|
|
2078 |
|
|
2079 |
// left neighbour, if allowed |
// left neighbour, if allowed |
2080 |
if (x != 0) |
if (x != 0) { |
2081 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2082 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[1].x = EVEN(pmv[1].x); |
|
{ pmv[1].x = EVEN(pmv[1].x); |
|
2083 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
2084 |
} |
} |
2085 |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV16_CANDIDATE(pmv[1].x,pmv[1].y); |
2086 |
} |
} |
|
|
|
2087 |
// top neighbour, if allowed |
// top neighbour, if allowed |
2088 |
if (y != 0) |
if (y != 0) { |
2089 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2090 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[2].x = EVEN(pmv[2].x); |
|
{ pmv[2].x = EVEN(pmv[2].x); |
|
2091 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
2092 |
} |
} |
2093 |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV16_CANDIDATE(pmv[2].x,pmv[2].y); |
2094 |
|
|
2095 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
2096 |
if ((uint32_t)x != (iWcount-1)) |
if ((uint32_t) x != (iWcount - 1)) { |
2097 |
{ |
if (!(MotionFlags & PMV_HALFPEL16)) { |
2098 |
if (!(MotionFlags & PMV_HALFPEL16 )) |
pmv[3].x = EVEN(pmv[3].x); |
|
{ pmv[3].x = EVEN(pmv[3].x); |
|
2099 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
2100 |
} |
} |
2101 |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV16_CANDIDATE(pmv[3].x,pmv[3].y); |
2107 |
*/ |
*/ |
2108 |
|
|
2109 |
if ( (iMinSAD <= thresh2) |
if ( (iMinSAD <= thresh2) |
2110 |
|| ( MVequal(*currMV,prevMB->mvs[0]) && ((uint32_t)iMinSAD <= prevMB->sad16) ) ) |
|| (MVequal(*currMV, prevMB->mvs[0]) && |
2111 |
{ |
((uint32_t) iMinSAD <= prevMB->sad16))) { |
2112 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2113 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2114 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2129 |
|
|
2130 |
// top neighbour |
// top neighbour |
2131 |
if (y != 0) |
if (y != 0) |
2132 |
CHECK_MV16_CANDIDATE((prevMB-iWcount)->mvs[0].x,(prevMB-iWcount)->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB - iWcount)->mvs[0].x, |
2133 |
|
(prevMB - iWcount)->mvs[0].y); |
2134 |
|
|
2135 |
// 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 |
2136 |
|
|
2139 |
|
|
2140 |
// bottom neighbour, dito |
// bottom neighbour, dito |
2141 |
if ((uint32_t)y != iHcount-1) |
if ((uint32_t)y != iHcount-1) |
2142 |
CHECK_MV16_CANDIDATE((prevMB+iWcount)->mvs[0].x,(prevMB+iWcount)->mvs[0].y); |
CHECK_MV16_CANDIDATE((prevMB + iWcount)->mvs[0].x, |
2143 |
|
(prevMB + iWcount)->mvs[0].y); |
2144 |
|
|
2145 |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
/* Terminate if MinSAD <= T_3 (here T_3 = T_2) */ |
2146 |
if (iMinSAD <= thresh2) |
if (iMinSAD <= thresh2) { |
|
{ |
|
2147 |
if (MotionFlags & PMV_QUICKSTOP16) |
if (MotionFlags & PMV_QUICKSTOP16) |
2148 |
goto EPZS16_Terminate_without_Refine; |
goto EPZS16_Terminate_without_Refine; |
2149 |
if (MotionFlags & PMV_EARLYSTOP16) |
if (MotionFlags & PMV_EARLYSTOP16) |
2154 |
|
|
2155 |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
backupMV = *currMV; /* save best prediction, actually only for EXTSEARCH */ |
2156 |
|
|
2157 |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
if (MotionFlags & PMV_USESQUARES8) |
2158 |
|
MainSearchPtr = Square16_MainSearch; |
2159 |
if (MotionFlags & PMV_USESQUARES16) |
else |
2160 |
EPZSMainSearchPtr = Square16_MainSearch; |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2161 |
