52 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
53 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
54 |
|
|
|
#define GET_REFERENCE(X, Y, REF) { \ |
|
|
switch ( (((X)&1)<<1) + ((Y)&1) ) \ |
|
|
{ \ |
|
|
case 0 : REF = (uint8_t *)data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ |
|
|
case 1 : REF = (uint8_t *)data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ |
|
|
case 2 : REF = (uint8_t *)data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \ |
|
|
default : REF = (uint8_t *)data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \ |
|
|
} \ |
|
|
} |
|
|
|
|
55 |
#define iDiamondSize 2 |
#define iDiamondSize 2 |
56 |
|
|
57 |
|
static VECTOR |
58 |
|
GlobalMotionEst(const MACROBLOCK * const pMBs, |
59 |
|
const MBParam * const pParam, const uint32_t iFcode); |
60 |
|
|
61 |
|
|
62 |
static __inline int |
static __inline int |
63 |
d_mv_bits(int x, int y, const uint32_t iFcode) |
d_mv_bits(int x, int y, const uint32_t iFcode) |
64 |
{ |
{ |
121 |
return sad; |
return sad; |
122 |
} |
} |
123 |
|
|
124 |
|
static __inline const uint8_t * |
125 |
|
GetReference(const int x, const int y, const int dir, const SearchData * const data) |
126 |
|
{ |
127 |
|
// dir : 0 = forward, 1 = backward |
128 |
|
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
129 |
|
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
130 |
|
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
131 |
|
case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
132 |
|
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
133 |
|
case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth); |
134 |
|
case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
135 |
|
case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
136 |
|
default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
137 |
|
|
138 |
/* CHECK_CANDIATE FUNCTIONS START */ |
} |
139 |
|
} |
140 |
|
|
141 |
|
static uint8_t * |
142 |
|
Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data) |
143 |
|
{ |
144 |
|
// create or find a qpel-precision reference picture; return pointer to it |
145 |
|
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
146 |
|
const int32_t iEdgedWidth = data->iEdgedWidth; |
147 |
|
const uint32_t rounding = data->rounding; |
148 |
|
const int halfpel_x = x/2; |
149 |
|
const int halfpel_y = y/2; |
150 |
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
151 |
|
|
152 |
|
ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases |
153 |
|
ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
154 |
|
switch( ((x&1)<<1) + (y&1) ) { |
155 |
|
case 0: // pure halfpel position |
156 |
|
Reference = (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data); |
157 |
|
Reference += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
158 |
|
break; |
159 |
|
|
160 |
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
161 |
|
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
162 |
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
163 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
164 |
|
break; |
165 |
|
|
166 |
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
167 |
|
ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
168 |
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
169 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
170 |
|
break; |
171 |
|
|
172 |
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
173 |
|
// bottom left/right) during qpel refinement |
174 |
|
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
175 |
|
ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
176 |
|
ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data); |
177 |
|
ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
178 |
|
ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
179 |
|
ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth; |
180 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
181 |
|
break; |
182 |
|
} |
183 |
|
return Reference; |
184 |
|
} |
185 |
|
|
186 |
|
static uint8_t * |
187 |
|
Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data) |
188 |
|
{ |
189 |
|
// create or find a qpel-precision reference picture; return pointer to it |
190 |
|
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
191 |
|
const int32_t iEdgedWidth = data->iEdgedWidth; |
192 |
|
const uint32_t rounding = data->rounding; |
193 |
|
const int halfpel_x = x/2; |
194 |
|
const int halfpel_y = y/2; |
195 |
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
196 |
|
|
197 |
|
ref1 = GetReference(halfpel_x, halfpel_y, dir, data); // this reference is used in all cases |
198 |
|
switch( ((x&1)<<1) + (y&1) ) { |
199 |
|
case 0: // pure halfpel position |
200 |
|
return (uint8_t *) GetReference(halfpel_x, halfpel_y, dir, data); |
201 |
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
202 |
|
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
203 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
204 |
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
205 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
206 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
207 |
|
break; |
208 |
|
|
209 |
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
210 |
|
ref2 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
211 |
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
212 |
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding, 8); |
213 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding, 8); |
214 |
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
215 |
|
break; |
216 |
|
|
217 |
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
218 |
|
// bottom left/right) during qpel refinement |
219 |
|
ref2 = GetReference(halfpel_x, y - halfpel_y, dir, data); |
220 |
|
ref3 = GetReference(x - halfpel_x, halfpel_y, dir, data); |
221 |
|
ref4 = GetReference(x - halfpel_x, y - halfpel_y, dir, data); |
222 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
223 |
|
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
224 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
225 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
226 |
|
break; |
227 |
|
} |
228 |
|
return Reference; |
229 |
|
} |
230 |
|
|
231 |
|
/* CHECK_CANDIATE FUNCTIONS START */ |
232 |
|
|
233 |
static void |
static void |
234 |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
235 |
{ |
{ |
236 |
int t; |
int t, xc, yc; |
237 |
const uint8_t * Reference; |
const uint8_t * Reference; |
238 |
|
VECTOR * current; |
239 |
|
|
240 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
241 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
242 |
|
|
243 |
|
if (data->qpel_precision) { // x and y are in 1/4 precision |
244 |
|
Reference = Interpolate16x16qpel(x, y, 0, data); |
245 |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
246 |
|
xc = x/2; yc = y/2; //for chroma sad |
247 |
|
current = data->currentQMV; |
248 |
|
} else { |
249 |
switch ( ((x&1)<<1) + (y&1) ) { |
switch ( ((x&1)<<1) + (y&1) ) { |
250 |
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
251 |
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
252 |
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
253 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
254 |
} |
} |
255 |
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
256 |
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
257 |
|
current = data->currentMV; |
258 |
|
xc = x; yc = y; |
