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 |
|
|
55 |
static __inline int |
static __inline uint32_t |
56 |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv) |
57 |
{ |
{ |
58 |
int xb, yb; |
int xb, yb; |
59 |
if (qpel) { x *= 2; y *= 2;} |
x += x * qpel; y += y * qpel; |
60 |
else if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
61 |
x -= pred.x; |
x -= pred.x; |
62 |
y -= pred.y; |
y -= pred.y; |
63 |
|
|
64 |
if (x) { |
if (x) { |
65 |
if (x < 0) x = -x; |
x = ABS(x); |
66 |
x += (1 << (iFcode - 1)) - 1; |
x += (1 << (iFcode - 1)) - 1; |
67 |
x >>= (iFcode - 1); |
x >>= (iFcode - 1); |
68 |
if (x > 32) x = 32; |
if (x > 32) x = 32; |
70 |
} else xb = 1; |
} else xb = 1; |
71 |
|
|
72 |
if (y) { |
if (y) { |
73 |
if (y < 0) y = -y; |
y = ABS(y); |
74 |
y += (1 << (iFcode - 1)) - 1; |
y += (1 << (iFcode - 1)) - 1; |
75 |
y >>= (iFcode - 1); |
y >>= (iFcode - 1); |
76 |
if (y > 32) y = 32; |
if (y > 32) y = 32; |
118 |
} |
} |
119 |
|
|
120 |
static __inline const uint8_t * |
static __inline const uint8_t * |
121 |
GetReferenceB(const int x, const int y, const int dir, const SearchData * const data) |
GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data) |
122 |
{ |
{ |
123 |
// dir : 0 = forward, 1 = backward |
// dir : 0 = forward, 1 = backward |
124 |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) { |
146 |
} |
} |
147 |
|
|
148 |
static uint8_t * |
static uint8_t * |
149 |
Interpolate8x8qpel(const int x, const int y, const int block, const int dir, const SearchData * const data) |
Interpolate8x8qpel(const int x, const int y, const uint32_t block, const uint32_t dir, const SearchData * const data) |
150 |
{ |
{ |
151 |
// create or find a qpel-precision reference picture; return pointer to it |
// create or find a qpel-precision reference picture; return pointer to it |
152 |
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
uint8_t * Reference = data->RefQ + 16*dir; |
153 |
const int32_t iEdgedWidth = data->iEdgedWidth; |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
154 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
155 |
const int halfpel_x = x/2; |
const int halfpel_x = x/2; |
156 |
const int halfpel_y = y/2; |
const int halfpel_y = y/2; |
190 |
} |
} |
191 |
|
|
192 |
static uint8_t * |
static uint8_t * |
193 |
Interpolate16x16qpel(const int x, const int y, const int dir, const SearchData * const data) |
Interpolate16x16qpel(const int x, const int y, const uint32_t dir, const SearchData * const data) |
194 |
{ |
{ |
195 |
// create or find a qpel-precision reference picture; return pointer to it |
// create or find a qpel-precision reference picture; return pointer to it |
196 |
uint8_t * Reference = (uint8_t *)data->RefQ + 16*dir; |
uint8_t * Reference = data->RefQ + 16*dir; |
197 |
const int32_t iEdgedWidth = data->iEdgedWidth; |
const uint32_t iEdgedWidth = data->iEdgedWidth; |
198 |
const uint32_t rounding = data->rounding; |
const uint32_t rounding = data->rounding; |
199 |
const int halfpel_x = x/2; |
const int halfpel_x = x/2; |
200 |
const int halfpel_y = y/2; |
const int halfpel_y = y/2; |
243 |
const uint8_t * Reference; |
const uint8_t * Reference; |
244 |
VECTOR * current; |
VECTOR * current; |
245 |
|
|
246 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if ( (x > data->max_dx) | (x < data->min_dx) |
247 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
248 |
|
|
249 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
250 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
255 |
current = data->currentMV; |
current = data->currentMV; |
256 |
xc = x; yc = y; |
xc = x; yc = y; |
257 |
} |
} |
258 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
259 |
|
|
260 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
261 |
|
|
262 |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
data->temp[0] += (data->lambda16 * t * data->temp[0])>>10; |
263 |
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))>>10; |
264 |
|
|
265 |
if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
266 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
284 |
static void |
static void |
285 |
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
286 |
{ |
{ |
287 |
int t; |
uint32_t t; |
288 |
const uint8_t * Reference; |
const uint8_t * Reference; |
289 |
|
|
290 |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value |
if ( (!(x&1) && x !=0) | (!(y&1) && y !=0) || //non-zero integer value |
291 |
( x > data->max_dx) || ( x < data->min_dx) |
(x > data->max_dx) | (x < data->min_dx) |
292 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
293 |
|
|
294 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
295 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
296 |
|
|
297 |
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
298 |
|
|
299 |
data->temp[0] += (data->lambda16 * t * data->temp[0])/1000; |
data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10; |
300 |
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))>>10; |
301 |
|
|
302 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (data->temp[0] < data->iMinSAD[0]) { |
303 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = data->temp[0]; |
319 |
{ |
{ |
320 |
int32_t sad; |
int32_t sad; |
321 |
const uint8_t * Reference; |
const uint8_t * Reference; |
322 |
int t; |
uint32_t t; |
323 |
VECTOR * current; |
VECTOR * current; |
324 |
|
|
325 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if ( (x > data->max_dx) | ( x < data->min_dx) |
326 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
327 |
|
|
328 |
if (data->rrv) |
if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; //non-zero even value |
|
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) ) return; //non-zero integer value |
|
329 |
|
|
330 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
331 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
335 |
current = data->currentMV; |
current = data->currentMV; |
336 |
} |
} |
337 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
338 |
data->qpel && !data->qpel_precision, data->rrv); |
data->qpel^data->qpel_precision, data->rrv); |
339 |
|
|
340 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
341 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)>>10; |