|
MainSearchPtr = AdvDiamond16_MainSearch; |
2162 |
else |
else |
2163 |
EPZSMainSearchPtr = Diamond16_MainSearch; |
MainSearchPtr = Diamond16_MainSearch; |
2164 |
|
|
2165 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
/* default: use best prediction as starting point for one call of PMVfast_MainSearch */ |
|
x, y, |
|
|
currMV->x, currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
|
|
2, iFcode, iQuant, 0); |
|
2166 |
|
|
2167 |
if (iSAD < iMinSAD) |
iSAD = |
2168 |
{ |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2169 |
|
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
2170 |
|
min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2171 |
|
|
2172 |
|
if (iSAD < iMinSAD) { |
2173 |
*currMV = newMV; |
*currMV = newMV; |
2174 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2175 |
} |
} |
2176 |
|
|
2177 |
|
|
2178 |
if (MotionFlags & PMV_EXTSEARCH16) |
if (MotionFlags & PMV_EXTSEARCH16) { |
|
{ |
|
2179 |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2180 |
|
|
2181 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
2182 |
{ |
iSAD = |
2183 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2184 |
x, y, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
2185 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2186 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
2, iFcode, iQuant, 0); |
2187 |
} |
} |
2188 |
|
|
2189 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2190 |
*currMV = newMV; |
*currMV = newMV; |
2191 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2192 |
} |
} |
2193 |
|
|
2194 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2195 |
{ |
iSAD = |
2196 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2197 |
x, y, |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
2198 |
0, 0, iMinSAD, &newMV, |
max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
|
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, 2, iFcode, iQuant, 0); |
|
2199 |
|
|
2200 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2201 |
*currMV = newMV; |
*currMV = newMV; |
2202 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2203 |
} |
} |
2208 |
|
|
2209 |
EPZS16_Terminate_with_Refine: |
EPZS16_Terminate_with_Refine: |
2210 |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE16) // perform final half-pel step |
2211 |
iMinSAD = Halfpel16_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
2212 |
x, y, |
Halfpel16_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2213 |
currMV, iMinSAD, |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
2214 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2215 |
|
|
2216 |
EPZS16_Terminate_without_Refine: |
EPZS16_Terminate_without_Refine: |
2217 |
|
|
2223 |
} |
} |
2224 |
|
|
2225 |
|
|
2226 |
int32_t EPZSSearch8( |
int32_t |
2227 |
const uint8_t * const pRef, |
EPZSSearch8(const uint8_t * const pRef, |
2228 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
2229 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
2230 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
2231 |
const IMAGE * const pCur, |
const IMAGE * const pCur, |
2232 |
const int x, const int y, |
const int x, |
2233 |
const int start_x, const int start_y, |
const int y, |
2234 |
|
const int start_x, |
2235 |
|
const int start_y, |
2236 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
2237 |
const uint32_t iQuant, |
const uint32_t iQuant, |
2238 |