259 |
|
} |
260 |
|
|
261 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
262 |
|
|
|
if (data->qpel) t = d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode); |
|
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
|
|
|
|
263 |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
264 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
265 |
|
|
266 |
if (data->chroma) data->temp[0] += ChromaSAD(x, y, data); |
if (data->chroma) data->temp[0] += ChromaSAD(xc, yc, data); |
267 |
|
|
268 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (data->temp[0] < data->iMinSAD[0]) { |
269 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = data->temp[0]; |
270 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
current[0].x = x; current[0].y = y; |
271 |
*dir = Direction; } |
*dir = Direction; } |
272 |
|
|
273 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[1] < data->iMinSAD[1]) { |
274 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; } |
275 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[2] < data->iMinSAD[2]) { |
276 |
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; } |
277 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[3] < data->iMinSAD[3]) { |
278 |
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; } |
279 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[4] < data->iMinSAD[4]) { |
280 |
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; } |
281 |
|
|
282 |
} |
} |
283 |
|
|
286 |
{ |
{ |
287 |
int32_t sad; |
int32_t sad; |
288 |
const uint8_t * Reference; |
const uint8_t * Reference; |
289 |
|
int t; |
290 |
|
VECTOR * current; |
291 |
|
|
292 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
293 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
294 |
|
|
295 |
switch ( ((x&1)<<1) + (y&1) ) |
if (data->qpel_precision) { // x and y are in 1/4 precision |
296 |
{ |
Reference = Interpolate16x16qpel(x, y, 0, data); |
297 |
|
t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
298 |
|
current = data->currentQMV; |
299 |
|
} else { |
300 |
|
switch ( ((x&1)<<1) + (y&1) ) { |
301 |
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
302 |
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
303 |
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
304 |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
305 |
} |
} |
306 |
|
if (data->qpel) t = d_mv_bits(2*x - data->predMV.x, 2*y - data->predMV.y, data->iFcode); |
307 |
|
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
308 |
|
current = data->currentMV; |
309 |
|
} |
310 |
|
|
311 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
312 |
sad += (data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode) * sad)/1000; |
sad += (data->lambda16 * t * sad)/1000; |
313 |
|
|
314 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
315 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
316 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
current->x = x; current->y = y; |
317 |
*dir = Direction; } |
*dir = Direction; } |
318 |
} |
} |
319 |
|
|
320 |
static void |
static void |
|
CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
|
|
|
|
|
// CheckCandidate16 variant which expects x and y in quarter pixel resolution |
|
|
// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) |
|
|
// around currentMV! |
|
|
{ |
|
|
int t; |
|
|
uint8_t * Reference = (uint8_t *)data->RefQ; |
|
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
|
|
VECTOR halfpelMV = *(data->currentMV); |
|
|
|
|
|
int32_t iEdgedWidth = data->iEdgedWidth; |
|
|
uint32_t rounding = data->rounding; |
|
|
|
|
|
if (( x > data->max_dx) || ( x < data->min_dx) |
|
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|
|
|
|
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases |
|
|
switch( ((x&1)<<1) + (y&1) ) |
|
|
{ |
|
|
case 0: // pure halfpel position - shouldn't happen during a refinement step |
|
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); |
|
|
break; |
|
|
|
|
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
|
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
|
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); |
|
|
break; |
|
|
|
|
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
|
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); |
|
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding); |
|
|
break; |
|
|
|
|
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
|
|
// bottom left/right) during qpel refinement |
|
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
|
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); |
|
|
GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); |
|
|
|
|
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
|
|
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
|
|
break; |
|
|
} |
|
|
|
|
|
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1); |
|
|
|
|
|
t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode); |
|
|
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
|
|
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))/100; |
|
|
|
|
|
if (data->chroma) |
|
|
data->temp[0] += ChromaSAD(x/2, y/2, data); |
|
|
|
|
|
if (data->temp[0] < data->iMinSAD[0]) { |
|
|
data->iMinSAD[0] = data->temp[0]; |
|
|
data->currentQMV[0].x = x; data->currentQMV[0].y = y; |
|
|
/* *dir = Direction;*/ } |
|
|
|
|
|
if (data->temp[1] < data->iMinSAD[1]) { |
|
|
data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; } |
|
|
if (data->temp[2] < data->iMinSAD[2]) { |
|
|
data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; } |
|
|
if (data->temp[3] < data->iMinSAD[3]) { |
|
|
data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; } |
|
|
if (data->temp[4] < data->iMinSAD[4]) { |
|
|
data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; } |
|
|
} |
|
|
|
|
|
static void |
|
321 |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
322 |
{ |
{ |
323 |
|
// maximum speed - for P/B/I decision |
324 |
int32_t sad; |
int32_t sad; |
325 |
|
|
326 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
340 |
CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data) |
341 |
{ |
{ |
342 |
int32_t sad; |
int32_t sad; |
343 |
const int xb = data->currentMV[1].x; |
int xb, yb, t; |
|
const int yb = data->currentMV[1].y; |
|
344 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
345 |
|
VECTOR *current; |
346 |
|
|
347 |
if (( xf > data->max_dx) || ( xf < data->min_dx) |
if (( xf > data->max_dx) || ( xf < data->min_dx) |
348 |
|| ( yf > data->max_dy) || (yf < data->min_dy)) return; |
|| ( yf > data->max_dy) || (yf < data->min_dy)) return; |
349 |
|
|
350 |
switch ( ((xf&1)<<1) + (yf&1) ) { |
if (data->qpel_precision) { |
351 |
case 0 : ReferenceF = data->Ref + xf/2 + (yf/2)*(data->iEdgedWidth); break; |
ReferenceF = Interpolate16x16qpel(xf, yf, 0, data); |
352 |
case 1 : ReferenceF = data->RefV + xf/2 + ((yf-1)/2)*(data->iEdgedWidth); break; |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
353 |
case 2 : ReferenceF = data->RefH + (xf-1)/2 + (yf/2)*(data->iEdgedWidth); break; |
current = data->currentQMV; |
354 |
default : ReferenceF = data->RefHV + (xf-1)/2 + ((yf-1)/2)*(data->iEdgedWidth); break; |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
355 |
} |
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
356 |
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
357 |
switch ( ((xb&1)<<1) + (yb&1) ) { |
} else { |
358 |
case 0 : ReferenceB = data->bRef + xb/2 + (yb/2)*(data->iEdgedWidth); break; |