342 |
|
|
343 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
344 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
351 |
{ |
{ |
352 |
// maximum speed - for P/B/I decision |
// maximum speed - for P/B/I decision |
353 |
|
|
354 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if ( (x > data->max_dx) | (x < data->min_dx) |
355 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
356 |
|
|
357 |
data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
358 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
359 |
|
|
360 |
if (data->temp[0] < *(data->iMinSAD)) { |
if (data->temp[0] < *(data->iMinSAD)) { |
361 |
*(data->iMinSAD) = data->temp[0]; |
*(data->iMinSAD) = data->temp[0]; |
362 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
375 |
static void |
static void |
376 |
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) |
377 |
{ |
{ |
378 |
int32_t sad; |
int32_t sad, xb, yb; |
379 |
int xb, yb, t; |
uint32_t t; |
380 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
381 |
VECTOR *current; |
VECTOR *current; |
382 |
|
|
383 |
if (( xf > data->max_dx) || ( xf < data->min_dx) |
if ( (xf > data->max_dx) | (xf < data->min_dx) |
384 |
|| ( yf > data->max_dy) || (yf < data->min_dy)) return; |
| (yf > data->max_dy) | (yf < data->min_dy) ) return; |
385 |
|
|
386 |
if (data->qpel_precision) { |
if (!data->qpel_precision) { |
|
ReferenceF = Interpolate16x16qpel(xf, yf, 0, data); |
|
|
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
|
|
current = data->currentQMV; |
|
|
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
|
|
} else { |
|
387 |
ReferenceF = GetReference(xf, yf, data); |
ReferenceF = GetReference(xf, yf, data); |
388 |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
389 |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
390 |
current = data->currentMV; |
current = data->currentMV; |
391 |
|
} else { |
392 |
|
ReferenceF = Interpolate16x16qpel(xf, yf, 0, data); |
393 |
|
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
394 |
|
current = data->currentQMV; |
395 |
|
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
396 |
} |
} |
397 |
|
|
398 |
t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0) |
t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0) |
399 |
+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
+ d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
400 |
|
|
401 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
402 |
sad += (data->lambda16 * t * sad)/1000; |
sad += (data->lambda16 * t * sad)>>10; |
403 |
|
|
404 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
405 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
411 |
CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
412 |
{ |
{ |
413 |
int32_t sad = 0; |
int32_t sad = 0; |
414 |
int k; |
uint32_t k; |
415 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
416 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
417 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
|
const VECTOR zeroMV={0,0}; |
|
418 |
|
|
419 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return; |
420 |
|
|
421 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
422 |
mvs.x = data->directmvF[k].x + x; |
mvs.x = data->directmvF[k].x + x; |
434 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
435 |
|| ( 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; |
436 |
|
|
437 |
if (!data->qpel) { |
|
438 |
mvs.x *= 2; mvs.y *= 2; |
mvs.x *= 2 - data->qpel; mvs.y *= 2 - data->qpel; |
439 |
b_mvs.x *= 2; b_mvs.y *= 2; //we move to qpel precision anyway |
b_mvs.x *= 2 - data->qpel; b_mvs.y *= 2 - data->qpel; //we move to qpel precision anyway |
440 |
} |
|
441 |
ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); |
ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); |
442 |
ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); |
ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); |
443 |
|
|
444 |
sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
sad += sad8bi(data->Cur + 8*(k&1) + 8*(k>>1)*(data->iEdgedWidth), |
445 |
ReferenceF, ReferenceB, |
ReferenceF, ReferenceB, data->iEdgedWidth); |
|
data->iEdgedWidth); |
|
446 |
if (sad > *(data->iMinSAD)) return; |
if (sad > *(data->iMinSAD)) return; |
447 |
} |
} |
448 |
|
|
449 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
450 |
|
|
451 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
452 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
461 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
462 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
463 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
|
const VECTOR zeroMV = {0,0}; |
|
464 |
|
|
465 |
if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return; |
if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return; |
466 |
|
|
467 |
mvs.x = data->directmvF[0].x + x; |
mvs.x = data->directmvF[0].x + x; |
468 |
b_mvs.x = ((x == 0) ? |
b_mvs.x = ((x == 0) ? |
479 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
480 |
|| ( 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; |
481 |
|
|
482 |
if (!data->qpel) { |
mvs.x *= 2 - data->qpel; mvs.y *= 2 - data->qpel; |
483 |
mvs.x *= 2; mvs.y *= 2; |
b_mvs.x *= 2 - data->qpel; b_mvs.y *= 2 - data->qpel; //we move to qpel precision anyway |
484 |
b_mvs.x *= 2; b_mvs.y *= 2; //we move to qpel precision anyway |
|
|
} |
|
485 |
ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data); |
ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data); |
486 |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
487 |
|
|
488 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
489 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)/1000; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
490 |
|
|
491 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
492 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
497 |
static void |
static void |
498 |
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
499 |
{ |
{ |
500 |
int32_t sad; int t; |
int32_t sad; uint32_t t; |
501 |
const uint8_t * Reference; |
const uint8_t * Reference; |
502 |
|
|
503 |
if (( x > data->max_dx) || ( x < data->min_dx) |
if ( (x > data->max_dx) | (x < data->min_dx) |
504 |
