50 |
#define MAX_SAD00_FOR_SKIP (20) |
#define MAX_SAD00_FOR_SKIP (20) |
51 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
52 |
|
|
53 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { CheckCandidate((X),(Y), (D), &iDirection, data ); } |
|
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
|
54 |
|
|
55 |
static __inline uint32_t |
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 |
x += x * qpel; y += y * qpel; |
x = qpel ? x<<1 : x; |
60 |
|
y = qpel ? y<<1 : y; |
61 |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
62 |
|
|
63 |
x -= pred.x; |
x -= pred.x; |
64 |
y -= pred.y; |
y -= pred.y; |
65 |
|
|
203 |
{ |
{ |
204 |
switch ( ((x&1)<<1) | (y&1) ) { |
switch ( ((x&1)<<1) | (y&1) ) { |
205 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
206 |
|
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
207 |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
208 |
case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); |
default : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); //case 2 |
|
default : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
|
209 |
} |
} |
210 |
} |
} |
211 |
|
|
266 |
|
|
267 |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
268 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
269 |
case 0: // pure halfpel position |
case 3: // x and y in qpel resolution - the "corners" (top left/right and |
270 |
return (uint8_t *) ref1; |
// bottom left/right) during qpel refinement |
271 |
|
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
272 |
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
273 |
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
274 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
275 |
|
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
276 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
277 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
278 |
|
break; |
279 |
|
|
280 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
281 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
282 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
293 |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8); |
294 |
break; |
break; |
295 |
|
|
296 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
case 0: // pure halfpel position |
297 |
// bottom left/right) during qpel refinement |
return (uint8_t *) ref1; |
|
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
|
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
|
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
|
|
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; |
|
298 |
} |
} |
299 |
return Reference; |
return Reference; |
300 |
} |
} |
304 |
static void |
static void |
305 |
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) |
306 |
{ |
{ |
307 |
int t, xc, yc; |
int xc, yc; |
308 |
const uint8_t * Reference; |
const uint8_t * Reference; |
309 |
VECTOR * current; |
VECTOR * current; |
310 |
|
int32_t sad; uint32_t t; |
311 |
|
|
312 |
if ( (x > data->max_dx) | (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
313 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
314 |
|
|
315 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (!data->qpel_precision) { |
|
Reference = Interpolate16x16qpel(x, y, 0, data); |
|
|
xc = x/2; yc = y/2; //for chroma sad |
|
|
current = data->currentQMV; |
|
|
} else { |
|
316 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
317 |
current = data->currentMV; |
current = data->currentMV; |
318 |
xc = x; yc = y; |
xc = x; yc = y; |
319 |
|
} else { // x and y are in 1/4 precision |
320 |
|
Reference = Interpolate16x16qpel(x, y, 0, data); |
321 |
|
xc = x/2; yc = y/2; //for chroma sad |
322 |
|
current = data->currentQMV; |
323 |
} |
} |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
|
324 |
|
|
325 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
326 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
327 |
|
|
328 |
data->temp[0] += (data->lambda16 * t * data->temp[0])>>10; |
sad += (data->lambda16 * t * sad)>>10; |
329 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
330 |
|
|
331 |
if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
332 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
333 |
|
|
334 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
335 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = sad; |
336 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
337 |
*dir = Direction; } |
*dir = Direction; |
338 |
|
} |
339 |
|
|
340 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[1] < data->iMinSAD[1]) { |
341 |
data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; } |
data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y= y; } |
349 |
} |
} |
350 |
|
|
351 |
static void |
static void |
352 |
|
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
353 |
|
{ |
354 |
|
int32_t sad; uint32_t t; |
355 |
|
const uint8_t * Reference; |
356 |
|
|
357 |
|
if ( (x > data->max_dx) || (x < data->min_dx) |
358 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
359 |
|
|
360 |
|
if (!data->qpel_precision) Reference = GetReference(x, y, data); |
361 |
|
else Reference = Interpolate8x8qpel(x, y, 0, 0, data); |
362 |
|
|
363 |
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
364 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
365 |
|
|
366 |
|
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
367 |
|
|
368 |
|
if (sad < *(data->iMinSAD)) { |
369 |
|
*(data->iMinSAD) = sad; |
370 |
|
data->currentMV->x = x; data->currentMV->y = y; |
371 |
|
*dir = Direction; |
372 |
|
} |
373 |
|
} |
374 |
|
|
375 |
|
|
376 |
|
static void |
377 |
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) |
378 |
{ |
{ |
379 |
uint32_t t; |
uint32_t t; |
380 |
const uint8_t * Reference; |
const uint8_t * Reference; |
381 |
|
|
382 |
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 |
383 |
(x > data->max_dx) | (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
384 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
385 |
|
|
386 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
387 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
449 |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
450 |
{ |
{ |
451 |
// maximum speed - for P/B/I decision |
// maximum speed - for P/B/I decision |
452 |
|
int32_t sad; |
453 |
|
|
454 |
if ( (x > data->max_dx) | (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
455 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
456 |
|
|
457 |
data->temp[0] = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth), |
458 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
459 |
|
|
460 |
if (data->temp[0] < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
461 |
*(data->iMinSAD) = data->temp[0]; |
*(data->iMinSAD) = sad; |
462 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
463 |
*dir = Direction; } |
*dir = Direction; |
464 |
|
} |
465 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[1] < data->iMinSAD[1]) { |
466 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
467 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[2] < data->iMinSAD[2]) { |
629 |
} |
} |
630 |
} |
} |
631 |
|
|
|
static void |
|
|
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
|
|
{ |
|
|
int32_t sad; uint32_t t; |
|
|
const uint8_t * Reference; |
|
|
|
|
|
if ( (x > data->max_dx) | (x < data->min_dx) |
|
|
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|
|
|
|
|
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
|
|
else Reference = GetReference(x, y, data); |
|
|
|
|
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
|
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
|
|
|
|
|
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
|
|
|
|
|
if (sad < *(data->iMinSAD)) { |
|
|
*(data->iMinSAD) = sad; |
|
|
data->currentMV->x = x; data->currentMV->y = y; |
|
|
*dir = Direction; } |
|
|
} |
|
|
|
|
632 |
/* CHECK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
633 |
|
|
634 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
1111 |
|
|
1112 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) { |
1113 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
1114 |
(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1115 |
if (Data->iMinSAD[0] <= threshA) break; |
if (Data->iMinSAD[0] <= threshA) break; |
1116 |
} |
} |
1117 |
|
|
1144 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1145 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1146 |
|
|
1147 |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
1148 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
1149 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1150 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1157 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1158 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1159 |
|
|
1160 |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
1161 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
1162 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1163 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1589 |
|
|
1590 |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
1591 |
|
|
1592 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
CheckCandidate(0, 0, 255, &k, Data); |
1593 |
|
|
1594 |
// initial (fast) skip decision |
// initial (fast) skip decision |
1595 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) { |