ReferenceF = Interpolate16x16qpel(2*xf, 2*yf, 0, data); |
359 |
case 1 : ReferenceB = data->bRefV + xb/2 + ((yb-1)/2)*(data->iEdgedWidth); break; |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
360 |
case 2 : ReferenceB = data->bRefH + (xb-1)/2 + (yb/2)*(data->iEdgedWidth); break; |
ReferenceB = Interpolate16x16qpel(2*xb, 2*yb, 1, data); |
361 |
default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break; |
current = data->currentMV; |
362 |
|
if (data->qpel) |
363 |
|
t = d_mv_bits(2*xf - data->predMV.x, 2*yf - data->predMV.y, data->iFcode) |
364 |
|
+ d_mv_bits(2*xb - data->bpredMV.x, 2*yb - data->bpredMV.y, data->iFcode); |
365 |
|
else |
366 |
|
t = d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) |
367 |
|
+ d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode); |
368 |
} |
} |
369 |
|
|
370 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
371 |
|
sad += (data->lambda16 * t * sad)/1000; |
|
sad += (data->lambda16 * |
|
|
( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) + |
|
|
d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode)) * sad)/1000; |
|
372 |
|
|
373 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
374 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
375 |
data->currentMV->x = xf; data->currentMV->y = yf; |
current->x = xf; current->y = yf; |
376 |
*dir = Direction; } |
*dir = Direction; } |
377 |
} |
} |
378 |
|
|
403 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
404 |
|| ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; |
|| ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; |
405 |
|
|
406 |
switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) { |
if (!data->qpel) { |
407 |
case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break; |
mvs.x *= 2; mvs.y *= 2; |
408 |
case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; |
b_mvs.x *= 2; b_mvs.y *= 2; //we move to qpel precision anyway |
|
case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break; |
|
|
default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
|
|
|
switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) { |
|
|
case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; |
|
|
default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
409 |
} |
} |
410 |
|
ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); |
411 |
|
ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); |
412 |
|
|
413 |
sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
414 |
ReferenceF + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
ReferenceF, ReferenceB, |
|
ReferenceB + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
|
415 |
data->iEdgedWidth); |
data->iEdgedWidth); |
416 |
if (sad > *(data->iMinSAD)) return; |
if (sad > *(data->iMinSAD)) return; |
417 |
} |
} |
449 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
450 |
|| ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; |
|| ( b_mvs.y > data->max_dy ) || ( b_mvs.y < data->min_dy )) return; |
451 |
|
|
452 |
switch ( ((mvs.x&1)<<1) + (mvs.y&1) ) { |
if (!data->qpel) { |
453 |
case 0 : ReferenceF = data->Ref + mvs.x/2 + (mvs.y/2)*(data->iEdgedWidth); break; |
mvs.x *= 2; mvs.y *= 2; |
454 |
case 1 : ReferenceF = data->RefV + mvs.x/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; |
b_mvs.x *= 2; b_mvs.y *= 2; //we move to qpel precision anyway |
|
case 2 : ReferenceF = data->RefH + (mvs.x-1)/2 + (mvs.y/2)*(data->iEdgedWidth); break; |
|
|
default : ReferenceF = data->RefHV + (mvs.x-1)/2 + ((mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
|
|
|
|
switch ( ((b_mvs.x&1)<<1) + (b_mvs.y&1) ) { |
|
|
case 0 : ReferenceB = data->bRef + b_mvs.x/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : ReferenceB = data->bRefV + b_mvs.x/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : ReferenceB = data->bRefH + (b_mvs.x-1)/2 + (b_mvs.y/2)*(data->iEdgedWidth); break; |
|
|
default : ReferenceB = data->bRefHV + (b_mvs.x-1)/2 + ((b_mvs.y-1)/2)*(data->iEdgedWidth); break; |
|
455 |
} |
} |
456 |
|
ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data); |
457 |
|
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
458 |
|
|
459 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
460 |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, 1) * sad)/1000; |
474 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if (( x > data->max_dx) || ( x < data->min_dx) |
475 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
476 |
|
|
477 |
switch ( ((x&1)<<1) + (y&1) ) |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
478 |
{ |
else Reference = Interpolate16x16qpel(2*x, 2*y, 0, data); |
|
case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break; |
|
|
default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break; |
|
|
} |
|
479 |
|
|
480 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
481 |
if (data->qpel) t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode); |
if (data->qpel) t = d_mv_bits(2 * x - data->predMV.x, 2 * y - data->predMV.y, data->iFcode); |
482 |
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
else t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode); |
483 |
|
|
484 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
489 |
*dir = Direction; } |
*dir = Direction; } |
490 |
} |
} |
491 |
|
|
|
static void |
|
|
CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
|
|
// CheckCandidate16no4v variant which expects x and y in quarter pixel resolution |
|
|
// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!) |
|
|
// around currentMV! |
|
|
|
|
|
{ |
|
|
int32_t sad; |
|
|
uint8_t *Reference = (uint8_t *) data->RefQ; |
|
|
const uint8_t *ref1, *ref2, *ref3, *ref4; |
|
|
VECTOR halfpelMV = *(data->currentMV); |
|
|
|
|
|
int32_t iEdgedWidth = data->iEdgedWidth; |
|
|
uint32_t rounding = data->rounding; |
|
|
|
|
|
if (( x > data->max_dx) || ( x < data->min_dx) |
|
|
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
|
|
|
|
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); |
|
|
switch( ((x&1)<<1) + (y&1) ) |
|
|
{ |
|
|
case 0: // pure halfpel position - shouldn't happen during a refinement step |
|
|
GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); |
|
|
break; |
|
|
|
|
|
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
|
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
|
|
|
|
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
|
|
break; |
|
|
|
|
|
case 2: // x qpel, y halfpel - left or right during qpel refinement |
|
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2); |
|
|
|
|
|
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding); |
|
|
break; |
|
|
|
|
|
default: // x and y in qpel resolution - the "corners" (top left/right and |
|
|
// bottom left/right) during qpel refinement |
|
|
GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2); |
|
|
GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3); |
|
|
GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4); |
|
|
|
|
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
|
|
break; |
|
|
} |
|
|
|
|
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
|
|
sad += (data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode) * (sad+NEIGH_8X8_BIAS))/100; |
|
|
|
|
|
if (sad < *(data->iMinSAD)) { |
|
|
*(data->iMinSAD) = sad; |
|
|
data->currentQMV->x = x; data->currentQMV->y = y; |
|
|
*dir = Direction; } |
|
|
} |
|
|
|
|
492 |
