17 |
#include "../bitstream/vlc_codes.h" |
#include "../bitstream/vlc_codes.h" |
18 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
19 |
#include "motion_inlines.h" |
#include "motion_inlines.h" |
|
#include "../plugins/HVS_base.h" |
|
|
|
|
20 |
|
|
21 |
/* rd = BITS_MULT*bits + LAMBDA*distortion */ |
/* rd = BITS_MULT*bits + LAMBDA*distortion */ |
22 |
#define LAMBDA ( (int)(BITS_MULT*1.0) ) |
#define LAMBDA ( (int)(BITS_MULT*1.0) ) |
31 |
const int block, |
const int block, |
32 |
const uint16_t * scan_table, |
const uint16_t * scan_table, |
33 |
const unsigned int lambda, |
const unsigned int lambda, |
34 |
const uint16_t * mpeg_quant_matrices) |
const uint16_t * mpeg_quant_matrices, |
35 |
|
const unsigned int quant_sq) |
36 |
{ |
{ |
37 |
int sum; |
int sum; |
38 |
int bits; |
int bits; |
67 |
distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t)); |
distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t)); |
68 |
} |
} |
69 |
|
|
70 |
return bits + (lambda*distortion)/(quant*quant); |
return bits + (lambda*distortion)/quant_sq; |
71 |
} |
} |
72 |
|
|
73 |
|
|
80 |
const int block, |
const int block, |
81 |
const uint16_t * scan_table, |
const uint16_t * scan_table, |
82 |
const unsigned int lambda, |
const unsigned int lambda, |
83 |
const uint16_t * mpeg_quant_matrices) |
const uint16_t * mpeg_quant_matrices, |
84 |
|
const unsigned int quant_sq) |
85 |
{ |
{ |
86 |
int sum; |
int sum; |
87 |
int bits; |
int bits; |
116 |
distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t)); |
distortion = sse8_16bit(data, zero_block, 8*sizeof(int16_t)); |
117 |
} |
} |
118 |
|
|
119 |
return bits + (lambda*distortion)/(quant*quant); |
return bits + (lambda*distortion)/quant_sq; |
|
} |
|
|
|
|
|
|
|
|
static void |
|
|
transfer_8to16sub2ro(int16_t * const dct, |
|
|
const uint8_t * const cur, |
|
|
const uint8_t * ref1, |
|
|
const uint8_t * ref2, |
|
|
const uint32_t stride) |
|
|
{ |
|
|
uint32_t i, j; |
|
|
|
|
|
for (j = 0; j < 8; j++) { |
|
|
for (i = 0; i < 8; i++) { |
|
|
uint8_t c = cur[j * stride + i]; |
|
|
int r = (ref1[j * stride + i] + ref2[j * stride + i] + 1) / 2; |
|
|
dct[j * 8 + i] = (int16_t) c - (int16_t) r; |
|
|
} |
|
|
} |
|
120 |
} |
} |
121 |
|
|
122 |
static void |
static void |
124 |
{ |
{ |
125 |
|
|
126 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
127 |
int32_t rd = 3*BITS_MULT; /* note to self: 3 bits minimum, but maybe 4 if it's forward mode */ |
int32_t rd = (3+2)*BITS_MULT; /* 3 bits for mode + 2 for vector (minimum) */ |
128 |
VECTOR * current; |
VECTOR * current; |
129 |
const uint8_t * ptr; |
const uint8_t * ptr; |
130 |
int i, xc, yc; |
int i, xc, yc; |
143 |
xc = x/2; yc = y/2; |
xc = x/2; yc = y/2; |
144 |
} |
} |
145 |
|
|
146 |
rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
rd += BITS_MULT*(d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision)-2); |
147 |
|
|
148 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
149 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
150 |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
151 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
152 |
|
&cbp, i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices, data->quant_sq); |
153 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
154 |
} |
} |
155 |
|
|
160 |
/* chroma U */ |
/* chroma U */ |
161 |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
162 |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
163 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
164 |
|
&cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices, data->quant_sq); |
165 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
166 |
|
|
167 |
/* chroma V */ |
/* chroma V */ |
168 |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); |
169 |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
170 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
171 |
|
&cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices, data->quant_sq); |
172 |
|
|
173 |
if (cbp) rd += BITS_MULT * 7; |
if (cbp) rd += BITS_MULT * 7; |
174 |
|
|
180 |
} |
} |
181 |
} |
} |
182 |
|
|
|
|
|
183 |
static void |
static void |
184 |
CheckCandidateRDDirect(const int x, const int y, SearchData * const data, const unsigned int Direction) |
CheckCandidateRDDirect(const int x, const int y, SearchData * const data, const unsigned int Direction) |
185 |
{ |
{ |
226 |
} |
} |
227 |
|
|
228 |
transfer_8to16sub2ro(in, data->Cur + s, ReferenceF, ReferenceB, data->iEdgedWidth); |
transfer_8to16sub2ro(in, data->Cur + s, ReferenceF, ReferenceB, data->iEdgedWidth); |