|| ( y > data->max_dy) || (y < data->min_dy)) return; |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
505 |
|
|
506 |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
507 |
else Reference = GetReference(x, y, data); |
else Reference = GetReference(x, y, data); |
509 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
510 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
511 |
|
|
512 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))/100; |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
513 |
|
|
514 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
515 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
538 |
|
|
539 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
540 |
|
|
541 |
if (iDirection) { //checking if anything found |
if (iDirection) { //if anything found |
542 |
bDirection = iDirection; |
bDirection = iDirection; |
543 |
iDirection = 0; |
iDirection = 0; |
544 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
664 |
|
|
665 |
/* MAINSEARCH FUNCTIONS END */ |
/* MAINSEARCH FUNCTIONS END */ |
666 |
|
|
|
/* HALFPELREFINE COULD BE A MAINSEARCH FUNCTION, BUT THERE IS NO NEED FOR IT */ |
|
|
|
|
667 |
static void |
static void |
668 |
SubpelRefine(const SearchData * const data) |
SubpelRefine(const SearchData * const data) |
669 |
{ |
{ |
670 |
/* Do a half-pel or q-pel refinement */ |
/* Do a half-pel or q-pel refinement */ |
671 |
VECTOR backupMV; |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
672 |
int iDirection; //not needed |
int iDirection; //only needed because macro expects it |
673 |
|
|
674 |
if (data->qpel_precision) |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
675 |
backupMV = *(data->currentQMV); |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
676 |
else backupMV = *(data->currentMV); |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y, 0); |
677 |
|
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y + 1, 0); |
678 |
CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x, centerMV.y + 1, 0); |
679 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y + 1, 0); |
680 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0); |
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y, 0); |
681 |
CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0); |
CHECK_CANDIDATE(centerMV.x - 1, centerMV.y - 1, 0); |
|
CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0); |
|
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0); |
|
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0); |
|
|
CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0); |
|
682 |
} |
} |
683 |
|
|
684 |
static __inline int |
static __inline int |
685 |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
SkipDecisionP(const IMAGE * current, const IMAGE * reference, |
686 |
const int x, const int y, |
const int x, const int y, |
687 |
const uint32_t iEdgedWidth, const uint32_t iQuant, int rrv) |
const uint32_t stride, const uint32_t iQuant, int rrv) |
688 |
|
|
689 |
{ |
{ |
690 |
/* keep repeating checks for all b-frames before this P frame, |
if(!rrv) { |
691 |
to make sure that SKIP is possible (todo) |
uint32_t sadC = sad8(current->u + x*8 + y*stride*8, |
692 |
how: if skip is not possible set sad00 to a very high value */ |
reference->u + x*8 + y*stride*8, stride); |
|
if(rrv) { |
|
|
uint32_t sadC = sad16(current->u + x*16 + y*(iEdgedWidth/2)*16, |
|
|
reference->u + x*16 + y*(iEdgedWidth/2)*16, iEdgedWidth/2, 256*4096); |
|
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
|
|
sadC += sad16(current->v + (x + y*(iEdgedWidth/2))*16, |
|
|
reference->v + (x + y*(iEdgedWidth/2))*16, iEdgedWidth/2, 256*4096); |
|
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
|
|
return 1; |
|
|
} else { |
|
|
uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8, |
|
|
reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2); |
|
693 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
694 |
sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8, |
sadC += sad8(current->v + (x + y*stride)*8, |
695 |
reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
reference->v + (x + y*stride)*8, stride); |
696 |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0; |
697 |
return 1; |
return 1; |
698 |
|
|
699 |
|
} else { |
700 |
|
uint32_t sadC = sad16(current->u + x*16 + y*stride*16, |
701 |
|
reference->u + x*16 + y*stride*16, stride, 256*4096); |
702 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
703 |
|
sadC += sad16(current->v + (x + y*stride)*16, |
704 |
|
reference->v + (x + y*stride)*16, stride, 256*4096); |
705 |
|
if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0; |
706 |
|
return 1; |
707 |
} |
} |
708 |
} |
} |
709 |
|
|
711 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
712 |
{ |
{ |
713 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_NOT_CODED; |
714 |
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
715 |
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0; |
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV; |
|
|
|
|
pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0; |
|
|
pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0; |
|
|
|
|
716 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
717 |
} |
} |
718 |
|
|
729 |
const IMAGE *const pCurrent = ¤t->image; |
const IMAGE *const pCurrent = ¤t->image; |
730 |
const IMAGE *const pRef = &reference->image; |
const IMAGE *const pRef = &reference->image; |
731 |
|
|
|
const VECTOR zeroMV = { 0, 0 }; |
|
|
|
|
732 |
uint32_t mb_width = pParam->mb_width; |
uint32_t mb_width = pParam->mb_width; |
733 |
uint32_t mb_height = pParam->mb_height; |
uint32_t mb_height = pParam->mb_height; |
734 |
|
const uint32_t iEdgedWidth = pParam->edged_width; |
735 |
|
|
736 |
uint32_t x, y; |
uint32_t x, y; |
737 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
744 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
745 |
SearchData Data; |
SearchData Data; |
746 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
747 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = iEdgedWidth; |
748 |
Data.currentMV = currentMV; |
Data.currentMV = currentMV; |
749 |
Data.currentQMV = currentQMV; |
Data.currentQMV = currentQMV; |
750 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
768 |
for (x = 0; x < mb_width; x++) { |