/* CHECK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
493 |
|
|
494 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
638 |
/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */ |
/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */ |
639 |
|
|
640 |
static void |
static void |
641 |
HalfpelRefine(const SearchData * const data) |
SubpelRefine(const SearchData * const data) |
642 |
{ |
{ |
643 |
/* Do a half-pel refinement (or rather a "smallest possible amount" refinement) */ |
/* Do a half-pel or q-pel refinement */ |
644 |
|
VECTOR backupMV; |
|
VECTOR backupMV = *(data->currentMV); |
|
645 |
int iDirection; //not needed |
int iDirection; //not needed |
646 |
|
|
647 |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
if (data->qpel_precision) |
648 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
backupMV = *(data->currentQMV); |
649 |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); |
else backupMV = *(data->currentMV); |
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); |
|
|
|
|
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
|
|
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); |
|
|
|
|
|
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
|
|
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
|
|
} |
|
|
|
|
|
|
|
|
static void |
|
|
QuarterpelRefine(const SearchData * const data) |
|
|
{ |
|
|
/* Perform quarter pixel refinement*/ |
|
|
|
|
|
VECTOR backupMV = *(data->currentQMV); |
|
|
int iDirection; //not needed |
|
650 |
|
|
651 |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
652 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
658 |
|
|
659 |
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
660 |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
|
|
|
661 |
} |
} |
662 |
|
|
663 |
static __inline int |
static __inline int |
706 |
const IMAGE *const pCurrent = ¤t->image; |
const IMAGE *const pCurrent = ¤t->image; |
707 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
708 |
|
|
|
FILE * debug; |
|
709 |
const VECTOR zeroMV = { 0, 0 }; |
const VECTOR zeroMV = { 0, 0 }; |
710 |
|
|
711 |
uint32_t x, y; |
uint32_t x, y; |
795 |
/* finally, intra decision */ |
/* finally, intra decision */ |
796 |
|
|
797 |
InterBias = MV16_INTER_BIAS; |
InterBias = MV16_INTER_BIAS; |
798 |
if (pMB->quant > 8) InterBias += 80 * (pMB->quant - 8); // to make high quants work |
if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work |
799 |
if (y != 0) |
if (y != 0) |
800 |
if ((pMB - pParam->mb_width)->mode == MODE_INTER ) InterBias -= 80; |
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
801 |
if (x != 0) |
if (x != 0) |
802 |
if ((pMB - 1)->mode == MODE_INTER ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
803 |
|
|
804 |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
805 |
|
|
907 |
|
|
908 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); //has to be changed to get_pmv(2)() |
909 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
910 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel); |
|
|
|
|
Data->predMV = pmv[0]; |
|
911 |
|
|
912 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
913 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
922 |
|
|
923 |
Data->lambda16 = lambda_vec16[iQuant]; |
Data->lambda16 = lambda_vec16[iQuant]; |
924 |
Data->lambda8 = lambda_vec8[iQuant]; |
Data->lambda8 = lambda_vec8[iQuant]; |
925 |
|
Data->qpel_precision = 0; |
926 |
|
|
927 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
928 |
Data->min_dx = EVEN(Data->min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
935 |
for(i = 0; i < 5; i++) |
for(i = 0; i < 5; i++) |
936 |
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
937 |
|
|
938 |
if (pParam->m_quarterpel) { |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
939 |
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
else Data->predMV = pmv[0]; |
|
i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode); |
|
|
} else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
|
940 |
|
|
941 |
|
i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode); |
942 |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
943 |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
Data->iMinSAD[1] = pMB->sad8[0] + (Data->lambda8 * i * (pMB->sad8[0]+NEIGH_8X8_BIAS))/100; |
944 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
954 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
955 |
prevMBs + x + y * pParam->mb_width); |
prevMBs + x + y * pParam->mb_width); |
956 |
|
|
957 |
if (inter4v || pParam->m_quarterpel || Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
958 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
959 |
|
|
960 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
961 |
|
|
1012 |
} |
} |
1013 |
} |
} |
1014 |
|
|
1015 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data); |
1016 |
|
|
1017 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1018 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
1021 |
|
|
1022 |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1023 |
|
|
1024 |
CheckCandidate = CheckCandidate16_qpel; |
Data->qpel_precision = 1; |
1025 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range_qpel(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1026 |
pParam->width, pParam->height, Data->iFcode, 0); |
pParam->width, pParam->height, Data->iFcode); |
1027 |
|
|
1028 |
QuarterpelRefine(Data); |
SubpelRefine(Data); |
1029 |
} |
} |
1030 |
|
|
1031 |
if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0; |
if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0; |
1042 |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1043 |
|
|
1044 |
if (Data->chroma) { |
if (Data->chroma) { |
1045 |
int sum, dx, dy; |
int sumx, sumy, dx, dy; |
|
|
|
|
if(pParam->m_quarterpel) { |
|
|
sum = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
|
|
} else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
|
1046 |
|
|
1047 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1048 |
sum = pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
1049 |
} else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
1050 |
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
} else { |
1051 |
|
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1052 |
|
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1053 |
|
} |
1054 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
1055 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
1056 |
|
|
1057 |
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
1058 |
} |
} |
1066 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] |
1067 |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1068 |
|
|
|
pMB->qmvs[0] = pMB->qmvs[1] |
|
|
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
|
|
|
|
1069 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
1070 |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1071 |
|
|
1072 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1073 |
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x; |
pMB->qmvs[0] = pMB->qmvs[1] |
1074 |
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y; |
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1075 |
|
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
1076 |
|
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
1077 |
} else { |
} else { |
1078 |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1079 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1101 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
