44 |
#include "motion.h" |
#include "motion.h" |
45 |
#include "sad.h" |
#include "sad.h" |
46 |
#include "../utils/emms.h" |
#include "../utils/emms.h" |
47 |
|
#include "../dct/fdct.h" |
48 |
|
|
49 |
#define INITIAL_SKIP_THRESH (10) |
#define INITIAL_SKIP_THRESH (10) |
50 |
#define FINAL_SKIP_THRESH (50) |
#define FINAL_SKIP_THRESH (50) |
52 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
53 |
|
|
54 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
55 |
(*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); } |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
56 |
|
|
57 |
static __inline uint32_t |
static __inline uint32_t |
58 |
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) |
59 |
{ |
{ |
60 |
int xb, yb; |
int xb, yb; |
61 |
x += x * qpel; y += y * qpel; |
x = qpel ? x<<1 : x; |
62 |
|
y = qpel ? y<<1 : y; |
63 |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); } |
64 |
|
|
65 |
x -= pred.x; |
x -= pred.x; |
66 |
y -= pred.y; |
y -= pred.y; |
67 |
|
|
83 |
return xb + yb; |
return xb + yb; |
84 |
} |
} |
85 |
|
|
86 |
|
static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data) |
87 |
|
{ |
88 |
|
int sad; |
89 |
|
const uint32_t stride = data->iEdgedWidth/2; |
90 |
|
uint8_t * f_refu = data->RefQ, |
91 |
|
* f_refv = data->RefQ + 8, |
92 |
|
* b_refu = data->RefQ + 16, |
93 |
|
* b_refv = data->RefQ + 24; |
94 |
|
|
95 |
|
switch (((fx & 1) << 1) | (fy & 1)) { |
96 |
|
case 0: |
97 |
|
fx = fx / 2; fy = fy / 2; |
98 |
|
f_refu = (uint8_t*)data->RefCU + fy * stride + fx, stride; |
99 |
|
f_refv = (uint8_t*)data->RefCV + fy * stride + fx, stride; |
100 |
|
break; |
101 |
|
case 1: |
102 |
|
fx = fx / 2; fy = (fy - 1) / 2; |
103 |
|
interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
104 |
|
interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
105 |
|
break; |
106 |
|
case 2: |
107 |
|
fx = (fx - 1) / 2; fy = fy / 2; |
108 |
|
interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
109 |
|
interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
110 |
|
break; |
111 |
|
default: |
112 |
|
fx = (fx - 1) / 2; fy = (fy - 1) / 2; |
113 |
|
interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding); |
114 |
|
interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding); |
115 |
|
break; |
116 |
|
} |
117 |
|
|
118 |
|
switch (((bx & 1) << 1) | (by & 1)) { |
119 |
|
case 0: |
120 |
|
bx = bx / 2; by = by / 2; |
121 |
|
b_refu = (uint8_t*)data->b_RefCU + by * stride + bx, stride; |
122 |
|
b_refv = (uint8_t*)data->b_RefCV + by * stride + bx, stride; |
123 |
|
break; |
124 |
|
case 1: |
125 |
|
bx = bx / 2; by = (by - 1) / 2; |
126 |
|
interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
127 |
|
interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
128 |
|
break; |
129 |
|
case 2: |
130 |
|
bx = (bx - 1) / 2; by = by / 2; |
131 |
|
interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
132 |
|
interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
133 |
|
break; |
134 |
|
default: |
135 |
|
bx = (bx - 1) / 2; by = (by - 1) / 2; |
136 |
|
interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding); |
137 |
|
interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding); |
138 |
|
break; |
139 |
|
} |
140 |
|
|
141 |
|
sad = sad8bi(data->CurU, b_refu, f_refu, stride); |
142 |
|
sad += sad8bi(data->CurV, b_refv, f_refv, stride); |
143 |
|
|
144 |
|
return sad; |
145 |
|
} |
146 |
|
|
147 |
|
|
148 |
static int32_t |
static int32_t |
149 |
ChromaSAD(int dx, int dy, const SearchData * const data) |
ChromaSAD(int dx, int dy, const SearchData * const data) |
150 |
{ |
{ |
205 |
{ |
{ |
206 |
switch ( ((x&1)<<1) | (y&1) ) { |
switch ( ((x&1)<<1) | (y&1) ) { |
207 |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth); |
208 |
|
case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); |
209 |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); |
210 |
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); |
|
211 |
} |
} |
212 |
} |
} |
213 |
|
|
268 |
|
|
269 |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data); |
270 |
switch( ((x&1)<<1) + (y&1) ) { |
switch( ((x&1)<<1) + (y&1) ) { |
271 |
case 0: // pure halfpel position |
case 3: // x and y in qpel resolution - the "corners" (top left/right and |
272 |
return (uint8_t *) ref1; |
// bottom left/right) during qpel refinement |
273 |
|
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
274 |
|
ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data); |
275 |
|
ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data); |
276 |
|
interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding); |
277 |
|
interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding); |
278 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding); |
279 |
|
interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding); |
280 |
|
break; |
281 |
|
|
282 |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
case 1: // x halfpel, y qpel - top or bottom during qpel refinement |
283 |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data); |
284 |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8); |
295 |
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); |
296 |
break; |
break; |
297 |
|
|
298 |
default: // x and y in qpel resolution - the "corners" (top left/right and |
case 0: // pure halfpel position |
299 |
// 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; |
|
300 |
} |
} |
301 |
return Reference; |
return Reference; |
302 |
} |
} |
306 |
static void |
static void |
307 |
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) |
308 |
{ |
{ |
309 |
int t, xc, yc; |
int xc, yc; |
310 |
const uint8_t * Reference; |
const uint8_t * Reference; |
311 |
VECTOR * current; |
VECTOR * current; |
312 |
|
int32_t sad; uint32_t t; |
313 |
|
|
314 |
if ( (x > data->max_dx) | (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
315 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
316 |
|
|
317 |
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 { |
|
318 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
319 |
current = data->currentMV; |
current = data->currentMV; |
320 |
xc = x; yc = y; |
xc = x; yc = y; |
321 |
|
} else { // x and y are in 1/4 precision |
322 |
|
Reference = Interpolate16x16qpel(x, y, 0, data); |
323 |
|
xc = x/2; yc = y/2; //for chroma sad |
324 |
|
current = data->currentQMV; |
325 |
} |
} |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
|
326 |
|
|
327 |
data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
328 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
329 |
|
|
330 |
data->temp[0] += (data->lambda16 * t * data->temp[0])>>10; |
sad += (data->lambda16 * t * sad)>>10; |
331 |
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; |
332 |
|
|
333 |
if (data->chroma) data->temp[0] += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
334 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
335 |
|
|
336 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
337 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = sad; |
338 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
339 |
*dir = Direction; } |
*dir = Direction; |
340 |
|
} |
341 |
|
|
342 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[1] < data->iMinSAD[1]) { |
343 |
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; } |
351 |
} |
} |
352 |
|
|
353 |
static void |
static void |
354 |
|
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
355 |
|
{ |
356 |
|
int32_t sad; uint32_t t; |
357 |
|
