1 |
/****************************************************************************** |
/***************************************************************************** |
2 |
* * |
* |
3 |
* This file is part of XviD, a free MPEG-4 video encoder/decoder * |
* XVID MPEG-4 VIDEO CODEC |
4 |
* * |
* - MB Transfert/Quantization functions - |
5 |
* XviD is an implementation of a part of one or more MPEG-4 Video tools * |
* |
6 |
* as specified in ISO/IEC 14496-2 standard. Those intending to use this * |
* Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
7 |
* software module in hardware or software products are advised that its * |
* 2001-2003 Michael Militzer <isibaar@xvid.org> |
8 |
* use may infringe existing patents or copyrights, and any such use * |
* 2003 Edouard Gomez <ed.gomez@free.fr> |
9 |
* would be at such party's own risk. The original developer of this * |
* |
10 |
* software module and his/her company, and subsequent editors and their * |
* This program is free software ; you can redistribute it and/or modify |
11 |
* companies, will have no liability for use of this software or * |
* it under the terms of the GNU General Public License as published by |
12 |
* modifications or derivatives thereof. * |
* the Free Software Foundation ; either version 2 of the License, or |
13 |
* * |
* (at your option) any later version. |
14 |
* XviD is free software; you can redistribute it and/or modify it * |
* |
15 |
* under the terms of the GNU General Public License as published by * |
* This program is distributed in the hope that it will be useful, |
16 |
* the Free Software Foundation; either version 2 of the License, or * |
* but WITHOUT ANY WARRANTY ; without even the implied warranty of |
17 |
* (at your option) any later version. * |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
18 |
* * |
* GNU General Public License for more details. |
19 |
* XviD is distributed in the hope that it will be useful, but * |
* |
20 |
* WITHOUT ANY WARRANTY; without even the implied warranty of * |
* You should have received a copy of the GNU General Public License |
21 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
* along with this program ; if not, write to the Free Software |
22 |
* GNU General Public License for more details. * |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
* * |
* |
24 |
* You should have received a copy of the GNU General Public License * |
* $Id$ |
25 |
* along with this program; if not, write to the Free Software * |
* |
26 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * |
****************************************************************************/ |
|
* * |
|
|
******************************************************************************/ |
|
|
|
|
|
/****************************************************************************** |
|
|
* * |
|
|
* mbtransquant.c * |
|
|
* * |
|
|
* Copyright (C) 2001 - Peter Ross <pross@cs.rmit.edu.au> * |
|
|
* Copyright (C) 2001 - Michael Militzer <isibaar@xvid.org> * |
|
|
* * |
|
|
* For more information visit the XviD homepage: http://www.xvid.org * |
|
|
* * |
|
|
******************************************************************************/ |
|
|
|
|
|
/****************************************************************************** |
|
|
* * |
|
|
* Revision history: * |
|
|
* * |
|
|
* 26.03.2002 interlacing support - moved transfers outside loops |
|
|
* 22.12.2001 get_dc_scaler() moved to common.h |
|
|
* 19.11.2001 introduced coefficient thresholding (Isibaar) * |
