1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* motion estimation |
* - Motion Estimation related code - |
5 |
* |
* |
6 |
* This program is an implementation of a part of one or more MPEG-4 |
* Copyright(C) 2002 Christoph Lampert <gruel@web.de> |
7 |
* Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
* 2002 Michael Militzer <michael@xvid.org> |
8 |
* to use this software module in hardware or software products are |
* 2002-2003 Radoslaw Czyz <xvid@syskin.cjb.net> |
|
* advised that its use may infringe existing patents or copyrights, and |
|
|
* any such use would be at such party's own risk. The original |
|
|
* developer of this software module and his/her company, and subsequent |
|
|
* editors and their companies, will have no liability for use of this |
|
|
* software or modifications or derivatives thereof. |
|
9 |
* |
* |
10 |
* This program is free software; you can redistribute it and/or modify |
* This program is free software; you can redistribute it and/or modify |
11 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
19 |
* |
* |
20 |
* You should have received a copy of the GNU General Public License |
* You should have received a copy of the GNU General Public License |
21 |
* along with this program; if not, write to the Free Software |
* along with this program; if not, write to the Free Software |
22 |
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
|
* |
24 |
|
* $Id: motion_est.c,v 1.58.2.30 2003-08-28 11:14:04 syskin Exp $ |
25 |
* |
* |
26 |
*************************************************************************/ |
****************************************************************************/ |
27 |
|
|
28 |
#include <assert.h> |
#include <assert.h> |
29 |
#include <stdio.h> |
#include <stdio.h> |
40 |
#include "motion_est.h" |
#include "motion_est.h" |
41 |
#include "motion.h" |
#include "motion.h" |
42 |
#include "sad.h" |
#include "sad.h" |
43 |
|
#include "gmc.h" |
44 |
#include "../utils/emms.h" |
#include "../utils/emms.h" |
45 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
46 |
|
|
70 |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
#define MAX_CHROMA_SAD_FOR_SKIP (22) |
71 |
|
|
72 |
#define CHECK_CANDIDATE(X,Y,D) { \ |
#define CHECK_CANDIDATE(X,Y,D) { \ |
73 |
CheckCandidate((X),(Y), (D), &iDirection, data ); } |
CheckCandidate((X),(Y), data, (D) ); } |
74 |
|
|
75 |
|
|
76 |
/***************************************************************************** |
/***************************************************************************** |
77 |
* Code |
* Code |
109 |
{ |
{ |
110 |
int sad; |
int sad; |
111 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
112 |
uint8_t * f_refu = data->RefQ, |
uint8_t *f_refu, *f_refv, *b_refu, *b_refv; |
113 |
* f_refv = data->RefQ + 8, |
|
114 |
* b_refu = data->RefQ + 16, |
const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = { |
115 |
* b_refv = data->RefQ + 24; |
NULL, |
116 |
|
interpolate8x8_halfpel_v, |
117 |
|
interpolate8x8_halfpel_h, |
118 |
|
interpolate8x8_halfpel_hv |
119 |
|
}; |
120 |
|
|
121 |
int offset = (fx>>1) + (fy>>1)*stride; |
int offset = (fx>>1) + (fy>>1)*stride; |
122 |
|
int filter = ((fx & 1) << 1) | (fy & 1); |
123 |
|
|
124 |
switch (((fx & 1) << 1) | (fy & 1)) { |
if (filter != 0) { |
125 |
case 0: |
f_refu = data->RefQ; |
126 |
|
f_refv = data->RefQ + 8; |
127 |
|
interpolate8x8_halfpel[filter](f_refu, data->RefP[4] + offset, stride, data->rounding); |
128 |
|
interpolate8x8_halfpel[filter](f_refv, data->RefP[5] + offset, stride, data->rounding); |
129 |
|
} else { |
130 |
f_refu = (uint8_t*)data->RefP[4] + offset; |
f_refu = (uint8_t*)data->RefP[4] + offset; |
131 |
f_refv = (uint8_t*)data->RefP[5] + offset; |
f_refv = (uint8_t*)data->RefP[5] + offset; |
|
break; |
|
|
case 1: |
|
|
interpolate8x8_halfpel_v(f_refu, data->RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_v(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
|
case 2: |
|
|
interpolate8x8_halfpel_h(f_refu, data->RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_h(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
|
default: |
|
|
interpolate8x8_halfpel_hv(f_refu, data->RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_hv(f_refv, data->RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
132 |
} |
} |
133 |
|
|
134 |
offset = (bx>>1) + (by>>1)*stride; |
offset = (bx>>1) + (by>>1)*stride; |
135 |
switch (((bx & 1) << 1) | (by & 1)) { |
filter = ((bx & 1) << 1) | (by & 1); |
136 |
case 0: |
|
137 |
|
if (filter != 0) { |
138 |
|
b_refu = data->RefQ + 16; |
139 |
|
b_refv = data->RefQ + 24; |
140 |
|
interpolate8x8_halfpel[filter](b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
141 |
|
interpolate8x8_halfpel[filter](b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
142 |
|
} else { |
143 |
b_refu = (uint8_t*)data->b_RefP[4] + offset; |
b_refu = (uint8_t*)data->b_RefP[4] + offset; |
144 |
b_refv = (uint8_t*)data->b_RefP[5] + offset; |
b_refv = (uint8_t*)data->b_RefP[5] + offset; |
|
break; |
|
|
case 1: |
|
|
interpolate8x8_halfpel_v(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_v(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
|
case 2: |
|
|
interpolate8x8_halfpel_h(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_h(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
|
default: |
|
|
interpolate8x8_halfpel_hv(b_refu, data->b_RefP[4] + offset, stride, data->rounding); |
|
|
interpolate8x8_halfpel_hv(b_refv, data->b_RefP[5] + offset, stride, data->rounding); |
|
|
break; |
|
145 |
} |
} |
146 |
|
|
147 |
sad = sad8bi(data->CurU, b_refu, f_refu, stride); |
sad = sad8bi(data->CurU, b_refu, f_refu, stride); |
156 |
int sad; |
int sad; |
157 |
const uint32_t stride = data->iEdgedWidth/2; |
const uint32_t stride = data->iEdgedWidth/2; |
158 |
int offset = (dx>>1) + (dy>>1)*stride; |
int offset = (dx>>1) + (dy>>1)*stride; |
159 |
|
int next = 1; |
160 |
|
|
161 |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */ |
if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; /* it has been checked recently */ |
162 |
data->temp[5] = dx; data->temp[6] = dy; /* backup */ |
data->temp[5] = dx; data->temp[6] = dy; /* backup */ |
167 |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
sad += sad8(data->CurV, data->RefP[5] + offset, stride); |
168 |
break; |
break; |
169 |
case 1: |
case 1: |
170 |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + stride, stride); |
next = stride; |
|
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + stride, stride); |
|
|
break; |
|
171 |
case 2: |
case 2: |
172 |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + 1, stride); |
sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + next, stride); |
173 |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + 1, stride); |
sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + next, stride); |
174 |
break; |
break; |
175 |
default: |
default: |
176 |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding); |
290 |
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); |
291 |
break; |
break; |
292 |
|
|
293 |
|
|
294 |
default: /* pure halfpel position */ |
default: /* pure halfpel position */ |
295 |
return (uint8_t *) ref1; |
return (uint8_t *) ref1; |
296 |
} |
} |
300 |
/* CHECK_CANDIATE FUNCTIONS START */ |
/* CHECK_CANDIATE FUNCTIONS START */ |
301 |
|
|
302 |
static void |
static void |
303 |
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 SearchData * const data, const int Direction) |
304 |
{ |
{ |
305 |
int xc, yc; |
int xc, yc; |
306 |
const uint8_t * Reference; |
const uint8_t * Reference; |
320 |
current = data->currentQMV; |
current = data->currentQMV; |
321 |
} |
} |
322 |
|
|
323 |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp); |
324 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); |
325 |
|
|
326 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
327 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
328 |
|
|
329 |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma && sad < data->iMinSAD[0]) |
330 |
|
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
331 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
332 |
|
|
333 |
if (sad < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
334 |
data->iMinSAD[0] = sad; |
data->iMinSAD[0] = sad; |
335 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
336 |
*dir = Direction; |
*data->dir = Direction; |
337 |
} |
} |
338 |
|
|
339 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[0] < data->iMinSAD[1]) { |
340 |
data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y = y; } |
data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; } |
341 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
342 |
data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
343 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
344 |
data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
345 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
346 |
data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
347 |
} |
} |
348 |
|
|
349 |
static void |
static void |
350 |
CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidate8(const int x, const int y, const SearchData * const data, const int Direction) |
351 |
{ |
{ |
352 |
int32_t sad; uint32_t t; |
int32_t sad; uint32_t t; |
353 |
const uint8_t * Reference; |
const uint8_t * Reference; |
372 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
373 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
374 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
375 |
*dir = Direction; |
*data->dir = Direction; |
376 |
} |
} |
377 |
} |
} |
378 |
|
|
379 |
static void |
static void |
380 |
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 SearchData * const data, const int Direction) |
381 |
{ |
{ |
382 |
uint32_t t; |
uint32_t t; |
383 |
const uint8_t * Reference; |
const uint8_t * Reference; |
384 |
|
int sad; |
385 |
|
|
386 |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */ |
387 |
(x > data->max_dx) || (x < data->min_dx) |
(x > data->max_dx) || (x < data->min_dx) |
390 |
Reference = GetReference(x, y, data); |
Reference = GetReference(x, y, data); |
391 |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1); |
392 |
|
|
393 |
data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1); |
sad = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp); |
394 |
|
|
395 |
data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10; |
sad += (data->lambda16 * t * sad) >> 10; |
396 |
data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10; |
data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10; |
397 |
|
|
398 |
if (data->temp[0] < data->iMinSAD[0]) { |
if (sad < data->iMinSAD[0]) { |
399 |
data->iMinSAD[0] = data->temp[0]; |
data->iMinSAD[0] = sad; |
400 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
401 |
*dir = Direction; } |
*data->dir = Direction; |
402 |
|
} |
403 |
|
|
404 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[0] < data->iMinSAD[1]) { |
405 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
406 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
407 |
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
408 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
409 |
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
410 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
411 |
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
412 |
} |
} |
413 |
|
|
414 |
static void |
static void |
415 |
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 SearchData * const data, const int Direction) |
416 |
{ |
{ |
417 |
int32_t sad, xc, yc; |
int32_t sad, xc, yc; |
418 |
const uint8_t * Reference; |
const uint8_t * Reference; |
439 |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
440 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
441 |
|
|
442 |
if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
if (data->chroma && sad < *data->iMinSAD) |
443 |
|
sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
444 |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
(yc >> 1) + roundtab_79[yc & 0x3], data); |
445 |
|
|
446 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
447 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
448 |
current->x = x; current->y = y; |
current->x = x; current->y = y; |
449 |
*dir = Direction; |
*data->dir = Direction; |
450 |
|
} |
451 |
|
} |
452 |
|
|
453 |
|
static void |
454 |
|
CheckCandidate16I(const int x, const int y, const SearchData * const data, const int Direction) |
455 |
|
{ |
456 |
|
int sad; |
457 |
|
// int xc, yc; |
458 |
|
const uint8_t * Reference; |
459 |
|
// VECTOR * current; |
460 |
|
|
461 |
|
if ( (x > data->max_dx) || ( x < data->min_dx) |
462 |
|
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
463 |
|
|
464 |
|
Reference = GetReference(x, y, data); |
465 |
|
// xc = x; yc = y; |
466 |
|
|
467 |
|
sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); |
468 |
|
// sad += d_mv_bits(x, y, data->predMV, data->iFcode, 0, 0); |
469 |
|
|
470 |
|
/* if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3], |
471 |
|
(yc >> 1) + roundtab_79[yc & 0x3], data); |
472 |
|
*/ |
473 |
|
|
474 |
|
if (sad < data->iMinSAD[0]) { |
475 |
|
data->iMinSAD[0] = sad; |
476 |
|
data->currentMV[0].x = x; data->currentMV[0].y = y; |
477 |
|
*data->dir = Direction; |
478 |
} |
} |
479 |
} |
} |
480 |
|
|
481 |
static void |
static void |
482 |
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 SearchData * const data, const int Direction) |
483 |
{ |
{ |
484 |
/* maximum speed - for P/B/I decision */ |
/* maximum speed - for P/B/I decision */ |
485 |
int32_t sad; |
int32_t sad; |
487 |
if ( (x > data->max_dx) || (x < data->min_dx) |
if ( (x > data->max_dx) || (x < data->min_dx) |
488 |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
|| (y > data->max_dy) || (y < data->min_dy) ) return; |
489 |