const uint8_t * Reference; |
358 |
|
|
359 |
|
if ( (x > data->max_dx) || (x < data->min_dx) |
360 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
361 |
|
|
362 |
|
if (!data->qpel_precision) Reference = GetReference(x, y, data); |
363 |
|
else Reference = Interpolate8x8qpel(x, y, 0, 0, data); |
364 |
|
|
365 |
|
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
366 |
|
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
367 |
|
|
368 |
|
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
369 |
|
|
370 |
|
if (sad < *(data->iMinSAD)) { |
371 |
|
*(data->iMinSAD) = sad; |
372 |
|
data->currentMV->x = x; data->currentMV->y = y; |
373 |
|
*dir = Direction; |
374 |
|
} |
375 |
|
} |
376 |
|
|
377 |
|
|
378 |
|
static void |
379 |
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) |
380 |
{ |
{ |
381 |
uint32_t t; |
uint32_t t; |
382 |
const uint8_t * Reference; |
const uint8_t * Reference; |
383 |
|
|
384 |
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 |
385 |
(x > data->max_dx) | (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
386 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
387 |
|
|
388 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
389 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
411 |
static void |
static void |
412 |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
413 |
{ |
{ |
414 |
int32_t sad; |
int32_t sad, xc, yc; |
415 |
const uint8_t * Reference; |
const uint8_t * Reference; |
416 |
uint32_t t; |
uint32_t t; |
417 |
VECTOR * current; |
VECTOR * current; |
424 |
if (data->qpel_precision) { // x and y are in 1/4 precision |
if (data->qpel_precision) { // x and y are in 1/4 precision |
425 |
Reference = Interpolate16x16qpel(x, y, 0, data); |
Reference = Interpolate16x16qpel(x, y, 0, data); |
426 |
current = data->currentQMV; |
current = data->currentQMV; |
427 |
|
xc = x/2; yc = y/2; |
428 |
} else { |
} else { |
429 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
430 |
current = data->currentMV; |
current = data->currentMV; |
431 |
|
xc = x; yc = y; |
432 |
} |
} |
433 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
t = d_mv_bits(x, y, data->predMV, data->iFcode, |
434 |
data->qpel^data->qpel_precision, data->rrv); |
data->qpel^data->qpel_precision, data->rrv); |
436 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
437 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
438 |
|
|
439 |
|
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
440 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
441 |
|
|
442 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
443 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
444 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
445 |
*dir = Direction; } |
*dir = Direction; |
446 |
|
} |
447 |
} |
} |
448 |
|
|
449 |
static void |
static void |
450 |
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) |
451 |
{ |
{ |
452 |
// maximum speed - for P/B/I decision |
// maximum speed - for P/B/I decision |
453 |
|
int32_t sad; |
454 |
|
|
455 |
if ( (x > data->max_dx) | (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
456 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
457 |
|
|
458 |
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), |
459 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp+1); |
460 |
|
|
461 |
if (data->temp[0] < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
462 |
*(data->iMinSAD) = data->temp[0]; |
*(data->iMinSAD) = sad; |
463 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
464 |
*dir = Direction; } |
*dir = Direction; |
465 |
|
} |
466 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[1] < data->iMinSAD[1]) { |
467 |
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; } |
468 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[2] < data->iMinSAD[2]) { |
477 |
static void |
static void |
478 |
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) |
479 |
{ |
{ |
480 |
int32_t sad, xb, yb; |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
481 |
uint32_t t; |
uint32_t t; |
482 |
const uint8_t *ReferenceF, *ReferenceB; |
const uint8_t *ReferenceF, *ReferenceB; |
483 |
VECTOR *current; |
VECTOR *current; |
490 |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
xb = data->currentMV[1].x; yb = data->currentMV[1].y; |
491 |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
ReferenceB = GetReferenceB(xb, yb, 1, data); |
492 |
current = data->currentMV; |
current = data->currentMV; |
493 |
|
xcf = xf; ycf = yf; |
494 |
|
xcb = xb; ycb = yb; |
495 |
} else { |
} else { |
496 |
ReferenceF = Interpolate16x16qpel(xf, yf, 0, data); |
ReferenceF = Interpolate16x16qpel(xf, yf, 0, data); |
497 |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
xb = data->currentQMV[1].x; yb = data->currentQMV[1].y; |
498 |
current = data->currentQMV; |
current = data->currentQMV; |
499 |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
ReferenceB = Interpolate16x16qpel(xb, yb, 1, data); |
500 |
|
xcf = xf/2; ycf = yf/2; |
501 |
|
xcb = xb/2; ycb = yb/2; |
502 |
} |
} |
503 |
|
|
504 |
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) |
507 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
508 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
509 |
|
|
510 |
|
if (data->chroma) sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], |
511 |
|
(ycf >> 1) + roundtab_79[ycf & 0x3], |
512 |
|
(xcb >> 1) + roundtab_79[xcb & 0x3], |
513 |
|
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
514 |
|
|
515 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
516 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
517 |
current->x = xf; current->y = yf; |
current->x = xf; current->y = yf; |
518 |
*dir = Direction; } |
*dir = Direction; |
519 |
|
} |
520 |
} |
} |
521 |
|
|
522 |
static void |
static void |
523 |
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) |
524 |
{ |
{ |
525 |
int32_t sad = 0; |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
526 |
uint32_t k; |
uint32_t k; |
527 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
528 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
541 |
data->directmvB[k].y |
data->directmvB[k].y |
542 |
: mvs.y - data->referencemv[k].y); |
: mvs.y - data->referencemv[k].y); |
543 |
|
|
544 |
if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx ) |
if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx) |
545 |
|| ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy ) |
| (mvs.y > data->max_dy) | (mvs.y < data->min_dy) |
546 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx) |
547 |
|| ( 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; |
548 |
|
|
549 |
|
if (data->qpel) { |
550 |
mvs.x *= 2 - data->qpel; mvs.y *= 2 - data->qpel; |
xcf += mvs.x/2; ycf += mvs.y/2; |
551 |
b_mvs.x *= 2 - data->qpel; b_mvs.y *= 2 - data->qpel; //we move to qpel precision anyway |
xcb += b_mvs.x/2; ycb += b_mvs.y/2; |
552 |
|
} else { |
553 |
|
xcf += mvs.x; ycf += mvs.y; |
554 |
|
xcb += b_mvs.x; ycb += b_mvs.y; |
555 |
|
mvs.x *= 2; mvs.y *= 2; //we move to qpel precision anyway |
556 |
|
b_mvs.x *= 2; b_mvs.y *= 2; |
557 |
|
} |
558 |
|
|
559 |
ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); |
ReferenceF = Interpolate8x8qpel(mvs.x, mvs.y, k, 0, data); |
560 |
ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); |
ReferenceB = Interpolate8x8qpel(b_mvs.x, b_mvs.y, k, 1, data); |
566 |
|
|
567 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
568 |
|
|
569 |
|
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