|
|
* 17.11.2001 initial version * |
|
|
* * |
|
|
******************************************************************************/ |
|
27 |
|
|
28 |
|
#include <stdio.h> |
29 |
|
#include <stdlib.h> |
30 |
#include <string.h> |
#include <string.h> |
31 |
|
|
32 |
#include "../portab.h" |
#include "../portab.h" |
35 |
#include "../global.h" |
#include "../global.h" |
36 |
#include "mem_transfer.h" |
#include "mem_transfer.h" |
37 |
#include "timer.h" |
#include "timer.h" |
38 |
|
#include "../bitstream/mbcoding.h" |
39 |
|
#include "../bitstream/zigzag.h" |
40 |
#include "../dct/fdct.h" |
#include "../dct/fdct.h" |
41 |
#include "../dct/idct.h" |
#include "../dct/idct.h" |
42 |
#include "../quant/quant_mpeg4.h" |
#include "../quant/quant.h" |
|
#include "../quant/quant_h263.h" |
|
43 |
#include "../encoder.h" |
#include "../encoder.h" |
44 |
|
|
45 |
#define MIN(X, Y) ((X)<(Y)?(X):(Y)) |
#include "../quant/quant_matrix.h" |
|
#define MAX(X, Y) ((X)>(Y)?(X):(Y)) |
|
46 |
|
|
47 |
#define TOOSMALL_LIMIT 1 /* skip blocks having a coefficient sum below this value */ |
MBFIELDTEST_PTR MBFieldTest; |
48 |
|
|
49 |
/* this isnt pretty, but its better than 20 ifdefs */ |
/* |
50 |
|
* Skip blocks having a coefficient sum below this value. This value will be |
51 |
void MBTransQuantIntra(const MBParam *pParam, |
* corrected according to the MB quantizer to avoid artifacts for quant==1 |
52 |
MACROBLOCK * pMB, |
*/ |
53 |
const uint32_t x_pos, |
#define PVOP_TOOSMALL_LIMIT 1 |
54 |
const uint32_t y_pos, |
#define BVOP_TOOSMALL_LIMIT 3 |
|
int16_t data[6*64], |
|
|
int16_t qcoeff[6*64], |
|
|
IMAGE * const pCurrent) |
|
55 |
|
|
56 |
|
/***************************************************************************** |
57 |
|
* Local functions |
58 |
|
****************************************************************************/ |
59 |
|
|
60 |
|
/* permute block and return field dct choice */ |
61 |
|
static __inline uint32_t |
62 |
|
MBDecideFieldDCT(int16_t data[6 * 64]) |
63 |
{ |
{ |
64 |
|
uint32_t field = MBFieldTest(data); |
65 |
|
|
66 |
const uint32_t stride = pParam->edged_width; |
if (field) |
67 |
uint32_t i; |
MBFrameToField(data); |
|
uint32_t iQuant = pParam->quant; |
|
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
|
|
|
|
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
|
|
pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
|
|
pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
|
68 |
|
|
69 |
start_timer(); |
return field; |
70 |
transfer_8to16copy(&data[0*64], pY_Cur, stride); |
} |
|
transfer_8to16copy(&data[1*64], pY_Cur + 8, stride); |
|
|
transfer_8to16copy(&data[2*64], pY_Cur + 8 * stride, stride); |
|
|
transfer_8to16copy(&data[3*64], pY_Cur + 8 * stride + 8, stride); |
|
|
transfer_8to16copy(&data[4*64], pU_Cur, stride / 2); |
|
|
transfer_8to16copy(&data[5*64], pV_Cur, stride / 2); |
|
|
stop_transfer_timer(); |
|
71 |
|
|
72 |
|
/* Performs Forward DCT on all blocks */ |
73 |
|
static __inline void |
74 |
|
MBfDCT(const MBParam * const pParam, |
75 |
|
const FRAMEINFO * const frame, |
76 |
|
MACROBLOCK * const pMB, |
77 |
|
uint32_t x_pos, |
78 |
|
uint32_t y_pos, |
79 |
|
int16_t data[6 * 64]) |
80 |
|
{ |
81 |
|
/* Handles interlacing */ |
82 |
start_timer(); |
start_timer(); |
83 |
pMB->field_dct = 0; |
pMB->field_dct = 0; |
84 |
if (pParam->global_flags & XVID_INTERLACING) |
if ((frame->vol_flags & XVID_VOL_INTERLACING) && |
85 |
{ |
(x_pos>0) && (x_pos<pParam->mb_width-1) && |
86 |
|