1102 |
|
|
1103 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1104 |
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1105 |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
if (block != 0) *(Data->iMinSAD) += (Data->lambda8 * |
1106 |
d_mv_bits( Data->currentQMV->x - Data->predQMV.x, |
d_mv_bits( Data->currentQMV->x - Data->predMV.x, |
1107 |
Data->currentQMV->y - Data->predQMV.y, |
Data->currentQMV->y - Data->predMV.y, |
1108 |
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
Data->iFcode) * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
1109 |
} else { |
} else { |
1110 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block); |
1122 |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1123 |
|
|
1124 |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1)); |
1125 |
|
Data->qpel_precision = 0; |
1126 |
|
|
1127 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1128 |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, OldData->iFcode - pParam->m_quarterpel); |
1129 |
CheckCandidate = CheckCandidate8; |
CheckCandidate = CheckCandidate8; |
1130 |
|
|
1131 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1147 |
if (MotionFlags & PMV_HALFPELREFINE8) { |
if (MotionFlags & PMV_HALFPELREFINE8) { |
1148 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1149 |
|
|
1150 |
HalfpelRefine(Data); // perform halfpel refine of current best vector |
SubpelRefine(Data); // perform halfpel refine of current best vector |
1151 |
|
|
1152 |
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
if(*(Data->iMinSAD) < temp_sad) { // we have found a better match |
1153 |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1158 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1159 |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
1160 |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
1161 |
CheckCandidate = CheckCandidate8_qpel; |
Data->qpel_precision = 1; |
1162 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range_qpel(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
1163 |
pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, OldData->iFcode); |
1164 |
QuarterpelRefine(Data); |
SubpelRefine(Data); |
1165 |
} |
} |
1166 |
} |
} |
1167 |
} |
} |
1168 |
|
|
1169 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1170 |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x; |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
1171 |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y; |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
1172 |
|
pMB->qmvs[block] = *(Data->currentQMV); |
1173 |
} |
} |
1174 |
else { |
else { |
1175 |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1177 |
} |
} |
1178 |
|
|
1179 |
pMB->mvs[block] = *(Data->currentMV); |
pMB->mvs[block] = *(Data->currentMV); |
|
pMB->qmvs[block] = *(Data->currentQMV); |
|
|
|
|
1180 |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
1181 |
} |
} |
1182 |
|
|
1254 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1255 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
1256 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
1257 |
|
Data->qpel_precision = 0; |
1258 |
|
|
1259 |
Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
1260 |
Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; |
Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; |
1264 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1265 |
|
|
1266 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1267 |
pParam->width, pParam->height, iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, iFcode - pParam->m_quarterpel); |
1268 |
|
|
1269 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1270 |
|
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
1271 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
1272 |
|
|
1273 |
Data->currentMV->x = Data->currentMV->y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
|
|
|
1274 |
CheckCandidate = CheckCandidate16no4v; |
CheckCandidate = CheckCandidate16no4v; |
1275 |
|
|
1276 |
// main loop. checking all predictions |
// main loop. checking all predictions |
1287 |
|
|
1288 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1289 |
|
|
1290 |
HalfpelRefine(Data); |
SubpelRefine(Data); |
1291 |
|
|
1292 |
|
if (Data->qpel) { |
1293 |
|
Data->currentQMV->x = 2*Data->currentMV->x; |
1294 |
|
Data->currentQMV->y = 2*Data->currentMV->y; |
1295 |
|
Data->qpel_precision = 1; |
1296 |
|
get_range_qpel(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1297 |
|
pParam->width, pParam->height, iFcode); |
1298 |
|
SubpelRefine(Data); |
1299 |
|
} |
1300 |
|
|
1301 |
// three bits are needed to code backward mode. four for forward |
// three bits are needed to code backward mode. four for forward |
1302 |
// we treat the bits just like they were vector's |
// we treat the bits just like they were vector's |
1306 |
if (*Data->iMinSAD < *best_sad) { |
if (*Data->iMinSAD < *best_sad) { |
1307 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1308 |
pMB->mode = mode_current; |
pMB->mode = mode_current; |
1309 |
|
if (Data->qpel) { |
1310 |
|
pMB->pmvs[0].x = Data->currentQMV->x - predMV->x; |
1311 |
|
pMB->pmvs[0].y = Data->currentQMV->y - predMV->y; |
1312 |
|
if (mode_current == MODE_FORWARD) |
1313 |
|
pMB->qmvs[0] = *Data->currentQMV; |
1314 |
|
else |
1315 |
|
pMB->b_qmvs[0] = *Data->currentQMV; |
1316 |
|
} else { |
1317 |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
pMB->pmvs[0].x = Data->currentMV->x - predMV->x; |
1318 |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
pMB->pmvs[0].y = Data->currentMV->y - predMV->y; |
1319 |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV; |
} |
1320 |
else pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
if (mode_current == MODE_FORWARD) |
1321 |
|
pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV; |
1322 |
|
else |
1323 |
|
pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
1324 |
|
|
1325 |
} |
} |
1326 |
|
|
1327 |
} |
} |
1352 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1353 |
|
|
1354 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
|
Data->referencemv = b_mb->mvs; |
|
1355 |
|
|
1356 |
Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
1357 |
Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; |
Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; |
1366 |
Data->max_dy = 2 * pParam->height - 2 * (y) * 16; |
Data->max_dy = 2 * pParam->height - 2 * (y) * 16; |
1367 |
Data->min_dx = -(2 * 16 + 2 * (x) * 16); |
Data->min_dx = -(2 * 16 + 2 * (x) * 16); |
1368 |
Data->min_dy = -(2 * 16 + 2 * (y) * 16); |
Data->min_dy = -(2 * 16 + 2 * (y) * 16); |
1369 |
|
if (Data->qpel) { //we measure in qpixels |
1370 |
|
Data->max_dx *= 2; |
1371 |
|
Data->max_dy *= 2; |
1372 |
|
Data->min_dx *= 2; |
1373 |
|
Data->min_dy *= 2; |
1374 |
|
Data->referencemv = b_mb->qmvs; |
1375 |
|
} else Data->referencemv = b_mb->mvs; |
1376 |
|
Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here |
1377 |
|
|
1378 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1379 |
pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD); |
1398 |
} |
} |
1399 |
} |
} |
1400 |
|
|
1401 |
if (b_mb->mode == MODE_INTER4V) |
|
1402 |
CheckCandidate = CheckCandidateDirect; |
if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect; |
1403 |
else CheckCandidate = CheckCandidateDirectno4v; |