for (x = 0; x < mb_width; x++) { |
769 |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width]; |
770 |
|
|
771 |
if (Data.rrv) pMB->sad16 = |
if (!Data.rrv) pMB->sad16 = |
772 |
sad32v_c(pCurrent->y + (x + y * pParam->edged_width) * 32, |
sad16v(pCurrent->y + (x + y * iEdgedWidth) * 16, |
773 |
pRef->y + (x + y * pParam->edged_width) * 32, |
pRef->y + (x + y * iEdgedWidth) * 16, |
774 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
775 |
|
|
776 |
else pMB->sad16 = |
else pMB->sad16 = |
777 |
sad16v(pCurrent->y + (x + y * pParam->edged_width) * 16, |
sad32v_c(pCurrent->y + (x + y * iEdgedWidth) * 32, |
778 |
pRef->y + (x + y * pParam->edged_width) * 16, |
pRef->y + (x + y * iEdgedWidth) * 32, |
779 |
pParam->edged_width, pMB->sad8 ); |
pParam->edged_width, pMB->sad8 ); |
780 |
|
|
781 |
if (Data.chroma) { |
if (Data.chroma) { |
782 |
pMB->sad16 += sad8(pCurrent->u + x*8 + y*(pParam->edged_width/2)*8, |
Data.temp[7] = sad8(pCurrent->u + x*8 + y*(iEdgedWidth/2)*8, |
783 |
pRef->u + x*8 + y*(pParam->edged_width/2)*8, pParam->edged_width/2); |
pRef->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2) |
784 |
|
+ sad8(pCurrent->v + (x + y*(iEdgedWidth/2))*8, |
785 |
pMB->sad16 += sad8(pCurrent->v + (x + y*(pParam->edged_width/2))*8, |
pRef->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2); |
786 |
pRef->v + (x + y*(pParam->edged_width/2))*8, pParam->edged_width/2); |
pMB->sad16 += Data.temp[7]; |
787 |
} |
} |
788 |
|
|
789 |
sad00 = pMB->sad16; //if no gmc; else sad00 = (..) |
sad00 = pMB->sad16; |
790 |
|
|
791 |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
if (!(current->global_flags & XVID_LUMIMASKING)) { |
792 |
pMB->dquant = NO_CHANGE; |
pMB->dquant = NO_CHANGE; |
|
pMB->quant = current->quant; |
|
793 |
} else { |
} else { |
794 |
if (pMB->dquant != NO_CHANGE) { |
if (pMB->dquant != NO_CHANGE) { |
795 |
quant += DQtab[pMB->dquant]; |
quant += DQtab[pMB->dquant]; |
796 |
if (quant > 31) quant = 31; |
if (quant > 31) quant = 31; |
797 |
else if (quant < 1) quant = 1; |
else if (quant < 1) quant = 1; |
798 |
} |
} |
|
pMB->quant = quant; |
|
799 |
} |
} |
800 |
|
pMB->quant = current->quant; |
801 |
|
|
802 |
//initial skip decision |
//initial skip decision |
803 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
804 |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
if (current->coding_type == P_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
805 |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
806 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
807 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
808 |
continue; |
continue; |
809 |
} |
} |
818 |
if (current->coding_type == P_VOP) { |
if (current->coding_type == P_VOP) { |
819 |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
820 |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1)) ) |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1)) ) |
821 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant, Data.rrv)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
822 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
823 |
continue; |
continue; |
824 |
} |
} |
834 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
835 |
|
|
836 |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
837 |
if (Data.rrv) InterBias *= 4; //?? |
if (Data.rrv) InterBias *= 4; |
838 |
|
|
839 |
if (InterBias < pMB->sad16) { |
if (InterBias < pMB->sad16) { |
840 |
int32_t deviation; |
int32_t deviation; |
841 |
if (Data.rrv) { |
if (!Data.rrv) |
842 |
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 32, |
deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 16, iEdgedWidth); |
843 |
pParam->edged_width) |
else { |
844 |
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16, |
deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 32, iEdgedWidth) |
845 |
pParam->edged_width) |
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16, iEdgedWidth) |
846 |
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * pParam->edged_width, |
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * iEdgedWidth, iEdgedWidth) |
847 |
pParam->edged_width) |
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * (iEdgedWidth+1), iEdgedWidth); |
848 |
+ dev16(pCurrent->y + (x + y * pParam->edged_width) * 32 + 16 * (pParam->edged_width+1), |
} |
|
pParam->edged_width); |
|
|
} else |
|
|
deviation = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16, |
|
|
pParam->edged_width); |
|
|
|
|
849 |
if (deviation < (pMB->sad16 - InterBias)) { |
if (deviation < (pMB->sad16 - InterBias)) { |
850 |
if (++iIntra >= iLimit) return 1; |
if (++iIntra >= iLimit) return 1; |
851 |
|
SkipMacroblockP(pMB, 0); //same thing |
852 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = |
|
|
pMB->mvs[3] = zeroMV; |
|
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = |
|
|
pMB->qmvs[3] = zeroMV; |
|
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = |
|
|
pMB->sad8[3] = 0; |
|
853 |
} |
} |
854 |
} |
} |
855 |
} |
} |
864 |
} |
} |
865 |
|
|
866 |
|
|
|
#define PMV_HALFPEL16 (PMV_HALFPELDIAMOND16|PMV_HALFPELREFINE16) |
|
|
|
|
867 |
static __inline int |
static __inline int |
868 |
make_mask(const VECTOR * const pmv, const int i) |
make_mask(const VECTOR * const pmv, const int i) |
869 |
{ |
{ |
871 |
for (j = 0; j < i; j++) { |
for (j = 0; j < i; j++) { |
872 |
if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already |
if (MVequal(pmv[i], pmv[j])) return 0; // same vector has been checked already |
873 |
if (pmv[i].x == pmv[j].x) { |
if (pmv[i].x == pmv[j].x) { |
874 |
if (pmv[i].y == pmv[j].y + iDiamondSize) { mask &= ~4; continue; } |
if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4; |
875 |
if (pmv[i].y == pmv[j].y - iDiamondSize) { mask &= ~8; continue; } |
else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8; |
876 |
} else |
} else |
877 |
if (pmv[i].y == pmv[j].y) { |
if (pmv[i].y == pmv[j].y) { |
878 |
if (pmv[i].x == pmv[j].x + iDiamondSize) { mask &= ~1; continue; } |
if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1; |
879 |
if (pmv[i].x == pmv[j].x - iDiamondSize) { mask &= ~2; continue; } |
else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2; |
880 |
} |
} |
881 |
} |
} |
882 |
return mask; |
return mask; |
917 |