|
|
490 |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*(data->iEdgedWidth), |
sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*((int)data->iEdgedWidth), |
491 |
data->iEdgedWidth, data->temp+1); |
data->iEdgedWidth, data->temp); |
492 |
|
|
493 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
494 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
495 |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
data->currentMV[0].x = x; data->currentMV[0].y = y; |
496 |
*dir = Direction; |
*data->dir = Direction; |
497 |
} |
} |
498 |
if (data->temp[1] < data->iMinSAD[1]) { |
if (data->temp[0] < data->iMinSAD[1]) { |
499 |
data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; } |
500 |
if (data->temp[2] < data->iMinSAD[2]) { |
if (data->temp[1] < data->iMinSAD[2]) { |
501 |
data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; } |
502 |
if (data->temp[3] < data->iMinSAD[3]) { |
if (data->temp[2] < data->iMinSAD[3]) { |
503 |
data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; } |
504 |
if (data->temp[4] < data->iMinSAD[4]) { |
if (data->temp[3] < data->iMinSAD[4]) { |
505 |
data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; } |
506 |
|
|
507 |
} |
} |
508 |
|
|
509 |
static void |
static void |
510 |
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 SearchData * const data, const int Direction) |
511 |
{ |
{ |
512 |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
int32_t sad, xb, yb, xcf, ycf, xcb, ycb; |
513 |
uint32_t t; |
uint32_t t; |
540 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
541 |
sad += (data->lambda16 * t * sad)>>10; |
sad += (data->lambda16 * t * sad)>>10; |
542 |
|
|
543 |
if (data->chroma) sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], |
if (data->chroma && sad < *data->iMinSAD) |
544 |
|
sad += ChromaSAD2((xcf >> 1) + roundtab_79[xcf & 0x3], |
545 |
(ycf >> 1) + roundtab_79[ycf & 0x3], |
(ycf >> 1) + roundtab_79[ycf & 0x3], |
546 |
(xcb >> 1) + roundtab_79[xcb & 0x3], |
(xcb >> 1) + roundtab_79[xcb & 0x3], |
547 |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
(ycb >> 1) + roundtab_79[ycb & 0x3], data); |
549 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
550 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
551 |
current->x = xf; current->y = yf; |
current->x = xf; current->y = yf; |
552 |
*dir = Direction; |
*data->dir = Direction; |
553 |
} |
} |
554 |
} |
} |
555 |
|
|
556 |
static void |
static void |
557 |
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 SearchData * const data, const int Direction) |
558 |
{ |
{ |
559 |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0; |
560 |
uint32_t k; |
uint32_t k; |
601 |
|
|
602 |
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; |
603 |
|
|
604 |
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
if (data->chroma && sad < *data->iMinSAD) |
605 |
|
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
606 |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
607 |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
608 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
610 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
611 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
612 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
613 |
*dir = Direction; |
*data->dir = Direction; |
614 |
} |
} |
615 |
} |
} |
616 |
|
|
617 |
static void |
static void |
618 |
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 SearchData * const data, const int Direction) |
619 |
{ |
{ |
620 |
int32_t sad, xcf, ycf, xcb, ycb; |
int32_t sad, xcf, ycf, xcb, ycb; |
621 |
const uint8_t *ReferenceF; |
const uint8_t *ReferenceF; |
654 |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); |
655 |
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; |
656 |
|
|
657 |
if (data->chroma) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
if (data->chroma && sad < *data->iMinSAD) |
658 |
|
sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], |
659 |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
(ycf >> 3) + roundtab_76[ycf & 0xf], |
660 |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
(xcb >> 3) + roundtab_76[xcb & 0xf], |
661 |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
(ycb >> 3) + roundtab_76[ycb & 0xf], data); |
663 |
if (sad < *(data->iMinSAD)) { |
if (sad < *(data->iMinSAD)) { |
664 |
*(data->iMinSAD) = sad; |
*(data->iMinSAD) = sad; |
665 |
data->currentMV->x = x; data->currentMV->y = y; |
data->currentMV->x = x; data->currentMV->y = y; |
666 |
*dir = Direction; |
*data->dir = Direction; |
667 |
} |
} |
668 |
} |
} |
669 |
|
|
670 |
|
|
671 |
static void |
static void |
672 |
CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD16(const int x, const int y, const SearchData * const data, const int Direction) |
673 |
{ |
{ |
674 |
|
|
675 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
676 |
int32_t bits = 0; |
int32_t rd = 0; |
677 |
VECTOR * current; |
VECTOR * current; |
678 |
const uint8_t * ptr; |
const uint8_t * ptr; |
679 |
int i, cbp = 0, t, xc, yc; |
int i, cbp = 0, t, xc, yc; |
694 |
for(i = 0; i < 4; i++) { |
for(i = 0; i < 4; i++) { |
695 |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); |
696 |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); |
697 |
bits += data->temp[i] = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, i); |
rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i); |
698 |
} |
} |
699 |
|
|
700 |
bits += 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, 0); |
701 |
|
|
702 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
if (data->temp[0] + t < data->iMinSAD[1]) { |
703 |
|
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); } |
704 |
|
if (data->temp[1] < data->iMinSAD[2]) { |
705 |
|
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); } |
706 |
|
if (data->temp[2] < data->iMinSAD[3]) { |
707 |
|
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); } |
708 |
|
if (data->temp[3] < data->iMinSAD[4]) { |
709 |
|
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); } |
710 |
|
|
711 |
if (bits >= data->iMinSAD[0]) return; |
rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
712 |
|
|
713 |
|
if (rd >= data->iMinSAD[0]) return; |
714 |
|
|
715 |
/* chroma */ |
/* chroma */ |
716 |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
xc = (xc >> 1) + roundtab_79[xc & 0x3]; |
717 |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
yc = (yc >> 1) + roundtab_79[yc & 0x3]; |
718 |
|
|
719 |
/* chroma U */ |
/* chroma U */ |
720 |
ptr = interpolate8x8_switch2(data->RefQ + 64, 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); |
721 |
transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); |
722 |
bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 4); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4); |
723 |
if (bits >= data->iMinSAD[0]) return; |
if (rd >= data->iMinSAD[0]) return; |
724 |
|
|
725 |
/* chroma V */ |
/* chroma V */ |
726 |
ptr = interpolate8x8_switch2(data->RefQ + 64, 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); |
727 |
transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2); |
transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); |
728 |
bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5); |
rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
729 |
|
|
730 |
bits += 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; |
731 |
|
|
732 |
if (bits < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
733 |
data->iMinSAD[0] = bits; |
data->iMinSAD[0] = rd; |
734 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
735 |
*dir = Direction; |
*data->dir = Direction; |
736 |
|
*data->cbp = cbp; |
737 |
} |
} |
|
|
|
|
if (data->temp[0] + t < data->iMinSAD[1]) { |
|
|
data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; } |
|
|
if (data->temp[1] < data->iMinSAD[2]) { |
|
|
data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; } |
|
|
if (data->temp[2] < data->iMinSAD[3]) { |
|
|
data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; } |
|
|
if (data->temp[3] < data->iMinSAD[4]) { |
|
|
data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; } |
|
|
|
|
738 |
} |
} |
739 |
|
|
740 |
static void |
static void |
741 |
CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data) |
CheckCandidateRD8(const int x, const int y, const SearchData * const data, const int Direction) |
742 |
{ |
{ |
743 |
|
|
744 |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; |
745 |
int32_t bits; |
int32_t rd; |
746 |
VECTOR * current; |
VECTOR * current; |
747 |
const uint8_t * ptr; |
const uint8_t * ptr; |
748 |
int cbp = 0; |
int cbp = 0; |
759 |
} |
} |
760 |
|
|
761 |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); |
762 |
bits = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5); |
rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5); |
763 |
bits += 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, 0); |
764 |
|
|
765 |
if (bits < data->iMinSAD[0]) { |
if (rd < data->iMinSAD[0]) { |
766 |
data->temp[0] = cbp; |
*data->cbp = cbp; |
767 |
data->iMinSAD[0] = bits; |
data->iMinSAD[0] = rd; |
768 |
current[0].x = x; current[0].y = y; |
current[0].x = x; current[0].y = y; |
769 |
*dir = Direction; |
*data->dir = Direction; |
770 |
} |
} |
771 |
} |
} |
772 |
|
|
775 |
/* MAINSEARCH FUNCTIONS START */ |
/* MAINSEARCH FUNCTIONS START */ |
776 |
|
|
777 |
static void |
static void |
778 |
AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection) |
AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
779 |
{ |
{ |
780 |
|
|
781 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
782 |
|
|
783 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
784 |
|
|
785 |
for(;;) { /* forever */ |
for(;;) { /* forever */ |
786 |
iDirection = 0; |
*iDirection = 0; |
787 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
788 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
789 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
791 |
|
|
792 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
793 |
|
|
794 |
if (iDirection) { /* if anything found */ |
if (*iDirection) { /* if anything found */ |
795 |
bDirection = iDirection; |
bDirection = *iDirection; |
796 |
iDirection = 0; |
*iDirection = 0; |
797 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
798 |
if (bDirection & 3) { /* our candidate is left or right */ |
if (bDirection & 3) { /* our candidate is left or right */ |
799 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
803 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
804 |
} |
} |
805 |
|
|
806 |
if (iDirection) { |
if (*iDirection) { |
807 |
bDirection += iDirection; |
bDirection += *iDirection; |
808 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
809 |
} |
} |
810 |
} else { /* about to quit, eh? not so fast.... */ |
} else { /* about to quit, eh? not so fast.... */ |
852 |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8); |
853 |
break; |
break; |
854 |
} |
} |
855 |
if (!iDirection) break; /* ok, the end. really */ |
if (!*iDirection) break; /* ok, the end. really */ |
856 |
bDirection = iDirection; |
bDirection = *iDirection; |
857 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
858 |
} |
} |
859 |
} |
} |
860 |
} |
} |
861 |
|
|
862 |
static void |
static void |
863 |
SquareSearch(int x, int y, const SearchData * const data, int bDirection) |
SquareSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
864 |
{ |
{ |
865 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
866 |
|
|
867 |
do { |
do { |
868 |
iDirection = 0; |
*iDirection = 0; |
869 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64); |
870 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128); |
871 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32); |
875 |
if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128); |
if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128); |
876 |
if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128); |
if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128); |
877 |
|
|
878 |
bDirection = iDirection; |
bDirection = *iDirection; |
879 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
880 |
} while (iDirection); |
} while (*iDirection); |
881 |
} |
} |
882 |
|
|
883 |
static void |
static void |
884 |
DiamondSearch(int x, int y, const SearchData * const data, int bDirection) |
DiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate) |
885 |
{ |
{ |
886 |
|
|
887 |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
/* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */ |
888 |
|
|
889 |
int iDirection; |
unsigned int * const iDirection = data->dir; |
890 |
|
|
891 |
do { |
do { |
892 |
iDirection = 0; |
*iDirection = 0; |
893 |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
894 |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
895 |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4); |
897 |
|
|
898 |
/* now we're doing diagonal checks near our candidate */ |
/* now we're doing diagonal checks near our candidate */ |
899 |
|
|
900 |
if (iDirection) { /* checking if anything found */ |
if (*iDirection) { /* checking if anything found */ |
901 |
bDirection = iDirection; |
bDirection = *iDirection; |
902 |
iDirection = 0; |
*iDirection = 0; |
903 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
904 |
if (bDirection & 3) { /* our candidate is left or right */ |
if (bDirection & 3) { /* our candidate is left or right */ |
905 |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
CHECK_CANDIDATE(x, y + iDiamondSize, 8); |
908 |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
CHECK_CANDIDATE(x + iDiamondSize, y, 2); |
909 |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
CHECK_CANDIDATE(x - iDiamondSize, y, 1); |
910 |
} |
} |
911 |
bDirection += iDirection; |
bDirection += *iDirection; |
912 |
x = data->currentMV->x; y = data->currentMV->y; |