else CheckCandidate = CheckCandidateDirectno4v; |
1404 |
|
|
1405 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1410 |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
//this is not full chroma compensation, only it's fullpel approximation. should work though |
1411 |
int sum, dx, dy, b_dx, b_dy; |
int sum, dx, dy, b_dx, b_dy; |
1412 |
|
|
1413 |
|
if (Data->qpel) { |
1414 |
|
sum = pMB->mvs[0].y/2 + pMB->mvs[1].y/2 + pMB->mvs[2].y/2 + pMB->mvs[3].y/2; |
1415 |
|
dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
1416 |
|
sum = pMB->mvs[0].x/2 + pMB->mvs[1].x/2 + pMB->mvs[2].x/2 + pMB->mvs[3].x/2; |
1417 |
|
dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
1418 |
|
|
1419 |
|
sum = pMB->b_mvs[0].y/2 + pMB->b_mvs[1].y/2 + pMB->b_mvs[2].y/2 + pMB->b_mvs[3].y/2; |
1420 |
|
b_dy = (sum >> 3) + roundtab_76[sum & 0xf]; |
1421 |
|
sum = pMB->b_mvs[0].x/2 + pMB->b_mvs[1].x/2 + pMB->b_mvs[2].x/2 + pMB->b_mvs[3].x/2; |
1422 |
|
b_dx = (sum >> 3) + roundtab_76[sum & 0xf]; |
1423 |
|
|
1424 |
|
} else { |
1425 |
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1426 |
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
1427 |
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1428 |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1429 |
|
|
1430 |
sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; |
1431 |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
|
|
|
1432 |
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; |
1433 |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); |
1434 |
|
} |
1435 |
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
1436 |
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
1437 |
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
1458 |
|
|
1459 |
(*MainSearchPtr)(0, 0, Data, 255); |
(*MainSearchPtr)(0, 0, Data, 255); |
1460 |
|
|
1461 |
HalfpelRefine(Data); |
SubpelRefine(Data); |
1462 |
|
|
1463 |
*Data->iMinSAD += 1 * Data->lambda16; // one bit is needed to code direct mode |
// *Data->iMinSAD += 1 * Data->lambda16; // one bit is needed to code direct mode |
1464 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1465 |
|
|
1466 |
if (b_mb->mode == MODE_INTER4V) |
if (b_mb->mode == MODE_INTER4V) |
1478 |
pMB->b_mvs[k].y = ((Data->currentMV->y == 0) |
pMB->b_mvs[k].y = ((Data->currentMV->y == 0) |
1479 |
? Data->directmvB[k].y |
? Data->directmvB[k].y |
1480 |
: pMB->mvs[k].y - Data->referencemv[k].y); |
: pMB->mvs[k].y - Data->referencemv[k].y); |
1481 |
|
if (Data->qpel) { |
1482 |
|
pMB->qmvs[k].x = pMB->mvs[k].x; pMB->mvs[k].x /= 2; |
1483 |
|
pMB->b_qmvs[k].x = pMB->b_mvs[k].x; pMB->b_mvs[k].x /= 2; |
1484 |
|
pMB->qmvs[k].y = pMB->mvs[k].y; pMB->mvs[k].y /= 2; |
1485 |
|
pMB->b_qmvs[k].y = pMB->b_mvs[k].y; pMB->b_mvs[k].y /= 2; |
1486 |
|
} |
1487 |
|
|
1488 |
if (b_mb->mode != MODE_INTER4V) { |
if (b_mb->mode != MODE_INTER4V) { |
1489 |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0]; |
1490 |
pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; |
pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0]; |
1491 |
|
pMB->qmvs[3] = pMB->qmvs[2] = pMB->qmvs[1] = pMB->qmvs[0]; |
1492 |
|
pMB->b_qmvs[3] = pMB->b_qmvs[2] = pMB->b_qmvs[1] = pMB->b_qmvs[0]; |
1493 |
break; |
break; |
1494 |
} |
} |
1495 |
} |
} |
1521 |
{ |
{ |
1522 |
|
|
1523 |
const int32_t iEdgedWidth = pParam->edged_width; |
const int32_t iEdgedWidth = pParam->edged_width; |
|
|
|
1524 |
int iDirection, i, j; |
int iDirection, i, j; |
1525 |
SearchData bData; |
SearchData bData; |
1526 |
|
|
1527 |
*(bData.iMinSAD = fData->iMinSAD) = 4096*256; |
*(bData.iMinSAD = fData->iMinSAD) = 4096*256; |
1528 |
bData.Cur = fData->Cur; |
bData.Cur = fData->Cur; |
1529 |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth; |
1530 |
bData.currentMV = fData->currentMV + 1; |
bData.currentMV = fData->currentMV + 1; bData.currentQMV = fData->currentQMV + 1; |
1531 |
bData.lambda16 = fData->lambda16; |
bData.lambda16 = fData->lambda16; |
1532 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1533 |
|
|
1539 |
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16; |
1540 |
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16; |
1541 |
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16; |
1542 |
|
bData.RefQ = fData->RefQ; |
1543 |
|
fData->qpel_precision = bData.qpel_precision = 0; bData.qpel = fData->qpel; |
1544 |
|
bData.rounding = 0; |
1545 |
|
|
1546 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1547 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
1548 |
|
|
1549 |
fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed |
fData->currentMV[0] = fData->currentMV[2]; |
1550 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - pParam->m_quarterpel); |
1551 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - pParam->m_quarterpel); |
1552 |
|
|
1553 |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1554 |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
1555 |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx; |
if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dy; |
1556 |
if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
if (fData->currentMV[0].y < fData->min_dy) fData->currentMV[0].y = fData->min_dy; |
1557 |
|
|
1558 |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx; |
1559 |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy; |
if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dx; |
1560 |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
1561 |
if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
1562 |
|
|
1563 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1564 |
|
|
1577 |
// backward MV moves |
// backward MV moves |
1578 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
1579 |
fData->currentMV[2] = fData->currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
|
|
|
1580 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
1581 |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
1582 |
CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); |
1584 |
|
|
1585 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1586 |
|
|
1587 |
|
if (fData->qpel) { |
1588 |
|
CheckCandidate = CheckCandidateInt; |
1589 |
|
fData->qpel_precision = bData.qpel_precision = 1; |
1590 |
|
get_range_qpel(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode); |
1591 |
|
get_range_qpel(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode); |
1592 |
|
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
1593 |
|
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
1594 |
|
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
1595 |
|
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
1596 |
|
SubpelRefine(fData); |
1597 |
|
fData->currentQMV[2] = fData->currentQMV[0]; |
1598 |
|
SubpelRefine(&bData); |
1599 |
|
} |
1600 |
|
|
1601 |
*fData->iMinSAD += 2 * fData->lambda16; // two bits are needed to code interpolate mode. |
*fData->iMinSAD += 2 * fData->lambda16; // two bits are needed to code interpolate mode. |
1602 |
|
|
1603 |
if (*fData->iMinSAD < *best_sad) { |
if (*fData->iMinSAD < *best_sad) { |
1605 |
pMB->mvs[0] = fData->currentMV[0]; |
pMB->mvs[0] = fData->currentMV[0]; |
1606 |
pMB->b_mvs[0] = fData->currentMV[1]; |
pMB->b_mvs[0] = fData->currentMV[1]; |
1607 |
pMB->mode = MODE_INTERPOLATE; |
pMB->mode = MODE_INTERPOLATE; |
1608 |
|
if (fData->qpel) { |
1609 |
|
pMB->qmvs[0] = fData->currentQMV[0]; |
1610 |
|
pMB->b_qmvs[0] = fData->currentQMV[1]; |
1611 |