const uint32_t iFcode, |
const uint32_t iFcode, |
2266 |
|
|
2267 |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
const int32_t iSubBlock = ((y&1)<<1) + (x&1); |
2268 |
|
|
2269 |
const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
// const MACROBLOCK * const pMB = pMBs + (x>>1) + (y>>1) * iWcount; |
2270 |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
const MACROBLOCK * const prevMB = prevMBs + (x>>1) + (y>>1) * iWcount; |
2271 |
|
|
2272 |
int32_t bPredEq; |
int32_t bPredEq; |
2273 |
int32_t iMinSAD,iSAD=9999; |
int32_t iMinSAD,iSAD=9999; |
2274 |
|
|
2275 |
MainSearch8FuncPtr EPZSMainSearchPtr; |
MainSearch8FuncPtr MainSearchPtr; |
2276 |
|
|
2277 |
/* Get maximum range */ |
/* Get maximum range */ |
2278 |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, |
get_range(&min_dx, &max_dx, &min_dy, &max_dy, x, y, 8, iWidth, iHeight, |
2279 |
x, y, 8, iWidth, iHeight, iFcode); |
iFcode); |
2280 |
|
|
2281 |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
/* we work with abs. MVs, not relative to prediction, so get_range is called relative to 0,0 */ |
2282 |
|
|
2283 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2284 |
{ min_dx = EVEN(min_dx); |
min_dx = EVEN(min_dx); |
2285 |
max_dx = EVEN(max_dx); |
max_dx = EVEN(max_dx); |
2286 |
min_dy = EVEN(min_dy); |
min_dy = EVEN(min_dy); |
2287 |
max_dy = EVEN(max_dy); |
max_dy = EVEN(max_dy); |
2288 |
} /* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
} |
2289 |
|
/* because we might use something like IF (dx>max_dx) THEN dx=max_dx; */ |
2290 |
bPredEq = get_pmvdata(pMBs, x>>1, y>>1, iWcount, iSubBlock, pmv, psad); |
bPredEq = get_pmvdata(pMBs, x>>1, y>>1, iWcount, iSubBlock, pmv, psad); |
2291 |
|
|
2292 |
|
|
2300 |
// Prepare for main loop |
// Prepare for main loop |
2301 |
|
|
2302 |
|
|
2303 |
if (!(MotionFlags & PMV_HALFPEL8)) |
if (!(MotionFlags & PMV_HALFPEL8)) { |
|
{ |
|
2304 |
currMV->x = EVEN(currMV->x); |
currMV->x = EVEN(currMV->x); |
2305 |
currMV->y = EVEN(currMV->y); |
currMV->y = EVEN(currMV->y); |
2306 |
} |
} |
2317 |
/***************** This is predictor SET A: only median prediction ******************/ |
/***************** This is predictor SET A: only median prediction ******************/ |
2318 |
|
|
2319 |
|
|
2320 |
iMinSAD = sad8( cur, |
iMinSAD = |
2321 |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, iEdgedWidth), |
sad8(cur, |
2322 |
iEdgedWidth); |
get_ref_mv(pRef, pRefH, pRefV, pRefHV, x, y, 8, currMV, |
2323 |
iMinSAD += calc_delta_8(currMV->x-pmv[0].x, currMV->y-pmv[0].y, (uint8_t)iFcode, iQuant); |
iEdgedWidth), iEdgedWidth); |
2324 |
|
iMinSAD += |
2325 |
|
calc_delta_8(currMV->x - pmv[0].x, currMV->y - pmv[0].y, |
2326 |
|
(uint8_t) iFcode, iQuant); |
2327 |
|
|
2328 |
|
|
2329 |
// thresh1 is fixed to 256 |
// thresh1 is fixed to 256 |
2330 |
if (iMinSAD < 256/4 ) |
if (iMinSAD < 256 / 4) { |
|
{ |
|
2331 |
if (MotionFlags & PMV_QUICKSTOP8) |
if (MotionFlags & PMV_QUICKSTOP8) |
2332 |
goto EPZS8_Terminate_without_Refine; |
goto EPZS8_Terminate_without_Refine; |
2333 |
if (MotionFlags & PMV_EARLYSTOP8) |
if (MotionFlags & PMV_EARLYSTOP8) |
2344 |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
CHECK_MV8_CANDIDATE(prevMB->mvs[iSubBlock].x,prevMB->mvs[iSubBlock].y); |
2345 |
|
|
2346 |
// left neighbour, if allowed |
// left neighbour, if allowed |
2347 |
if (psad[1] != MV_MAX_ERROR) |
if (psad[1] != MV_MAX_ERROR) { |