if (rrv) { |
if (rrv) { |
918 |
int i; |
int i; |
919 |
for (i = 0; i < 7; i++) { |
for (i = 0; i < 7; i++) { |
920 |
pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); // halfzero->halfpel |
pmv[i].x = RRV_MV_SCALEUP(pmv[i].x); |
921 |
pmv[i].y = RRV_MV_SCALEUP(pmv[i].y); |
pmv[i].y = RRV_MV_SCALEUP(pmv[i].y); |
922 |
} |
} |
923 |
} |
} |
947 |
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, |
948 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
949 |
|
|
950 |
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); |
951 |
|
|
952 |
Data->temp[5] = Data->temp[7] = 256*4096; // to reset chroma-sad cache |
Data->temp[5] = Data->temp[6] = 0; // chroma-sad cache |
953 |
if (Data->rrv) i = 2; else i = 1; |
i = Data->rrv ? 2 : 1; |
954 |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16*i; |
955 |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
956 |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i; |
971 |
for(i = 0; i < 5; i++) |
for(i = 0; i < 5; i++) |
972 |
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
Data->currentMV[i].x = Data->currentMV[i].y = 0; |
973 |
|
|
974 |
if (pParam->m_quarterpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
975 |
else Data->predMV = pmv[0]; |
else Data->predMV = pmv[0]; |
976 |
|
|
977 |
i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0); |
i = d_mv_bits(0, 0, Data->predMV, Data->iFcode, 0, 0); |
978 |
Data->iMinSAD[0] = pMB->sad16 + (Data->lambda16 * i * pMB->sad16)/1000; |
Data->iMinSAD[0] = pMB->sad16 + ((Data->lambda16 * i * pMB->sad16)>>10); |
979 |
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)) >> 10); |
980 |
Data->iMinSAD[2] = pMB->sad8[1]; |
Data->iMinSAD[2] = pMB->sad8[1]; |
981 |
Data->iMinSAD[3] = pMB->sad8[2]; |
Data->iMinSAD[3] = pMB->sad8[2]; |
982 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
983 |
|
|
984 |
if ((x == 0) && (y == 0)) threshA = 512; |
if (x | y) { |
|
else { |
|
985 |
threshA = Data->temp[0]; // that's when we keep this SAD atm |
threshA = Data->temp[0]; // that's when we keep this SAD atm |
986 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
987 |
if (threshA > 1024) threshA = 1024; } |
else if (threshA > 1024) threshA = 1024; |
988 |
|
} else threshA = 512; |
989 |
|
|
990 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
991 |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
992 |
|
|
993 |
if (Data->rrv) CheckCandidate = CheckCandidate32; |
if (!Data->rrv) { |
994 |
else if (inter4v || Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16; |
995 |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
else CheckCandidate = CheckCandidate16no4v; //for extra speed |
996 |
|
} else CheckCandidate = CheckCandidate32; |
997 |
|
|
998 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
999 |
|
|
1000 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) { |
1001 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
1005 |
|
|
1006 |
if ((Data->iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
1007 |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
1008 |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) |
1009 |
inter4v = 0; |
inter4v = 0; |
1010 |
} else { |
else { |
1011 |
|
|
1012 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1013 |
if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; |
1040 |
} |
} |
1041 |
|
|
1042 |
backupMV = Data->currentMV[0]; |
backupMV = Data->currentMV[0]; |
1043 |
if (!MotionFlags & PMV_HALFPELREFINE16 || Data->rrv) startMV.x = startMV.y = 0; |
if (MotionFlags & PMV_HALFPELREFINE16 && !Data->rrv) startMV.x = startMV.y = 1; |
1044 |
else startMV.x = startMV.y = 1; |
else startMV.x = startMV.y = 0; |
1045 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1046 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1047 |
|
|
1061 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1062 |
} |
} |
1063 |
|
|
1064 |
if((Data->qpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) { |
if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE16) { |
|
|
|
1065 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1066 |
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, |
1067 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1072 |
if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0; |
if (Data->iMinSAD[0] < (int32_t)iQuant * 30 ) inter4v = 0; |
1073 |
if (inter4v) { |
if (inter4v) { |
1074 |
SearchData Data8; |
SearchData Data8; |
1075 |
Data8.iFcode = Data->iFcode; |
memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data |
1076 |
Data8.lambda8 = Data->lambda8; |
|
|
Data8.iEdgedWidth = Data->iEdgedWidth; |
|
|
Data8.RefQ = Data->RefQ; |
|
|
Data8.qpel = Data->qpel; |
|
|
Data8.rrv = Data->rrv; |
|
1077 |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8); |
1078 |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8); |
1079 |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8); |
1080 |
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); |
1081 |
|
|
1082 |
if (Data->chroma) { |
if (Data->chroma) { |
1083 |
int sumx, sumy; |
int sumx = 0, sumy = 0; |
1084 |
|
const int div = 1 + Data->qpel; |
1085 |
|
const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs; |
1086 |
|
|
1087 |
if(pParam->m_quarterpel) { |
for (i = 0; i < 4; i++) { |
1088 |
sumx = pMB->qmvs[0].x/2 + pMB->qmvs[1].x/2 + pMB->qmvs[2].x/2 + pMB->qmvs[3].x/2; |
sumx += mv[i].x / div; |
1089 |
sumy = pMB->qmvs[0].y/2 + pMB->qmvs[1].y/2 + pMB->qmvs[2].y/2 + pMB->qmvs[3].y/2; |
sumy += mv[i].y / div; |
|
} else { |
|
|
sumx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; |
|
|
sumy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; |
|
1090 |
} |
} |
1091 |
|
|
1092 |
Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf], |
Data->iMinSAD[1] += ChromaSAD( (sumx >> 3) + roundtab_76[sumx & 0xf], |
1104 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
1105 |
// INTER MODE |
// INTER MODE |
1106 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
1107 |
pMB->mvs[0] = pMB->mvs[1] |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1108 |
= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1109 |
|
|
1110 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = |
if(Data->qpel) { |
|
pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
|