|
pMB->pmvs[1].x = pMB->qmvs[0].x - f_predMV->x; |
1612 |
|
pMB->pmvs[1].y = pMB->qmvs[0].y - f_predMV->y; |
1613 |
|
pMB->pmvs[0].x = pMB->b_qmvs[0].x - b_predMV->x; |
1614 |
|
pMB->pmvs[0].y = pMB->b_qmvs[0].y - b_predMV->y; |
1615 |
|
} else { |
1616 |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x; |
1617 |
pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y; |
pMB->pmvs[1].y = pMB->mvs[0].y - f_predMV->y; |
1618 |
pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x; |
pMB->pmvs[0].x = pMB->b_mvs[0].x - b_predMV->x; |
1619 |
pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y; |
pMB->pmvs[0].y = pMB->b_mvs[0].y - b_predMV->y; |
1620 |
} |
} |
1621 |
} |
} |
1622 |
|
} |
1623 |
|
|
1624 |
void |
void |
1625 |
MotionEstimationBVOP(MBParam * const pParam, |
MotionEstimationBVOP(MBParam * const pParam, |
1633 |
const IMAGE * const f_refV, |
const IMAGE * const f_refV, |
1634 |
const IMAGE * const f_refHV, |
const IMAGE * const f_refHV, |
1635 |
// backward (future) reference |
// backward (future) reference |
1636 |
const MACROBLOCK * const b_mbs, |
const FRAMEINFO * const b_reference, |
1637 |
const IMAGE * const b_ref, |
const IMAGE * const b_ref, |
1638 |
const IMAGE * const b_refH, |
const IMAGE * const b_refH, |
1639 |
const IMAGE * const b_refV, |
const IMAGE * const b_refV, |
1643 |
int32_t best_sad, skip_sad; |
int32_t best_sad, skip_sad; |
1644 |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
int f_count = 0, b_count = 0, i_count = 0, d_count = 0, n_count = 0; |
1645 |
static const VECTOR zeroMV={0,0}; |
static const VECTOR zeroMV={0,0}; |
1646 |
|
const MACROBLOCK * const b_mbs = b_reference->mbs; |
1647 |
|
|
1648 |
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
1649 |
|
|
1650 |
const int32_t TRB = time_pp - time_bp; |
const int32_t TRB = time_pp - time_bp; |
1651 |
const int32_t TRD = time_pp; |
const int32_t TRD = time_pp; |
1652 |
|
uint8_t * qimage; |
1653 |
|
|
1654 |
// some pre-inintialized data for the rest of the search |
// some pre-inintialized data for the rest of the search |
1655 |
|
|
1656 |
SearchData Data; |
SearchData Data; |
1657 |
int32_t iMinSAD; |
int32_t iMinSAD; |
1658 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
1659 |
|
VECTOR currentQMV[3]; |
1660 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
1661 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
1662 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
1663 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant] + 2; |
1664 |
|
Data.qpel = pParam->m_quarterpel; |
1665 |
|
Data.rounding = 0; |
1666 |
|
|
1667 |
// note: i==horizontal, j==vertical |
if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL) |
1668 |
|
return; // allocate some mem for qpel interpolated blocks |
1669 |
|
// somehow this is dirty since I think we shouldn't use malloc outside |
1670 |
|
// encoder_create() - so please fix me! |
1671 |
|
Data.RefQ = qimage; |
1672 |
|
|
1673 |
|
// note: i==horizontal, j==vertical |
1674 |
for (j = 0; j < pParam->mb_height; j++) { |
for (j = 0; j < pParam->mb_height; j++) { |
1675 |
|
|
1676 |
f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ |
f_predMV = b_predMV = zeroMV; /* prediction is reset at left boundary */ |
1680 |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
const MACROBLOCK * const b_mb = b_mbs + i + j * pParam->mb_width; |
1681 |
|
|
1682 |
/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */ |
/* special case, if collocated block is SKIPed in P-VOP: encoding is forward (0,0), cpb=0 without further ado */ |
1683 |
|
if (b_reference->coding_type != S_VOP) |
1684 |
if (b_mb->mode == MODE_NOT_CODED) { |
if (b_mb->mode == MODE_NOT_CODED) { |
1685 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
1686 |
continue; |
continue; |
1736 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1737 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1738 |
f_count++; |
f_count++; |
1739 |
f_predMV = pMB->mvs[0]; |
if (pParam->m_quarterpel) f_predMV = pMB->qmvs[0]; |
1740 |
|
else f_predMV = pMB->mvs[0]; |
1741 |
break; |
break; |
1742 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
1743 |
b_count++; |
b_count++; |
1744 |
b_predMV = pMB->b_mvs[0]; |
if (pParam->m_quarterpel) b_predMV = pMB->b_qmvs[0]; |
1745 |
|
else b_predMV = pMB->b_mvs[0]; |
1746 |
break; |
break; |
1747 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
1748 |
i_count++; |
i_count++; |
1749 |
|
if (pParam->m_quarterpel) { |
1750 |
|
f_predMV = pMB->qmvs[0]; |
1751 |
|
b_predMV = pMB->b_qmvs[0]; |
1752 |
|
} else { |
1753 |
f_predMV = pMB->mvs[0]; |
f_predMV = pMB->mvs[0]; |
1754 |
b_predMV = pMB->b_mvs[0]; |
b_predMV = pMB->b_mvs[0]; |
1755 |
|
} |
1756 |
break; |
break; |
1757 |
case MODE_DIRECT: |
case MODE_DIRECT: |
1758 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
1759 |
d_count++; |
d_count++; |
|
break; |
|
1760 |
default: |
default: |
1761 |
break; |
break; |
1762 |
} |
} |
1763 |
} |
} |
1764 |
} |
} |
1765 |
|
free(qimage); |
1766 |
} |
} |
1767 |
|
|
1768 |
/* Hinted ME starts here */ |
/* Hinted ME starts here */ |
1787 |
int i, t; |
int i, t; |
1788 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1789 |
|
|
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
|
Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
1790 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1791 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel); |
1792 |
|
|
1793 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16; |
1794 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
1800 |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16; |
1801 |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8; |
1802 |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8; |
1803 |
|
Data->qpel_precision = 0; |
1804 |
|
|
1805 |
if (!(MotionFlags & PMV_HALFPEL16)) { |
if (!(MotionFlags & PMV_HALFPEL16)) { |
1806 |
Data->min_dx = EVEN(Data->min_dx); |
Data->min_dx = EVEN(Data->min_dx); |
1808 |
Data->min_dy = EVEN(Data->min_dy); |
Data->min_dy = EVEN(Data->min_dy); |
1809 |
Data->max_dy = EVEN(Data->max_dy); |
Data->max_dy = EVEN(Data->max_dy); |
1810 |
} |
} |
1811 |
|
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1812 |
|
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
1813 |
|
|
1814 |
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR; |
1815 |
|
|
1816 |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
if (pMB->dquant != NO_CHANGE) inter4v = 0; |
1817 |
|
|
1818 |
if (inter4v || pParam->m_quarterpel || Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
1819 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1820 |
|
|
1821 |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
pMB->mvs[0].x = EVEN(pMB->mvs[0].x); |
1843 |
|
|
1844 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1845 |
|
|
1846 |
if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data); |
1847 |
|
|
1848 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1849 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
1851 |
} |
} |
1852 |
|
|
1853 |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
1854 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range_qpel(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
1855 |
pParam->width, pParam->height, Data->iFcode, 0); |
pParam->width, pParam->height, Data->iFcode); |
1856 |
CheckCandidate = CheckCandidate16_qpel; |
Data->qpel_precision = 1; |
1857 |
QuarterpelRefine(Data); |