2348 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2349 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[1].x = EVEN(pmv[1].x); |
|
{ pmv[1].x = EVEN(pmv[1].x); |
|
2350 |
pmv[1].y = EVEN(pmv[1].y); |
pmv[1].y = EVEN(pmv[1].y); |
2351 |
} |
} |
2352 |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
CHECK_MV8_CANDIDATE(pmv[1].x,pmv[1].y); |
2353 |
} |
} |
|
|
|
2354 |
// top neighbour, if allowed |
// top neighbour, if allowed |
2355 |
if (psad[2] != MV_MAX_ERROR) |
if (psad[2] != MV_MAX_ERROR) { |
2356 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2357 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[2].x = EVEN(pmv[2].x); |
|
{ pmv[2].x = EVEN(pmv[2].x); |
|
2358 |
pmv[2].y = EVEN(pmv[2].y); |
pmv[2].y = EVEN(pmv[2].y); |
2359 |
} |
} |
2360 |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
CHECK_MV8_CANDIDATE(pmv[2].x,pmv[2].y); |
2361 |
|
|
2362 |
// top right neighbour, if allowed |
// top right neighbour, if allowed |
2363 |
if (psad[3] != MV_MAX_ERROR) |
if (psad[3] != MV_MAX_ERROR) { |
2364 |
{ |
if (!(MotionFlags & PMV_HALFPEL8)) { |
2365 |
if (!(MotionFlags & PMV_HALFPEL8 )) |
pmv[3].x = EVEN(pmv[3].x); |
|
{ pmv[3].x = EVEN(pmv[3].x); |
|
2366 |
pmv[3].y = EVEN(pmv[3].y); |
pmv[3].y = EVEN(pmv[3].y); |
2367 |
} |
} |
2368 |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
CHECK_MV8_CANDIDATE(pmv[3].x,pmv[3].y); |
2380 |
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] |
2381 |
*/ |
*/ |
2382 |
|
|
2383 |
if (iMinSAD < 512/4) /* T_2 == 512/4 hardcoded */ |
if (iMinSAD < 512 / 4) { /* T_2 == 512/4 hardcoded */ |
|
{ |
|
2384 |
if (MotionFlags & PMV_QUICKSTOP8) |
if (MotionFlags & PMV_QUICKSTOP8) |
2385 |
goto EPZS8_Terminate_without_Refine; |
goto EPZS8_Terminate_without_Refine; |
2386 |
if (MotionFlags & PMV_EARLYSTOP8) |
if (MotionFlags & PMV_EARLYSTOP8) |
2396 |
|
|
2397 |
/* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
/* default: use best prediction as starting point for one call of EPZS_MainSearch */ |
2398 |
|
|
2399 |
/* // there is no EPZS^2 for inter4v at the moment |
// there is no EPZS^2 for inter4v at the moment |
2400 |
|
|
2401 |
if (MotionFlags & PMV_USESQUARES8) |
// if (MotionFlags & PMV_USESQUARES8) |
2402 |
EPZSMainSearchPtr = Square8_MainSearch; |
// MainSearchPtr = Square8_MainSearch; |
2403 |
else |
// else |
|
*/ |
|
2404 |
|
|
2405 |
EPZSMainSearchPtr = Diamond8_MainSearch; |
if (MotionFlags & PMV_ADVANCEDDIAMOND8) |
2406 |
|
MainSearchPtr = AdvDiamond8_MainSearch; |
2407 |
|
else |
2408 |
|
MainSearchPtr = Diamond8_MainSearch; |
2409 |
|
|
2410 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
iSAD = |
2411 |
x, y, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV->x, |
2412 |
currMV->x, currMV->y, iMinSAD, &newMV, |
currMV->y, iMinSAD, &newMV, pmv, min_dx, max_dx, |
2413 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2414 |
iDiamondSize, iFcode, iQuant, 0); |
iQuant, 0); |
2415 |
|
|
2416 |
|
|
2417 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2418 |
*currMV = newMV; |
*currMV = newMV; |
2419 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2420 |
} |
} |
2421 |
|
|
2422 |
if (MotionFlags & PMV_EXTSEARCH8) |
if (MotionFlags & PMV_EXTSEARCH8) { |
|
{ |
|
2423 |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
/* extended mode: search (up to) two more times: orignal prediction and (0,0) */ |
2424 |
|
|
2425 |
if (!(MVequal(pmv[0],backupMV)) ) |
if (!(MVequal(pmv[0], backupMV))) { |
2426 |
{ |
iSAD = |