|
|
|
if(pParam->m_quarterpel) { |
|
1111 |
pMB->qmvs[0] = pMB->qmvs[1] |
pMB->qmvs[0] = pMB->qmvs[1] |
1112 |
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0]; |
1113 |
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x; |
1139 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1140 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
1141 |
|
|
1142 |
if(pParam->m_quarterpel) { |
if(Data->qpel) { |
1143 |
Data->predMV = 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); |
1144 |
if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y, |
if (block != 0) i = d_mv_bits( Data->currentQMV->x, Data->currentQMV->y, |
1145 |
Data->predMV, Data->iFcode, 0, 0); |
Data->predMV, Data->iFcode, 0, 0); |
|
|
|
1146 |
} else { |
} else { |
1147 |
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); |
|
if (block != 0) { |
|
1148 |
if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y, |
if (block != 0) i = d_mv_bits( Data->currentMV->x, Data->currentMV->y, |
1149 |
Data->predMV, Data->iFcode, 0, Data->rrv); |
Data->predMV, Data->iFcode, 0, Data->rrv); |
1150 |
} |
} |
|
} |
|
1151 |
|
|
1152 |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))/100; |
*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10; |
1153 |
|
|
1154 |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) { |
if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8|PMV_QUARTERPELREFINE8)) { |
1155 |
if (Data->rrv) i = 2; else i = 1; |
if (Data->rrv) i = 2; else i = 1; |
1156 |
|
|
1157 |
Data->Ref = OldData->Ref + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1158 |
Data->RefH = OldData->RefH + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1159 |
Data->RefV = OldData->RefV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1160 |
Data->RefHV = OldData->RefHV + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1161 |
|
|
1162 |
Data->Cur = OldData->Cur + i*8 * ((block&1) + pParam->edged_width*(block>>1)); |
Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1163 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1164 |
|
|
1165 |
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, |
1166 |
pParam->width, pParam->height, OldData->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
1167 |
|
|
1168 |
if (Data->rrv) CheckCandidate = CheckCandidate16no4v; |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
1169 |
else CheckCandidate = CheckCandidate8; |
else CheckCandidate = CheckCandidate16no4v; |
1170 |
|
|
1171 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8) { |
1172 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1195 |
} |
} |
1196 |
} |
} |
1197 |
|
|
1198 |
if(Data->qpel) { |
if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE8) { |
|
if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) && |
|
|
(MotionFlags & PMV_QUARTERPELREFINE8)) { |
|
1199 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1200 |
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, |
1201 |
pParam->width, pParam->height, OldData->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1202 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1203 |
} |
} |
1204 |
} |
} |
|
} |
|
1205 |
|
|
1206 |
if (Data->rrv) { |
if (Data->rrv) { |
1207 |
Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x); |
Data->currentMV->x = RRV_MV_SCALEDOWN(Data->currentMV->x); |
1211 |
if(Data->qpel) { |
if(Data->qpel) { |
1212 |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentQMV->x - Data->predMV.x; |
1213 |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentQMV->y - Data->predMV.y; |
1214 |
pMB->qmvs[block] = *(Data->currentQMV); |
pMB->qmvs[block] = *Data->currentQMV; |
1215 |
} else { |
} else { |
1216 |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x; |
1217 |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y; |
1218 |
} |
} |
1219 |
|
|
1220 |
pMB->mvs[block] = *(Data->currentMV); |
pMB->mvs[block] = *Data->currentMV; |
1221 |
pMB->sad8[block] = 4 * (*Data->iMinSAD); |
pMB->sad8[block] = 4 * *Data->iMinSAD; |
1222 |
} |
} |
1223 |
|
|
1224 |
/* B-frames code starts here */ |
/* motion estimation for B-frames */ |
1225 |
|
|
1226 |
static __inline VECTOR |
static __inline VECTOR |
1227 |
ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) |
ChoosePred(const MACROBLOCK * const pMB, const uint32_t mode) |
1228 |
{ |
{ |
1229 |
/* the stupidiest function ever */ |
/* the stupidiest function ever */ |
1230 |
if (mode == MODE_FORWARD) return pMB->mvs[0]; |
return (mode == MODE_FORWARD ? pMB->mvs[0] : pMB->b_mvs[0]); |
|
else return pMB->b_mvs[0]; |
|
1231 |
} |
} |
1232 |
|
|
1233 |
static void __inline |
static void __inline |
1260 |
pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); |
pmv[5].x = EVEN(pmv[5].x); pmv[5].y = EVEN(pmv[5].y); |
1261 |
} else pmv[5].x = pmv[5].y = 0; |
} else pmv[5].x = pmv[5].y = 0; |
1262 |
|
|
1263 |
if ((x != 0)&&(y != 0)) { |
if (x != 0 && y != 0) { |
1264 |
pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); |
pmv[6] = ChoosePred(pMB-1-iWcount, mode_curr); |
1265 |
pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y); |
pmv[6].x = EVEN(pmv[6].x); pmv[6].y = EVEN(pmv[6].y); |
1266 |
} else pmv[6].x = pmv[6].y = 0; |
} else pmv[6].x = pmv[6].y = 0; |
|
|
|
|
// more? |
|
1267 |
} |
} |
1268 |
|
|
1269 |
|
|
1270 |
/* search backward or forward, for b-frames */ |
/* search backward or forward */ |
1271 |
static void |
static void |
1272 |
SearchBF( const uint8_t * const pRef, |
SearchBF( const uint8_t * const pRef, |
1273 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1285 |
SearchData * const Data) |
SearchData * const Data) |
1286 |
{ |
{ |
1287 |
|
|
|
const int32_t iEdgedWidth = pParam->edged_width; |
|
|
|
|
1288 |
int i, iDirection = 255, mask; |
int i, iDirection = 255, mask; |
1289 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1290 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1292 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
1293 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1294 |
|
|
1295 |
Data->Ref = pRef + (x + y * iEdgedWidth) * 16; |
Data->Ref = pRef + (x + y * Data->iEdgedWidth) * 16; |
1296 |
Data->RefH = pRefH + (x + y * iEdgedWidth) * 16; |
Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16; |
1297 |