(y_pos>0) && (y_pos<pParam->mb_height-1)) { |
87 |
pMB->field_dct = MBDecideFieldDCT(data); |
pMB->field_dct = MBDecideFieldDCT(data); |
88 |
} |
} |
89 |
stop_interlacing_timer(); |
stop_interlacing_timer(); |
90 |
|
|
91 |
for(i = 0; i < 6; i++) |
/* Perform DCT */ |
|
{ |
|
|
uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); |
|
|
|
|
92 |
start_timer(); |
start_timer(); |
93 |
fdct(&data[i*64]); |
fdct((short * const)&data[0 * 64]); |
94 |
|
fdct((short * const)&data[1 * 64]); |
95 |
|
fdct((short * const)&data[2 * 64]); |
96 |
|
fdct((short * const)&data[3 * 64]); |
97 |
|
fdct((short * const)&data[4 * 64]); |
98 |
|
fdct((short * const)&data[5 * 64]); |
99 |
stop_dct_timer(); |
stop_dct_timer(); |
100 |
|
} |
101 |
|
|
102 |
if (pParam->quant_type == H263_QUANT) |
/* Performs Inverse DCT on all blocks */ |
103 |
|
static __inline void |
104 |
|
MBiDCT(int16_t data[6 * 64], |
105 |
|
const uint8_t cbp) |
106 |
{ |
{ |
107 |
start_timer(); |
start_timer(); |
108 |
quant_intra(&qcoeff[i*64], &data[i*64], iQuant, iDcScaler); |
if(cbp & (1 << (5 - 0))) idct((short * const)&data[0 * 64]); |
109 |
stop_quant_timer(); |
if(cbp & (1 << (5 - 1))) idct((short * const)&data[1 * 64]); |
110 |
|
if(cbp & (1 << (5 - 2))) idct((short * const)&data[2 * 64]); |
111 |
start_timer(); |
if(cbp & (1 << (5 - 3))) idct((short * const)&data[3 * 64]); |
112 |
dequant_intra(&data[i*64], &qcoeff[i*64], iQuant, iDcScaler); |
if(cbp & (1 << (5 - 4))) idct((short * const)&data[4 * 64]); |
113 |
stop_iquant_timer(); |
if(cbp & (1 << (5 - 5))) idct((short * const)&data[5 * 64]); |
114 |
|
stop_idct_timer(); |
115 |
} |
} |
116 |
else |
|
117 |
|
/* Quantize all blocks -- Intra mode */ |
118 |
|
static __inline void |
119 |
|
MBQuantIntra(const MBParam * pParam, |
120 |
|
const FRAMEINFO * const frame, |
121 |
|
const MACROBLOCK * pMB, |
122 |
|
int16_t qcoeff[6 * 64], |
123 |
|
int16_t data[6*64]) |
124 |
{ |
{ |
125 |
start_timer(); |
int mpeg; |
126 |
quant4_intra(&qcoeff[i*64], &data[i*64], iQuant, iDcScaler); |
int scaler_lum, scaler_chr; |
|
stop_quant_timer(); |
|
127 |
|
|
128 |
start_timer(); |
quant_intraFuncPtr const quant[2] = |
129 |
dequant4_intra(&data[i*64], &qcoeff[i*64], iQuant, iDcScaler); |
{ |
130 |
stop_iquant_timer(); |
quant_h263_intra, |
131 |
} |
quant_mpeg_intra |
132 |
|
}; |
133 |
|
|
134 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
135 |
|
scaler_lum = get_dc_scaler(pMB->quant, 1); |
136 |
|
scaler_chr = get_dc_scaler(pMB->quant, 0); |
137 |
|
|
138 |
|
/* Quantize the block */ |
139 |
start_timer(); |
start_timer(); |
140 |
idct(&data[i*64]); |
quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
141 |
stop_idct_timer(); |
quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
142 |
|
quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
143 |
|
quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices); |
144 |
|
quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
145 |
|
quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices); |
146 |
|
stop_quant_timer(); |
147 |
} |
} |
148 |
|
|
149 |
start_timer(); |
/* DeQuantize all blocks -- Intra mode */ |
150 |
if (pMB->field_dct) |
static __inline void |
151 |
|
MBDeQuantIntra(const MBParam * pParam, |
152 |
|
const int iQuant, |
153 |
|
int16_t qcoeff[6 * 64], |
154 |
|
int16_t data[6*64]) |
155 |
{ |
{ |
156 |
MBFieldToFrame(data); |
int mpeg; |
157 |
} |
int scaler_lum, scaler_chr; |
|
stop_interlacing_timer(); |
|
158 |
|
|
159 |
start_timer(); |
quant_intraFuncPtr const dequant[2] = |