570 |
|
(ycf >> 3) + roundtab_76[ycf & 0xf], |
571 |
|
(xcb >> 3) + roundtab_76[xcb & 0xf], |
572 |
|
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
573 |
|
|
574 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
575 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
576 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
577 |
*dir = Direction; } |
*dir = Direction; |
578 |
|
} |
579 |
} |
} |
580 |
|
|
581 |
static void |
static void |
582 |
CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
583 |
{ |
{ |
584 |
int32_t sad; |
int32_t sad, xcf, ycf, xcb, ycb; |
585 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
586 |
const uint8_t *ReferenceB; |
const uint8_t *ReferenceB; |
587 |
VECTOR mvs, b_mvs; |
VECTOR mvs, b_mvs; |
598 |
data->directmvB[0].y |
data->directmvB[0].y |
599 |
: mvs.y - data->referencemv[0].y); |
: mvs.y - data->referencemv[0].y); |
600 |
|
|
601 |
if (( mvs.x > data->max_dx ) || ( mvs.x < data->min_dx ) |
if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx) |
602 |
|| ( mvs.y > data->max_dy ) || ( mvs.y < data->min_dy ) |
| (mvs.y > data->max_dy) | (mvs.y < data->min_dy) |
603 |
|| ( b_mvs.x > data->max_dx ) || ( b_mvs.x < data->min_dx ) |
| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx) |
604 |
|| ( 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; |
605 |
|
|
606 |
mvs.x *= 2 - data->qpel; mvs.y *= 2 - data->qpel; |
if (data->qpel) { |
607 |
b_mvs.x *= 2 - data->qpel; b_mvs.y *= 2 - data->qpel; //we move to qpel precision anyway |
xcf = 4*(mvs.x/2); ycf = 4*(mvs.y/2); |
608 |
|
xcb = 4*(b_mvs.x/2); ycb = 4*(b_mvs.y/2); |
609 |
ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data); |
ReferenceF = Interpolate16x16qpel(mvs.x, mvs.y, 0, data); |
610 |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
ReferenceB = Interpolate16x16qpel(b_mvs.x, b_mvs.y, 1, data); |
611 |
|
} else { |
612 |
|
xcf = 4*mvs.x; ycf = 4*mvs.y; |
613 |
|
xcb = 4*b_mvs.x; ycb = 4*b_mvs.y; |
614 |
|
ReferenceF = GetReference(mvs.x, mvs.y, data); |
615 |
|
ReferenceB = GetReferenceB(b_mvs.x, b_mvs.y, 1, data); |
616 |
|
} |
617 |
|
|
618 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
619 |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; |
620 |
|
|
621 |
|
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
622 |
|
(ycf >> 3) + roundtab_76[ycf & 0xf], |
623 |
|
(xcb >> 3) + roundtab_76[xcb & 0xf], |
624 |
|
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
625 |
|
|
626 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
627 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
628 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
629 |
*dir = Direction; } |
*dir = Direction; |
630 |
} |
} |
631 |
|
} |
632 |
|
|
633 |
|
|
634 |
static void |
static void |
635 |
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
636 |
{ |
{ |
|
int32_t sad; uint32_t t; |
|
|
const uint8_t * Reference; |
|
637 |
|
|
638 |
if ( (x > data->max_dx) | (x < data->min_dx) |
static int16_t in[64], coeff[64]; |
639 |
| (y > data->max_dy) | (y < data->min_dy) ) return; |
int32_t bits = 0, sum; |
640 |
|
VECTOR * current; |
641 |
|
const uint8_t * ptr; |
642 |
|
int i, cbp = 0, t, xc, yc; |
643 |
|
|
644 |
if (data->qpel) Reference = Interpolate16x16qpel(x, y, 0, data); |
if ( (x > data->max_dx) || (x < data->min_dx) |
645 |
else Reference = GetReference(x, y, data); |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
646 |
|
|
647 |
sad = sad8(data->Cur, Reference, data->iEdgedWidth); |
if (!data->qpel_precision) { |
648 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel && !data->qpel_precision, 0); |
ptr = GetReference(x, y, data); |
649 |
|
current = data->currentMV; |
650 |
|
xc = x; yc = y; |
651 |
|
} else { // x and y are in 1/4 precision |
652 |
|
ptr = Interpolate16x16qpel(x, y, 0, data); |
653 |
|
current = data->currentQMV; |
654 |
|
xc = x/2; yc = y/2; |
655 |
|
} |
656 |
|
|
657 |
sad += (data->lambda8 * t * (sad+NEIGH_8X8_BIAS))>>10; |
for(i = 0; i < 4; i++) { |
658 |
|
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
659 |
|
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
660 |
|
fdct(in); |
661 |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
662 |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
663 |
|
if (sum > 0) { |
664 |
|
cbp |= 1 << (5 - i); |
665 |
|
bits += data->temp[i] = CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
666 |
|
} else data->temp[i] = 0; |
667 |
|
} |
668 |
|
|
669 |
|
bits += t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
670 |
|
|
671 |
|
if (bits < data->iMinSAD[0]) { // there is still a chance, adding chroma |
672 |
|
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
673 |
|
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
674 |
|
|
675 |
|
//chroma U |
676 |
|
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCU, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
677 |
|
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
678 |
|
fdct(in); |
679 |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
680 |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
681 |
|
if (sum > 0) { |
682 |
|
cbp |= 1 << (5 - 4); |
683 |
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
684 |
|
} |
685 |
|
|
686 |
|
if (bits < data->iMinSAD[0]) { |
687 |
|
//chroma V |
688 |
|
ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCV, 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
689 |
|
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
690 |
|
fdct(in); |
691 |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
692 |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
693 |
|
if (sum > 0) { |
694 |
|
cbp |= 1 << (5 - 5); |
695 |
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
696 |
|
} |
697 |
|
} |
698 |
|
} |
699 |
|
|
700 |
if (sad < *(data->iMinSAD)) { |
bits += cbpy_tab[15-(cbp>>2)].len; |
701 |
*(data->iMinSAD) = sad; |
bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
702 |
data->currentMV->x = x; data->currentMV->y = y; |
|
703 |
*dir = Direction; } |
if (bits < data->iMinSAD[0]) { |
704 |
|
data->iMinSAD[0] = bits; |
705 |
|
current[0].x = x; current[0].y = y; |
706 |
|
*dir = Direction; |
707 |
|
} |
708 |
|
|
709 |
|
if (data->temp[0] + t < data->iMinSAD[1]) { |
710 |
|
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; } |
711 |
|
if (data->temp[1] < data->iMinSAD[2]) { |
712 |
|
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
713 |
|
if (data->temp[2] < data->iMinSAD[3]) { |
714 |
|
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
715 |
|
if (data->temp[3] < data->iMinSAD[4]) { |
716 |
|
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
717 |
|
|
718 |
|
} |
719 |
|
static void |
720 |
|
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
721 |
|
{ |
722 |
|
|
723 |
|
static int16_t in[64], coeff[64]; |
724 |
|
int32_t sum, bits; |
725 |
|
VECTOR * current; |
726 |
|
const uint8_t * ptr; |
727 |
|
int cbp; |
728 |
|
|
729 |
|
if ( (x > data->max_dx) || (x < data->min_dx) |
730 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
731 |
|
|
732 |
|
if (!data->qpel_precision) { |
733 |
|
ptr = GetReference(x, y, data); |
734 |
|
current = data->currentMV; |
735 |
|
} else { // x and y are in 1/4 precision |
736 |
|
ptr = Interpolate8x8qpel(x, y, 0, 0, data); |
737 |
|
current = data->currentQMV; |
738 |
|
} |
739 |
|
|
740 |
|
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
741 |
|
fdct(in); |
742 |
|
if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16); |
743 |
|