Data->RefV = pRefV + (x + y * iEdgedWidth) * 16; |
Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16; |
1298 |
Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16; |
Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16; |
1299 |
|
|
1300 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1301 |
|
|
1304 |
|
|
1305 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1306 |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
1307 |
|
|
1308 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
1309 |
|
|
1310 |
Data->currentMV->x = Data->currentMV->y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
1316 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1317 |
} |
} |
1318 |
|
|
1319 |
if (MotionFlags & PMV_USESQUARES16) |
if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch; |
1320 |
MainSearchPtr = SquareSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
|
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) |
|
|
MainSearchPtr = AdvDiamondSearch; |
|
1321 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1322 |
|
|
1323 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
1355 |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
if (mode_current == MODE_FORWARD) pMB->mvs[0] = *Data->currentMV; |
1356 |
else pMB->b_mvs[0] = *Data->currentMV; |
else pMB->b_mvs[0] = *Data->currentMV; |
1357 |
} |
} |
1358 |
|
|
1359 |
if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; |
if (mode_current == MODE_FORWARD) *(Data->currentMV+2) = *Data->currentMV; |
1360 |
else *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
else *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search |
|
|
|
1361 |
} |
} |
1362 |
|
|
1363 |
static void |
static void |
1369 |
const SearchData * const Data) |
const SearchData * const Data) |
1370 |
{ |
{ |
1371 |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
int dx = 0, dy = 0, b_dx = 0, b_dy = 0; |
1372 |
uint32_t sum; |
int32_t sum; |
1373 |
const int div = 1 + Data->qpel; |
const int div = 1 + Data->qpel; |
1374 |
int k; |
int k; |
1375 |
const uint32_t quant = pMB->quant; |
const uint32_t stride = Data->iEdgedWidth/2; |
1376 |
//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 |
1377 |
|
|
1378 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1387 |
b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf]; |
b_dy = (b_dy >> 3) + roundtab_76[b_dy & 0xf]; |
1388 |
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; |
1389 |
|
|
1390 |
sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2), |
sum = sad8bi(pCur->u + 8 * x + 8 * y * stride, |
1391 |
f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
f_Ref->u + (y*8 + dy/2) * stride + x*8 + dx/2, |
1392 |
b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1393 |
Data->iEdgedWidth/2); |
stride); |
1394 |
|
|
1395 |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * quant) return; //no skip |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; //no skip |
1396 |
|
|
1397 |
sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2), |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
1398 |
f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
1399 |
b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1400 |
Data->iEdgedWidth/2); |
stride); |
1401 |
|
|
1402 |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped |
1403 |
} |
} |
1404 |
|
|
|
|
|
|
|
|
1405 |
static __inline uint32_t |
static __inline uint32_t |
1406 |
SearchDirect(const IMAGE * const f_Ref, |
SearchDirect(const IMAGE * const f_Ref, |
1407 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1423 |
|
|
1424 |
{ |
{ |
1425 |
int32_t skip_sad; |
int32_t skip_sad; |
1426 |
int k; |
int k = (x + Data->iEdgedWidth*y) * 16; |
|
|
|
1427 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1428 |
|
|
1429 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
1430 |
|
Data->Ref = f_Ref->y + k; |
1431 |
|
Data->RefH = f_RefH + k; |
1432 |
|
Data->RefV = f_RefV + k; |
1433 |
|
Data->RefHV = f_RefHV + k; |
1434 |
|
Data->bRef = b_Ref->y + k; |
1435 |
|
Data->bRefH = b_RefH + k; |
1436 |
|
Data->bRefV = b_RefV + k; |
1437 |
|
Data->bRefHV = b_RefHV + k; |
1438 |
|
|
1439 |
|
k = Data->qpel ? 4 : 2; |
1440 |
|
Data->max_dx = k * (pParam->width - x * 16); |
1441 |
|
Data->max_dy = k * (pParam->height - y * 16); |
1442 |
|
Data->min_dx = -k * (16 + x * 16); |
1443 |
|
Data->min_dy = -k * (16 + y * 16); |
1444 |
|
|
1445 |
Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
Data->referencemv = Data->qpel ? b_mb->qmvs : b_mb->mvs; |
1446 |
Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16; |
Data->qpel_precision = 0; |
|
Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16; |
|
|
Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16; |
|
|
|
|
|
Data->max_dx = 2 * pParam->width - 2 * (x) * 16; |
|
|
Data->max_dy = 2 * pParam->height - 2 * (y) * 16; |
|
|
Data->min_dx = -(2 * 16 + 2 * (x) * 16); |
|
|
Data->min_dy = -(2 * 16 + 2 * (y) * 16); |
|
|
if (Data->qpel) { //we measure in qpixels |
|
|
Data->max_dx *= 2; |
|
|
Data->max_dy *= 2; |
|
|
Data->min_dx *= 2; |
|
|
Data->min_dy *= 2; |
|
|
Data->referencemv = b_mb->qmvs; |
|
|
} else Data->referencemv = b_mb->mvs; |
|
|
Data->qpel_precision = 0; // it's a trick. it's 1 not 0, but we need 0 here |
|
1447 |
|
|
1448 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1449 |
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); |
1451 |
pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD); |
pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD); |
1452 |
pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD; |
pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD; |
1453 |
|
|
1454 |
if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx ) |
if ( (pMB->b_mvs[k].x > Data->max_dx) | (pMB->b_mvs[k].x < Data->min_dx) |
1455 |
|| ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) { |
| (pMB->b_mvs[k].y > Data->max_dy) | (pMB->b_mvs[k].y < Data->min_dy) ) { |
1456 |
|
|
1457 |
*best_sad = 256*4096; // in that case, we won't use direct mode |
*best_sad = 256*4096; // in that case, we won't use direct mode |
1458 |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV" |
1468 |
} |
} |
1469 |
} |
} |
1470 |
|
|
1471 |
if (b_mb->mode == MODE_INTER4V) CheckCandidate = CheckCandidateDirect; |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