160 |
transfer_16to8copy(pY_Cur, &data[0*64], stride); |
{ |
161 |
transfer_16to8copy(pY_Cur + 8, &data[1*64], stride); |
dequant_h263_intra, |
162 |
transfer_16to8copy(pY_Cur + 8 * stride, &data[2*64], stride); |
dequant_mpeg_intra |
163 |
transfer_16to8copy(pY_Cur + 8 + 8 * stride, &data[3*64], stride); |
}; |
164 |
transfer_16to8copy(pU_Cur, &data[4*64], stride / 2); |
|
165 |
transfer_16to8copy(pV_Cur, &data[5*64], stride / 2); |
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
166 |
stop_transfer_timer(); |
scaler_lum = get_dc_scaler(iQuant, 1); |
167 |
|
scaler_chr = get_dc_scaler(iQuant, 0); |
168 |
|
|
169 |
|
start_timer(); |
170 |
|
dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
171 |
|
dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
172 |
|
dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
173 |
|
dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices); |
174 |
|
dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices); |
175 |
|
dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices); |
176 |
|
stop_iquant_timer(); |
177 |
} |
} |
178 |
|
|
179 |
|
static int |
180 |
uint8_t MBTransQuantInter(const MBParam *pParam, |
dct_quantize_trellis_c(int16_t *const Out, |
181 |
MACROBLOCK * pMB, |
const int16_t *const In, |
182 |
const uint32_t x_pos, const uint32_t y_pos, |
int Q, |
183 |
|
const uint16_t * const Zigzag, |
184 |
|
const uint16_t * const QuantMatrix, |
185 |
|
int Non_Zero, |
186 |
|
int Sum, |
187 |
|
int Lambda_Mod); |
188 |
|
|
189 |
|
/* Quantize all blocks -- Inter mode */ |
190 |
|
static __inline uint8_t |
191 |
|
MBQuantInter(const MBParam * pParam, |
192 |
|
const FRAMEINFO * const frame, |
193 |
|
const MACROBLOCK * pMB, |
194 |
int16_t data[6*64], |
int16_t data[6*64], |
195 |
int16_t qcoeff[6*64], |
int16_t qcoeff[6*64], |
196 |
IMAGE * const pCurrent) |
int bvop, |
197 |
|
int limit) |
198 |
{ |
{ |
199 |
|
|
200 |
const uint32_t stride = pParam->edged_width; |
int i; |
|
uint32_t i; |
|
|
uint32_t iQuant = pParam->quant; |
|
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
|
201 |
uint8_t cbp = 0; |
uint8_t cbp = 0; |
202 |
uint32_t sum; |
int sum; |
203 |
|
int code_block, mpeg; |
204 |
|
|
205 |
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
quant_interFuncPtr const quant[2] = |
206 |
pU_Cur = pCurrent->u + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
{ |
207 |
pV_Cur = pCurrent->v + (y_pos << 3) * (stride >> 1) + (x_pos << 3); |
quant_h263_inter, |
208 |
|
quant_mpeg_inter |
209 |
|
}; |
210 |
|
|
211 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
212 |
|
|
213 |
|
for (i = 0; i < 6; i++) { |
214 |
|
|
215 |
|
/* Quantize the block */ |
216 |
start_timer(); |
start_timer(); |
217 |
pMB->field_dct = 0; |
|
218 |
if (pParam->global_flags & XVID_INTERLACING) |
sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices); |
219 |
|
|
220 |
|
if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) { |
221 |
|
const uint16_t *matrix; |
222 |
|
const static uint16_t h263matrix[] = |
223 |
{ |
{ |
224 |
pMB->field_dct = MBDecideFieldDCT(data); |
16, 16, 16, 16, 16, 16, 16, 16, |
225 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
226 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
227 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
228 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
229 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
230 |
|
16, 16, 16, 16, 16, 16, 16, 16, |
231 |
|
16, 16, 16, 16, 16, 16, 16, 16 |