else sum = quant4_inter(coeff, in, data->lambda16); |
744 |
|
if (sum > 0) { |
745 |
|
bits = CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
746 |
|
cbp = 1; |
747 |
|
} else cbp = bits = 0; |
748 |
|
|
749 |
|
bits += sum = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
750 |
|
|
751 |
|
if (bits < data->iMinSAD[0]) { |
752 |
|
data->temp[0] = cbp; |
753 |
|
data->iMinSAD[0] = bits; |
754 |
|
current[0].x = x; current[0].y = y; |
755 |
|
*dir = Direction; |
756 |
|
} |
757 |
} |
} |
758 |
|
|
759 |
/* CHECK_CANDIATE FUNCTIONS END */ |
/* CHECK_CANDIATE FUNCTIONS END */ |
971 |
uint32_t mb_width = pParam->mb_width; |
uint32_t mb_width = pParam->mb_width; |
972 |
uint32_t mb_height = pParam->mb_height; |
uint32_t mb_height = pParam->mb_height; |
973 |
const uint32_t iEdgedWidth = pParam->edged_width; |
const uint32_t iEdgedWidth = pParam->edged_width; |
974 |
|
const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->global_flags); |
975 |
|
|
976 |
uint32_t x, y; |
uint32_t x, y; |
977 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
978 |
int32_t InterBias, quant = current->quant, sad00; |
int32_t quant = current->quant, sad00; |
979 |
|
|
980 |
// some pre-initialized thingies for SearchP |
// some pre-initialized thingies for SearchP |
981 |
int32_t temp[8]; |
int32_t temp[8]; |
992 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
993 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
994 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
995 |
Data.chroma = current->global_flags & ( PMV_CHROMA16 | PMV_CHROMA8 ); |
Data.chroma = MotionFlags & PMV_CHROMA16; |
996 |
Data.rrv = current->global_flags & XVID_REDUCED; |
Data.rrv = current->global_flags & XVID_REDUCED; |
997 |
|
|
998 |
if ((current->global_flags & XVID_REDUCED)) { |
if ((current->global_flags & XVID_REDUCED)) { |
999 |
mb_width = (pParam->width + 31) / 32; |
mb_width = (pParam->width + 31) / 32; |
1000 |
mb_height = (pParam->height + 31) / 32; |
mb_height = (pParam->height + 31) / 32; |
1001 |
Data.qpel = Data.chroma = 0; |
Data.qpel = 0; |
1002 |
} |
} |
1003 |
|
|
1004 |
Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose) |
Data.RefQ = pRefV->u; // a good place, also used in MC (for similar purpose) |
1050 |
} |
} |
1051 |
|
|
1052 |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1053 |
y, current->motion_flags, pMB->quant, |
y, MotionFlags, current->global_flags, pMB->quant, |
1054 |
&Data, pParam, pMBs, reference->mbs, |
&Data, pParam, pMBs, reference->mbs, |
1055 |
current->global_flags & XVID_INTER4V, pMB); |
current->global_flags & XVID_INTER4V, pMB); |
1056 |
|
|
1057 |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
/* final skip decision, a.k.a. "the vector you found, really that good?" */ |
1058 |
if (!(current->global_flags & XVID_GMC)) { |
if (!(current->global_flags & XVID_GMC)) { |
1059 |
if ( (pMB->dquant == NO_CHANGE) && (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) |
if ( pMB->dquant == NO_CHANGE && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) { |
1060 |
&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1)) ) |
if (!(current->global_flags & XVID_MODEDECISION_BITS)) { |
1061 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) ) |
1062 |
|
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) |
1063 |
SkipMacroblockP(pMB, sad00); |
SkipMacroblockP(pMB, sad00); |
1064 |
continue; |
} else { // BITS mode decision |
1065 |
} |
if (pMB->sad16 > 10) |
1066 |
} |
SkipMacroblockP(pMB, sad00); // more than 10 bits would be used for this MB - skip |
1067 |
|
|
|
/* finally, intra decision */ |
|
|
|
|
|
InterBias = MV16_INTER_BIAS; |
|
|
if (pMB->quant > 8) InterBias += 100 * (pMB->quant - 8); // to make high quants work |
|
|
if (y != 0) |
|
|
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
|
|
if (x != 0) |
|
|
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
|
|
|
|
|
if (Data.chroma) InterBias += 50; // to compensate bigger SAD |
|
|
if (Data.rrv) InterBias *= 4; |
|
|
|
|
|
if (InterBias < pMB->sad16) { |
|
|
int32_t deviation; |
|
|
if (!Data.rrv) |
|
|
deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 16, iEdgedWidth); |
|
|
else { |
|
|
deviation = dev16(pCurrent->y + (x + y * iEdgedWidth) * 32, iEdgedWidth) |
|
|
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16, iEdgedWidth) |
|
|
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * iEdgedWidth, iEdgedWidth) |
|
|
+ dev16(pCurrent->y + (x + y * iEdgedWidth) * 32 + 16 * (iEdgedWidth+1), iEdgedWidth); |
|
|
} |
|
|
if (deviation < (pMB->sad16 - InterBias)) { |
|
|
if (++iIntra >= iLimit) return 1; |
|
|
SkipMacroblockP(pMB, 0); //same thing |
|
|
pMB->mode = MODE_INTRA; |
|
1068 |
} |
} |
1069 |
} |
} |
1070 |
} |
} |
1071 |
|
if (pMB->mode == MODE_INTRA) |
1072 |
|
if (++iIntra > iLimit) return 1; |
1073 |
|
} |
1074 |
} |
} |
1075 |
|
|
1076 |
if (current->global_flags & XVID_GMC ) /* GMC only for S(GMC)-VOPs */ |
if (current->global_flags & XVID_GMC ) /* GMC only for S(GMC)-VOPs */ |
1077 |
{ |
{ |
1078 |
current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV); |
current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV); |
1079 |
} |
} |
|
|
|
1080 |
return 0; |
return 0; |
1081 |
} |
} |
1082 |
|
|
1140 |
} |
} |
1141 |
} |
} |
1142 |
|
|
1143 |
|
static int |
1144 |
|
ModeDecision(const uint32_t iQuant, SearchData * const Data, |
1145 |
|
int inter4v, |
1146 |
|
MACROBLOCK * const pMB, |
1147 |
|
const MACROBLOCK * const pMBs, |
1148 |
|
const int x, const int y, |
1149 |
|
const MBParam * const pParam, |
1150 |
|
const uint32_t MotionFlags, |
1151 |
|
const uint32_t GlobalFlags) |
1152 |
|
{ |
1153 |
|
|
1154 |
|
int mode = MODE_INTER; |
1155 |
|
|
1156 |
|
if (!(GlobalFlags & XVID_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision |
1157 |
|
int intra = 0; |
1158 |
|
int sad; |
1159 |
|
int InterBias = MV16_INTER_BIAS; |
1160 |
|
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1161 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) { |
1162 |
|
mode = 0; //inter |
1163 |
|
sad = Data->iMinSAD[0]; |
1164 |
|
} else { |
1165 |
|
mode = MODE_INTER4V; |
1166 |
|
sad = Data->iMinSAD[1] + Data->iMinSAD[2] + |
1167 |
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant; |
1168 |
|
Data->iMinSAD[0] = sad; |
1169 |
|
} |
1170 |
|
|
1171 |
|
/* intra decision */ |
1172 |
|
|
1173 |
|
if (iQuant > 8) InterBias += 100 * (iQuant - 8); // to make high quants work |
1174 |
|
if (y != 0) |
1175 |
|
if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80; |
1176 |
|
if (x != 0) |
1177 |
|
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
1178 |
|
|
1179 |
|
if (Data->chroma) InterBias += 50; // to compensate bigger SAD |
1180 |
|
if (Data->rrv) InterBias *= 4; |
1181 |
|
|
1182 |
|
if (InterBias < pMB->sad16) { |
1183 |
|
int32_t deviation; |
1184 |
|
if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth); |
1185 |
|
else deviation = dev16(Data->Cur, Data->iEdgedWidth) + |
1186 |
|
dev16(Data->Cur+8, Data->iEdgedWidth) + |
1187 |
|
dev16(Data->Cur + 8*Data->iEdgedWidth, Data->iEdgedWidth) + |
1188 |
|
dev16(Data->Cur+8+8*Data->iEdgedWidth, Data->iEdgedWidth); |
1189 |
|
|
1190 |
|
if (deviation < (sad - InterBias)) return MODE_INTRA;// intra |
1191 |
|
} |
1192 |
|
return mode; |
1193 |
|
|
1194 |
|
} else { |
1195 |
|
|
1196 |
|
int bits, intra, i; |
1197 |
|
VECTOR backup[5], *v; |
1198 |
|
Data->lambda16 = iQuant; |
1199 |
|
Data->lambda8 = pParam->m_quant_type; |
1200 |
|
|
1201 |
|