|
else CheckCandidate = CheckCandidateDirectno4v; |
|
1472 |
|
|
1473 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
(*CheckCandidate)(0, 0, 255, &k, Data); |
1474 |
|
|
1475 |
// initial (fast) skip decision |
// initial (fast) skip decision |
1476 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH*2) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH*2) { |
1477 |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); //possible skip - checking chroma |
//possible skip - checking chroma |
1478 |
|
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
1479 |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
1480 |
} |
} |
1481 |
|
|
1494 |
|
|
1495 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1496 |
|
|
1497 |
if (b_mb->mode == MODE_INTER4V || Data->qpel) pMB->mode = MODE_DIRECT; |
if (Data->qpel || b_mb->mode == MODE_INTER4V) pMB->mode = MODE_DIRECT; |
1498 |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation |
1499 |
|
|
1500 |
pMB->pmvs[3] = *Data->currentMV; |
pMB->pmvs[3] = *Data->currentMV; |
1526 |
return skip_sad; |
return skip_sad; |
1527 |
} |
} |
1528 |
|
|
|
|
|
1529 |
static void |
static void |
1530 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const uint8_t * const f_Ref, |
1531 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1587 |
|
|
1588 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
1589 |
|
|
1590 |
//diamond. I wish we could use normal mainsearch functions (square, advdiamond) |
//diamond |
|
|
|
1591 |
do { |
do { |
1592 |
iDirection = 255; |
iDirection = 255; |
1593 |
// forward MV moves |
// forward MV moves |
1608 |
|
|
1609 |
} while (!(iDirection)); |
} while (!(iDirection)); |
1610 |
|
|
1611 |
|
//qpel refinement |
1612 |
if (fData->qpel) { |
if (fData->qpel) { |
1613 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*fData->iMinSAD > *best_sad + 500) return; |
1614 |
CheckCandidate = CheckCandidateInt; |
CheckCandidate = CheckCandidateInt; |
1670 |
int32_t best_sad; |
int32_t best_sad; |
1671 |
uint32_t skip_sad; |
uint32_t skip_sad; |
1672 |
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; |
|
static const VECTOR zeroMV={0,0}; |
|
1673 |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
const MACROBLOCK * const b_mbs = b_reference->mbs; |
1674 |
|
|
1675 |
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
VECTOR f_predMV, b_predMV; /* there is no prediction for direct mode*/ |
1763 |
switch (pMB->mode) { |
switch (pMB->mode) { |
1764 |
case MODE_FORWARD: |
case MODE_FORWARD: |
1765 |
f_count++; |
f_count++; |
1766 |
if (Data.qpel) f_predMV = pMB->qmvs[0]; |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
|
else f_predMV = pMB->mvs[0]; |
|
1767 |
break; |
break; |
1768 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
1769 |
b_count++; |
b_count++; |
1770 |
if (Data.qpel) b_predMV = pMB->b_qmvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
|
else b_predMV = pMB->b_mvs[0]; |
|
1771 |
break; |
break; |
1772 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
1773 |
i_count++; |
i_count++; |
1774 |
if (Data.qpel) { |
f_predMV = Data.qpel ? pMB->qmvs[0] : pMB->mvs[0]; |
1775 |
f_predMV = pMB->qmvs[0]; |
b_predMV = Data.qpel ? pMB->b_qmvs[0] : pMB->b_mvs[0]; |
|
b_predMV = pMB->b_qmvs[0]; |
|
|
} else { |
|
|
f_predMV = pMB->mvs[0]; |
|
|
b_predMV = pMB->b_mvs[0]; |
|
|
} |
|
1776 |
break; |
break; |
1777 |
case MODE_DIRECT: |
case MODE_DIRECT: |
1778 |
case MODE_DIRECT_NO4V: |
case MODE_DIRECT_NO4V: |
1834 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
1835 |
|
|
1836 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
1837 |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1) * pParam->mb_width)]; |
MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width]; |
1838 |
MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i]; |
MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i]; |
1839 |
MB->mode = MODE_INTER; |
MB->mode = MODE_INTER; |
1840 |
MB->sad16 = Data->iMinSAD[i+1]; |
MB->sad16 = Data->iMinSAD[i+1]; |
1846 |
#define INTRA_THRESH 1500 |
#define INTRA_THRESH 1500 |
1847 |
#define INTER_THRESH 1400 |
#define INTER_THRESH 1400 |
1848 |
|
|
|
|
|
1849 |
int |
int |
1850 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
1851 |
FRAMEINFO * const Current, |
FRAMEINFO * const Current, |
1852 |
MBParam * const pParam, |
MBParam * const pParam, |
1853 |
int maxIntra, //maximum number if non-I frames |
int maxIntra, //maximum number if non-I frames |
1854 |
int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
1855 |
int bCount) // number if B frames in a row |
int bCount) // number of B frames in a row |
1856 |
{ |
{ |
1857 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
1858 |
int sSAD = 0; |
int sSAD = 0; |
1859 |
MACROBLOCK * const pMBs = Current->mbs; |
MACROBLOCK * const pMBs = Current->mbs; |
1860 |
const IMAGE * const pCurrent = &Current->image; |
const IMAGE * const pCurrent = &Current->image; |
1861 |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH; |
|
const VECTOR zeroMV = {0,0}; |
|
1862 |
|
|
1863 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
1864 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
1871 |
Data.temp = temp; |
Data.temp = temp; |
1872 |
CheckCandidate = CheckCandidate32I; |
CheckCandidate = CheckCandidate32I; |
1873 |
|
|
1874 |
if (intraCount < 10) // we're right after an I frame |
if (intraCount != 0 && intraCount < 10) // we're right after an I frame |
1875 |
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
IntraThresh += 4 * (intraCount - 10) * (intraCount - 10); |
1876 |
else |
else |
1877 |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
1892 |
|
|
1893 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
1894 |
int dev; |
int dev; |
1895 |
MACROBLOCK *pMB = &pMBs[x+(i&1) + y+(i>>1) * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x+(i&1) + (y+(i>>1)) * pParam->mb_width]; |
1896 |
if (pMB->sad16 > IntraThresh) { |
if (pMB->sad16 > IntraThresh) { |
1897 |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1))* pParam->edged_width) * 16, |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1))* pParam->edged_width) * 16, |
1898 |
pParam->edged_width); |
pParam->edged_width); |