56 |
uint32_t * cbp, |
uint32_t * cbp, |
57 |
const int block, |
const int block, |
58 |
const uint16_t * scan_table, |
const uint16_t * scan_table, |
59 |
const uint16_t * mpeg_quant_matrices) |
const unsigned int lambda, |
60 |
|
const uint16_t * mpeg_quant_matrices, |
61 |
|
const unsigned int quant_sq) |
62 |
{ |
{ |
63 |
int sum; |
int sum; |
64 |
int bits; |
int bits; |
94 |
} |
} |
95 |
|
|
96 |
|
|
97 |
return bits + (LAMBDA*distortion)/(quant*quant); |
return bits + (lambda*distortion)/quant_sq; |
98 |
} |
} |
99 |
|
|
100 |
static __inline unsigned int |
static __inline unsigned int |
111 |
const int quant_type, |
const int quant_type, |
112 |
unsigned int bits[2], |
unsigned int bits[2], |
113 |
unsigned int cbp[2], |
unsigned int cbp[2], |
114 |
const uint16_t * mpeg_quant_matrices) |
unsigned int lambda, |
115 |
|
const uint16_t * mpeg_quant_matrices, |
116 |
|
const unsigned int quant_sq) |
117 |
{ |
{ |
118 |
int direction; |
int direction; |
119 |
int16_t *pCurrent; |
int16_t *pCurrent; |
148 |
/* dc prediction */ |
/* dc prediction */ |
149 |
qcoeff[0] = qcoeff[0] - predictors[0]; |
qcoeff[0] = qcoeff[0] - predictors[0]; |
150 |
|
|
151 |
if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len; |
if (block < 4) bits[1] = bits[0] = dcy_tab[qcoeff[0] + 255].len - 3; /* 3 bits added before (4 times) */ |
152 |
else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len; |
else bits[1] = bits[0] = dcc_tab[qcoeff[0] + 255].len - 2; /* 2 bits added before (2 times)*/ |
153 |
|
|
154 |
/* calc cost before ac prediction */ |
/* calc cost before ac prediction */ |
155 |
bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]); |
bits[0] += coded = CodeCoeffIntra_CalcBits(qcoeff, scan_tables[0]); |
173 |
|
|
174 |
distortion = sse8_16bit(coeff, dqcoeff, 8*sizeof(int16_t)); |
distortion = sse8_16bit(coeff, dqcoeff, 8*sizeof(int16_t)); |
175 |
|
|
176 |
return (LAMBDA*distortion)/(quant*quant); |
return (lambda*distortion)/quant_sq; |
177 |
} |
} |
178 |
|
|
179 |
|
|
183 |
{ |
{ |
184 |
|
|
185 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
186 |
int32_t rd = 0; |
/* minimum nuber of bits INTER can take is 1 (mcbpc) + 2 (cby) + 2 (vector) */ |
187 |
|
int32_t rd = BITS_MULT * (1+2+2); |
188 |
VECTOR * current; |
VECTOR * current; |
189 |
const uint8_t * ptr; |
const uint8_t * ptr; |
190 |
int i, t, xc, yc; |
int i, t, xc, yc; |
206 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
207 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
208 |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
209 |
rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table, data->mpeg_quant_matrices); |
rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, |
210 |
|
data->quant_type, &cbp, i, data->scan_table, data->lambda[i], |
211 |
|
data->mpeg_quant_matrices, data->quant_sq); |
212 |
} |
} |
213 |
|
|
214 |
rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
rd += t = BITS_MULT * (d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision) - 2); |
215 |
|
|
216 |
if (data->temp[0] + t < data->iMinSAD[1]) { |
if (data->temp[0] + t < data->iMinSAD[1]) { |
217 |
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); } |
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); } |
222 |
if (data->temp[3] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
223 |
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); } |
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); } |
224 |
|
|
225 |
rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
rd += BITS_MULT * (xvid_cbpy_tab[15-(cbp>>2)].len - 2); |
226 |
|
|
227 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
228 |
|
|
233 |
/* chroma U */ |
/* chroma U */ |
234 |
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); |
235 |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
236 |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->mpeg_quant_matrices); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
237 |
|
&cbp, 4, data->scan_table, data->lambda[4], |
238 |
|
data->mpeg_quant_matrices, data->quant_sq); |
239 |
if (rd >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
240 |
|
|
241 |
/* chroma V */ |
/* chroma V */ |
242 |