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1202 |
|
for (i = 0; i < 5; i++) { |
1203 |
|
Data->iMinSAD[i] = 256*4096; |
1204 |
|
backup[i] = v[i]; |
1205 |
|
} |
1206 |
|
|
1207 |
|
bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags); |
1208 |
|
if (bits == 0) return MODE_INTER; // quick stop |
1209 |
|
|
1210 |
|
if (inter4v) { |
1211 |
|
int inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1212 |
|
if (inter4v < bits) { Data->iMinSAD[0] = bits = inter4v; mode = MODE_INTER4V; } |
1213 |
|
} |
1214 |
|
|
1215 |
|
|
1216 |
|
intra = CountMBBitsIntra(Data); |
1217 |
|
|
1218 |
|
if (intra < bits) { *Data->iMinSAD = bits = intra; return MODE_INTRA; } |
1219 |
|
|
1220 |
|
return mode; |
1221 |
|
} |
1222 |
|
} |
1223 |
|
|
1224 |
static void |
static void |
1225 |
SearchP(const IMAGE * const pRef, |
SearchP(const IMAGE * const pRef, |
1226 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1230 |
const int x, |
const int x, |
1231 |
const int y, |
const int y, |
1232 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
1233 |
|
const uint32_t GlobalFlags, |
1234 |
const uint32_t iQuant, |
const uint32_t iQuant, |
1235 |
SearchData * const Data, |
SearchData * const Data, |
1236 |
const MBParam * const pParam, |
const MBParam * const pParam, |
1280 |
Data->iMinSAD[3] = pMB->sad8[2]; |
Data->iMinSAD[3] = pMB->sad8[2]; |
1281 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
1282 |
|
|
1283 |
if (x | y) { |
if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) || (x | y)) { |
1284 |
threshA = Data->temp[0]; // that's when we keep this SAD atm |
threshA = Data->temp[0]; // that's where we keep this SAD atm |
1285 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
1286 |
else if (threshA > 1024) threshA = 1024; |
else if (threshA > 1024) threshA = 1024; |
1287 |
} else threshA = 512; |
} else |
1288 |
|
threshA = 512; |
1289 |
|
|
1290 |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height, |
1291 |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
1299 |
|
|
1300 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) { |
1301 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!(mask = make_mask(pmv, i)) ) continue; |
1302 |
(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
1303 |
if (Data->iMinSAD[0] <= threshA) break; |
if (Data->iMinSAD[0] <= threshA) break; |
1304 |
} |
} |
1305 |
|
|
1306 |
if ((Data->iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
1307 |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) && |
1308 |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) |
(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) { |
1309 |
inter4v = 0; |
if (!(GlobalFlags & XVID_MODEDECISION_BITS)) inter4v = 0; } |
1310 |
else { |
else { |
1311 |
|
|
1312 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1314 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1315 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1316 |
|
|
1317 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
1318 |
|
|
1319 |
/* extended search, diamond starting in 0,0 and in prediction. |
/* extended search, diamond starting in 0,0 and in prediction. |
1320 |
note that this search is/might be done in halfpel positions, |
note that this search is/might be done in halfpel positions, |
1326 |
if (Data->rrv) { |
if (Data->rrv) { |
1327 |
startMV.x = RRV_MV_SCALEUP(startMV.x); |
startMV.x = RRV_MV_SCALEUP(startMV.x); |
1328 |
startMV.y = RRV_MV_SCALEUP(startMV.y); |
startMV.y = RRV_MV_SCALEUP(startMV.y); |
1329 |
} else |
} |
|
if (!(MotionFlags & PMV_HALFPELREFINE16)) // who's gonna use extsearch and no halfpel? |
|
|
startMV.x = EVEN(startMV.x); startMV.y = EVEN(startMV.y); |
|
1330 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1331 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1332 |
|
|
1333 |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
1334 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
1335 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1336 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1337 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
1338 |
} |
} |
1339 |
|
|
1340 |
backupMV = Data->currentMV[0]; |
backupMV = Data->currentMV[0]; |
1341 |
if (MotionFlags & PMV_HALFPELREFINE16 && !Data->rrv) startMV.x = startMV.y = 1; |
startMV.x = startMV.y = 1; |
|
else startMV.x = startMV.y = 0; |
|
1342 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1343 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1344 |
|
|
1345 |
(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
1346 |
(*MainSearchPtr)(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
1347 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1348 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1349 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
1351 |
} |
} |
1352 |
} |
} |
1353 |
|
|
1354 |
if (MotionFlags & PMV_HALFPELREFINE16) SubpelRefine(Data); |
if (MotionFlags & PMV_HALFPELREFINE16) |
1355 |
|
if ((!(MotionFlags & HALFPELREFINE16_BITS)) || Data->iMinSAD[0] < 200*(int)iQuant) |
1356 |
|
SubpelRefine(Data); |
1357 |
|
|
1358 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1359 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors |
1360 |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
1361 |
} |
} |
1362 |
|
|
1363 |
if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE16) { |
if (MotionFlags & PMV_QUARTERPELREFINE16) |
1364 |
|
if ((!(MotionFlags & QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) { |
1365 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1366 |
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, |
1367 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
1369 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1370 |
} |
} |
1371 |
|
|
1372 |
if (Data->iMinSAD[0] < (int32_t)iQuant * 30) inter4v = 0; |
if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0; |
1373 |
if (inter4v) { |
|
1374 |
|
if (inter4v && (!(GlobalFlags & XVID_MODEDECISION_BITS) || |
1375 |
|
(!(MotionFlags & QUARTERPELREFINE8_BITS)) || (!(MotionFlags & HALFPELREFINE8_BITS)) || |
1376 |
|
((!(MotionFlags & EXTSEARCH_BITS)) && (!(MotionFlags&PMV_EXTSEARCH8)) ))) { |
1377 |
|
// if decision is BITS-based and all refinement steps will be done in BITS domain, there is no reason to call this loop |
1378 |
|
|
1379 |
SearchData Data8; |
SearchData Data8; |
1380 |
memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data |
memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data |
1381 |
|
|
1384 |
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); |
1385 |
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); |
1386 |
|
|
1387 |
if (Data->chroma) { |
if ((Data->chroma) && (!(GlobalFlags & XVID_MODEDECISION_BITS))) { |
1388 |
|
// chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, there is no reason to compute it |
1389 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
1390 |
const int div = 1 + Data->qpel; |
const int div = 1 + Data->qpel; |
1391 |
const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs; |
const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs; |
1400 |
} |
} |
1401 |
} |
} |
1402 |
|
|
1403 |
|
inter4v = ModeDecision(iQuant, Data, inter4v, pMB, pMBs, x, y, pParam, MotionFlags, GlobalFlags); |
1404 |
|
|
1405 |
if (Data->rrv) { |
if (Data->rrv) { |
1406 |
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
1407 |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1408 |
} |
} |
1409 |
|
|
1410 |
if (!(inter4v) || |
if (inter4v == MODE_INTER) { |
|