x = data->currentMV->x; y = data->currentMV->y; |
913 |
} |
} |
914 |
} |
} |
915 |
while (iDirection); |
while (*iDirection); |
916 |
} |
} |
917 |
|
|
918 |
/* MAINSEARCH FUNCTIONS END */ |
/* MAINSEARCH FUNCTIONS END */ |
919 |
|
|
920 |
static void |
static void |
921 |
SubpelRefine(const SearchData * const data) |
SubpelRefine(const SearchData * const data, CheckFunc * const CheckCandidate) |
922 |
{ |
{ |
923 |
/* Do a half-pel or q-pel refinement */ |
/* Do a half-pel or q-pel refinement */ |
924 |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV; |
|
int iDirection; /* only needed because macro expects it */ |
|
925 |
|
|
926 |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0); |
927 |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0); |
961 |
} |
} |
962 |
|
|
963 |
static __inline void |
static __inline void |
964 |
SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
ZeroMacroblockP(MACROBLOCK *pMB, const int32_t sad) |
965 |
{ |
{ |
966 |
pMB->mode = MODE_NOT_CODED; |
pMB->mode = MODE_INTER; |
967 |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = zeroMV; |
968 |
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV; |
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = zeroMV; |
969 |
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; |
979 |
const uint32_t VopFlags, |
const uint32_t VopFlags, |
980 |
const uint32_t VolFlags, |
const uint32_t VolFlags, |
981 |
const IMAGE * const pCurrent, |
const IMAGE * const pCurrent, |
982 |
const IMAGE * const pRef) |
const IMAGE * const pRef, |
983 |
|
const IMAGE * const vGMC, |
984 |
|
const int coding_type) |
985 |
{ |
{ |
986 |
int mode = MODE_INTER; |
int mode = MODE_INTER; |
987 |
|
int mcsel = 0; |
988 |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
989 |
const uint32_t iQuant = pMB->quant; |
const uint32_t iQuant = pMB->quant; |
990 |
|
|
991 |
const int skip_possible = (!(VolFlags & XVID_VOL_GMC)) && (pMB->dquant == 0); |
const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0); |
992 |
|
|
993 |
|
pMB->mcsel = 0; |
994 |
|
|
995 |
if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { /* normal, fast, SAD-based mode decision */ |
if (!(VopFlags & XVID_VOP_MODEDECISION_RD)) { /* normal, fast, SAD-based mode decision */ |
996 |
int sad; |
int sad; |
997 |
int InterBias = MV16_INTER_BIAS; |
int InterBias = MV16_INTER_BIAS; |
998 |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + |
1014 |
sad = 0; |
sad = 0; |
1015 |
} |
} |
1016 |
|
|
1017 |
|
/* mcsel */ |
1018 |
|
if (coding_type == S_VOP) { |
1019 |
|
|
1020 |
|
int32_t iSAD = sad16(Data->Cur, |
1021 |
|
vGMC->y + 16*y*Data->iEdgedWidth + 16*x, Data->iEdgedWidth, 65536); |
1022 |
|
|
1023 |
|
if (Data->chroma) { |
1024 |
|
iSAD += sad8(Data->CurU, vGMC->u + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1025 |
|
iSAD += sad8(Data->CurV, vGMC->v + 8*y*(Data->iEdgedWidth/2) + 8*x, Data->iEdgedWidth/2); |
1026 |
|
} |
1027 |
|
|
1028 |
|
if (iSAD <= sad) { /* mode decision GMC */ |
1029 |
|
mode = MODE_INTER; |
1030 |
|
mcsel = 1; |
1031 |
|
sad = iSAD; |
1032 |
|
} |
1033 |
|
|
1034 |
|
} |
1035 |
|
|
1036 |
/* intra decision */ |
/* intra decision */ |
1037 |
|
|
1038 |
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
if (iQuant > 8) InterBias += 100 * (iQuant - 8); /* to make high quants work */ |
1041 |
if (x != 0) |
if (x != 0) |
1042 |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80; |
1043 |
|
|
1044 |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? */ |
if (Data->chroma) InterBias += 50; /* dev8(chroma) ??? <-- yes, we need dev8 (no big difference though) */ |
1045 |
if (Data->rrv) InterBias *= 4; |
if (Data->rrv) InterBias *= 4; |
1046 |
|
|
1047 |
if (InterBias < pMB->sad16) { |
if (InterBias < sad) { |
1048 |
int32_t deviation; |
int32_t deviation; |
1049 |
if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth); |
if (!Data->rrv) |
1050 |
else deviation = dev16(Data->Cur, Data->iEdgedWidth) + |
deviation = dev16(Data->Cur, Data->iEdgedWidth); |
1051 |
|
else |
1052 |
|
deviation = dev16(Data->Cur, Data->iEdgedWidth) + /* dev32() */ |
1053 |
dev16(Data->Cur+16, Data->iEdgedWidth) + |
dev16(Data->Cur+16, Data->iEdgedWidth) + |
1054 |
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) + |
1055 |
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth); |
1057 |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
if (deviation < (sad - InterBias)) mode = MODE_INTRA; |
1058 |
} |
} |
1059 |
|
|
1060 |
} else { /* BITS */ |
pMB->cbp = 63; |
1061 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad; |
1062 |
|
|
1063 |
|
} else { /* Rate-Distortion */ |
1064 |
|
|
1065 |
int bits, intra, i; |
int min_rd, intra_rd, i, cbp, c[2] = {0, 0}; |
1066 |
VECTOR backup[5], *v; |
VECTOR backup[5], *v; |
1067 |
Data->iQuant = iQuant; |
Data->iQuant = iQuant; |
1068 |
|
Data->cbp = c; |
1069 |
|
|
1070 |
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
v = Data->qpel ? Data->currentQMV : Data->currentMV; |
1071 |
for (i = 0; i < 5; i++) { |
for (i = 0; i < 5; i++) { |
1073 |
backup[i] = v[i]; |
backup[i] = v[i]; |
1074 |
} |
} |
1075 |
|
|
1076 |
bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags); |
min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags); |
1077 |
if (bits == 0) |
cbp = *Data->cbp; |
1078 |
mode = MODE_INTER; /* quick stop */ |
|
1079 |
else { |
if (coding_type == S_VOP) { |
1080 |
if (inter4v) { |
int gmc_rd; |
1081 |
int bits_inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
*Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */ |
1082 |
if (bits_inter4v < bits) { Data->iMinSAD[0] = bits = bits_inter4v; mode = MODE_INTER4V; } |
gmc_rd = findRDgmc(Data, vGMC, x, y); |
1083 |
|
if (gmc_rd < min_rd) { |
1084 |
|
mcsel = 1; |
1085 |
|
*Data->iMinSAD = min_rd = gmc_rd; |
1086 |
|
mode = MODE_INTER; |
1087 |
|
cbp = *Data->cbp; |
1088 |
|
} |
1089 |
} |
} |
1090 |
|
|
1091 |
intra = CountMBBitsIntra(Data); |
if (inter4v) { |
1092 |
|
int v4_rd; |
1093 |
|
v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup); |
1094 |
|
if (v4_rd < min_rd) { |
1095 |
|
Data->iMinSAD[0] = min_rd = v4_rd; |
1096 |
|
mode = MODE_INTER4V; |
1097 |
|
cbp = *Data->cbp; |
1098 |
|
} |
1099 |
|
} |
1100 |
|
|
1101 |
if (intra < bits) { *Data->iMinSAD = bits = intra; mode = MODE_INTRA; } |
intra_rd = findRDintra(Data); |
1102 |
|
if (intra_rd < min_rd) { |
1103 |
|
*Data->iMinSAD = min_rd = intra_rd; |
1104 |
|
mode = MODE_INTRA; |
1105 |
} |
} |
1106 |
|
|
1107 |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = 0; |
1108 |
|
pMB->cbp = cbp; |
1109 |
} |
} |
1110 |
|
|
1111 |
if (Data->rrv) { |
if (Data->rrv) { |
1113 |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y); |
1114 |
} |
} |
1115 |
|
|
1116 |
if (mode == MODE_INTER) { |
if (mode == MODE_INTER && mcsel == 0) { |
1117 |
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]; |
|
pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0]; |
|
1118 |
|
|
1119 |
if(Data->qpel) { |
if(Data->qpel) { |
1120 |
pMB->qmvs[0] = pMB->qmvs[1] |
pMB->qmvs[0] = pMB->qmvs[1] |
1126 |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y; |
1127 |
} |
} |
1128 |
|
|
1129 |
} else if (mode == MODE_INTER4V) |
} else if (mode == MODE_INTER ) { // but mcsel == 1 |
1130 |
pMB->sad16 = Data->iMinSAD[0]; |
|
1131 |
|
pMB->mcsel = 1; |
1132 |
|
if (Data->qpel) { |
1133 |
|
pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] = pMB->qmvs[3] = pMB->amv; |
1134 |
|
pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = pMB->amv.x/2; |
1135 |
|
pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = pMB->amv.y/2; |
1136 |
|
} else |
1137 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; |
1138 |
|
|
1139 |
|
} else |
1140 |
|
if (mode == MODE_INTER4V) ; /* anything here? */ |
1141 |
else /* INTRA, NOT_CODED */ |
else /* INTRA, NOT_CODED */ |
1142 |
SkipMacroblockP(pMB, 0); |
ZeroMacroblockP(pMB, 0); |
1143 |
|
|
1144 |
pMB->mode = mode; |
pMB->mode = mode; |
1145 |
} |
} |
1151 |
const IMAGE * const pRefH, |
const IMAGE * const pRefH, |
1152 |
const IMAGE * const pRefV, |
const IMAGE * const pRefV, |
1153 |
const IMAGE * const pRefHV, |
const IMAGE * const pRefHV, |
1154 |
|
const IMAGE * const pGMC, |
1155 |
const uint32_t iLimit) |
const uint32_t iLimit) |
1156 |
{ |
{ |
1157 |
MACROBLOCK *const pMBs = current->mbs; |
MACROBLOCK *const pMBs = current->mbs; |
1165 |
|
|
1166 |
uint32_t x, y; |
uint32_t x, y; |
1167 |
uint32_t iIntra = 0; |
uint32_t iIntra = 0; |
1168 |
int32_t quant = current->quant, sad00; |
int32_t sad00; |
1169 |
int skip_thresh = \ |
int skip_thresh = INITIAL_SKIP_THRESH * \ |
|
INITIAL_SKIP_THRESH * \ |
|
1170 |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \ |
1171 |
(current->vop_flags & XVID_VOP_MODEDECISION_BITS ? 2:1); |
(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1); |
1172 |
|
|
1173 |
/* some pre-initialized thingies for SearchP */ |
/* some pre-initialized thingies for SearchP */ |
1174 |
int32_t temp[8]; |
int32_t temp[8]; uint32_t dir; |
1175 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
1176 |
VECTOR currentQMV[5]; |
VECTOR currentQMV[5]; |
1177 |
int32_t iMinSAD[5]; |
int32_t iMinSAD[5]; |
1178 |
DECLARE_ALIGNED_MATRIX(dct_space, 2, 64, int16_t, CACHE_LINE); |
DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); |
1179 |
SearchData Data; |
SearchData Data; |
1180 |
memset(&Data, 0, sizeof(SearchData)); |
memset(&Data, 0, sizeof(SearchData)); |
1181 |
Data.iEdgedWidth = iEdgedWidth; |
Data.iEdgedWidth = iEdgedWidth; |
1183 |
Data.currentQMV = currentQMV; |
Data.currentQMV = currentQMV; |
1184 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
1185 |
Data.temp = temp; |
Data.temp = temp; |
1186 |
|
Data.dir = &dir; |
1187 |
Data.iFcode = current->fcode; |
Data.iFcode = current->fcode; |
1188 |
Data.rounding = pParam->m_rounding_type; |
Data.rounding = pParam->m_rounding_type; |
1189 |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0); |
1190 |
Data.chroma = MotionFlags & XVID_ME_CHROMA16; |
Data.chroma = MotionFlags & XVID_ME_CHROMA_PVOP; |
1191 |
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED ? 1:0); |
Data.rrv = (current->vop_flags & XVID_VOP_REDUCED) ? 1:0; |
1192 |
Data.dctSpace = dct_space; |
Data.dctSpace = dct_space; |
1193 |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
1194 |
|
|
1225 |
|
|
1226 |
sad00 = pMB->sad16; |
sad00 = pMB->sad16; |
1227 |
|
|
|
if (pMB->dquant != 0) { |
|
|
quant += DQtab[pMB->dquant]; |
|
|
if (quant > 31) quant = 31; |
|
|
else if (quant < 1) quant = 1; |
|
|
} |
|
|
pMB->quant = quant; |
|
|
|
|
1228 |
/* initial skip decision */ |
/* initial skip decision */ |
1229 |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
/* no early skip for GMC (global vector = skip vector is unknown!) */ |
1230 |
if (!(current->vol_flags & XVID_VOL_GMC)) { /* no fast SKIP for S(GMC)-VOPs */ |
if (current->coding_type != S_VOP) { /* no fast SKIP for S(GMC)-VOPs */ |
1231 |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh) |
1232 |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) { |
1233 |
SkipMacroblockP(pMB, sad00); |
ZeroMacroblockP(pMB, sad00); |
1234 |
|
pMB->mode = MODE_NOT_CODED; |
1235 |
continue; |
continue; |
1236 |
} |
} |
1237 |
} |
} |
1238 |
|
|
1239 |
|
if ((current->vop_flags & XVID_VOP_CARTOON) && |
1240 |
|
(sad00 < pMB->quant * 4 * skip_thresh)) { /* favorize (0,0) vector for cartoons */ |
1241 |
|
ZeroMacroblockP(pMB, sad00); |
1242 |
|
continue; |
1243 |
|
} |
1244 |
|
|
1245 |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x, |
1246 |
y, MotionFlags, current->vop_flags, current->vol_flags, |
y, MotionFlags, current->vop_flags, current->vol_flags, |
1247 |
&Data, pParam, pMBs, reference->mbs, pMB); |
&Data, pParam, pMBs, reference->mbs, pMB); |
1248 |
|
|
1249 |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
ModeDecision(&Data, pMB, pMBs, x, y, pParam, |
1250 |
MotionFlags, current->vop_flags, current->vol_flags, |
MotionFlags, current->vop_flags, current->vol_flags, |
1251 |
pCurrent, pRef); |
pCurrent, pRef, pGMC, current->coding_type); |
1252 |
|
|
1253 |
if (pMB->mode == MODE_INTRA) |
if (pMB->mode == MODE_INTRA) |
1254 |
if (++iIntra > iLimit) return 1; |
if (++iIntra > iLimit) return 1; |
1255 |
} |
} |
1256 |
} |
} |
1257 |
|
|
|
if (current->vol_flags & XVID_VOL_GMC ) /* GMC only for S(GMC)-VOPs */ |
|
|
{ |
|
|
current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV); |
|
|
} |
|
1258 |
return 0; |
return 0; |
1259 |
} |
} |
1260 |
|
|
1261 |
|
/* check if given vector is equal to any vector checked before */ |
1262 |
|
static __inline int |
1263 |
|
vector_repeats(const VECTOR * const pmv, const int i) |
1264 |
|
{ |
1265 |
|
unsigned int j; |
1266 |
|
for (j = 0; j < i; j++) |
1267 |
|
if (MVequal(pmv[i], pmv[j])) return 1; /* same vector has been checked already */ |
1268 |
|
return 0; |
1269 |
|
} |
1270 |
|
|
1271 |
|
/* make a binary mask that prevents diamonds/squares |
1272 |
|
from checking a vector which has been checked as a prediction */ |
1273 |
static __inline int |
static __inline int |
1274 |
make_mask(const VECTOR * const pmv, const int i) |
make_mask(const VECTOR * const pmv, const int i, const int current) |
1275 |
{ |
{ |
1276 |
int mask = 255, j; |
unsigned int mask = 255, j; |
1277 |
for (j = 0; j < i; j++) { |
for (j = 0; j < i; j++) { |
1278 |
if (MVequal(pmv[i], pmv[j])) return 0; /* same vector has been checked already */ |
if (pmv[current].x == pmv[j].x) { |
1279 |
if (pmv[i].x == pmv[j].x) { |
if (pmv[current].y == pmv[j].y + iDiamondSize) mask &= ~4; |
1280 |
if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4; |
else if (pmv[current].y == pmv[j].y - iDiamondSize) mask &= ~8; |
|
else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8; |
|
1281 |
} else |
} else |
1282 |
if (pmv[i].y == pmv[j].y) { |
if (pmv[current].y == pmv[j].y) { |
1283 |