229 |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, k, data->scan_table, data->lambda[k], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
230 |
|
&cbp, k, data->scan_table, data->lambda[k], data->mpeg_quant_matrices, data->quant_sq); |
231 |
if (rd > *(data->iMinSAD)) return; |
if (rd > *(data->iMinSAD)) return; |
232 |
} |
} |
233 |
|
|
241 |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
242 |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[4], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[4], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
243 |
transfer_8to16sub2ro(in, data->CurU, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
transfer_8to16sub2ro(in, data->CurU, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
244 |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
245 |
|
&cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices, data->quant_sq); |
246 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
247 |
|
|
248 |
/* chroma V */ |
/* chroma V */ |
249 |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
250 |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[5], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[5], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
251 |
transfer_8to16sub2ro(in, data->CurV, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
transfer_8to16sub2ro(in, data->CurV, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
252 |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP_direct(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
253 |
|
&cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices, data->quant_sq); |
254 |
|
|
255 |
if (cbp) |
if (cbp) |
256 |
rd += BITS_MULT * 6; |
rd += BITS_MULT * 6; |
257 |
if (cbp || x != 0 || y != 0) |
if (cbp || x != 0 || y != 0) |
258 |
rd += BITS_MULT * d_mv_bits(x, y, zeroMV, 1, 0, 0); |
rd += BITS_MULT * d_mv_bits(x, y, zeroMV, 1, 0); |
259 |
|
|
260 |
if (rd < *(data->iMinSAD)) { |
if (rd < *(data->iMinSAD)) { |
261 |
*data->iMinSAD = rd; |
*data->iMinSAD = rd; |
265 |
} |
} |
266 |
} |
} |
267 |
|
|
|
|
|
|
|
|
|
|
|
268 |
static void |
static void |
269 |
CheckCandidateRDInt(const int x, const int y, SearchData * const data, const unsigned int Direction) |
CheckCandidateRDInt(const int x, const int y, SearchData * const data, const unsigned int Direction) |
270 |
{ |
{ |
307 |
xcb = xb/2; ycb = yb/2; |
xcb = xb/2; ycb = yb/2; |
308 |
} |
} |
309 |
|
|
310 |
rd += BITS_MULT * (d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0) |
rd += BITS_MULT * (d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision) |
311 |
+ 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)); |
|
|
|
312 |
|
|
313 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
314 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
315 |
if (rd >= *data->iMinSAD) return; |
if (rd >= *data->iMinSAD) return; |
316 |
transfer_8to16sub2ro(in, data->Cur + s, ReferenceF + s, ReferenceB + s, data->iEdgedWidth); |
transfer_8to16sub2ro(in, data->Cur + s, ReferenceF + s, ReferenceB + s, data->iEdgedWidth); |
317 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, |
318 |
|
i, data->scan_table, data->lambda[i], data->mpeg_quant_matrices, data->quant_sq); |
319 |
} |
} |
320 |
|
|
321 |
/* chroma */ |
/* chroma */ |
328 |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
329 |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[4], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[4], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
330 |
transfer_8to16sub2ro(in, data->CurU, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
transfer_8to16sub2ro(in, data->CurU, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
331 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, |
332 |
|
4, data->scan_table, data->lambda[4], data->mpeg_quant_matrices, data->quant_sq); |
333 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
334 |
|
|
335 |
|
|
337 |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
ReferenceF = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xcf, ycf, data->iEdgedWidth/2, data->rounding); |
338 |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[5], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
ReferenceB = interpolate8x8_switch2(data->RefQ + 16, data->b_RefP[5], 0, 0, xcb, ycb, data->iEdgedWidth/2, data->rounding); |
339 |
transfer_8to16sub2ro(in, data->CurV, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
transfer_8to16sub2ro(in, data->CurV, ReferenceF, ReferenceB, data->iEdgedWidth/2); |