232 |
|
}; |
233 |
|
|
234 |
|
matrix = (mpeg)?get_inter_matrix(pParam->mpeg_quant_matrices):h263matrix; |
235 |
|
sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64], |
236 |
|
pMB->quant, &scan_tables[0][0], |
237 |
|
matrix, |
238 |
|
63, |
239 |
|
sum, |
240 |
|
pMB->lambda[i]); |
241 |
} |
} |
242 |
stop_interlacing_timer(); |
stop_quant_timer(); |
243 |
|
|
|
for(i = 0; i < 6; i++) |
|
|
{ |
|
244 |
/* |
/* |
245 |
* no need to transfer 8->16-bit |
* We code the block if the sum is higher than the limit and if the first |
246 |
* (this is performed already in motion compensation) |
* two AC coefficients in zig zag order are not zero. |
247 |
*/ |
*/ |
248 |
start_timer(); |
code_block = 0; |
249 |
fdct(&data[i*64]); |
if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) { |
250 |
stop_dct_timer(); |
code_block = 1; |
251 |
|
} else { |
252 |
|
|
253 |
if (pParam->quant_type == 0) |
if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) { |
254 |
{ |
/* dark blocks prevention for direct mode */ |
255 |
start_timer(); |
if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0)) |
256 |
sum = quant_inter(&qcoeff[i*64], &data[i*64], iQuant); |
code_block = 1; |
257 |
stop_quant_timer(); |
} else { |
258 |
|
/* not direct mode */ |
259 |
|
if (qcoeff[i*64] != 0) |
260 |
|
code_block = 1; |
261 |
} |
} |
|
else |
|
|
{ |
|
|
start_timer(); |
|
|
sum = quant4_inter(&qcoeff[i*64], &data[i*64], iQuant); |
|
|
stop_quant_timer(); |
|
262 |
} |
} |
263 |
|
|
264 |
if(sum >= TOOSMALL_LIMIT) { // skip block ? |
/* Set the corresponding cbp bit */ |
265 |
|
cbp |= code_block << (5 - i); |
266 |
|
} |
267 |
|
|
268 |
if (pParam->quant_type == H263_QUANT) |
return(cbp); |
|
{ |
|
|
start_timer(); |
|
|
dequant_inter(&data[i*64], &qcoeff[i*64], iQuant); |
|
|
stop_iquant_timer(); |
|
269 |
} |
} |
270 |
else |
|
271 |
|
/* DeQuantize all blocks -- Inter mode */ |
272 |
|
static __inline void |
273 |
|
MBDeQuantInter(const MBParam * pParam, |
274 |
|
const int iQuant, |
275 |
|
int16_t data[6 * 64], |
276 |
|
int16_t qcoeff[6 * 64], |
277 |
|
const uint8_t cbp) |
278 |
|
{ |
279 |
|
int mpeg; |
280 |
|
|
281 |
|
quant_interFuncPtr const dequant[2] = |
282 |
{ |
{ |
283 |
|
dequant_h263_inter, |
284 |
|
dequant_mpeg_inter |
285 |
|
}; |
286 |
|
|
287 |
|
mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT); |
288 |
|
|
289 |
start_timer(); |
start_timer(); |
290 |
dequant4_inter(&data[i*64], &qcoeff[i*64], iQuant); |
if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant, pParam->mpeg_quant_matrices); |
291 |
|
if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant, pParam->mpeg_quant_matrices); |
292 |
|
if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant, pParam->mpeg_quant_matrices); |
293 |
|
if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant, pParam->mpeg_quant_matrices); |
294 |
|
if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant, pParam->mpeg_quant_matrices); |
295 |
|
if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant, pParam->mpeg_quant_matrices); |
296 |
stop_iquant_timer(); |
stop_iquant_timer(); |
297 |
} |
} |
298 |
|
|
299 |
cbp |= 1 << (5 - i); |
typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS); |
300 |
|
typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS); |
301 |
|
|
302 |
|
|
303 |
|
static __inline void |
304 |
|
MBTrans8to16(const MBParam * const pParam, |
305 |
|
const FRAMEINFO * const frame, |
306 |
|
const MACROBLOCK * const pMB, |
307 |
|
const uint32_t x_pos, |
