1 |
/************************************************************************** |
/***************************************************************************** |
2 |
* |
* |
3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Motion estimation header - |
* - Motion Estimation related header - |
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 |
23 |
* |
* |
24 |
* $Id$ |
* $Id$ |
25 |
* |
* |
26 |
***************************************************************************/ |
****************************************************************************/ |
27 |
|
|
28 |
#ifndef _MOTION_EST_H_ |
#ifndef _MOTION_EST_H_ |
29 |
#define _MOTION_EST_H_ |
#define _MOTION_EST_H_ |
34 |
|
|
35 |
/* hard coded motion search parameters for motion_est and smp_motion_est */ |
/* hard coded motion search parameters for motion_est and smp_motion_est */ |
36 |
|
|
37 |
// very large value |
/* very large value */ |
38 |
#define MV_MAX_ERROR (4096 * 256) |
#define MV_MAX_ERROR (4096 * 256) |
39 |
|
|
40 |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
/* INTER bias for INTER/INTRA decision; mpeg4 spec suggests 2*nb */ |
86 |
(int)(21.2656*NEIGH_TEND_8X8 + 0.5), (int)(24.8580*NEIGH_TEND_8X8 + 0.5), |
(int)(21.2656*NEIGH_TEND_8X8 + 0.5), (int)(24.8580*NEIGH_TEND_8X8 + 0.5), |
87 |
(int)(29.6436*NEIGH_TEND_8X8 + 0.5), (int)(36.4949*NEIGH_TEND_8X8 + 0.5) }; |
(int)(29.6436*NEIGH_TEND_8X8 + 0.5), (int)(36.4949*NEIGH_TEND_8X8 + 0.5) }; |
88 |
|
|
89 |
// mv.length table |
/* mv.length table */ |
90 |
static const int mvtab[64] = { |
static const int mvtab[64] = { |
91 |
1, 2, 3, 4, 6, 7, 7, 7, |
1, 2, 3, 4, 6, 7, 7, 7, |
92 |
9, 9, 9, 10, 10, 10, 10, 10, |
9, 9, 9, 10, 10, 10, 10, 10, |
104 |
|
|
105 |
typedef struct |
typedef struct |
106 |
{ |
{ |
107 |
// general fields |
/* general fields */ |
108 |
int max_dx, min_dx, max_dy, min_dy; |
int max_dx, min_dx, max_dy, min_dy; |
109 |
uint32_t rounding; |
uint32_t rounding; |
110 |
VECTOR predMV; |
VECTOR predMV; |
111 |
VECTOR * currentMV; |
VECTOR * currentMV; |
112 |
VECTOR * currentQMV; |
VECTOR * currentQMV; |
113 |
int32_t * iMinSAD; |
int32_t * iMinSAD; |
114 |
const uint8_t * RefP[6]; // N, V, H, HV, cU, cV |
const uint8_t * RefP[6]; /* N, V, H, HV, cU, cV */ |
115 |
const uint8_t * CurU; |
const uint8_t * CurU; |
116 |
const uint8_t * CurV; |
const uint8_t * CurV; |
117 |
uint8_t * RefQ; |
uint8_t * RefQ; |
124 |
int qpel, qpel_precision; |
int qpel, qpel_precision; |
125 |
int chroma; |
int chroma; |
126 |
int rrv; |
int rrv; |
127 |
//fields for interpolate and direct modes |
|
128 |
const uint8_t * b_RefP[6]; // N, V, H, HV, cU, cV |
/* fields for interpolate and direct modes */ |
129 |
|
const uint8_t * b_RefP[6]; /* N, V, H, HV, cU, cV */ |
130 |
VECTOR bpredMV; |
VECTOR bpredMV; |
131 |
uint32_t bFcode; |
uint32_t bFcode; |
132 |
// fields for direct mode |
|
133 |
|
/* fields for direct mode */ |
134 |
VECTOR directmvF[4]; |
VECTOR directmvF[4]; |
135 |
VECTOR directmvB[4]; |
VECTOR directmvB[4]; |
136 |
const VECTOR * referencemv; |
const VECTOR * referencemv; |
137 |
// BITS/R-D stuff |
|
138 |
|
/* BITS/R-D stuff */ |
139 |
int16_t * dctSpace; |
int16_t * dctSpace; |
140 |
uint32_t iQuant; |
uint32_t iQuant; |
141 |
uint32_t quant_type; |
uint32_t quant_type; |
302 |
int CodeCoeffIntra_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); |
int CodeCoeffIntra_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); |
303 |
int CodeCoeffInter_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); |
int CodeCoeffInter_CalcBits(const int16_t qcoeff[64], const uint16_t * zigzag); |