if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1; |
if (pmv[current].x == pmv[j].x + iDiamondSize) mask &= ~1; |
1284 |
else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2; |
else if (pmv[current].x == pmv[j].x - iDiamondSize) mask &= ~2; |
1285 |
} |
} |
1286 |
} |
} |
1287 |
return mask; |
return mask; |
1345 |
MACROBLOCK * const pMB) |
MACROBLOCK * const pMB) |
1346 |
{ |
{ |
1347 |
|
|
1348 |
int i, iDirection = 255, mask, threshA; |
int i, threshA; |
1349 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1350 |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); |
1351 |
|
CheckFunc * CheckCandidate; |
1352 |
|
|
1353 |
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, 4, |
1354 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
1355 |
|
|
1356 |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); |
get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp); |
1357 |
|
|
1371 |
Data->lambda16 = lambda_vec16[pMB->quant]; |
Data->lambda16 = lambda_vec16[pMB->quant]; |
1372 |
Data->lambda8 = lambda_vec8[pMB->quant]; |
Data->lambda8 = lambda_vec8[pMB->quant]; |
1373 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1374 |
|
*Data->dir = 0; |
1375 |
|
|
1376 |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
memset(Data->currentMV, 0, 5*sizeof(VECTOR)); |
1377 |
|
|
1385 |
Data->iMinSAD[3] = pMB->sad8[2]; |
Data->iMinSAD[3] = pMB->sad8[2]; |
1386 |
Data->iMinSAD[4] = pMB->sad8[3]; |
Data->iMinSAD[4] = pMB->sad8[3]; |
1387 |
|
|
1388 |
if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) || (x | y)) { |
if ((!(VopFlags & XVID_VOP_MODEDECISION_RD)) && (x | y)) { |
1389 |
threshA = Data->temp[0]; /* that's where we keep this SAD atm */ |
threshA = Data->temp[0]; /* that's where we keep this SAD atm */ |
1390 |
if (threshA < 512) threshA = 512; |
if (threshA < 512) threshA = 512; |
1391 |
else if (threshA > 1024) threshA = 1024; |
else if (threshA > 1024) threshA = 1024; |
1396 |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
prevMBs + x + y * pParam->mb_width, Data->rrv); |
1397 |
|
|
1398 |
if (!Data->rrv) { |
if (!Data->rrv) { |
1399 |
if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16; |
if (inter4v) CheckCandidate = CheckCandidate16; |
1400 |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
else CheckCandidate = CheckCandidate16no4v; /* for extra speed */ |
1401 |
} else CheckCandidate = CheckCandidate32; |
} else CheckCandidate = CheckCandidate32; |
1402 |
|
|
1403 |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
/* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/ |
1404 |
|
|
1405 |
for (i = 1; i < 7; i++) { |
for (i = 1; i < 7; i++) |
1406 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!vector_repeats(pmv, i)) { |
1407 |
CheckCandidate(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate(pmv[i].x, pmv[i].y, Data, i); |
1408 |
if (Data->iMinSAD[0] <= threshA) break; |
if (Data->iMinSAD[0] <= threshA) { i++; break; } |
1409 |
} |
} |
1410 |
|
|
1411 |
if ((Data->iMinSAD[0] <= threshA) || |
if ((Data->iMinSAD[0] <= threshA) || |
1415 |
else { |
else { |
1416 |
|
|
1417 |
MainSearchFunc * MainSearchPtr; |
MainSearchFunc * MainSearchPtr; |
1418 |
|
int mask = make_mask(pmv, i, *Data->dir); // all vectors pmv[0..i-1] have been checked |
1419 |
|
|
1420 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
1421 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1422 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1423 |
|
|
1424 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate); |
1425 |
|
|
1426 |
/* extended search, diamond starting in 0,0 and in prediction. |
/* extended search, diamond starting in 0,0 and in prediction. |
1427 |
note that this search is/might be done in halfpel positions, |
note that this search is/might be done in halfpel positions, |
1437 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1438 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1439 |
|
|
1440 |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, Data, 255); |
1441 |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate); |
1442 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1443 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1444 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; } |
1449 |
if (!(MVequal(startMV, backupMV))) { |
if (!(MVequal(startMV, backupMV))) { |
1450 |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR; |
1451 |
|
|
1452 |
CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data); |
CheckCandidate(startMV.x, startMV.y, Data, 255); |
1453 |
MainSearchPtr(startMV.x, startMV.y, Data, 255); |
MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate); |
1454 |
if (bSAD < Data->iMinSAD[0]) { |
if (bSAD < Data->iMinSAD[0]) { |
1455 |
Data->currentMV[0] = backupMV; |
Data->currentMV[0] = backupMV; |
1456 |
Data->iMinSAD[0] = bSAD; } |
Data->iMinSAD[0] = bSAD; |
1457 |
|
} |
1458 |
} |
} |
1459 |
} |
} |
1460 |
} |
} |
1461 |
|
|
1462 |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
if (MotionFlags & XVID_ME_HALFPELREFINE16) |
1463 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
1464 |
|
|
1465 |
for(i = 0; i < 5; i++) { |
for(i = 0; i < 5; i++) { |
1466 |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */ |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */ |
1468 |
} |
} |
1469 |
|
|
1470 |
if (Data->qpel) { |
if (Data->qpel) { |
1471 |
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, 4, |
1472 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
1473 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1474 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
if (MotionFlags & XVID_ME_QUARTERPELREFINE16) |
1475 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
1476 |
} |
} |
1477 |
|
|
1478 |
if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30) |
if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30) |
1487 |
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); |
1488 |
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); |
1489 |
|
|
1490 |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_BITS))) { |
if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_RD))) { |
1491 |
/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */ |
/* chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, it will not be used */ |
1492 |
int sumx = 0, sumy = 0; |
int sumx = 0, sumy = 0; |
1493 |
|
|
1519 |
SearchData * const Data) |
SearchData * const Data) |
1520 |
{ |
{ |
1521 |
int i = 0; |
int i = 0; |
1522 |
|
CheckFunc * CheckCandidate; |
1523 |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
Data->iMinSAD = OldData->iMinSAD + 1 + block; |
1524 |
Data->currentMV = OldData->currentMV + 1 + block; |
Data->currentMV = OldData->currentMV + 1 + block; |
1525 |
Data->currentQMV = OldData->currentQMV + 1 + block; |
Data->currentQMV = OldData->currentQMV + 1 + block; |
1548 |
Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1)); |
1549 |
Data->qpel_precision = 0; |
Data->qpel_precision = 0; |
1550 |
|
|
1551 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
1552 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
1553 |
|
|
1554 |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
if (!Data->rrv) CheckCandidate = CheckCandidate8; |
1555 |
else CheckCandidate = CheckCandidate16no4v; |
else CheckCandidate = CheckCandidate16no4v; |
1556 |
|
|
1557 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_BITS))) { |
if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_RD))) { |
1558 |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
1559 |
|
|
1560 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1562 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch; |
1563 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1564 |
|
|
1565 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate); |
1566 |
|
|
1567 |
if(*(Data->iMinSAD) < temp_sad) { |
if(*(Data->iMinSAD) < temp_sad) { |
1568 |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
1573 |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
if (MotionFlags & XVID_ME_HALFPELREFINE8) { |
1574 |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */ |
1575 |
|
|
1576 |
SubpelRefine(Data); /* perform halfpel refine of current best vector */ |
SubpelRefine(Data, CheckCandidate); /* perform halfpel refine of current best vector */ |
1577 |
|
|
1578 |
if(*(Data->iMinSAD) < temp_sad) { /* we have found a better match */ |
if(*(Data->iMinSAD) < temp_sad) { /* we have found a better match */ |
1579 |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */ |
1583 |
|
|
1584 |
if (Data->qpel && MotionFlags & XVID_ME_QUARTERPELREFINE8) { |
if (Data->qpel && MotionFlags & XVID_ME_QUARTERPELREFINE8) { |
1585 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1586 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8, |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3, |
1587 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
pParam->width, pParam->height, Data->iFcode, 2, 0); |
1588 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
1589 |
} |
} |
1590 |
} |
} |
1591 |
|
|
1671 |
SearchData * const Data) |
SearchData * const Data) |
1672 |
{ |
{ |
1673 |
|
|
1674 |
int i, iDirection = 255, mask; |
int i, mask; |
1675 |
VECTOR pmv[7]; |
VECTOR pmv[7]; |
1676 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1677 |
*Data->iMinSAD = MV_MAX_ERROR; |
*Data->iMinSAD = MV_MAX_ERROR; |
1688 |
|
|
1689 |
Data->predMV = *predMV; |
Data->predMV = *predMV; |
1690 |
|
|
1691 |
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, 4, |
1692 |
pParam->width, pParam->height, iFcode - Data->qpel, 0, 0); |
pParam->width, pParam->height, iFcode - Data->qpel, 1, 0); |
1693 |
|
|
1694 |
pmv[0] = Data->predMV; |
pmv[0] = Data->predMV; |
1695 |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; } |
1697 |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current); |
1698 |
|
|
1699 |
Data->currentMV->x = Data->currentMV->y = 0; |
Data->currentMV->x = Data->currentMV->y = 0; |
|
CheckCandidate = CheckCandidate16no4v; |
|
1700 |
|
|
1701 |
/* main loop. checking all predictions */ |
/* main loop. checking all predictions */ |
1702 |
for (i = 0; i < 7; i++) { |
for (i = 0; i < 7; i++) |
1703 |
if (!(mask = make_mask(pmv, i)) ) continue; |
if (!vector_repeats(pmv, i) ) |
1704 |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data); |
CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i); |
|
} |
|
1705 |
|
|
1706 |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch; |
1707 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1708 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1709 |
|
|
1710 |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
mask = make_mask(pmv, 7, *Data->dir); |
1711 |
|
|
1712 |
SubpelRefine(Data); |
MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v); |
1713 |
|
|
1714 |
|
SubpelRefine(Data, CheckCandidate16no4v); |
1715 |
|
|
1716 |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
if (Data->qpel && *Data->iMinSAD < *best_sad + 300) { |
1717 |
Data->currentQMV->x = 2*Data->currentMV->x; |
Data->currentQMV->x = 2*Data->currentMV->x; |
1718 |
Data->currentQMV->y = 2*Data->currentMV->y; |
Data->currentQMV->y = 2*Data->currentMV->y; |
1719 |
Data->qpel_precision = 1; |
Data->qpel_precision = 1; |
1720 |
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, 4, |
1721 |
pParam->width, pParam->height, iFcode, 1, 0); |
pParam->width, pParam->height, iFcode, 2, 0); |
1722 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate16no4v); |
1723 |
} |
} |
1724 |
|
|
1725 |
/* three bits are needed to code backward mode. four for forward */ |
/* three bits are needed to code backward mode. four for forward */ |
1781 |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1782 |
stride); |
stride); |
1783 |
|
|
1784 |
if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
if (sum >= MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */ |
1785 |
|
|
1786 |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
sum += sad8bi(pCur->v + 8*x + 8 * y * stride, |
1787 |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2, |
1788 |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2, |
1789 |
stride); |
stride); |
1790 |
|
|
1791 |
if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) { |
1792 |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
pMB->mode = MODE_DIRECT_NONE_MV; /* skipped */ |
1793 |
for (k = 0; k < 4; k++) { |
for (k = 0; k < 4; k++) { |
1794 |
pMB->qmvs[k] = pMB->mvs[k]; |
pMB->qmvs[k] = pMB->mvs[k]; |
1820 |
int32_t skip_sad; |
int32_t skip_sad; |
1821 |
int k = (x + Data->iEdgedWidth*y) * 16; |
int k = (x + Data->iEdgedWidth*y) * 16; |
1822 |
MainSearchFunc *MainSearchPtr; |
MainSearchFunc *MainSearchPtr; |
1823 |
|
CheckFunc * CheckCandidate; |
1824 |
|
|
1825 |
*Data->iMinSAD = 256*4096; |
*Data->iMinSAD = 256*4096; |
1826 |
Data->RefP[0] = f_Ref->y + k; |
Data->RefP[0] = f_Ref->y + k; |
1870 |
|
|
1871 |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v; |
1872 |
|
|
1873 |
CheckCandidate(0, 0, 255, &k, Data); |
CheckCandidate(0, 0, Data, 255); |
1874 |
|
|
1875 |
/* initial (fast) skip decision */ |
/* initial (fast) skip decision */ |
1876 |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) { |
1896 |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch; |
1897 |
else MainSearchPtr = DiamondSearch; |
else MainSearchPtr = DiamondSearch; |
1898 |
|
|
1899 |
MainSearchPtr(0, 0, Data, 255); |
MainSearchPtr(0, 0, Data, 255, CheckCandidate); |
1900 |
|
|
1901 |
SubpelRefine(Data); |
SubpelRefine(Data, CheckCandidate); |
1902 |
|
|
1903 |
*best_sad = *Data->iMinSAD; |
*best_sad = *Data->iMinSAD; |
1904 |
|
|
1957 |
|
|
1958 |
{ |
{ |
1959 |
|
|
1960 |
int iDirection, i, j; |
int i, j; |
1961 |
SearchData bData; |
SearchData bData; |
1962 |
|
|
1963 |
fData->qpel_precision = 0; |
fData->qpel_precision = 0; |
1980 |