340 |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices); |
rd += Block_CalcBits_BVOP(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, |
341 |
|
5, data->scan_table, data->lambda[5], data->mpeg_quant_matrices, data->quant_sq); |
342 |
|
|
343 |
if (cbp) rd += BITS_MULT * 7; |
if (cbp) rd += BITS_MULT * 7; |
344 |
|
|
362 |
Data->iMinSAD[0] = *best_sad; |
Data->iMinSAD[0] = *best_sad; |
363 |
|
|
364 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
365 |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 1 + Data->qpel, 0); |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 1 + Data->qpel); |
366 |
|
|
367 |
Data->qpel_precision = Data->qpel; |
Data->qpel_precision = Data->qpel; |
368 |
|
|
405 |
Data->iMinSAD[0] = *best_sad; |
Data->iMinSAD[0] = *best_sad; |
406 |
|
|
407 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, |
408 |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 1 + Data->qpel, 0); |
x, y, 4, pParam->width, pParam->height, Data->iFcode, 1 + Data->qpel); |
409 |
|
|
410 |
Data->qpel_precision = Data->qpel; |
Data->qpel_precision = Data->qpel; |
411 |
|
|
447 |
VECTOR * b_predMV, |
VECTOR * b_predMV, |
448 |
const uint32_t MotionFlags, |
const uint32_t MotionFlags, |
449 |
const MBParam * const pParam, |
const MBParam * const pParam, |
450 |
int x, int y) |
int x, int y, |
451 |
|
int best_sad) |
452 |
{ |
{ |
453 |
int mode = MODE_DIRECT, k; |
int mode = MODE_DIRECT, k; |
454 |
int f_rd, b_rd, i_rd, d_rd, best_rd; |
int f_rd, b_rd, i_rd, d_rd, best_rd; |
456 |
const uint32_t iQuant = Data_d->iQuant; |
const uint32_t iQuant = Data_d->iQuant; |
457 |
int i; |
int i; |
458 |
int ref_quant = b_mb->quant; |
int ref_quant = b_mb->quant; |
459 |
|
int no_of_checks = 0; |
460 |
|
|
461 |
int order[4] = {MODE_DIRECT, MODE_FORWARD, MODE_BACKWARD, MODE_INTERPOLATE}; |
int order[4] = {MODE_DIRECT, MODE_FORWARD, MODE_BACKWARD, MODE_INTERPOLATE}; |
462 |
|
|
463 |
Data_d->scan_table = Data_b->scan_table = Data_f->scan_table = Data_i->scan_table |
Data_d->scan_table = Data_b->scan_table = Data_f->scan_table = Data_i->scan_table |
464 |
= /*VopFlags & XVID_VOP_ALTERNATESCAN ? scan_tables[2] : */scan_tables[0]; |
= /*VopFlags & XVID_VOP_ALTERNATESCAN ? scan_tables[2] : */scan_tables[0]; |
465 |
|
*Data_f->cbp = *Data_b->cbp = *Data_i->cbp = *Data_d->cbp = 63; |
466 |
|
|
467 |
f_rd = b_rd = i_rd = d_rd = best_rd = 256*4096; |
f_rd = b_rd = i_rd = d_rd = best_rd = 256*4096; |
468 |
|
|
490 |
} |
} |
491 |
} |
} |
492 |
|
|
493 |
|
for(i = 0; i < 4; i++) |
494 |
|
if (get_sad_for_mode(order[i], Data_d, Data_b, Data_f, Data_i) < 2*best_sad) |
495 |
|
no_of_checks++; |
496 |
|
|
497 |
|
if (no_of_checks > 1) { |
498 |
/* evaluate cost of all modes */ |
/* evaluate cost of all modes */ |
499 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < no_of_checks; i++) { |
500 |
int rd; |
int rd; |
501 |
|
if (2*best_sad < get_sad_for_mode(order[i], Data_d, Data_b, Data_f, Data_i)) |
502 |
|
break; /* further SADs are too big */ |
503 |
|
|
504 |
switch (order[i]) { |
switch (order[i]) { |
505 |
case MODE_DIRECT: |
case MODE_DIRECT: |
511 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
512 |
rd = b_rd = SearchBF_RD(x, y, MotionFlags, pParam, &best_rd, Data_b); |
rd = b_rd = SearchBF_RD(x, y, MotionFlags, pParam, &best_rd, Data_b); |
513 |
break; |
break; |
514 |
|
default: |
515 |
case MODE_INTERPOLATE: |
case MODE_INTERPOLATE: |
516 |
rd = i_rd = SearchInterpolate_RD(x, y, MotionFlags, pParam, &best_rd, Data_i); |
rd = i_rd = SearchInterpolate_RD(x, y, MotionFlags, pParam, &best_rd, Data_i); |
517 |
break; |
break; |
521 |
best_rd = rd; |
best_rd = rd; |
522 |
} |
} |
523 |
} |
} |
524 |
|
} else { |
525 |
|
/* only 1 mode is below the threshold */ |
526 |
|
mode = order[0]; |
527 |
|
best_rd = 0; |
528 |
|
} |
529 |
|
|
530 |
pMB->sad16 = best_rd; |
pMB->sad16 = best_rd; |
531 |
pMB->mode = mode; |
pMB->mode = mode; |
578 |
} |
} |
579 |
pMB->mvs[0] = *Data_f->currentMV; |
pMB->mvs[0] = *Data_f->currentMV; |
580 |
pMB->cbp = *Data_f->cbp; |
pMB->cbp = *Data_f->cbp; |
581 |
|
pMB->b_mvs[0] = *Data_b->currentMV; /* hint for future searches */ |
582 |
break; |
break; |
583 |
|
|
584 |
case MODE_BACKWARD: |
case MODE_BACKWARD: |
594 |
} |
} |
595 |
pMB->b_mvs[0] = *Data_b->currentMV; |
pMB->b_mvs[0] = *Data_b->currentMV; |
596 |
pMB->cbp = *Data_b->cbp; |
pMB->cbp = *Data_b->cbp; |
597 |
|
pMB->mvs[0] = *Data_f->currentMV; /* hint for future searches */ |
598 |
break; |
break; |
599 |
|
|
600 |
|
|