308 |
|
const uint32_t y_pos, |
309 |
|
int16_t data[6 * 64]) |
310 |
|
{ |
311 |
|
uint32_t stride = pParam->edged_width; |
312 |
|
uint32_t stride2 = stride / 2; |
313 |
|
uint32_t next_block = stride * 8; |
314 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
315 |
|
const IMAGE * const pCurrent = &frame->image; |
316 |
|
|
317 |
|
/* Image pointers */ |
318 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
319 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
320 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
321 |
|
|
322 |
|
/* Do the transfer */ |
323 |
start_timer(); |
start_timer(); |
324 |
idct(&data[i*64]); |
transfer_8to16copy(&data[0 * 64], pY_Cur, stride); |
325 |
stop_idct_timer(); |
transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride); |
326 |
} |
transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride); |
327 |
|
transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride); |
328 |
|
transfer_8to16copy(&data[4 * 64], pU_Cur, stride2); |
329 |
|
transfer_8to16copy(&data[5 * 64], pV_Cur, stride2); |
330 |
|
stop_transfer_timer(); |
331 |
} |
} |
332 |
|
|
333 |
start_timer(); |
static __inline void |
334 |
if (pMB->field_dct) |
MBTrans16to8(const MBParam * const pParam, |
335 |
|
const FRAMEINFO * const frame, |
336 |
|
const MACROBLOCK * const pMB, |
337 |
|
const uint32_t x_pos, |
338 |
|
const uint32_t y_pos, |
339 |
|
int16_t data[6 * 64], |
340 |
|
const uint32_t add, /* Must be 1 or 0 */ |
341 |
|
const uint8_t cbp) |
342 |
|
{ |
343 |
|
uint8_t *pY_Cur, *pU_Cur, *pV_Cur; |
344 |
|
uint32_t stride = pParam->edged_width; |
345 |
|
uint32_t stride2 = stride / 2; |
346 |
|
uint32_t next_block = stride * 8; |
347 |
|
const IMAGE * const pCurrent = &frame->image; |
348 |
|
|
349 |
|
/* Array of function pointers, indexed by [add] */ |
350 |
|
transfer_operation_16to8_t * const functions[2] = |
351 |
{ |
{ |
352 |
MBFieldToFrame(data); |
(transfer_operation_16to8_t*)transfer_16to8copy, |
353 |
|
(transfer_operation_16to8_t*)transfer_16to8add, |
354 |
|
}; |
355 |
|
|
356 |
|
transfer_operation_16to8_t *transfer_op = NULL; |
357 |
|
|
358 |
|
/* Image pointers */ |
359 |
|
pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4); |
360 |
|
pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3); |
361 |
|
pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3); |
362 |
|
|
363 |
|
if (pMB->field_dct) { |
364 |
|
next_block = stride; |
365 |
|
stride *= 2; |
366 |
} |
} |
|
stop_interlacing_timer(); |
|
367 |
|
|
368 |
|
/* Operation function */ |
369 |
|
transfer_op = functions[add]; |
370 |
|
|
371 |
|
/* Do the operation */ |
372 |
start_timer(); |
start_timer(); |
373 |
if (cbp & 32) |
if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride); |
374 |
transfer_16to8add(pY_Cur, &data[0*64], stride); |
if (cbp&16) transfer_op(pY_Cur + 8, &data[1 * 64], stride); |
375 |
if (cbp & 16) |
if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride); |
376 |
transfer_16to8add(pY_Cur + 8, &data[1*64], stride); |
if (cbp& 4) transfer_op(pY_Cur + next_block + 8, &data[3 * 64], stride); |
377 |
if (cbp & 8) |
if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2); |
378 |
transfer_16to8add(pY_Cur + 8 * stride, &data[2*64], stride); |
if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2); |
|
if (cbp & 4) |
|
|
transfer_16to8add(pY_Cur + 8 + 8 * stride, &data[3*64], stride); |
|
|
if (cbp & 2) |
|
|
transfer_16to8add(pU_Cur, &data[4*64], stride / 2); |
|
|
if (cbp & 1) |
|
|