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1981 |
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8; |
1982 |
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8; |
1983 |
|
bData.dir = fData->dir; |
1984 |
|
|
1985 |
bData.bpredMV = fData->predMV = *f_predMV; |
bData.bpredMV = fData->predMV = *f_predMV; |
1986 |
fData->bpredMV = bData.predMV = *b_predMV; |
fData->bpredMV = bData.predMV = *b_predMV; |
1987 |
fData->currentMV[0] = fData->currentMV[2]; |
fData->currentMV[0] = fData->currentMV[2]; |
1988 |
|
|
1989 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode - fData->qpel, 1, 0); |
1990 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode - fData->qpel, 1, 0); |
1991 |
|
|
1992 |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx; |
1993 |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dx; |
1999 |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy; |
2000 |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy; |
2001 |
|
|
2002 |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData); |
CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255); |
2003 |
|
|
2004 |
/* diamond */ |
/* diamond */ |
2005 |
do { |
do { |
2006 |
iDirection = 255; |
*fData->dir = 255; |
2007 |
/* forward MV moves */ |
/* forward MV moves */ |
2008 |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
i = fData->currentMV[0].x; j = fData->currentMV[0].y; |
2009 |
|
|
2010 |
CheckCandidateInt(i + 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i + 1, j, fData, 0); |
2011 |
CheckCandidateInt(i, j + 1, 0, &iDirection, fData); |
CheckCandidateInt(i, j + 1, fData, 0); |
2012 |
CheckCandidateInt(i - 1, j, 0, &iDirection, fData); |
CheckCandidateInt(i - 1, j, fData, 0); |
2013 |
CheckCandidateInt(i, j - 1, 0, &iDirection, fData); |
CheckCandidateInt(i, j - 1, fData, 0); |
2014 |
|
|
2015 |
/* backward MV moves */ |
/* backward MV moves */ |
2016 |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
i = fData->currentMV[1].x; j = fData->currentMV[1].y; |
2017 |
fData->currentMV[2] = fData->currentMV[0]; |
fData->currentMV[2] = fData->currentMV[0]; |
2018 |
CheckCandidateInt(i + 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i + 1, j, &bData, 0); |
2019 |
CheckCandidateInt(i, j + 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j + 1, &bData, 0); |
2020 |
CheckCandidateInt(i - 1, j, 0, &iDirection, &bData); |
CheckCandidateInt(i - 1, j, &bData, 0); |
2021 |
CheckCandidateInt(i, j - 1, 0, &iDirection, &bData); |
CheckCandidateInt(i, j - 1, &bData, 0); |
2022 |
|
|
2023 |
} while (!(iDirection)); |
} while (!(*fData->dir)); |
2024 |
|
|
2025 |
/* qpel refinement */ |
/* qpel refinement */ |
2026 |
if (fData->qpel) { |
if (fData->qpel) { |
2027 |
if (*fData->iMinSAD > *best_sad + 500) return; |
if (*fData->iMinSAD > *best_sad + 500) return; |
|
CheckCandidate = CheckCandidateInt; |
|
2028 |
fData->qpel_precision = bData.qpel_precision = 1; |
fData->qpel_precision = bData.qpel_precision = 1; |
2029 |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0); |
get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0); |
2030 |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0); |
get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0); |
2031 |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x; |
2032 |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y; |
2033 |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
fData->currentQMV[1].x = 2 * fData->currentMV[1].x; |
2034 |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
fData->currentQMV[1].y = 2 * fData->currentMV[1].y; |
2035 |
SubpelRefine(fData); |
SubpelRefine(fData, CheckCandidateInt); |
2036 |
if (*fData->iMinSAD > *best_sad + 300) return; |
if (*fData->iMinSAD > *best_sad + 300) return; |
2037 |
fData->currentQMV[2] = fData->currentQMV[0]; |
fData->currentQMV[2] = fData->currentQMV[0]; |
2038 |
SubpelRefine(&bData); |
SubpelRefine(&bData, CheckCandidateInt); |
2039 |
} |
} |
2040 |
|
|
2041 |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */ |
2094 |
|
|
2095 |
SearchData Data; |
SearchData Data; |
2096 |
int32_t iMinSAD; |
int32_t iMinSAD; |
2097 |
|
uint32_t dir; |
2098 |
VECTOR currentMV[3]; |
VECTOR currentMV[3]; |
2099 |
VECTOR currentQMV[3]; |
VECTOR currentQMV[3]; |
2100 |
int32_t temp[8]; |
int32_t temp[8]; |
2103 |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
Data.currentMV = currentMV; Data.currentQMV = currentQMV; |
2104 |
Data.iMinSAD = &iMinSAD; |
Data.iMinSAD = &iMinSAD; |
2105 |
Data.lambda16 = lambda_vec16[frame->quant]; |
Data.lambda16 = lambda_vec16[frame->quant]; |
2106 |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL; |
Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0; |
2107 |
Data.rounding = 0; |
Data.rounding = 0; |
2108 |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA8; |
Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP; |
2109 |
Data.temp = temp; |
Data.temp = temp; |
2110 |
|
Data.dir = &dir; |
2111 |
|
|
2112 |
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) */ |
2113 |
|
|
2214 |
SearchData * const Data) |
SearchData * const Data) |
2215 |
{ |
{ |
2216 |
|
|
2217 |
int i, mask; |
int i; |
|
int quarterpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0; |
|
2218 |
VECTOR pmv[3]; |
VECTOR pmv[3]; |
2219 |
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
2220 |
|
|
2229 |
Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */ |
Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */ |
2230 |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */ |
else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); /* else median */ |
2231 |
|
|
2232 |
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, 4, |
2233 |
pParam->width, pParam->height, Data->iFcode - quarterpel, 0, 0); |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0); |
2234 |
|
|
2235 |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
Data->Cur = pCur + (x + y * pParam->edged_width) * 16; |
2236 |
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16; |
2241 |
pmv[2].y = EVEN(Data->predMV.y); |
pmv[2].y = EVEN(Data->predMV.y); |
2242 |
pmv[0].x = pmv[0].y = 0; |
pmv[0].x = pmv[0].y = 0; |
2243 |
|
|
2244 |
CheckCandidate32I(0, 0, 255, &i, Data); |
CheckCandidate32I(0, 0, Data, 0); |
2245 |
|
|
2246 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { |
2247 |
|
|
2248 |
if (!(mask = make_mask(pmv, 1))) |
if (!vector_repeats(pmv, 1)) |
2249 |
CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data); |
CheckCandidate32I(pmv[1].x, pmv[1].y, Data, 1); |
2250 |
if (!(mask = make_mask(pmv, 2))) |
if (!vector_repeats(pmv, 2)) |
2251 |
CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data); |
CheckCandidate32I(pmv[2].x, pmv[2].y, Data, 2); |
2252 |
|
|
2253 |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) /* diamond only if needed */ |
if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { /* diamond only if needed */ |
2254 |
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i); |
unsigned int mask = make_mask(pmv, 3, *Data->dir); |
2255 |
|
DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, 255/*mask*/, CheckCandidate32I); |
2256 |
|
} |
2257 |
} |
} |
2258 |
|
|
2259 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
2264 |
} |
} |
2265 |
} |
} |
2266 |
|
|
2267 |
#define INTRA_THRESH 1700 |
#define INTRA_THRESH 2200 |
2268 |
#define INTER_THRESH 1200 |
#define INTER_THRESH 50 |
2269 |
|
#define INTRA_THRESH2 95 |
2270 |
|
|
2271 |
int |
int |
2272 |
MEanalysis( const IMAGE * const pRef, |
MEanalysis( const IMAGE * const pRef, |
2273 |
const FRAMEINFO * const Current, |
const FRAMEINFO * const Current, |
2274 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2275 |
const int maxIntra, /* maximum number if non-I frames */ |
const int maxIntra, //maximum number if non-I frames |
2276 |
const int intraCount, /* number of non-I frames after last I frame; 0 if we force P/B frame */ |
const int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame |
2277 |
const int bCount, /* number of B frames in a row */ |
const int bCount, // number of B frames in a row |
2278 |
const int b_thresh) |
const int b_thresh) |
2279 |
{ |
{ |
2280 |
uint32_t x, y, intra = 0; |
uint32_t x, y, intra = 0; |
2281 |
int sSAD = 0; |
int sSAD = 0; |
2282 |
MACROBLOCK * const pMBs = Current->mbs; |
MACROBLOCK * const pMBs = Current->mbs; |
2283 |
const IMAGE * const pCurrent = &Current->image; |
const IMAGE * const pCurrent = &Current->image; |
2284 |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + 10*b_thresh; |
int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + b_thresh; |
2285 |
int s = 0, blocks = 0; |
int blocks = 0; |
2286 |
|
int complexity = 0; |
2287 |
|
|
2288 |
int32_t iMinSAD[5], temp[5]; |
int32_t iMinSAD[5], temp[5]; |
2289 |
|
uint32_t dir; |
2290 |
VECTOR currentMV[5]; |
VECTOR currentMV[5]; |
2291 |
SearchData Data; |
SearchData Data; |
2292 |
Data.iEdgedWidth = pParam->edged_width; |
Data.iEdgedWidth = pParam->edged_width; |
2294 |
Data.iMinSAD = iMinSAD; |
Data.iMinSAD = iMinSAD; |
2295 |
Data.iFcode = Current->fcode; |
Data.iFcode = Current->fcode; |
2296 |
Data.temp = temp; |
Data.temp = temp; |
2297 |
CheckCandidate = CheckCandidate32I; |
Data.dir = &dir; |
2298 |
|
Data.qpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0; |
2299 |
|
|
2300 |
if (intraCount != 0) { |
if (intraCount != 0) { |
2301 |
if (intraCount < 10) /* we're right after an I frame */ |
if (intraCount < 10) // we're right after an I frame |
2302 |
IntraThresh += 15* (intraCount - 10) * (intraCount - 10); |
IntraThresh += 15* (intraCount - 10) * (intraCount - 10); |
2303 |
else |
else |
2304 |
if ( 5*(maxIntra - intraCount) < maxIntra) /* we're close to maximum. 2 sec when max is 10 sec */ |
if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec |
2305 |
IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra; |
IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra; |
2306 |
} |
} |
2307 |
|
|
2308 |
InterThresh -= (350 - 8*b_thresh) * bCount; |
InterThresh -= 12 * bCount; |
2309 |
if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh; |
if (InterThresh < 15 + b_thresh) InterThresh = 15 + b_thresh; |
2310 |
|
|
2311 |
if (sadInit) (*sadInit) (); |
if (sadInit) (*sadInit) (); |
2312 |
|
|
2313 |
for (y = 1; y < pParam->mb_height-1; y += 2) { |
for (y = 1; y < pParam->mb_height-1; y += 2) { |
2314 |
for (x = 1; x < pParam->mb_width-1; x += 2) { |
for (x = 1; x < pParam->mb_width-1; x += 2) { |
2315 |
int i; |
int i; |
2316 |
blocks += 4; |
blocks += 10; |
2317 |
|
|
2318 |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV; |
2319 |
else { /* extrapolation of the vector found for last frame */ |
else { //extrapolation of the vector found for last frame |
2320 |
pMBs[x + y * pParam->mb_width].mvs[0].x = |
pMBs[x + y * pParam->mb_width].mvs[0].x = |
2321 |
(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount; |
(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount; |
2322 |
pMBs[x + y * pParam->mb_width].mvs[0].y = |
pMBs[x + y * pParam->mb_width].mvs[0].y = |
2328 |
for (i = 0; i < 4; i++) { |
for (i = 0; i < 4; i++) { |
2329 |
int dev; |
int dev; |
2330 |
MACROBLOCK *pMB = &pMBs[x+(i&1) + (y+(i>>1)) * pParam->mb_width]; |
MACROBLOCK *pMB = &pMBs[x+(i&1) + (y+(i>>1)) * pParam->mb_width]; |
|
if (pMB->sad16 > IntraThresh) { |
|
2331 |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16, |
dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16, |
2332 |
pParam->edged_width); |
pParam->edged_width); |
2333 |
|
|
2334 |
|
complexity += MAX(dev, 300); |
2335 |
if (dev + IntraThresh < pMB->sad16) { |
if (dev + IntraThresh < pMB->sad16) { |
2336 |
pMB->mode = MODE_INTRA; |
pMB->mode = MODE_INTRA; |
2337 |
if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP; |
if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP; |
2338 |
} |
} |
|
} |
|
|
if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) s++; |
|
2339 |
|
|
2340 |
sSAD += pMB->sad16; |
if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) |
2341 |
|
if (dev > 500 && pMB->sad16 < 1000) |
2342 |
|
sSAD += 1000; |
2343 |
|
|
2344 |
|
sSAD += (dev < 3000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */ |
2345 |
} |
} |
2346 |
} |
} |
2347 |
} |
} |
2348 |
|
complexity >>= 7; |
2349 |
|
|
2350 |
sSAD /= blocks; |
sSAD /= complexity + 4*blocks; |
|
|
|
|
if (b_thresh < 20) { |
|
|
s = (10*s) / blocks; |
|
|
if (s > 4) sSAD += (s - 2) * (40 - 2*b_thresh); /* static block - looks bad when in bframe... */ |
|
|
} |
|
2351 |
|
|
2352 |
|
if (intraCount > 80 && sSAD > INTRA_THRESH2 ) return I_VOP; |
2353 |
if (sSAD > InterThresh ) return P_VOP; |
if (sSAD > InterThresh ) return P_VOP; |
2354 |
emms(); |
emms(); |
2355 |
return B_VOP; |
return B_VOP; |
2356 |
} |
} |
2357 |
|
|
2358 |
|
|
2359 |
static WARPPOINTS |
/* functions which perform BITS-based search/bitcount */ |
2360 |
GlobalMotionEst(const MACROBLOCK * const pMBs, |
|
2361 |
|
static int |
2362 |
|
findRDinter(SearchData * const Data, |
2363 |
|
const MACROBLOCK * const pMBs, const int x, const int y, |
2364 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2365 |
const FRAMEINFO * const current, |
const uint32_t MotionFlags) |
|
const FRAMEINFO * const reference, |
|
|
const IMAGE * const pRefH, |
|
|
const IMAGE * const pRefV, |
|
|
const IMAGE * const pRefHV ) |
|
2366 |
{ |
{ |
2367 |
|
int i; |
2368 |
|
int32_t bsad[5]; |
2369 |
|
|
2370 |
const int deltax=8; /* upper bound for difference between a MV and it's neighbour MVs */ |
if (Data->qpel) { |
2371 |
const int deltay=8; |
for(i = 0; i < 5; i++) { |
2372 |
const int grad=512; /* lower bound for deviation in MB */ |
Data->currentMV[i].x = Data->currentQMV[i].x/2; |
2373 |
|
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
2374 |
WARPPOINTS gmc; |
} |
2375 |
|
Data->qpel_precision = 1; |
2376 |
|
CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255); |