(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
|
|
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) { |
|
|
// INTER MODE |
|
1411 |
pMB->mode = MODE_INTER; |
pMB->mode = MODE_INTER; |
1412 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
1413 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
1421 |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x; |
1422 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1423 |
} |
} |
1424 |
} else { |
|
1425 |
// INTER4V MODE; all other things are already set in Search8 |
} else if (inter4v == MODE_INTER4V) { |
1426 |
pMB->mode = MODE_INTER4V; |
pMB->mode = MODE_INTER4V; |
1427 |
pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + |
pMB->sad16 = Data->iMinSAD[0]; |
1428 |
Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant; |
} else { // INTRA mode |
1429 |
|
SkipMacroblockP(pMB, 0); // not skip, but similar enough |
1430 |
|
pMB->mode = MODE_INTRA; |
1431 |
} |
} |
1432 |
|
|
1433 |
} |
} |
1434 |
|
|
1435 |
static void |
static void |
1476 |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
1477 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1478 |
|
|
1479 |
if (MotionFlags & PMV_EXTSEARCH8) { |
if (MotionFlags & PMV_EXTSEARCH8 && (!(MotionFlags & EXTSEARCH_BITS))) { |
1480 |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD |
1481 |
|
|
1482 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1484 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1485 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1486 |
|
|
1487 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255); |
1488 |
|
|
1489 |
if(*(Data->iMinSAD) < temp_sad) { |
if(*(Data->iMinSAD) < temp_sad) { |
1490 |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector |
1577 |
|
|
1578 |
/* search backward or forward */ |
/* search backward or forward */ |
1579 |
static void |
static void |
1580 |
SearchBF( const uint8_t * const pRef, |
SearchBF( const IMAGE * const pRef, |
1581 |
const uint8_t * const pRefH, |
const uint8_t * const pRefH, |
1582 |
const uint8_t * const pRefV, |
const uint8_t * const pRefV, |
1583 |
const uint8_t * const pRefHV, |
const uint8_t * const pRefHV, |
1599 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
1600 |
Data->iFcode = iFcode; |
Data->iFcode = iFcode; |
1601 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1602 |
|
Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache |
1603 |
|
|
1604 |
Data->Ref = pRef + (x + y * Data->iEdgedWidth) * 16; |
Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16; |
1605 |
Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16; |
Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16; |
1606 |
Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16; |
1607 |
Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16; |
Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16; |
1608 |
|
Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8; |
1609 |
|
Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8; |
1610 |
|
|
1611 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1612 |
|
|
1631 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1632 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1633 |
|
|
1634 |
(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
1635 |
|
|
1636 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1637 |
|
|
1746 |
Data->bRefH = b_RefH + k; |
Data->bRefH = b_RefH + k; |
1747 |
Data->bRefV = b_RefV + k; |
Data->bRefV = b_RefV + k; |
1748 |
Data->bRefHV = b_RefHV + k; |
Data->bRefHV = b_RefHV + k; |
1749 |
|
Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1750 |
|
Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1751 |
|
Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8; |
1752 |
|
Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8; |
1753 |
|
|
1754 |
k = Data->qpel ? 4 : 2; |
k = Data->qpel ? 4 : 2; |
1755 |
Data->max_dx = k * (pParam->width - x * 16); |
Data->max_dx = k * (pParam->width - x * 16); |
1785 |
|
|
1786 |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
1787 |
|
|
1788 |
(*CheckCandidate)(0, 0, 255, &k, Data); |
CheckCandidate(0, 0, 255, &k, Data); |
1789 |
|
|
1790 |
// initial (fast) skip decision |
// initial (fast) skip decision |
1791 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * 2) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (2 + Data->chroma?1:0)) { |
1792 |
//possible skip - checking chroma |
//possible skip |
1793 |
|
if (Data->chroma) { |
1794 |
|
pMB->mode = MODE_DIRECT_NONE_MV; |
1795 |
|
return *Data->iMinSAD; // skip. |
1796 |
|
} else { |
1797 |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
SkipDecisionB(pCur, f_Ref, b_Ref, pMB, x, y, Data); |
1798 |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
if (pMB->mode == MODE_DIRECT_NONE_MV) return *Data->iMinSAD; // skip. |
1799 |
} |
} |
1800 |
|
} |
1801 |
|
|
1802 |
skip_sad = *Data->iMinSAD; |
skip_sad = *Data->iMinSAD; |
1803 |
|
|
1808 |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1809 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1810 |
|
|
1811 |
(*MainSearchPtr)(0, 0, Data, 255); |
MainSearchPtr(0, 0, Data, 255); |
1812 |
|
|
1813 |
SubpelRefine(Data); |
SubpelRefine(Data); |
1814 |
|
|
1847 |
} |
} |
1848 |
|
|
1849 |
static void |
static void |
1850 |
SearchInterpolate(const uint8_t * const f_Ref, |
SearchInterpolate(const IMAGE * const f_Ref, |
1851 |
const uint8_t * const f_RefH, |
const uint8_t * const f_RefH, |
1852 |
const uint8_t * const f_RefV, |
const uint8_t * const f_RefV, |
1853 |
const uint8_t * const f_RefHV, |
const uint8_t * const f_RefHV, |
1854 |
const uint8_t * const b_Ref, |
const IMAGE * const b_Ref, |
1855 |
const uint8_t * const b_RefH, |
const uint8_t * const b_RefH, |
1856 |
const uint8_t * const b_RefV, |
const uint8_t * const b_RefV, |
1857 |
const uint8_t * const b_RefHV, |
const uint8_t * const b_RefHV, |
1879 |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode; |
1880 |
|
|
1881 |
i = (x + y * fData->iEdgedWidth) * 16; |
i = (x + y * fData->iEdgedWidth) * 16; |
1882 |
bData.bRef = fData->Ref = f_Ref + i; |
bData.bRef = fData->Ref = f_Ref->y + i; |
1883 |
bData.bRefH = fData->RefH = f_RefH + i; |
bData.bRefH = fData->RefH = f_RefH + i; |
1884 |
bData.bRefV = fData->RefV = f_RefV + i; |
bData.bRefV = fData->RefV = f_RefV + i; |
1885 |
bData.bRefHV = fData->RefHV = f_RefHV + i; |
bData.bRefHV = fData->RefHV = f_RefHV + i; |
1886 |
bData.Ref = fData->bRef = b_Ref + i; |
bData.Ref = fData->bRef = b_Ref->y + i; |
1887 |
bData.RefH = fData->bRefH = b_RefH + i; |
bData.RefH = fData->bRefH = b_RefH + i; |
1888 |
bData.RefV = fData->bRefV = b_RefV + i; |
bData.RefV = fData->bRefV = b_RefV + i; |
1889 |
bData.RefHV = fData->bRefHV = b_RefHV + i; |
bData.RefHV = fData->bRefHV = b_RefHV + i; |
1890 |
|
bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1891 |
|
bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1892 |
|
bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1893 |
|
bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1894 |
|
|
1895 |
|
|
1896 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1897 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
2008 |
int32_t iMinSAD; |
int32_t iMinSAD; |
2009 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
2010 |
VECTOR currentQMV[3]; |
VECTOR currentQMV[3]; |
2011 |
|
int32_t temp[8]; |
2012 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
2013 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2014 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