2427 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, |
2428 |
x, y, |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, pmv, |
2429 |
pmv[0].x, pmv[0].y, iMinSAD, &newMV, |
min_dx, max_dx, min_dy, max_dy, iEdgedWidth, |
2430 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
iDiamondSize, iFcode, iQuant, 0); |
2431 |
|
|
2432 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2433 |
*currMV = newMV; |
*currMV = newMV; |
2434 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2435 |
} |
} |
2436 |
} |
} |
2437 |
|
|
2438 |
if ( (!(MVzero(pmv[0]))) && (!(MVzero(backupMV))) ) |
if ((!(MVzero(pmv[0]))) && (!(MVzero(backupMV)))) { |
2439 |
{ |
iSAD = |
2440 |
iSAD = (*EPZSMainSearchPtr)(pRef, pRefH, pRefV, pRefHV, cur, |
(*MainSearchPtr) (pRef, pRefH, pRefV, pRefHV, cur, x, y, 0, 0, |
2441 |
x, y, |
iMinSAD, &newMV, pmv, min_dx, max_dx, min_dy, |
2442 |
0, 0, iMinSAD, &newMV, |
max_dy, iEdgedWidth, iDiamondSize, iFcode, |
2443 |
pmv, min_dx, max_dx, min_dy, max_dy, iEdgedWidth, iDiamondSize, iFcode, iQuant, 0); |
iQuant, 0); |
2444 |
|
|
2445 |
if (iSAD < iMinSAD) |
if (iSAD < iMinSAD) { |
|
{ |
|
2446 |
*currMV = newMV; |
*currMV = newMV; |
2447 |
iMinSAD = iSAD; |
iMinSAD = iSAD; |
2448 |
} |
} |
2453 |
|
|
2454 |
EPZS8_Terminate_with_Refine: |
EPZS8_Terminate_with_Refine: |
2455 |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
if (MotionFlags & PMV_HALFPELREFINE8) // perform final half-pel step |
2456 |
iMinSAD = Halfpel8_Refine( pRef, pRefH, pRefV, pRefHV, cur, |
iMinSAD = |
2457 |
x, y, |
Halfpel8_Refine(pRef, pRefH, pRefV, pRefHV, cur, x, y, currMV, |
2458 |
currMV, iMinSAD, |
iMinSAD, pmv, min_dx, max_dx, min_dy, max_dy, |
2459 |
pmv, min_dx, max_dx, min_dy, max_dy, iFcode, iQuant, iEdgedWidth); |
iFcode, iQuant, iEdgedWidth); |
2460 |
|
|
2461 |
EPZS8_Terminate_without_Refine: |
EPZS8_Terminate_without_Refine: |
2462 |
|
|
2475 |
***************************************************************/ |
***************************************************************/ |
2476 |
|
|
2477 |
|
|
2478 |
void MotionEstimationBVOP( |
void |
2479 |
MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
2480 |
FRAMEINFO * const frame, |
FRAMEINFO * const frame, |
|
|
|
2481 |
// forward (past) reference |
// forward (past) reference |
2482 |
const MACROBLOCK * const f_mbs, |
const MACROBLOCK * const f_mbs, |
2483 |
const IMAGE * const f_ref, |
const IMAGE * const f_ref, |
2506 |
VECTOR pmv_dontcare; |
VECTOR pmv_dontcare; |
2507 |
|
|
2508 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
2509 |
for (j = 0; j < mb_height; j++) |
for (j = 0; j < mb_height; j++) { |
2510 |
{ |
for (i = 0; i < mb_width; i++) { |
|
for (i = 0; i < mb_width; i++) |
|
|
{ |
|
2511 |
MACROBLOCK *mb = &frame->mbs[i + j*mb_width]; |
MACROBLOCK *mb = &frame->mbs[i + j*mb_width]; |
2512 |
const MACROBLOCK *f_mb = &f_mbs[i + j*mb_width]; |
const MACROBLOCK *f_mb = &f_mbs[i + j*mb_width]; |
2513 |
const MACROBLOCK *b_mb = &b_mbs[i + j*mb_width]; |
const MACROBLOCK *b_mb = &b_mbs[i + j*mb_width]; |
2514 |
|
|
2515 |
if (b_mb->mode == MODE_INTER |
if (b_mb->mode == MODE_INTER && b_mb->cbp == 0 && |
2516 |
&& b_mb->cbp == 0 |
b_mb->mvs[0].x == 0 && b_mb->mvs[0].y == 0) { |
|
&& b_mb->mvs[0].x == 0 |
|
|
&& b_mb->mvs[0].y == 0) |
|
|
{ |
|
2517 |
mb->mode = MODE_NOT_CODED; |
mb->mode = MODE_NOT_CODED; |
2518 |
mb->mvs[0].x = 0; |
mb->mvs[0].x = 0; |
2519 |
mb->mvs[0].y = 0; |
mb->mvs[0].y = 0; |