2377 |
|
|
2378 |
uint32_t mx, my; |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* we have to prepare for halfpixel-precision search */ |
2379 |
|
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
2380 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2381 |
|
pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv); |
2382 |
|
Data->qpel_precision = 0; |
2383 |
|
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
2384 |
|
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
2385 |
|
} |
2386 |
|
|
2387 |
int MBh = pParam->mb_height; |
} else { /* not qpel */ |
|
int MBw = pParam->mb_width; |
|
2388 |
|
|
2389 |
int *MBmask= calloc(MBh*MBw,sizeof(int)); |
CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255); |
2390 |
double DtimesF[4] = { 0.,0., 0., 0. }; |
} |
|
double sol[4] = { 0., 0., 0., 0. }; |
|
|
double a,b,c,n,denom; |
|
|
double meanx,meany; |
|
|
int num,oldnum; |
|
2391 |
|
|
2392 |
if (!MBmask) { fprintf(stderr,"Mem error\n"); |
if (MotionFlags&XVID_ME_EXTSEARCH_RD) |
2393 |
gmc.duv[0].x= gmc.duv[0].y = |
SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16); |
|
gmc.duv[1].x= gmc.duv[1].y = |
|
|
gmc.duv[2].x= gmc.duv[2].y = 0; |
|
|
return gmc; } |
|
2394 |
|
|
2395 |
/* filter mask of all blocks */ |
if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) |
2396 |
|
SubpelRefine(Data, CheckCandidateRD16); |
2397 |
|
|
2398 |
for (my = 1; my < (uint32_t)MBh-1; my++) |
if (Data->qpel) { |
2399 |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ |
2400 |
{ |
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
2401 |
const int mbnum = mx + my * MBw; |
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ |
2402 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
2403 |
const VECTOR mv = pMB->mvs[0]; |
} |
2404 |
|
|
2405 |
if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED) |
/* preparing for qpel-precision search */ |
2406 |
continue; |
Data->qpel_precision = 1; |
2407 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2408 |
|
pParam->width, pParam->height, Data->iFcode, 2, 0); |
2409 |
|
} |
2410 |
|
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) |
2411 |
|
SubpelRefine(Data, CheckCandidateRD16); |
2412 |
|
} |
2413 |
|
|
2414 |
if ( ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) ) |
if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ |
2415 |
&& ( (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) ) |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
2416 |
&& ( (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) ) |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
2417 |
&& ( (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) ) |
CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255); |
2418 |
MBmask[mbnum]=1; |
} |
2419 |
|
return Data->iMinSAD[0]; |
2420 |
} |
} |
2421 |
|
|
2422 |
for (my = 1; my < (uint32_t)MBh-1; my++) |
static int |
2423 |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
findRDinter4v(const SearchData * const Data, |
2424 |
|
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
2425 |
|
const int x, const int y, |
2426 |
|
const MBParam * const pParam, const uint32_t MotionFlags, |
2427 |
|
const VECTOR * const backup) |
2428 |
{ |
{ |
|
const uint8_t *const pCur = current->image.y + 16*my*pParam->edged_width + 16*mx; |
|
2429 |
|
|
2430 |
const int mbnum = mx + my * MBw; |
int cbp = 0, bits = 0, t = 0, i; |
2431 |
if (!MBmask[mbnum]) |
SearchData Data2, *Data8 = &Data2; |
2432 |
continue; |
int sumx = 0, sumy = 0; |
2433 |
|
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
2434 |
|
uint8_t * ptr; |
2435 |
|
|
2436 |
|
memcpy(Data8, Data, sizeof(SearchData)); |
2437 |
|
|
2438 |
|
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
2439 |
|
|
2440 |
if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= (uint32_t)grad ) |
Data8->iMinSAD = Data->iMinSAD + i + 1; |
2441 |
MBmask[mbnum] = 0; |
Data8->currentMV = Data->currentMV + i + 1; |
2442 |
if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= (uint32_t)grad ) |
Data8->currentQMV = Data->currentQMV + i + 1; |
2443 |
MBmask[mbnum] = 0; |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2444 |
|
Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2445 |
|
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2446 |
|
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2447 |
|
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2448 |
|
*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; // copy corresponding cbp bit |
2449 |
|
|
2450 |
|
if(Data->qpel) { |
2451 |
|
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2452 |
|
if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, |
2453 |
|
Data8->predMV, Data8->iFcode, 0, 0); |
2454 |
|
} else { |
2455 |
|
Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); |
2456 |
|
if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, |
2457 |
|
Data8->predMV, Data8->iFcode, 0, 0); |
2458 |
} |
} |
2459 |
|
|
2460 |
emms(); |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
2461 |
|
pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0); |
2462 |
|
|
2463 |
do { /* until convergence */ |
*Data8->iMinSAD += BITS_MULT*t; |
2464 |
|
|
2465 |
a = b = c = n = 0; |
Data8->qpel_precision = Data8->qpel; |
2466 |
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
/* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */ |
|
for (my = 0; my < (uint32_t)MBh; my++) |
|
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
|
2467 |
{ |
{ |
2468 |
const int mbnum = mx + my * MBw; |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
2469 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
if (!MVequal (*v, backup[i+1]) ) |
2470 |
const VECTOR mv = pMB->mvs[0]; |
CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255); |
2471 |
|
} |
2472 |
|
|
2473 |
if (!MBmask[mbnum]) |
if (Data8->qpel) { |
2474 |
continue; |
if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ |
2475 |
|
int32_t s = *Data8->iMinSAD; |
2476 |
|
Data8->currentMV->x = Data8->currentQMV->x/2; |
2477 |
|
Data8->currentMV->y = Data8->currentQMV->y/2; |
2478 |
|
Data8->qpel_precision = 0; |
2479 |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
2480 |
|
pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); |
2481 |
|
|
2482 |
n++; |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
2483 |
a += 16*mx+8; |
CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255); |
2484 |
b += 16*my+8; |
|
2485 |
c += (16*mx+8)*(16*mx+8)+(16*my+8)*(16*my+8); |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) |
2486 |
|
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); |
2487 |
|
|
2488 |
|
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2489 |
|
SubpelRefine(Data8, CheckCandidateRD8); |
2490 |
|
|
2491 |
|
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
2492 |
|
Data8->currentQMV->x = 2*Data8->currentMV->x; |
2493 |
|
Data8->currentQMV->y = 2*Data8->currentMV->y; |
2494 |
|
} |
2495 |
|
|
2496 |
|
Data8->qpel_precision = 1; |
2497 |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, |
2498 |
|
pParam->width, pParam->height, Data8->iFcode, 2, 0); |
2499 |
|
|
2500 |
|
} |
2501 |
|
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) |
2502 |
|
SubpelRefine(Data8, CheckCandidateRD8); |
2503 |
|
|
2504 |
|
} else { /* not qpel */ |
2505 |
|
|
2506 |
|
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */ |
2507 |
|
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); |
2508 |
|
|
2509 |
|
if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) |
2510 |
|
SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */ |
2511 |
|
} |
2512 |
|
|
2513 |
|
/* checking vector equal to predicion */ |
2514 |
|
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) { |
2515 |
|
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
2516 |
|
if (!MVequal(*v, Data8->predMV)) |
2517 |
|
CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255); |
2518 |
|
} |
2519 |
|
|
2520 |
|
bits += *Data8->iMinSAD; |
2521 |
|
if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */ |
2522 |
|
|
2523 |
|
/* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */ |
2524 |
|
if(Data->qpel) { |
2525 |
|
pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; |
2526 |
|
pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; |
2527 |
|
pMB->qmvs[i] = *Data8->currentQMV; |
2528 |
|
sumx += Data8->currentQMV->x/2; |
2529 |
|
sumy += Data8->currentQMV->y/2; |
2530 |
|
} else { |
2531 |
|
pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x; |
2532 |
|
pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y; |
2533 |
|
sumx += Data8->currentMV->x; |
2534 |
|
sumy += Data8->currentMV->y; |
2535 |
|
} |
2536 |
|
pMB->mvs[i] = *Data8->currentMV; |
2537 |
|
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
2538 |
|
if (Data8->cbp[0]) cbp |= 1 << (5 - i); |
2539 |
|
|
2540 |
|
} /* end - for all luma blocks */ |
2541 |
|
|
2542 |
|
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
2543 |
|
|
2544 |
|
/* let's check chroma */ |
2545 |
|
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
2546 |
|
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
2547 |
|
|
2548 |
|
/* chroma U */ |
2549 |
|
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2550 |
|
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
2551 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4); |
2552 |
|
|
2553 |
|
if (bits >= *Data->iMinSAD) return bits; |
2554 |
|
|
2555 |
|
/* chroma V */ |
2556 |
|
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
2557 |
|
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
2558 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5); |
2559 |
|
|
2560 |
|
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
2561 |
|
|
2562 |
|
*Data->cbp = cbp; |
2563 |
|
return bits; |
2564 |
|
} |
2565 |
|
|
2566 |
|
static int |
2567 |
|
findRDintra(const SearchData * const Data) |
2568 |
|
{ |
2569 |
|
int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
2570 |
|
int cbp = 0, i, dc = 0; |
2571 |
|
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
2572 |
|
|
2573 |
|
for(i = 0; i < 4; i++) { |
2574 |
|
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2575 |
|
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
2576 |
|
bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, &dc); |
2577 |
|
|
2578 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2579 |
|
} |
2580 |
|
|
2581 |
|
bits += BITS_MULT*xvid_cbpy_tab[cbp>>2].len; |
2582 |
|
|
2583 |
|
/*chroma U */ |
2584 |
|
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
2585 |
|
bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, &dc); |
2586 |
|
|
2587 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2588 |
|
|
2589 |
|
/* chroma V */ |
2590 |
|
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
2591 |
|
bits += Block_CalcBitsIntra(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, &dc); |
2592 |
|
|
2593 |
|
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
2594 |
|
|
2595 |
|
return bits; |
2596 |
|
} |
2597 |
|
|
2598 |
|
static int |
2599 |
|
findRDgmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) |
2600 |
|
{ |
2601 |
|
int bits = BITS_MULT*1; /* this one is mcsel */ |
2602 |
|
int cbp = 0, i; |
2603 |
|
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
2604 |
|
|
2605 |
|
for(i = 0; i < 4; i++) { |
2606 |
|
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
2607 |
|
transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); |
2608 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i); |
2609 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2610 |
|
} |
2611 |
|
|
2612 |
|
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
2613 |
|
|
2614 |
|
/*chroma U */ |
2615 |
|
transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
2616 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4); |
2617 |
|
|
2618 |
|
if (bits >= Data->iMinSAD[0]) return bits; |
2619 |
|
|
2620 |
|
/* chroma V */ |
2621 |
|
transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); |
2622 |
|
bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5); |
2623 |
|
|
2624 |
|
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; |
2625 |
|
|
2626 |
|
*Data->cbp = cbp; |
2627 |
|
|
2628 |
|
return bits; |
2629 |
|
} |
2630 |
|
|
2631 |
|
|
2632 |
|
|
2633 |
|
|
2634 |
|
static __inline void |
2635 |
|
GMEanalyzeMB ( const uint8_t * const pCur, |
2636 |
|
const uint8_t * const pRef, |
2637 |
|
const uint8_t * const pRefH, |
2638 |
|
const uint8_t * const pRefV, |
2639 |
|
const uint8_t * const pRefHV, |
2640 |
|
const int x, |
2641 |
|
const int y, |
2642 |
|
const MBParam * const pParam, |
2643 |
|
MACROBLOCK * const pMBs, |
2644 |
|
SearchData * const Data) |
2645 |
|
{ |
2646 |
|
|
2647 |
|
int i=0; |
2648 |
|
MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width]; |
2649 |
|
|
2650 |
|
Data->iMinSAD[0] = MV_MAX_ERROR; |
2651 |
|
|
2652 |
|
Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); |
2653 |
|
|
2654 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2655 |
|
pParam->width, pParam->height, 16, 1, 0); |
2656 |
|
|
2657 |
|
Data->Cur = pCur + 16*(x + y * pParam->edged_width); |
2658 |
|
Data->RefP[0] = pRef + 16*(x + y * pParam->edged_width); |
2659 |
|
Data->RefP[1] = pRefV + 16*(x + y * pParam->edged_width); |
2660 |
|
Data->RefP[2] = pRefH + 16*(x + y * pParam->edged_width); |
2661 |
|
Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width); |
2662 |
|
|
2663 |
|
Data->currentMV[0].x = Data->currentMV[0].y = 0; |
2664 |
|
CheckCandidate16I(0, 0, Data, 255); |
2665 |
|
|
2666 |
|
if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) ) |
2667 |
|
CheckCandidate16I(Data->predMV.x, Data->predMV.y, Data, 255); |
2668 |
|
|
2669 |
|
DiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255, CheckCandidate16I); |
2670 |
|
|
2671 |
|
SubpelRefine(Data, CheckCandidate16I); |