2015 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2016 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant]; |
|
Data.chroma = frame->quant; |
|
2017 |
Data.qpel = pParam->m_quarterpel; |
Data.qpel = pParam->m_quarterpel; |
2018 |
Data.rounding = 0; |
Data.rounding = 0; |
2019 |
|
Data.chroma = frame->motion_flags & PMV_CHROMA8; |
2020 |
|
Data.temp = temp; |
2021 |
|
|
2022 |
Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose) |
Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose) |
2023 |
// note: i==horizontal, j==vertical |
// note: i==horizontal, j==vertical |
2037 |
} |
} |
2038 |
|
|
2039 |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16; |
2040 |
|
Data.CurU = frame->image.u + (j * Data.iEdgedWidth/2 + i) * 8; |
2041 |
|
Data.CurV = frame->image.v + (j * Data.iEdgedWidth/2 + i) * 8; |
2042 |
pMB->quant = frame->quant; |
pMB->quant = frame->quant; |
2043 |
|
|
2044 |
/* direct search comes first, because it (1) checks for SKIP-mode |
/* direct search comes first, because it (1) checks for SKIP-mode |
2057 |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; } |
2058 |
|
|
2059 |
// forward search |
// forward search |
2060 |
SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
SearchBF(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
2061 |
&frame->image, i, j, |
&frame->image, i, j, |
2062 |
frame->motion_flags, |
frame->motion_flags, |
2063 |
frame->fcode, pParam, |
frame->fcode, pParam, |
2065 |
MODE_FORWARD, &Data); |
MODE_FORWARD, &Data); |
2066 |
|
|
2067 |
// backward search |
// backward search |
2068 |
SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
SearchBF(b_ref, b_refH->y, b_refV->y, b_refHV->y, |
2069 |
&frame->image, i, j, |
&frame->image, i, j, |
2070 |
frame->motion_flags, |
frame->motion_flags, |
2071 |
frame->bcode, pParam, |
frame->bcode, pParam, |
2073 |
MODE_BACKWARD, &Data); |
MODE_BACKWARD, &Data); |
2074 |
|
|
2075 |
// interpolate search comes last, because it uses data from forward and backward as prediction |
// interpolate search comes last, because it uses data from forward and backward as prediction |
2076 |
SearchInterpolate(f_ref->y, f_refH->y, f_refV->y, f_refHV->y, |
SearchInterpolate(f_ref, f_refH->y, f_refV->y, f_refHV->y, |
2077 |
b_ref->y, b_refH->y, b_refV->y, b_refHV->y, |
b_ref, b_refH->y, b_refV->y, b_refHV->y, |
2078 |
&frame->image, |
&frame->image, |
2079 |
i, j, |
i, j, |
2080 |
frame->fcode, frame->bcode, |
frame->fcode, frame->bcode, |
2235 |
} |
} |
2236 |
} |
} |
2237 |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2); |
2238 |
if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP; |
// if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP; |
2239 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2240 |
emms(); |
emms(); |
2241 |
return B_VOP; |
return B_VOP; |
2420 |
|
|
2421 |
return gmc; |
return gmc; |
2422 |
} |
} |
2423 |
|
|
2424 |
|
// functions which perform BITS-based search/bitcount |
2425 |
|
|
2426 |
|
static int |
2427 |
|
CountMBBitsInter(SearchData * const Data, |
2428 |
|
const MACROBLOCK * const pMBs, const int x, const int y, |
2429 |
|
const MBParam * const pParam, |
2430 |
|
const uint32_t MotionFlags) |
2431 |
|
{ |
2432 |
|
int i, iDirection; |
2433 |
|
int32_t bsad[5]; |
2434 |
|
|
2435 |
|
CheckCandidate = CheckCandidateBits16; |
2436 |
|
|
2437 |
|
if (Data->qpel) { |
2438 |
|
for(i = 0; i < 5; i++) { |
2439 |
|
Data->currentMV[i].x = Data->currentQMV[i].x/2; |
2440 |
|
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
2441 |
|
} |
2442 |
|
Data->qpel_precision = 1; |
2443 |
|
CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
2444 |
|
|
2445 |
|
//checking if this vector is perfect. if it is, we stop. |
2446 |
|
if (Data->temp[0] == 0 && Data->temp[1] == 0 && Data->temp[2] == 0 && Data->temp[3] == 0) |
2447 |
|
return 0; //quick stop |
2448 |
|
|
2449 |
|
if (MotionFlags & (HALFPELREFINE16_BITS | EXTSEARCH_BITS)) { //we have to prepare for halfpixel-precision search |
2450 |
|
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
2451 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2452 |
|
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
2453 |
|
Data->qpel_precision = 0; |
2454 |
|
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
2455 |
|
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
2456 |
|
} |
2457 |
|
|
2458 |
|
} else { // not qpel |
2459 |
|
|
2460 |
|
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
2461 |
|
//checking if this vector is perfect. if it is, we stop. |
2462 |
|
if (Data->temp[0] == 0 && Data->temp[1] == 0 && Data->temp[2] == 0 && Data->temp[3] == 0) { |
2463 |
|
return 0; //inter |
2464 |
|
} |
2465 |
|
} |
2466 |
|
|
2467 |
|
if (MotionFlags&EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
2468 |
|
|
2469 |
|
if (MotionFlags&HALFPELREFINE16_BITS) SubpelRefine(Data); |
2470 |
|
|
2471 |
|
if (Data->qpel) { |
2472 |
|
if (MotionFlags&(EXTSEARCH_BITS | HALFPELREFINE16_BITS)) { // there was halfpel-precision search |
2473 |
|
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
2474 |
|
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // we have found a better match |
2475 |
|
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
2476 |
|
} |
2477 |
|
|
2478 |
|
// preparing for qpel-precision search |
2479 |
|
Data->qpel_precision = 1; |
2480 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
2481 |
|
pParam->width, pParam->height, Data->iFcode, 1, 0); |
2482 |
|
} |
2483 |
|
if (MotionFlags&QUARTERPELREFINE16_BITS) SubpelRefine(Data); |
2484 |
|
} |
2485 |
|
|
2486 |
|
if (MotionFlags&CHECKPREDICTION_BITS) { //let's check vector equal to prediction |
2487 |
|
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
2488 |
|
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
2489 |
|
CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
2490 |
|
} |
2491 |
|
return Data->iMinSAD[0]; |
2492 |
|
} |
2493 |
|
|
2494 |
|
|
2495 |
|
static int |
2496 |
|
CountMBBitsInter4v(const SearchData * const Data, |
2497 |
|
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
2498 |
|
const int x, const int y, |
2499 |
|
const MBParam * const pParam, const uint32_t MotionFlags, |
2500 |
|
const VECTOR * const backup) |
2501 |
|
{ |
2502 |
|
|
2503 |
|
int cbp = 0, bits = 0, t = 0, i, iDirection; |
2504 |
|
SearchData Data2, *Data8 = &Data2; |
2505 |
|
int sumx = 0, sumy = 0; |
2506 |
|
int16_t in[64], coeff[64]; |
2507 |
|
|
2508 |
|
memcpy(Data8, Data, sizeof(SearchData)); |
2509 |
|
CheckCandidate = CheckCandidateBits8; |
2510 |
|
|
2511 |
|
for (i = 0; i < 4; i++) { |
2512 |
|
Data8->iMinSAD = Data->iMinSAD + i + 1; |
2513 |
|
Data8->currentMV = Data->currentMV + i + 1; |
2514 |
|
Data8->currentQMV = Data->currentQMV + i + 1; |
2515 |
|
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2516 |
|
Data8->Ref = Data->Ref + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2517 |
|
Data8->RefH = Data->RefH + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2518 |
|
Data8->RefV = Data->RefV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2519 |
|
Data8->RefHV = Data->RefHV + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2520 |
|
|
2521 |
|
if(Data->qpel) { |
2522 |
|
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2523 |
|
if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, |
2524 |
|
Data8->predMV, Data8->iFcode, 0, 0); |
2525 |
|
} else { |
2526 |
|
Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2527 |
|
if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, |