2521 |
mb->b_mvs[0].y = 0; |
mb->b_mvs[0].y = 0; |
2522 |
continue; |
continue; |
2523 |
} |
} |
2524 |
|
/* force F_SAD16 |
2525 |
|
f_sad16 = 100; |
2526 |
|
b_sad16 = 65535; |
2527 |
|
|
2528 |
|
mb->mode = MODE_FORWARD; |
2529 |
|
mb->mvs[0].x = 1; |
2530 |
|
mb->mvs[0].y = 1; |
2531 |
|
mb->b_mvs[0].x = 1; |
2532 |
|
mb->b_mvs[0].y = 1; |
2533 |
|
continue; |
2534 |
|
^^ force F_SAD16 */ |
2535 |
|
|
2536 |
|
|
2537 |
// forward search |
// forward search |
2538 |
f_sad16 = SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
f_sad16 = |
2539 |
&frame->image, |
SEARCH16(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2540 |
i, j, |
&frame->image, i, j, frame->motion_flags, |
2541 |
frame->motion_flags, frame->quant, frame->fcode, |
frame->quant, frame->fcode, pParam, |
2542 |
pParam, |
f_mbs, f_mbs, /* todo */ |
|
f_mbs, f_mbs /* todo */, |
|
2543 |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
&mb->mvs[0], &pmv_dontcare); // ignore pmv |
2544 |
|
|
2545 |
// backward search |
// backward search |
2546 |
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
b_sad16 = SEARCH16(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
2547 |
&frame->image, |
&frame->image, i, j, frame->motion_flags, |
2548 |
i, j, |
frame->quant, frame->bcode, pParam, |
|
frame->motion_flags, frame->quant, frame->bcode, |
|
|
pParam, |
|
2549 |
b_mbs, b_mbs, /* todo */ |
b_mbs, b_mbs, /* todo */ |
2550 |
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
&mb->b_mvs[0], &pmv_dontcare); // ignore pmv |
2551 |
|
|
2552 |
// interpolate search (simple, but effective) |
// interpolate search (simple, but effective) |
2553 |
i_sad16 = sad16bi_c( |
i_sad16 = 65535; |
2554 |
frame->image.y + i*16 + j*16*edged_width, |
|
2555 |
|
/* |
2556 |
|
x/y range somewhat buggy |
2557 |
|
i_sad16 = |
2558 |
|
sad16bi_c(frame->image.y + i * 16 + j * 16 * edged_width, |
2559 |
get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
get_ref(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
2560 |
i, j, 16, mb->mvs[0].x, mb->mvs[0].y, edged_width), |
i, j, 16, mb->mvs[0].x, mb->mvs[0].y, |
2561 |
get_ref(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
edged_width), get_ref(b_ref->y, b_refH->y, |
2562 |
i, j, 16, mb->b_mvs[0].x, mb->b_mvs[0].x, edged_width), |
b_refV->y, b_refHV->y, |
2563 |
|
i, j, 16, |
2564 |
|
mb->b_mvs[0].x, |
2565 |
|
mb->b_mvs[0].x, |
2566 |
|
edged_width), |
2567 |
edged_width); |
edged_width); |
2568 |
|
*/ |
2569 |
|
|
2570 |
// TODO: direct search |
// TODO: direct search |
2571 |
// predictor + range of [-32,32] |
// predictor + range of [-32,32] |
2572 |
d_sad16 = 65535; |
d_sad16 = 65535; |
2573 |
|
|
2574 |
|
|
2575 |
if (f_sad16 < b_sad16) |
if (f_sad16 < b_sad16) { |
|
{ |
|
2576 |
best_sad = f_sad16; |
best_sad = f_sad16; |
2577 |
mb->mode = MODE_FORWARD; |
mb->mode = MODE_FORWARD; |
2578 |
} |
} else { |
|
else |
|
|
{ |
|
2579 |
best_sad = b_sad16; |
best_sad = b_sad16; |
2580 |
mb->mode = MODE_BACKWARD; |
mb->mode = MODE_BACKWARD; |
2581 |
} |
} |
2582 |
|
|
2583 |
if (i_sad16 < best_sad) |
if (i_sad16 < best_sad) { |
|
{ |
|
2584 |
best_sad = i_sad16; |
best_sad = i_sad16; |
2585 |
mb->mode = MODE_INTERPOLATE; |
mb->mode = MODE_INTERPOLATE; |
2586 |
} |
} |
2587 |
|
|
2588 |
if (d_sad16 < best_sad) |
if (d_sad16 < best_sad) { |
|
{ |
|
2589 |
best_sad = d_sad16; |
best_sad = d_sad16; |
2590 |
mb->mode = MODE_DIRECT; |
mb->mode = MODE_DIRECT; |
2591 |
} |
} |