2672 |
|
|
2673 |
|
|
2674 |
|
/* for QPel halfpel positions are worse than in halfpel mode :( */ |
2675 |
|
/* if (Data->qpel) { |
2676 |
|
Data->currentQMV->x = 2*Data->currentMV->x; |
2677 |
|
Data->currentQMV->y = 2*Data->currentMV->y; |
2678 |
|
Data->qpel_precision = 1; |
2679 |
|
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, |
2680 |
|
pParam->width, pParam->height, iFcode, 2, 0); |
2681 |
|
SubpelRefine(Data); |
2682 |
|
} |
2683 |
|
*/ |
2684 |
|
|
2685 |
|
pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0]; |
2686 |
|
pMB->sad16 = Data->iMinSAD[0]; |
2687 |
|
pMB->mode = MODE_INTER; |
2688 |
|
pMB->sad16 += 10*d_mv_bits(pMB->mvs[0].x, pMB->mvs[0].y, Data->predMV, Data->iFcode, 0, 0); |
2689 |
|
return; |
2690 |
|
} |
2691 |
|
|
2692 |
|
void |
2693 |
|
GMEanalysis(const MBParam * const pParam, |
2694 |
|
const FRAMEINFO * const current, |
2695 |
|
const FRAMEINFO * const reference, |
2696 |
|
const IMAGE * const pRefH, |
2697 |
|
const IMAGE * const pRefV, |
2698 |
|
const IMAGE * const pRefHV) |
2699 |
|
{ |
2700 |
|
uint32_t x, y; |
2701 |
|
MACROBLOCK * const pMBs = current->mbs; |
2702 |
|
const IMAGE * const pCurrent = ¤t->image; |
2703 |
|
const IMAGE * const pReference = &reference->image; |
2704 |
|
|
2705 |
|
int32_t iMinSAD[5], temp[5]; |
2706 |
|
VECTOR currentMV[5]; |
2707 |
|
uint32_t dir; |
2708 |
|
SearchData Data; |
2709 |
|
memset(&Data, 0, sizeof(SearchData)); |
2710 |
|
|
2711 |
|
Data.iEdgedWidth = pParam->edged_width; |
2712 |
|
Data.rounding = pParam->m_rounding_type; |
2713 |
|
|
2714 |
|
Data.currentMV = ¤tMV[0]; |
2715 |
|
Data.iMinSAD = &iMinSAD[0]; |
2716 |
|
Data.iFcode = current->fcode; |
2717 |
|
Data.temp = temp; |
2718 |
|
Data.dir = &dir; |
2719 |
|
|
2720 |
|
if (sadInit) (*sadInit) (); |
2721 |
|
|
2722 |
|
for (y = 0; y < pParam->mb_height; y ++) { |
2723 |
|
for (x = 0; x < pParam->mb_width; x ++) { |
2724 |
|
GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data); |
2725 |
|
} |
2726 |
|
} |
2727 |
|
return; |
2728 |
|
} |
2729 |
|
|
2730 |
|
|
2731 |
|
WARPPOINTS |
2732 |
|
GlobalMotionEst(MACROBLOCK * const pMBs, |
2733 |
|
const MBParam * const pParam, |
2734 |
|
const FRAMEINFO * const current, |
2735 |
|
const FRAMEINFO * const reference, |
2736 |
|
const IMAGE * const pRefH, |
2737 |
|
const IMAGE * const pRefV, |
2738 |
|
const IMAGE * const pRefHV) |
2739 |
|
{ |
2740 |
|
|
2741 |
|
const int deltax=8; // upper bound for difference between a MV and it's neighbour MVs |
2742 |
|
const int deltay=8; |
2743 |
|
const unsigned int gradx=512; // lower bound for gradient in MB (ignore "flat" blocks) |
2744 |
|
const unsigned int grady=512; |
2745 |
|
|
2746 |
|
double sol[4] = { 0., 0., 0., 0. }; |
2747 |
|
|
2748 |
|
WARPPOINTS gmc; |
2749 |
|
|
2750 |
|
uint32_t mx, my; |
2751 |
|
|
2752 |
|
int MBh = pParam->mb_height; |
2753 |
|
int MBw = pParam->mb_width; |
2754 |
|
const int minblocks = 9; //MBh*MBw/32+3; /* just some reasonable number 3% + 3 */ |
2755 |
|
const int maxblocks = MBh*MBw/4; /* just some reasonable number 3% + 3 */ |
2756 |
|
|
2757 |
|
int num=0; |
2758 |
|
int oldnum; |
2759 |
|
|
2760 |
|
gmc.duv[0].x = gmc.duv[0].y = gmc.duv[1].x = gmc.duv[1].y = gmc.duv[2].x = gmc.duv[2].y = 0; |
2761 |
|
|
2762 |
|
GMEanalysis(pParam,current, reference, pRefH, pRefV, pRefHV); |
2763 |
|
|
2764 |
|
/* block based ME isn't done, yet, so do a quick presearch */ |
2765 |
|
|
2766 |
|
// filter mask of all blocks |
2767 |
|
|
2768 |
|
for (my = 0; my < (uint32_t)MBh; my++) |
2769 |
|
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2770 |
|
{ |
2771 |
|
const int mbnum = mx + my * MBw; |
2772 |
|
pMBs[mbnum].mcsel = 0; |
2773 |
|
} |
2774 |
|
|
2775 |
|
|
2776 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ |
2777 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ |
2778 |
|
{ |
2779 |
|
const int mbnum = mx + my * MBw; |
2780 |
|
MACROBLOCK *const pMB = &pMBs[mbnum]; |
2781 |
|
const VECTOR mv = pMB->mvs[0]; |
2782 |
|
|
2783 |
|
/* don't use object boundaries */ |
2784 |
|
if ( (abs(mv.x - (pMB-1)->mvs[0].x) < deltax) |
2785 |
|
&& (abs(mv.y - (pMB-1)->mvs[0].y) < deltay) |
2786 |
|
&& (abs(mv.x - (pMB+1)->mvs[0].x) < deltax) |
2787 |
|
&& (abs(mv.y - (pMB+1)->mvs[0].y) < deltay) |
2788 |
|
&& (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) |
2789 |
|
&& (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) |
2790 |
|
&& (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) |
2791 |
|
&& (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) |
2792 |
|
{ const int iEdgedWidth = pParam->edged_width; |
2793 |
|
const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); |
2794 |
|
if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) |
2795 |
|
&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) |
2796 |
|
{ pMB->mcsel = 1; |
2797 |
|
num++; |
2798 |
|
} |
2799 |
|
|
2800 |
|
/* only use "structured" blocks */ |
2801 |
|
} |
2802 |
|
} |
2803 |
|
emms(); |
2804 |
|
|
2805 |
|
/* further filtering would be possible, but during iteration, remaining |
2806 |
|
outliers usually are removed, too */ |
2807 |
|
|
2808 |
|
if (num>= minblocks) |
2809 |
|
do { /* until convergence */ |
2810 |
|
double DtimesF[4]; |
2811 |
|
double a,b,c,n,invdenom; |
2812 |
|
double meanx,meany; |
2813 |
|
|
2814 |
|
a = b = c = n = 0; |
2815 |
|
DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.; |
2816 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
2817 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2818 |
|
{ |
2819 |
|
const int mbnum = mx + my * MBw; |
2820 |
|
const VECTOR mv = pMBs[mbnum].mvs[0]; |
2821 |
|
|
2822 |
|
if (!pMBs[mbnum].mcsel) |
2823 |
|
continue; |
2824 |
|
|
2825 |
|
n++; |
2826 |
|
a += 16*mx+8; |
2827 |
|
b += 16*my+8; |
2828 |
|
c += (16*mx+8)*(16*mx+8)+(16*my+8)*(16*my+8); |
2829 |
|
|
2830 |
DtimesF[0] += (double)mv.x; |
DtimesF[0] += (double)mv.x; |
2831 |
DtimesF[1] += (double)mv.x*(16*mx+8) + (double)mv.y*(16*my+8); |
DtimesF[1] += (double)mv.x*(16*mx+8) + (double)mv.y*(16*my+8); |
2833 |
DtimesF[3] += (double)mv.y; |
DtimesF[3] += (double)mv.y; |
2834 |
} |
} |
2835 |
|
|
2836 |
denom = a*a+b*b-c*n; |
invdenom = a*a+b*b-c*n; |
2837 |
|
|
2838 |
/* Solve the system: sol = (D'*E*D)^{-1} D'*E*F */ |
/* Solve the system: sol = (D'*E*D)^{-1} D'*E*F */ |
2839 |
/* D'*E*F has been calculated in the same loop as matrix */ |
/* D'*E*F has been calculated in the same loop as matrix */ |
2843 |
sol[2] = b*DtimesF[0] - n*DtimesF[2] - a*DtimesF[3]; |
sol[2] = b*DtimesF[0] - n*DtimesF[2] - a*DtimesF[3]; |
2844 |
sol[3] = b*DtimesF[1] - a*DtimesF[2] - c*DtimesF[3]; |
sol[3] = b*DtimesF[1] - a*DtimesF[2] - c*DtimesF[3]; |
2845 |
|
|
2846 |
sol[0] /= denom; |
sol[0] /= invdenom; |
2847 |
sol[1] /= denom; |
sol[1] /= invdenom; |
2848 |
sol[2] /= denom; |
sol[2] /= invdenom; |
2849 |
sol[3] /= denom; |
sol[3] /= invdenom; |
2850 |
|
|
2851 |
meanx = meany = 0.; |
meanx = meany = 0.; |
2852 |
oldnum = 0; |
oldnum = 0; |
2853 |
for (my = 0; my < (uint32_t)MBh; my++) |
for (my = 1; my < (uint32_t)MBh-1; my++) |
2854 |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2855 |
{ |
{ |
2856 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2857 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const VECTOR mv = pMBs[mbnum].mvs[0]; |
|
const VECTOR mv = pMB->mvs[0]; |
|
2858 |
|
|
2859 |
if (!MBmask[mbnum]) |
if (!pMBs[mbnum].mcsel) |
2860 |
continue; |
continue; |
2861 |
|
|
2862 |
oldnum++; |
oldnum++; |
2863 |
meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ); |
meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ); |
2864 |
meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ); |
meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ); |
2865 |
} |
} |
2866 |
|
|
2867 |
if (4*meanx > oldnum) /* better fit than 0.25 is useless */ |
if (4*meanx > oldnum) /* better fit than 0.25 (=1/4pel) is useless */ |
2868 |
meanx /= oldnum; |
meanx /= oldnum; |
2869 |
else |
else |
2870 |
meanx = 0.25; |
meanx = 0.25; |
2874 |
else |
else |
2875 |
meany = 0.25; |
meany = 0.25; |
2876 |
|
|
|
/* fprintf(stderr,"sol = (%8.5f, %8.5f, %8.5f, %8.5f)\n",sol[0],sol[1],sol[2],sol[3]); |
|
|
fprintf(stderr,"meanx = %8.5f meany = %8.5f %d\n",meanx,meany, oldnum); |
|
|
*/ |
|
2877 |
num = 0; |
num = 0; |
2878 |
for (my = 0; my < (uint32_t)MBh; my++) |
for (my = 0; my < (uint32_t)MBh; my++) |
2879 |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
for (mx = 0; mx < (uint32_t)MBw; mx++) |
2880 |
{ |
{ |
2881 |
const int mbnum = mx + my * MBw; |
const int mbnum = mx + my * MBw; |
2882 |
const MACROBLOCK *pMB = &pMBs[mbnum]; |
const VECTOR mv = pMBs[mbnum].mvs[0]; |
|
const VECTOR mv = pMB->mvs[0]; |
|
2883 |
|
|
2884 |
if (!MBmask[mbnum]) |
if (!pMBs[mbnum].mcsel) |
2885 |
continue; |
continue; |
2886 |
|
|
2887 |
if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx ) |
if ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - (double)mv.x ) > meanx ) |
2888 |
|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) ) |
|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - (double)mv.y ) > meany ) ) |
2889 |
MBmask[mbnum]=0; |
pMBs[mbnum].mcsel=0; |
2890 |
else |
else |
2891 |
num++; |
num++; |
2892 |
} |
} |
2893 |
|
|
2894 |
} while ( (oldnum != num) && (num>=4) ); |
} while ( (oldnum != num) && (num>= minblocks) ); |
2895 |
|
|
2896 |
if (num < 4) |
if (num < minblocks) |
2897 |
{ |
{ |
2898 |
|
const int iEdgedWidth = pParam->edged_width; |
2899 |
|
num = 0; |
2900 |
|
|
2901 |
|
/* fprintf(stderr,"Warning! Unreliable GME (%d/%d blocks), falling back to translation.\n",num,MBh*MBw); |
2902 |
|
*/ |
2903 |
gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0; |
gmc.duv[0].x= gmc.duv[0].y= gmc.duv[1].x= gmc.duv[1].y= gmc.duv[2].x= gmc.duv[2].y=0; |
2904 |
|
|
2905 |
|
if (!(current->motion_flags & XVID_ME_GME_REFINE)) |
2906 |
|
return gmc; |
2907 |
|
|
2908 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) /* ignore boundary blocks */ |
2909 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) /* theirs MVs are often wrong */ |
2910 |
|
{ |
2911 |
|
const int mbnum = mx + my * MBw; |
2912 |
|
MACROBLOCK *const pMB = &pMBs[mbnum]; |
2913 |
|
const uint8_t *const pCur = current->image.y + 16*(my*iEdgedWidth + mx); |
2914 |
|
if ( (sad16 ( pCur, pCur+1 , iEdgedWidth, 65536) >= gradx ) |
2915 |
|
&& (sad16 ( pCur, pCur+iEdgedWidth, iEdgedWidth, 65536) >= grady ) ) |
2916 |
|
{ pMB->mcsel = 1; |
2917 |
|
gmc.duv[0].x += pMB->mvs[0].x; |
2918 |
|
gmc.duv[0].y += pMB->mvs[0].y; |
2919 |
|
num++; |
2920 |
|
} |
2921 |
|
} |
2922 |
|
|
2923 |
|
if (gmc.duv[0].x) |
2924 |
|
gmc.duv[0].x /= num; |
2925 |
|
if (gmc.duv[0].y) |
2926 |
|
gmc.duv[0].y /= num; |
2927 |
} else { |
} else { |
2928 |
|
|
2929 |
gmc.duv[0].x=(int)(sol[0]+0.5); |
gmc.duv[0].x=(int)(sol[0]+0.5); |
2932 |
gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5); |
gmc.duv[1].x=(int)(sol[1]*pParam->width+0.5); |
2933 |
gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5); |
gmc.duv[1].y=(int)(-sol[2]*pParam->width+0.5); |
2934 |
|
|
2935 |
gmc.duv[2].x=0; |
gmc.duv[2].x=-gmc.duv[1].y; /* two warp points only */ |
2936 |
gmc.duv[2].y=0; |
gmc.duv[2].y=gmc.duv[1].x; |
2937 |
|
} |
2938 |
|
if (num>maxblocks) |
2939 |
|
{ for (my = 1; my < (uint32_t)MBh-1; my++) |
2940 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2941 |
|
{ |
2942 |
|
const int mbnum = mx + my * MBw; |
2943 |
|
if (pMBs[mbnum-1].mcsel) |
2944 |
|
pMBs[mbnum].mcsel=0; |
2945 |
|
else |
2946 |
|
if (pMBs[mbnum-MBw].mcsel) |
2947 |
|
pMBs[mbnum].mcsel=0; |
2948 |
|
} |
2949 |
} |
} |
|
/* fprintf(stderr,"wp1 = ( %4d, %4d) wp2 = ( %4d, %4d) \n", gmc.duv[0].x, gmc.duv[0].y, gmc.duv[1].x, gmc.duv[1].y); */ |
|
|
|
|
|
free(MBmask); |
|
|
|
|
2950 |
return gmc; |
return gmc; |
2951 |
} |
} |
2952 |
|
|
2953 |
/* functions which perform BITS-based search/bitcount */ |
int |
2954 |
|
GlobalMotionEstRefine( |
2955 |
static int |
WARPPOINTS *const startwp, |
2956 |
CountMBBitsInter(SearchData * const Data, |
MACROBLOCK * const pMBs, |
|
const MACROBLOCK * const pMBs, const int x, const int y, |
|
2957 |
const MBParam * const pParam, |
const MBParam * const pParam, |
2958 |
const uint32_t MotionFlags) |
const FRAMEINFO * const current, |
2959 |
|
const FRAMEINFO * const reference, |
2960 |
|
const IMAGE * const pCurr, |
2961 |
|
const IMAGE * const pRef, |
2962 |
|
const IMAGE * const pRefH, |
2963 |
|
const IMAGE * const pRefV, |
2964 |
|
const IMAGE * const pRefHV) |
2965 |
{ |
{ |
2966 |
int i, iDirection; |
uint8_t* GMCblock = (uint8_t*)malloc(16*pParam->edged_width); |
2967 |
int32_t bsad[5]; |
WARPPOINTS bestwp=*startwp; |
2968 |
|
WARPPOINTS centerwp,currwp; |
2969 |
|
int gmcminSAD=0; |
2970 |
|
int gmcSAD=0; |
2971 |
|
int direction; |
2972 |
|
// int mx,my; |
2973 |
|
|
2974 |
|
/* use many blocks... */ |
2975 |
|
/* for (my = 0; my < (uint32_t)pParam->mb_height; my++) |
2976 |
|
for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) |
2977 |
|
{ |
2978 |
|
const int mbnum = mx + my * pParam->mb_width; |
2979 |
|
pMBs[mbnum].mcsel=1; |
2980 |
|
} |
2981 |
|
*/ |
2982 |
|
|
2983 |
CheckCandidate = CheckCandidateBits16; |
/* or rather don't use too many blocks... */ |
2984 |
|
/* |
2985 |
|
for (my = 1; my < (uint32_t)MBh-1; my++) |
2986 |
|
for (mx = 1; mx < (uint32_t)MBw-1; mx++) |
2987 |
|
{ |
2988 |
|
const int mbnum = mx + my * MBw; |
2989 |
|
if (MBmask[mbnum-1]) |
2990 |
|
MBmask[mbnum-1]=0; |
2991 |
|
else |
2992 |
|
if (MBmask[mbnum-MBw]) |
2993 |
|
MBmask[mbnum-1]=0; |
2994 |
|
|
|
if (Data->qpel) { |
|
|
for(i = 0; i < 5; i++) { |
|
|
Data->currentMV[i].x = Data->currentQMV[i].x/2; |
|
|
Data->currentMV[i].y = Data->currentQMV[i].y/2; |
|
2995 |
} |
} |
2996 |
Data->qpel_precision = 1; |
*/ |
2997 |
CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data); |
gmcminSAD = globalSAD(&bestwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