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); |
243 |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
244 |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->mpeg_quant_matrices); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
245 |
|
&cbp, 5, data->scan_table, data->lambda[5], |
246 |
|
data->mpeg_quant_matrices, data->quant_sq); |
247 |
|
|
248 |
rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
rd += BITS_MULT * (mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len - 1); /* one was added before */ |
249 |
|
|
250 |
if (rd < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
251 |
data->iMinSAD[0] = rd; |
data->iMinSAD[0] = rd; |
277 |
} |
} |
278 |
|
|
279 |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
280 |
rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->mpeg_quant_matrices); |
rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, |
281 |
rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
&cbp, 5, data->scan_table, data->lambda[0], |
282 |
|
data->mpeg_quant_matrices, data->quant_sq); |
283 |
|
/* we took 2 bits into account before */ |
284 |
|
rd += BITS_MULT * (d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision) - 2); |
285 |
|
|
286 |
|
|
287 |
if (rd < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
288 |
*data->cbp = cbp; |
*data->cbp = cbp; |
313 |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
314 |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
315 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
316 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); |
317 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
318 |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
319 |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
340 |
/* preparing for qpel-precision search */ |
/* preparing for qpel-precision search */ |
341 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
342 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
343 |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
pParam->width, pParam->height, Data->iFcode, 2); |
344 |
} |
} |
345 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16_RD) { |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16_RD) { |
346 |
if (MotionFlags & XVID_ME_FASTREFINE16) |
if (MotionFlags & XVID_ME_FASTREFINE16) |
366 |
const VECTOR * const backup) |
const VECTOR * const backup) |
367 |
{ |
{ |
368 |
|
|
369 |
unsigned int cbp = 0, bits = 0, t = 0, i; |
unsigned int cbp = 0, t = 0, i; |
370 |
|
|
371 |
|
/* minimum number of bits INTER4V can take is 2 (cbpy) + 3 (mcbpc) + 4*2 (vectors)*/ |
372 |
|
int bits = (2+3+4*2)*BITS_MULT; |
373 |
SearchData Data2, *Data8 = &Data2; |
SearchData Data2, *Data8 = &Data2; |
374 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
375 |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
388 |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
389 |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
390 |
*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; /* copy corresponding cbp bit */ |
*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; /* copy corresponding cbp bit */ |
391 |
|
Data8->lambda[0] = Data->lambda[i]; |
392 |
|
|
393 |
if(Data->qpel) { |
if(Data->qpel) { |
394 |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
395 |
if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, |
if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, |
396 |
Data8->predMV, Data8->iFcode, 0, 0); |
Data8->predMV, Data8->iFcode, 0) - 2; |
397 |
} else { |
} else { |
398 |
Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); |
Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); |
399 |
if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, |
if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, |
400 |
Data8->predMV, Data8->iFcode, 0, 0); |
Data8->predMV, Data8->iFcode, 0) - 2; |
401 |
} |
} |
402 |
|
|
403 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
404 |
pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0); |
pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1); |
405 |
|
|
406 |
*Data8->iMinSAD += BITS_MULT*t; |
*Data8->iMinSAD += BITS_MULT*t; |
407 |
|
|
424 |