304 |
|
|
305 |
#define LAMBDA ( (int)(1.0*BITS_MULT) ) |
/* one over lambda for R-D mode decision and motion search */ |
306 |
|
#define LAMBDA ( (int)(BITS_MULT/1.0) ) |
307 |
|
|
308 |
static __inline unsigned int |
static __inline unsigned int |
309 |
Block_CalcBits( int16_t * const coeff, |
Block_CalcBits( int16_t * const coeff, |
310 |
int16_t * const data, |
int16_t * const data, |
311 |
|
int16_t * const dqcoeff, |
312 |
const uint32_t quant, const int quant_type, |
const uint32_t quant, const int quant_type, |
313 |
uint32_t * cbp, |
uint32_t * cbp, |
314 |
const int block) |
const int block) |
315 |
{ |
{ |
316 |
int sum; |
int sum; |
317 |
int bits; |
int bits; |
|
const int lambda = LAMBDA*quant*quant; |
|
318 |
int distortion = 0; |
int distortion = 0; |
319 |
int i; |
int i; |
320 |
|
|
328 |
bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_tables[0]); |
329 |
} else bits = 0; |
} else bits = 0; |
330 |
|
|
331 |
if (quant_type == 0) dequant_inter(coeff, coeff, quant); |
if (quant_type == 0) dequant_inter(dqcoeff, coeff, quant); |
332 |
else dequant4_inter(coeff, coeff, quant); |
else dequant4_inter(dqcoeff, coeff, quant); |
333 |
|
|
334 |
for (i = 0; i < 64; i++) { |
for (i = 0; i < 64; i++) { |
335 |
distortion += (data[i] - coeff[i])*(data[i] - coeff[i]); |
distortion += (data[i] - dqcoeff[i])*(data[i] - dqcoeff[i]); |
336 |
} |
} |
|
bits += (BITS_MULT*BITS_MULT*distortion)/lambda; |
|
337 |
|
|
338 |
|
bits += (LAMBDA*distortion)/(quant*quant); |
339 |
|
|
340 |
return bits; |
return bits; |
341 |
} |
} |
343 |
static __inline unsigned int |
static __inline unsigned int |
344 |
Block_CalcBitsIntra(int16_t * const coeff, |
Block_CalcBitsIntra(int16_t * const coeff, |
345 |
int16_t * const data, |
int16_t * const data, |
346 |
|
int16_t * const dqcoeff, |
347 |
const uint32_t quant, const int quant_type, |
const uint32_t quant, const int quant_type, |
348 |
uint32_t * cbp, |
uint32_t * cbp, |
349 |
const int block, |
const int block, |
350 |
int * dcpred) |
int * dcpred) |
351 |
{ |
{ |
352 |
int bits, i; |
int bits, i; |
|
const int lambda = LAMBDA*quant*quant; |
|
353 |
int distortion = 0; |
int distortion = 0; |
354 |
uint32_t iDcScaler = get_dc_scaler(quant, block > 3); |
uint32_t iDcScaler = get_dc_scaler(quant, block < 4); |
355 |
int b_dc; |
int b_dc; |
356 |
|
|
357 |
fdct(data); |
fdct(data); |
366 |
*dcpred = b_dc; |
*dcpred = b_dc; |
367 |
} |
} |
368 |
|
|
|
*cbp |= 1 << (5 - block); |
|
369 |
bits = BITS_MULT*CodeCoeffIntra_CalcBits(coeff, scan_tables[0]); |
bits = BITS_MULT*CodeCoeffIntra_CalcBits(coeff, scan_tables[0]); |
|
bits += BITS_MULT*dcy_tab[coeff[0] + 255].len; |
|
370 |
if (bits != 0) *cbp |= 1 << (5 - block); |
if (bits != 0) *cbp |= 1 << (5 - block); |
371 |
|
|
372 |
|
if (block < 4) bits += BITS_MULT*dcy_tab[coeff[0] + 255].len; |
373 |
|
else bits += BITS_MULT*dcc_tab[coeff[0] + 255].len; |
374 |
|
|
375 |
coeff[0] = b_dc; |
coeff[0] = b_dc; |
376 |
if (quant_type == 0) dequant_intra(coeff, coeff, quant, iDcScaler); |
if (quant_type == 0) dequant_intra(dqcoeff, coeff, quant, iDcScaler); |
377 |
else dequant4_intra(coeff, coeff, quant, iDcScaler); |
else dequant4_intra(dqcoeff, coeff, quant, iDcScaler); |
378 |
|
|
379 |
for (i = 0; i < 64; i++) { |
for (i = 0; i < 64; i++) { |
380 |
distortion += (data[i] - coeff[i])*(data[i] - coeff[i]); |
distortion += (data[i] - dqcoeff[i])*(data[i] - dqcoeff[i]); |
381 |
} |
} |
382 |
|
|
383 |
bits += (BITS_MULT*BITS_MULT*distortion)/lambda; |
bits += (LAMBDA*distortion)/(quant*quant); |
384 |
|
|
385 |
return bits; |
return bits; |
386 |
} |
} |