SubpelRefine(Data); |
1858 |
} |
} |
1859 |
|
|
1860 |
if (inter4v) { |
if (inter4v) { |
1870 |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8); |
1871 |
|
|
1872 |
if (Data->chroma) { |
if (Data->chroma) { |
1873 |
int sum, dx, dy; |
int sumx, sumy, dx, dy; |
1874 |
|
|
1875 |
|
if(pParam->m_quarterpel) { |
1876 |
|
sumx= pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
1877 |
|
sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
1878 |
|
} else { |
1879 |
|
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
1880 |
|
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
1881 |
|
} |
1882 |
|
dx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
1883 |
|
dy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
1884 |
|
|
|
if(pParam->m_quarterpel) |
|
|
sum = (pMB->qmvs[0].y + pMB->qmvs[1].y + pMB->qmvs[2].y + pMB->qmvs[3].y)/2; |
|
|
else sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
|
dy = (sum ? SIGN(sum) * |
|
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
|
|
|
|
if(pParam->m_quarterpel) |
|
|
sum = (pMB->qmvs[0].x + pMB->qmvs[1].x + pMB->qmvs[2].x + pMB->qmvs[3].x)/2; |
|
|
else sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
dx = (sum ? SIGN(sum) * |
|
|
(roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0); |
|
1885 |
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
Data->iMinSAD[1] += ChromaSAD(dx, dy, Data); |
1886 |
} |
} |
1887 |
} |
} |
1901 |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1902 |
|
|
1903 |
if(pParam->m_quarterpel) { |
if(pParam->m_quarterpel) { |
1904 |
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x; |
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
1905 |
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.y; |
pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y; |
1906 |
} else { |
} else { |
1907 |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1908 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1996 |
|
|
1997 |
int i = 255, mask; |
int i = 255, mask; |
1998 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
|
|
|
1999 |
*(Data->iMinSAD) = MV_MAX_ERROR; |
*(Data->iMinSAD) = MV_MAX_ERROR; |
2000 |
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
|
2001 |
|
//median is only used as prediction. it doesn't have to be real |
2002 |
|
if (x == 1 && y == 1) Data->predMV.x = Data->predMV.y = 0; |
2003 |
|
else |
2004 |
|
if (x == 1) //left macroblock does not have any vector now |
2005 |
|
Data->predMV = (pMB - pParam->mb_width)->mvs[0]; // top instead of median |
2006 |
|
else if (y == 1) // top macroblock don't have it's vector |
2007 |
|
Data->predMV = (pMB - 1)->mvs[0]; // left instead of median |
2008 |
|
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median |
2009 |
|
|
2010 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2011 |
pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel); |
pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel); |
2012 |
|
|
2013 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2014 |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
Data->Ref = pRef + (x + y * pParam->edged_width) * 16; |
2015 |
|
|
|
CheckCandidate = CheckCandidate16no4vI; |
|
|
|
|
2016 |
pmv[1].x = EVEN(pMB->mvs[0].x); |
pmv[1].x = EVEN(pMB->mvs[0].x); |
2017 |
pmv[1].y = EVEN(pMB->mvs[0].y); |
pmv[1].y = EVEN(pMB->mvs[0].y); |
2018 |
pmv[0].x = EVEN(Data->predMV.x); |
pmv[2].x = EVEN(Data->predMV.x); |
2019 |
pmv[0].y = EVEN(Data->predMV.y); |
pmv[2].y = EVEN(Data->predMV.y); |
2020 |
pmv[2].x = pmv[2].y = 0; |
pmv[0].x = pmv[0].y = 0; |
2021 |
|
|
2022 |
|
(*CheckCandidate)(0, 0, 255, &i, Data); |
2023 |
|
|
2024 |
|
//early skip for 0,0 |
2025 |
|
if (*Data->iMinSAD < MAX_SAD00_FOR_SKIP * 4) { |
2026 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
2027 |
|
pMB->mode = MODE_NOT_CODED; |
2028 |
|
return 0; |
2029 |
|
} |
2030 |
|
|
|
CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data); |
|
2031 |
if (!(mask = make_mask(pmv, 1))) |
if (!(mask = make_mask(pmv, 1))) |
2032 |
CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data); |
(*CheckCandidate)(pmv[1].x, pmv[1].y, mask, &i, Data); |
2033 |
if (!(mask = make_mask(pmv, 2))) |
if (!(mask = make_mask(pmv, 2))) |
2034 |
CheckCandidate16no4vI(0, 0, mask, &i, Data); |
(*CheckCandidate)(pmv[2].x, pmv[2].y, mask, &i, Data); |
2035 |
|
|
2036 |
|
if (*Data->iMinSAD > MAX_SAD00_FOR_SKIP * 4) // diamond only if needed |
2037 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
2038 |
|
|
2039 |
pMB->mvs[0] = *Data->currentMV; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
2040 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
|
|
|
2041 |
return *(Data->iMinSAD); |
return *(Data->iMinSAD); |
2042 |
} |
} |
2043 |
|
|
2044 |
#define INTRA_THRESH 1350 |
#define INTRA_THRESH 1350 |
2045 |
#define INTER_THRESH 900 |
#define INTER_THRESH 1200 |
2046 |
|
|
2047 |
|
|
2048 |
int |
int |
2049 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
2050 |
const IMAGE * const pCurrent, |
FRAMEINFO * const Current, |
2051 |
MBParam * const pParam, |
MBParam * const pParam, |
2052 |
MACROBLOCK * const pMBs, |
int maxIntra, //maximum number if non-I frames |
2053 |
const uint32_t iFcode) |
int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
2054 |
|
int bCount) // number if B frames in a row |
2055 |
{ |
{ |
2056 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
2057 |
int sSAD = 0; |
int sSAD = 0; |
2058 |
|
MACROBLOCK * const pMBs = Current->mbs; |
2059 |
|
const IMAGE * const pCurrent = &Current->image; |
2060 |
|
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
2061 |
|
|
2062 |
VECTOR currentMV; |
VECTOR currentMV; |
2063 |
int32_t iMinSAD; |
int32_t iMinSAD; |
2065 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2066 |
Data.currentMV = ¤tMV; |
Data.currentMV = ¤tMV; |
2067 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2068 |
Data.iFcode = iFcode; |
Data.iFcode = Current->fcode; |
2069 |
|
CheckCandidate = CheckCandidate16no4vI; |
2070 |
|
|
2071 |
|
if (intraCount < 10) // we're right after an I frame |
2072 |
|
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
2073 |
|
else |
2074 |
|
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
2075 |
|
IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra; |
2076 |
|
|
2077 |
|
|
2078 |
|
InterThresh += 400 * (1 - bCount); |
2079 |
|
if (InterThresh < 200) InterThresh = 200; |
2080 |
|
|
2081 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2082 |
|
|
2088 |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y, |
2089 |
pParam, pMBs, pMB, &Data); |
pParam, pMBs, pMB, &Data); |
2090 |
|
|
2091 |
if (sad > INTRA_THRESH) { |
if (sad > IntraThresh) { |
2092 |
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
2093 |
pParam->edged_width); |
pParam->edged_width); |
2094 |
if (dev + INTRA_THRESH < sad) { intra++; pMB->mode = MODE_INTRA; } |
if (dev + IntraThresh < sad) { |
2095 |
if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame |
pMB->mode = MODE_INTRA; |
2096 |
|
if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2; // I frame |
2097 |
|
} |
2098 |
} |
} |
2099 |
sSAD += sad; |
sSAD += sad; |
2100 |
} |
} |
2101 |
} |
} |
2102 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
2103 |
if (sSAD > INTER_THRESH ) return 1; //P frame |
if (sSAD > InterThresh ) return 1; //P frame |
2104 |
emms(); |
emms(); |
2105 |
return 0; // B frame |
return 0; // B frame |
2106 |
|
|