if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ |
if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ |
425 |
Data8->qpel_precision = 0; |
Data8->qpel_precision = 0; |
426 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
427 |
pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); |
pParam->width, pParam->height, Data8->iFcode - 1, 1); |
428 |
|
|
429 |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
430 |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255); |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255); |
443 |
|
|
444 |
Data8->qpel_precision = 1; |
Data8->qpel_precision = 1; |
445 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
446 |
pParam->width, pParam->height, Data8->iFcode, 2, 0); |
pParam->width, pParam->height, Data8->iFcode, 2); |
447 |
|
|
448 |
} |
} |
449 |
|
|
498 |
|
|
499 |
} /* end - for all luma blocks */ |
} /* end - for all luma blocks */ |
500 |
|
|
501 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
bits += BITS_MULT * (xvid_cbpy_tab[15-(cbp>>2)].len - 2); /* 2 were added before */ |
502 |
|
|
503 |
/* let's check chroma */ |
/* let's check chroma */ |
504 |
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
507 |
/* chroma U */ |
/* chroma U */ |
508 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
509 |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
510 |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->mpeg_quant_matrices); |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, |
511 |
|
Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq); |
512 |
|
|
513 |
if (bits >= *Data->iMinSAD) return bits; |
if (bits >= *Data->iMinSAD) return bits; |
514 |
|
|
515 |
/* chroma V */ |
/* chroma V */ |
516 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
517 |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
518 |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->mpeg_quant_matrices); |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, |
519 |
|
Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq); |
520 |
|
|
521 |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
bits += BITS_MULT*(mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len - 3); /* 3 were added before */ |
522 |
|
|
523 |
*Data->cbp = cbp; |
*Data->cbp = cbp; |
524 |
return bits; |
return bits; |
529 |
const int x, const int y, const int mb_width) |
const int x, const int y, const int mb_width) |
530 |
{ |
{ |
531 |
unsigned int cbp[2] = {0, 0}, bits[2], i; |
unsigned int cbp[2] = {0, 0}, bits[2], i; |
532 |
unsigned int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
/* minimum number of bits that WILL be coded in intra - mcbpc 5, cby 2 acdc flag - 1 and DC coeffs - 4*3+2*2 */ |
533 |
|
int bits1 = BITS_MULT*(5+2+1+4*3+2*2), bits2 = BITS_MULT*(5+2+1+4*3+2*2); |
534 |
unsigned int distortion = 0; |
unsigned int distortion = 0; |
535 |
|
|
536 |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128; |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128; |
543 |
|
|
544 |
|
|
545 |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff, |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff, |
546 |
predictors[i], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); |
predictors[i], iQuant, Data->quant_type, bits, cbp, |
547 |
|
Data->lambda[i], Data->mpeg_quant_matrices, Data->quant_sq); |
548 |
bits1 += distortion + BITS_MULT * bits[0]; |
bits1 += distortion + BITS_MULT * bits[0]; |
549 |
bits2 += distortion + BITS_MULT * bits[1]; |
bits2 += distortion + BITS_MULT * bits[1]; |
550 |
|
|
552 |
return bits1; |
return bits1; |
553 |
} |
} |
554 |
|
|
555 |
bits1 += BITS_MULT*xvid_cbpy_tab[cbp[0]>>2].len; |
bits1 += BITS_MULT * (xvid_cbpy_tab[cbp[0]>>2].len - 2); /* two bits were added before */ |
556 |
bits2 += BITS_MULT*xvid_cbpy_tab[cbp[1]>>2].len; |
bits2 += BITS_MULT * (xvid_cbpy_tab[cbp[1]>>2].len - 2); |
557 |
|
|
558 |
/*chroma U */ |
/*chroma U */ |
559 |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
560 |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff, |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff, |
561 |
predictors[4], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); |