2998 |
|
|
2999 |
if (MotionFlags & (XVID_ME_HALFPELREFINE16_BITS | XVID_ME_EXTSEARCH_BITS)) { /* we have to prepare for halfpixel-precision search */ |
if ( (reference->coding_type == S_VOP) |
3000 |
for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i]; |
&& ( (reference->warp.duv[1].x != bestwp.duv[1].x) |
3001 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
|| (reference->warp.duv[1].y != bestwp.duv[1].y) |
3002 |
pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv); |
|| (reference->warp.duv[0].x != bestwp.duv[0].x) |
3003 |
Data->qpel_precision = 0; |
|| (reference->warp.duv[0].y != bestwp.duv[0].y) |
3004 |
if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1) |
|| (reference->warp.duv[2].x != bestwp.duv[2].x) |
3005 |
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
|| (reference->warp.duv[2].y != bestwp.duv[2].y) ) ) |
3006 |
|
{ |
3007 |
|
gmcSAD = globalSAD(&reference->warp, pParam, pMBs, |
3008 |
|
current, pRef, pCurr, GMCblock); |
3009 |
|
|
3010 |
|
if (gmcSAD < gmcminSAD) |
3011 |
|
{ bestwp = reference->warp; |
3012 |
|
gmcminSAD = gmcSAD; |
3013 |
} |
} |
|
|
|
|
} else { /* not qpel */ |
|
|
|
|
|
CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data); |
|
3014 |
} |
} |
3015 |
|
|
3016 |
if (MotionFlags&XVID_ME_EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection); |
do { |
3017 |
|
direction = 0; |
3018 |
if (MotionFlags&XVID_ME_HALFPELREFINE16_BITS) SubpelRefine(Data); |
centerwp = bestwp; |
3019 |
|
|
3020 |
if (Data->qpel) { |
currwp = centerwp; |
|
if (MotionFlags&(XVID_ME_EXTSEARCH_BITS | XVID_ME_HALFPELREFINE16_BITS)) { /* there was halfpel-precision search */ |
|
|
for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) { |
|
|
Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* we have found a better match */ |
|
|
Data->currentQMV[i].y = 2 * Data->currentMV[i].y; |
|
|
} |
|
3021 |
|
|
3022 |
/* preparing for qpel-precision search */ |
currwp.duv[0].x--; |
3023 |
Data->qpel_precision = 1; |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3024 |
get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16, |
if (gmcSAD < gmcminSAD) |
3025 |
pParam->width, pParam->height, Data->iFcode, 1, 0); |
{ bestwp = currwp; |
3026 |
|
gmcminSAD = gmcSAD; |
3027 |
|
direction = 1; |
3028 |
} |
} |
3029 |
if (MotionFlags&XVID_ME_QUARTERPELREFINE16_BITS) SubpelRefine(Data); |
else |
3030 |
} |
{ |
3031 |
|
currwp = centerwp; currwp.duv[0].x++; |
3032 |
if (MotionFlags&XVID_ME_CHECKPREDICTION_BITS) { /* let's check vector equal to prediction */ |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3033 |
VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV; |
if (gmcSAD < gmcminSAD) |
3034 |
if (!(Data->predMV.x == v->x && Data->predMV.y == v->y)) |
{ bestwp = currwp; |
3035 |
CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data); |
gmcminSAD = gmcSAD; |
3036 |
} |
direction = 2; |
3037 |
return Data->iMinSAD[0]; |
} |
3038 |
|
} |
3039 |
|
if (direction) continue; |
3040 |
|
|
3041 |
|
currwp = centerwp; currwp.duv[0].y--; |
3042 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3043 |
|
if (gmcSAD < gmcminSAD) |
3044 |
|
{ bestwp = currwp; |
3045 |
|
gmcminSAD = gmcSAD; |
3046 |
|
direction = 4; |
3047 |
} |
} |
3048 |
|
else |
|
static int |
|
|
CountMBBitsInter4v(const SearchData * const Data, |
|
|
MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, |
|
|
const int x, const int y, |
|
|
const MBParam * const pParam, const uint32_t MotionFlags, |
|
|
const VECTOR * const backup) |
|
3049 |
{ |
{ |
3050 |
|
currwp = centerwp; currwp.duv[0].y++; |
3051 |
int cbp = 0, bits = 0, t = 0, i, iDirection; |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3052 |
SearchData Data2, *Data8 = &Data2; |
if (gmcSAD < gmcminSAD) |
3053 |
int sumx = 0, sumy = 0; |
{ bestwp = currwp; |
3054 |
int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; |
gmcminSAD = gmcSAD; |
3055 |
uint8_t * ptr; |
direction = 8; |
3056 |
|
} |
3057 |
memcpy(Data8, Data, sizeof(SearchData)); |
} |
3058 |
CheckCandidate = CheckCandidateBits8; |
if (direction) continue; |
3059 |
|
|
3060 |
for (i = 0; i < 4; i++) { /* for all luma blocks */ |
currwp = centerwp; currwp.duv[1].x++; |
3061 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3062 |
Data8->iMinSAD = Data->iMinSAD + i + 1; |
if (gmcSAD < gmcminSAD) |
3063 |
Data8->currentMV = Data->currentMV + i + 1; |
{ bestwp = currwp; |
3064 |
Data8->currentQMV = Data->currentQMV + i + 1; |
gmcminSAD = gmcSAD; |
3065 |
Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
direction = 32; |
3066 |
Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
} |
3067 |
Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
currwp.duv[2].y++; |
3068 |
Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3069 |
Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
if (gmcSAD < gmcminSAD) |
3070 |
|
{ bestwp = currwp; |
3071 |
if(Data->qpel) { |
gmcminSAD = gmcSAD; |
3072 |
Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i); |
direction = 1024; |
3073 |
if (i != 0) t = d_mv_bits( Data8->currentQMV->x, Data8->currentQMV->y, |
} |
3074 |
Data8->predMV, Data8->iFcode, 0, 0); |
|
3075 |
} else { |
currwp = centerwp; currwp.duv[1].x--; |
3076 |
Data8->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, i); |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3077 |
if (i != 0) t = d_mv_bits( Data8->currentMV->x, Data8->currentMV->y, |
if (gmcSAD < gmcminSAD) |
3078 |
Data8->predMV, Data8->iFcode, 0, 0); |
{ bestwp = currwp; |
3079 |
|
gmcminSAD = gmcSAD; |
3080 |
|
direction = 16; |
3081 |
} |
} |
3082 |
|
else |
|
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
|
|
pParam->width, pParam->height, Data8->iFcode, Data8->qpel, 0); |
|
|
|
|
|
*Data8->iMinSAD += BITS_MULT*t; |
|
|
|
|
|
Data8->qpel_precision = Data8->qpel; |
|
|
/* checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far) */ |
|
3083 |
{ |
{ |
3084 |
VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV; |
currwp = centerwp; currwp.duv[1].x++; |
3085 |
if (!MVequal (*v, backup[i+1]) ) |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3086 |
CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8); |
if (gmcSAD < gmcminSAD) |
3087 |
|
{ bestwp = currwp; |
3088 |
|
gmcminSAD = gmcSAD; |
3089 |
|
direction = 32; |
3090 |
|
} |
3091 |
|
} |
3092 |
|
if (direction) continue; |
3093 |
|
|
3094 |
|
|
3095 |
|
currwp = centerwp; currwp.duv[1].y--; |
3096 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3097 |
|
if (gmcSAD < gmcminSAD) |
3098 |
|
{ bestwp = currwp; |
3099 |
|
gmcminSAD = gmcSAD; |
3100 |
|
direction = 64; |
3101 |
} |
} |
3102 |
|
else |
3103 |
if (Data8->qpel) { |
{ |
3104 |
if (MotionFlags&XVID_ME_HALFPELREFINE8_BITS || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_BITS)) { /* halfpixel motion search follows */ |
currwp = centerwp; currwp.duv[1].y++; |
3105 |
int32_t s = *Data8->iMinSAD; |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3106 |
Data8->currentMV->x = Data8->currentQMV->x/2; |
if (gmcSAD < gmcminSAD) |
3107 |
Data8->currentMV->y = Data8->currentQMV->y/2; |
{ bestwp = currwp; |
3108 |
Data8->qpel_precision = 0; |
gmcminSAD = gmcSAD; |
3109 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
direction = 128; |
3110 |
pParam->width, pParam->height, Data8->iFcode - 1, 0, 0); |
} |
3111 |
|
} |
3112 |
if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1) |
if (direction) continue; |
3113 |
CheckCandidateBits8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8); |
|
3114 |
|
currwp = centerwp; currwp.duv[2].x--; |
3115 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3116 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
if (gmcSAD < gmcminSAD) |
3117 |
|
{ bestwp = currwp; |
3118 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
gmcminSAD = gmcSAD; |
3119 |
SubpelRefine(Data8); |
direction = 256; |
|
|
|
|
if(s > *Data8->iMinSAD) { /* we have found a better match */ |
|
|
Data8->currentQMV->x = 2*Data8->currentMV->x; |
|
|
Data8->currentQMV->y = 2*Data8->currentMV->y; |
|
3120 |
} |
} |
3121 |
|
else |
3122 |
Data8->qpel_precision = 1; |
{ |
3123 |
get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8, |
currwp = centerwp; currwp.duv[2].x++; |
3124 |
pParam->width, pParam->height, Data8->iFcode, 1, 0); |
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3125 |
|
if (gmcSAD < gmcminSAD) |
3126 |
|
{ bestwp = currwp; |
3127 |
|
gmcminSAD = gmcSAD; |
3128 |
|
direction = 512; |
3129 |
|
} |
3130 |
|
} |
3131 |
|
if (direction) continue; |
3132 |
|
|
3133 |
|
currwp = centerwp; currwp.duv[2].y--; |
3134 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3135 |
|
if (gmcSAD < gmcminSAD) |
3136 |
|
{ bestwp = currwp; |
3137 |
|
gmcminSAD = gmcSAD; |
3138 |
|
direction = 1024; |
3139 |
} |
} |
3140 |
if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8); |
else |
3141 |
|
{ |
3142 |
} else { /* not qpel */ |
currwp = centerwp; currwp.duv[2].y++; |
3143 |
|
gmcSAD = globalSAD(&currwp, pParam, pMBs, current, pRef, pCurr, GMCblock); |
3144 |
if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) /* extsearch */ |
if (gmcSAD < gmcminSAD) |
3145 |
SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255); |
{ bestwp = currwp; |
3146 |
|
gmcminSAD = gmcSAD; |
3147 |
if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) |
direction = 2048; |
|
SubpelRefine(Data8); /* halfpel refinement */ |
|
3148 |
} |
} |
|
|
|
|
/* checking vector equal to predicion */ |
|
|
if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_BITS) { |
|
|
const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV; |
|
|
if (!MVequal(*v, Data8->predMV)) |
|
|
CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8); |
|
3149 |
} |
} |
3150 |
|
} while (direction); |
3151 |
|
free(GMCblock); |
3152 |
|
|
3153 |
bits += *Data8->iMinSAD; |
*startwp = bestwp; |
|
if (bits >= Data->iMinSAD[0]) return bits; /* no chances for INTER4V */ |
|
3154 |
|
|
3155 |
/* MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else */ |
return gmcminSAD; |
|
if(Data->qpel) { |
|
|
pMB->pmvs[i].x = Data8->currentQMV->x - Data8->predMV.x; |
|
|
pMB->pmvs[i].y = Data8->currentQMV->y - Data8->predMV.y; |
|
|
pMB->qmvs[i] = *Data8->currentQMV; |
|
|
sumx += Data8->currentQMV->x/2; |
|
|
sumy += Data8->currentQMV->y/2; |
|
|
} else { |
|
|
pMB->pmvs[i].x = Data8->currentMV->x - Data8->predMV.x; |
|
|
pMB->pmvs[i].y = Data8->currentMV->y - Data8->predMV.y; |
|
|
sumx += Data8->currentMV->x; |
|
|
sumy += Data8->currentMV->y; |
|
3156 |
} |
} |
|
pMB->mvs[i] = *Data8->currentMV; |
|
|
pMB->sad8[i] = 4 * *Data8->iMinSAD; |
|
|
if (Data8->temp[0]) cbp |= 1 << (5 - i); |
|
3157 |
|
|
3158 |
} /* /for all luma blocks */ |
int |
3159 |
|
globalSAD(const WARPPOINTS *const wp, |
3160 |
bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len; |
const MBParam * const pParam, |
3161 |
|
const MACROBLOCK * const pMBs, |
3162 |
|
const FRAMEINFO * const current, |
3163 |
|
const IMAGE * const pRef, |
3164 |
|
const IMAGE * const pCurr, |
3165 |
|
uint8_t *const GMCblock) |
3166 |
|
{ |
3167 |
|
NEW_GMC_DATA gmc_data; |
3168 |
|
int iSAD, gmcSAD=0; |
3169 |
|
int num=0; |
3170 |
|
unsigned int mx, my; |
3171 |
|
|
3172 |
/* let's check chroma */ |
generate_GMCparameters( 3, 3, wp, pParam->width, pParam->height, &gmc_data); |
|
sumx = (sumx >> 3) + roundtab_76[sumx & 0xf]; |
|
|
sumy = (sumy >> 3) + roundtab_76[sumy & 0xf]; |
|
3173 |
|
|
3174 |
/* chroma U */ |
for (my = 0; my < (uint32_t)pParam->mb_height; my++) |
3175 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
for (mx = 0; mx < (uint32_t)pParam->mb_width; mx++) { |
|
transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); |
|
|
bits += Block_CalcBits(coeff, in, Data->iQuant, Data->quant_type, &cbp, 4); |
|
3176 |
|
|
3177 |
if (bits >= *Data->iMinSAD) return bits; |
const int mbnum = mx + my * pParam->mb_width; |
3178 |
|
const int iEdgedWidth = pParam->edged_width; |
3179 |
|
|
3180 |
/* chroma V */ |
if (!pMBs[mbnum].mcsel) |
3181 |
ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); |
continue; |
|
transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); |
|
|
bits += Block_CalcBits(coeff, in, Data->iQuant, Data->quant_type, &cbp, 5); |
|
3182 |
|
|
3183 |
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; |
gmc_data.predict_16x16(&gmc_data, GMCblock, |
3184 |
|
pRef->y, |
3185 |
|
iEdgedWidth, |
3186 |
|
iEdgedWidth, |
3187 |
|
mx, my, |
3188 |
|
pParam->m_rounding_type); |
3189 |
|
|
3190 |
|
iSAD = sad16 ( pCurr->y + 16*(my*iEdgedWidth + mx), |
3191 |
|
GMCblock , iEdgedWidth, 65536); |
3192 |
|
iSAD -= pMBs[mbnum].sad16; |
3193 |
|
|
3194 |
return bits; |
if (iSAD<0) |
3195 |
|
gmcSAD += iSAD; |
3196 |
|
num++; |
3197 |
} |
} |
3198 |
|
return gmcSAD; |
|
static int |
|
|
CountMBBitsIntra(const SearchData * const Data) |
|
|
{ |
|
|
int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ |
|
|
int cbp = 0, i, dc = 0; |
|
|
int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; |
|
|
|
|
|
for(i = 0; i < 4; i++) { |
|
|
int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); |
|
|
transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth); |
|
|
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, i, &dc); |
|
|
|
|
|
if (bits >= Data->iMinSAD[0]) return bits; |
|
3199 |
} |
} |
3200 |
|
|
|
bits += BITS_MULT*xvid_cbpy_tab[cbp>>2].len; |
|
|
|
|
|
/*chroma U */ |
|
|
transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); |
|
|
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, 4, &dc); |
|
|
|
|
|
if (bits >= Data->iMinSAD[0]) return bits; |
|
|
|
|
|
/* chroma V */ |
|
|
transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); |
|
|
bits += Block_CalcBitsIntra(coeff, in, Data->iQuant, Data->quant_type, &cbp, 5, &dc); |
|
|
|
|
|
bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len; |
|
|
|
|
|
return bits; |
|
|
} |
|