2528 |
|
Data8->predMV, Data8->iFcode, 0, 0); |
2529 |
|
} |
2530 |
|
|
2531 |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
2532 |
|
pParam->width, pParam->height, Data8->iFcode, Data8->qpel, 0); |
2533 |
|
|
2534 |
|
*Data8->iMinSAD += t; |
2535 |
|
|
2536 |
|
Data8->qpel_precision = Data8->qpel; |
2537 |
|
// checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) |
2538 |
|
if (Data8->qpel) { |
2539 |
|
if (!(Data8->currentQMV->x == backup[i+1].x && Data8->currentQMV->y == backup[i+1].y)) |
2540 |
|
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
2541 |
|
} else { |
2542 |
|
if (!(Data8->currentMV->x == backup[i+1].x && Data8->currentMV->y == backup[i+1].y)) |
2543 |
|
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
2544 |
|
} |
2545 |
|
|
2546 |
|
if (Data8->qpel) { |
2547 |
|
if (MotionFlags&HALFPELREFINE8_BITS || (MotionFlags&PMV_EXTSEARCH8 && MotionFlags&EXTSEARCH_BITS)) { // halfpixel motion search follows |
2548 |
|
int32_t s = *Data8->iMinSAD; |
2549 |
|
Data8->currentMV->x = Data8->currentQMV->x/2; |
2550 |
|
Data8->currentMV->y = Data8->currentQMV->y/2; |
2551 |
|
Data8->qpel_precision = 0; |
2552 |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
2553 |
|
pParam->width, pParam->height, Data8->iFcode - 1, 0, 0); |
2554 |
|
|
2555 |
|
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
2556 |
|
CheckCandidateBits8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8); |
2557 |
|
|
2558 |
|
if (MotionFlags & PMV_EXTSEARCH8 && MotionFlags & EXTSEARCH_BITS) |
2559 |
|
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
2560 |
|
|
2561 |
|
if (MotionFlags & HALFPELREFINE8_BITS) SubpelRefine(Data8); |
2562 |
|
|
2563 |
|
if(s > *Data8->iMinSAD) { //we have found a better match |
2564 |
|
Data8->currentQMV->x = 2*Data8->currentMV->x; |
2565 |
|
Data8->currentQMV->y = 2*Data8->currentMV->y; |
2566 |
|
} |
2567 |
|
|
2568 |
|
Data8->qpel_precision = 1; |
2569 |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
2570 |
|
pParam->width, pParam->height, Data8->iFcode, 1, 0); |
2571 |
|
|
2572 |
|
} |
2573 |
|
if (MotionFlags & QUARTERPELREFINE8_BITS) SubpelRefine(Data8); |
2574 |
|
|
2575 |
|
} else // not qpel |
2576 |
|
if (MotionFlags & HALFPELREFINE8_BITS) SubpelRefine(Data8); //halfpel mode, halfpel refinement |
2577 |
|
|
2578 |
|
//checking vector equal to predicion |
2579 |
|
if (i != 0 && MotionFlags & CHECKPREDICTION_BITS) { |
2580 |
|
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
2581 |
|
if (!(Data8->predMV.x == v->x && Data8->predMV.y == v->y)) |
2582 |
|
CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
2583 |
|
} |
2584 |
|
|
2585 |
|
bits += *Data8->iMinSAD; |
2586 |
|
if (bits >= Data->iMinSAD[0]) break; // no chances for INTER4V |
2587 |
|
|
2588 |
|
// MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else |
2589 |
|
if(Data->qpel) { |
2590 |
|
pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; |
2591 |
|
pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; |
2592 |
|
pMB->qmvs[i] = *Data8->currentQMV; |
2593 |
|
sumx += Data8->currentQMV->x/2; |
2594 |
|
sumy += Data8->currentQMV->y/2; |
2595 |
|
} else { |
2596 |
|
pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x; |
2597 |
|
pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y; |
2598 |
|
sumx += Data8->currentMV->x; |
2599 |
|
sumy += Data8->currentMV->y; |
2600 |
|
} |
2601 |
|
pMB->mvs[i] = *Data8->currentMV; |
2602 |
|
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
2603 |
|
if (Data8->temp[0]) cbp |= 1 << (5 - i); |
2604 |
|
} |
2605 |
|
|
2606 |
|
if (bits < *Data->iMinSAD) { // there is still a chance for inter4v mode. let's check chroma |
2607 |
|
const uint8_t * ptr; |
2608 |
|
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
2609 |
|
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
2610 |
|
|
2611 |
|
//chroma U |
2612 |
|
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCU, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2613 |
|
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
2614 |
|
fdct(in); |
2615 |
|
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
2616 |
|
else i = quant4_inter(coeff, in, Data->lambda16); |
2617 |
|
if (i > 0) { |
2618 |
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
2619 |
|
cbp |= 1 << (5 - 4); |
2620 |
|
} |
2621 |
|
|
2622 |
|
if (bits < *Data->iMinSAD) { // still possible |
2623 |
|
//chroma V |
2624 |
|
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCV, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2625 |
|
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
2626 |
|
fdct(in); |
2627 |
|
if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16); |
2628 |
|
else i = quant4_inter(coeff, in, Data->lambda16); |
2629 |
|
if (i > 0) { |
2630 |
|
bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
2631 |
|
cbp |= 1 << (5 - 5); |
2632 |
|
} |
2633 |
|
bits += cbpy_tab[15-(cbp>>2)].len; |
2634 |
|
bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
2635 |
|
} |
2636 |
|
} |
2637 |
|
|
2638 |
|
return bits; |
2639 |
|
} |
2640 |
|
|
2641 |
|
|
2642 |
|
static int |
2643 |
|
CountMBBitsIntra(const SearchData * const Data) |
2644 |
|
{ |
2645 |
|
int bits = 1; //this one is ac/dc prediction flag. always 1. |
2646 |
|
int cbp = 0, i, t, dc = 0, b_dc = 1024; |
2647 |
|
const uint32_t iQuant = Data->lambda16; |
2648 |
|
int16_t in[64], coeff[64]; |
2649 |
|
|
2650 |
|
for(i = 0; i < 4; i++) { |
2651 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 1); |
2652 |
|
|
2653 |
|
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2654 |
|
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
2655 |
|
fdct(in); |
2656 |
|
b_dc = dc; |
2657 |
|
dc = in[0]; |
2658 |
|
in[0] -= b_dc; |
2659 |
|
if (Data->lambda8 == 0) quant_intra_c(coeff, in, iQuant, iDcScaler); |
2660 |
|
else quant4_intra_c(coeff, in, iQuant, iDcScaler); |
2661 |
|
|
2662 |
|
b_dc = dc; |
2663 |
|
dc = coeff[0]; |
2664 |
|
if (i != 0) coeff[0] -= b_dc; |
2665 |
|
|
2666 |
|
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;; |
2667 |
|
Data->temp[i] = t; |
2668 |
|
if (t != 0) cbp |= 1 << (5 - i); |
2669 |
|
if (bits >= Data->iMinSAD[0]) break; |
2670 |
|
} |
2671 |
|
|
2672 |
|
if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma |
2673 |
|
uint32_t iDcScaler = get_dc_scaler(iQuant, 0); |
2674 |
|
//chroma U |
2675 |
|
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
2676 |
|
fdct(in); |
2677 |
|
in[0] -= 1024; |
2678 |
|
if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler); |
2679 |
|
else quant4_intra(coeff, in, iQuant, iDcScaler); |
2680 |
|
|
2681 |
|
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
2682 |
|
if (t != 0) cbp |= 1 << (5 - 4); |
2683 |
|
Data->temp[4] = t; |
2684 |
|
|
2685 |
|
if (bits < Data->iMinSAD[0]) { |
2686 |
|
//chroma V |
2687 |
|
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
2688 |
|
fdct(in); |
2689 |
|
in[0] -= 1024; |
2690 |
|
if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler); |
2691 |
|
else quant4_intra(coeff, in, iQuant, iDcScaler); |
2692 |
|
|
2693 |
|
bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len; |
2694 |
|
if (t != 0) cbp |= 1 << (5 - 5); |
2695 |
|
|
2696 |
|
Data->temp[5] = t; |
2697 |
|
|
2698 |
|
bits += t = cbpy_tab[cbp>>2].len; |
2699 |
|
Data->temp[6] = t; |
2700 |
|
|
2701 |
|
bits += t = mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
2702 |
|
Data->temp[7] = t; |
2703 |
|
|
2704 |
|
} |
2705 |
|
} |
2706 |
|
|
2707 |
|
return bits; |
2708 |
|
} |