transfer_16to8add(pV_Cur, &data[5*64], stride / 2); |
|
379 |
stop_transfer_timer(); |
stop_transfer_timer(); |
380 |
|
} |
381 |
|
|
382 |
return cbp; |
/***************************************************************************** |
383 |
|
* Module functions |
384 |
|
****************************************************************************/ |
385 |
|
|
386 |
|
void |
387 |
|
MBTransQuantIntra(const MBParam * const pParam, |
388 |
|
const FRAMEINFO * const frame, |
389 |
|
MACROBLOCK * const pMB, |
390 |
|
const uint32_t x_pos, |
391 |
|
const uint32_t y_pos, |
392 |
|
int16_t data[6 * 64], |
393 |
|
int16_t qcoeff[6 * 64]) |
394 |
|
{ |
395 |
|
|
396 |
|
/* Transfer data */ |
397 |
|
MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data); |
398 |
|
|
399 |
|
/* Perform DCT (and field decision) */ |
400 |
|
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
401 |
|
|
402 |
|
/* Quantize the block */ |
403 |
|
MBQuantIntra(pParam, frame, pMB, data, qcoeff); |
404 |
|
|
405 |
|
/* DeQuantize the block */ |
406 |
|
MBDeQuantIntra(pParam, pMB->quant, data, qcoeff); |
407 |
|
|
408 |
|
/* Perform inverse DCT*/ |
409 |
|
MBiDCT(data, 0x3F); |
410 |
|
|
411 |
|
/* Transfer back the data -- Don't add data */ |
412 |
|
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F); |
413 |
} |
} |
414 |
|
|
415 |
|
|
416 |
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
uint8_t |
417 |
|
MBTransQuantInter(const MBParam * const pParam, |
418 |
|
const FRAMEINFO * const frame, |
419 |
|
MACROBLOCK * const pMB, |
420 |
|
const uint32_t x_pos, |
421 |
|
const uint32_t y_pos, |
422 |
|
int16_t data[6 * 64], |
423 |
|
int16_t qcoeff[6 * 64]) |
424 |
|
{ |
425 |
|
uint8_t cbp; |
426 |
|
uint32_t limit; |
427 |
|
|
428 |
|
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
429 |
|
* already */ |
430 |
|
|
431 |
|
/* Perform DCT (and field decision) */ |
432 |
|
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
433 |
|
|
434 |
|
/* Set the limit threshold */ |
435 |
|
limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0); |
436 |
|
|
437 |
|
if (frame->vop_flags & XVID_VOP_CARTOON) |
438 |
|
limit *= 3; |
439 |
|
|
440 |
|
/* Quantize the block */ |
441 |
|
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit); |
442 |
|
|
443 |
#define ABS(X) (X)<0 ? -(X) : (X) |
/* DeQuantize the block */ |
444 |
|
MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); |
445 |
|
|
446 |
uint32_t MBDecideFieldDCT(int16_t data[6*64]) |
/* Perform inverse DCT*/ |
447 |
|
MBiDCT(data, cbp); |
448 |
|
|
449 |
|
/* Transfer back the data -- Add the data */ |
450 |
|
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); |
451 |
|
|
452 |
|
return(cbp); |
453 |
|
} |
454 |
|
|
455 |
|
uint8_t |
456 |
|
MBTransQuantInterBVOP(const MBParam * pParam, |
457 |
|
FRAMEINFO * frame, |
458 |
|
MACROBLOCK * pMB, |
459 |
|
const uint32_t x_pos, |
460 |
|
const uint32_t y_pos, |
461 |
|
int16_t data[6 * 64], |
462 |
|
int16_t qcoeff[6 * 64]) |
463 |
{ |
{ |
464 |
|
uint8_t cbp; |
465 |
|
uint32_t limit; |
466 |
|
|
467 |
const uint8_t blocks[] = {0*64, 0*64, 0*64, 0*64, 2*64, 2*64, 2*64, 2*64}; |
/* There is no MBTrans8to16 for Inter block, that's done in motion compensation |
468 |
|
* already */ |
469 |
|
|
470 |
|
/* Perform DCT (and field decision) */ |
471 |
|
MBfDCT(pParam, frame, pMB, x_pos, y_pos, data); |
472 |
|
|
473 |
|
/* Set the limit threshold */ |
474 |
|
limit = BVOP_TOOSMALL_LIMIT; |
475 |
|
|
476 |
|
if (frame->vop_flags & XVID_VOP_CARTOON) |
477 |