predictors[4], iQuant, Data->quant_type, bits, cbp, |
562 |
|
Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq); |
563 |
bits1 += distortion + BITS_MULT * bits[0]; |
bits1 += distortion + BITS_MULT * bits[0]; |
564 |
bits2 += distortion + BITS_MULT * bits[1]; |
bits2 += distortion + BITS_MULT * bits[1]; |
565 |
|
|
569 |
/* chroma V */ |
/* chroma V */ |
570 |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
571 |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff, |
distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff, |
572 |
predictors[5], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); |
predictors[5], iQuant, Data->quant_type, bits, cbp, |
573 |
|
Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq); |
574 |
|
|
575 |
bits1 += distortion + BITS_MULT * bits[0]; |
bits1 += distortion + BITS_MULT * bits[0]; |
576 |
bits2 += distortion + BITS_MULT * bits[1]; |
bits2 += distortion + BITS_MULT * bits[1]; |
577 |
|
|
578 |
bits1 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[0] & 3) << 3)].len; |
bits1 += BITS_MULT * (mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[0] & 3) << 3)].len - 5); /* 5 bits were added before */ |
579 |
bits2 += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[1] & 3) << 3)].len; |
bits2 += BITS_MULT * (mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp[1] & 3) << 3)].len - 5); |
580 |
|
|
581 |
*Data->cbp = bits1 <= bits2 ? cbp[0] : cbp[1]; |
*Data->cbp = bits1 <= bits2 ? cbp[0] : cbp[1]; |
582 |
|
|
587 |
static int |
static int |
588 |
findRD_gmc(SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) |
findRD_gmc(SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) |
589 |
{ |
{ |
590 |
int bits = BITS_MULT*1; /* this one is mcsel */ |
/* minimum nubler of bits - 1 (mcbpc) + 2 (cby) + 1 (mcsel) */ |
591 |
|
int bits = BITS_MULT * (1+2+1); |
592 |
unsigned int cbp = 0, i; |
unsigned int cbp = 0, i; |
593 |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
594 |
|
|
595 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
596 |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
597 |
transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); |
transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); |
598 |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, Data->scan_table, Data->mpeg_quant_matrices); |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, |
599 |
|
Data->scan_table, Data->lambda[i], Data->mpeg_quant_matrices, Data->quant_sq); |
600 |
if (bits >= Data->iMinSAD[0]) return bits; |
if (bits >= Data->iMinSAD[0]) return bits; |
601 |
} |
} |
602 |
|
|
603 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
bits += BITS_MULT * (xvid_cbpy_tab[15-(cbp>>2)].len - 2); |
604 |
|
|
605 |
/*chroma U */ |
/*chroma U */ |
606 |
transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
607 |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->mpeg_quant_matrices); |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, |
608 |
|
Data->scan_table, Data->lambda[4], Data->mpeg_quant_matrices, Data->quant_sq); |
609 |
|
|
610 |
if (bits >= Data->iMinSAD[0]) return bits; |
if (bits >= Data->iMinSAD[0]) return bits; |
611 |
|
|
612 |
/* chroma V */ |
/* chroma V */ |
613 |
transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
614 |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->mpeg_quant_matrices); |
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, |
615 |
|
Data->scan_table, Data->lambda[5], Data->mpeg_quant_matrices, Data->quant_sq); |
616 |
|
|
617 |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
bits += BITS_MULT * (mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len - 1); |
618 |
|
|
619 |
*Data->cbp = cbp; |
*Data->cbp = cbp; |
620 |
|
|
639 |
int mcsel = 0; |
int mcsel = 0; |
640 |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
641 |
const uint32_t iQuant = pMB->quant; |
const uint32_t iQuant = pMB->quant; |
|
|
|
642 |
int min_rd, intra_rd, i, cbp; |
int min_rd, intra_rd, i, cbp; |
643 |
VECTOR backup[5], *v; |
VECTOR backup[5], *v; |
644 |
Data->iQuant = iQuant; |
Data->iQuant = iQuant; |
645 |
|
Data->quant_sq = iQuant*iQuant; |
646 |
Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ? |
Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ? |
647 |
scan_tables[2] : scan_tables[0]; |
scan_tables[2] : scan_tables[0]; |
648 |
|
|
654 |
backup[i] = v[i]; |
backup[i] = v[i]; |
655 |
} |
} |
656 |
|
|
657 |
|
for (i = 0; i < 6; i++) { |
658 |
|
/* HVS models, anyone ? */ |
659 |
|
Data->lambda[i] = LAMBDA; |
660 |
|
} |
661 |
|
|
662 |
min_rd = findRD_inter(Data, x, y, pParam, MotionFlags); |
min_rd = findRD_inter(Data, x, y, pParam, MotionFlags); |
663 |
cbp = *Data->cbp; |
cbp = *Data->cbp; |
664 |
|
|
684 |
} |
} |
685 |
} |
} |
686 |
|
|
687 |
|
/* there is no way for INTRA to take less than 24 bits - go to findRD_intra() for calculations */ |
688 |
|
if (min_rd > 24*BITS_MULT) { |
689 |
intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width); |
intra_rd = findRD_intra(Data, pMB, x, y, pParam->mb_width); |
690 |
if (intra_rd < min_rd) { |
if (intra_rd < min_rd) { |
691 |
*Data->iMinSAD = min_rd = intra_rd; |
*Data->iMinSAD = min_rd = intra_rd; |
692 |
mode = MODE_INTRA; |
mode = MODE_INTRA; |
693 |
cbp = *Data->cbp; |
cbp = *Data->cbp; |
694 |
} |
} |
695 |
|
} |
696 |
|
|
697 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
698 |
pMB->cbp = cbp; |
pMB->cbp = cbp; |
699 |
|
|
|
|
|
|
if (Data->rrv) { |
|
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
|
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
|
|
} |
|
|
|
|
700 |
if (mode == MODE_INTER && mcsel == 0) { |
if (mode == MODE_INTER && mcsel == 0) { |
701 |
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]; |
702 |
|
|
758 |
scan_tables[2] : scan_tables[0]; |
scan_tables[2] : scan_tables[0]; |
759 |
|
|
760 |
pMB->mcsel = 0; |
pMB->mcsel = 0; |
761 |
|
Data->iQuant = iQuant; |
762 |
|
Data->quant_sq = iQuant*iQuant; |
763 |
|
|
764 |
|
for (i = 0; i < 6; i++) { |
765 |
|
/* HVS models, anyone ? */ |
766 |
|
Data->lambda[i] = LAMBDA; |
767 |
|
} |
768 |
|
|
769 |
/* INTER <-> INTER4V decision */ |
/* INTER <-> INTER4V decision */ |
770 |
if ((Data->iMinSAD[0] + 75 < Data->iMinSAD[1] + |
if ((Data->iMinSAD[0] + 75 < Data->iMinSAD[1] + |
783 |
/* 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?" */ |
784 |
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
785 |
if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
786 |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) { |
787 |
mode = MODE_NOT_CODED; |
mode = MODE_NOT_CODED; |
788 |
sad = 0; /* Compiler warning */ |
sad = 0; /* Compiler warning */ |
789 |
goto early_out; |
goto early_out; |
814 |
/* 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?" */ |
815 |
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP)) |
816 |
if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
if ( (100*Data->iMinSAD[0])/(pMB->sad16+1) > FINAL_SKIP_THRESH) |
817 |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) { |
if (Data->chroma || xvid_me_SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant)) { |
818 |
mode = MODE_NOT_CODED; |
mode = MODE_NOT_CODED; |
819 |
sad = 0; /* Compiler warning */ |
sad = 0; /* Compiler warning */ |
820 |
goto early_out; |
goto early_out; |
890 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
891 |
|
|
892 |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
|
if (Data->rrv) InterBias *= 4; |
|
893 |
|
|
894 |
if (InterBias < sad) { |
if (InterBias < sad) { |
895 |
int32_t deviation; |
int32_t deviation = dev16(Data->Cur, Data->iEdgedWidth); |
|
if (!Data->rrv) |
|
|
deviation = dev16(Data->Cur, Data->iEdgedWidth); |
|
|
else |
|
|
deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ |
|
|
dev16(Data->Cur+16, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
|
|
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
|
|
|
|
896 |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
897 |
} |
} |
898 |
|
|
948 |
early_out: |
early_out: |
949 |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
950 |
|
|
|
if (Data->rrv) { |
|
|
Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x); |
|
|
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
|
|
} |
|
|
|
|
951 |
if (mode == MODE_INTER && mcsel == 0) { |
if (mode == MODE_INTER && mcsel == 0) { |
952 |
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]; |
953 |
|
|