|
limit *= 2; |
478 |
|
|
479 |
|
/* Quantize the block */ |
480 |
|
cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit); |
481 |
|
|
482 |
|
/* |
483 |
|
* History comment: |
484 |
|
* We don't have to DeQuant, iDCT and Transfer back data for B-frames. |
485 |
|
* |
486 |
|
* BUT some plugins require the rebuilt original frame to be passed so we |
487 |
|
* have to take care of that here |
488 |
|
*/ |
489 |
|
if((pParam->plugin_flags & XVID_REQORIGINAL)) { |
490 |
|
|
491 |
|
/* DeQuantize the block */ |
492 |
|
MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp); |
493 |
|
|
494 |
|
/* Perform inverse DCT*/ |
495 |
|
MBiDCT(data, cbp); |
496 |
|
|
497 |
|
/* Transfer back the data -- Add the data */ |
498 |
|
MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp); |
499 |
|
} |
500 |
|
|
501 |
|
return(cbp); |
502 |
|
} |
503 |
|
|
504 |
|
/* if sum(diff between field lines) < sum(diff between frame lines), use field dct */ |
505 |
|
uint32_t |
506 |
|
MBFieldTest_c(int16_t data[6 * 64]) |
507 |
|
{ |
508 |
|
const uint8_t blocks[] = |
509 |
|
{ 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 }; |
510 |
const uint8_t lines[] = {0, 16, 32, 48, 0, 16, 32, 48}; |
const uint8_t lines[] = {0, 16, 32, 48, 0, 16, 32, 48}; |
511 |
|
|
512 |
int frame = 0, field = 0; |
int frame = 0, field = 0; |
513 |
int i, j; |
int i, j; |
514 |
|
|
515 |
for (i=0 ; i<7 ; ++i) |
for (i = 0; i < 7; ++i) { |
516 |
{ |
for (j = 0; j < 8; ++j) { |
517 |
for (j=0 ; j<8 ; ++j) |
frame += |
518 |
{ |
abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]); |
519 |
frame += ABS(data[0*64 + (i+1)*8 + j] - data[0*64 + i*8 + j]); |
frame += |
520 |
frame += ABS(data[1*64 + (i+1)*8 + j] - data[1*64 + i*8 + j]); |
abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]); |
521 |
frame += ABS(data[2*64 + (i+1)*8 + j] - data[2*64 + i*8 + j]); |
frame += |
522 |
frame += ABS(data[3*64 + (i+1)*8 + j] - data[3*64 + i*8 + j]); |
abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]); |
523 |
|
frame += |
524 |
|
abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]); |
525 |
|
|
526 |
field += ABS(data[blocks[i+1] + lines[i+1] + j] -\ |
field += |
527 |
|
abs(data[blocks[i + 1] + lines[i + 1] + j] - |
528 |
data[blocks[i ] + lines[i ] + j]); |
data[blocks[i ] + lines[i ] + j]); |
529 |
field += ABS(data[blocks[i+1] + lines[i+1] + 8 + j] -\ |
field += |
530 |
|
abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] - |
531 |
data[blocks[i ] + lines[i ] + 8 + j]); |
data[blocks[i ] + lines[i ] + 8 + j]); |
532 |
field += ABS(data[blocks[i+1] + 64 + lines[i+1] + j] -\ |
field += |
533 |
|
abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] - |
534 |
data[blocks[i ] + 64 + lines[i ] + j]); |
data[blocks[i ] + 64 + lines[i ] + j]); |
535 |
field += ABS(data[blocks[i+1] + 64 + lines[i+1] + 8 + j] -\ |
field += |
536 |
|
abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] - |
537 |
data[blocks[i ] + 64 + lines[i ] + 8 + j]); |
data[blocks[i ] + 64 + lines[i ] + 8 + j]); |
538 |
} |
} |
539 |
} |
} |
540 |
|
|
541 |
if (frame > field) |
return (frame >= (field + 350)); |
|
{ |
|
|
MBFrameToField(data); |
|
|
} |
|
|
|
|
|
return (frame > field); |
|
542 |
} |
} |
543 |
|
|
544 |
|
|
547 |
#define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp)) |
#define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp)) |
548 |
#define LINE(X,Y) &data[X*64 + Y*8] |
#define LINE(X,Y) &data[X*64 + Y*8] |
549 |
|
|
550 |
void MBFrameToField(int16_t data[6*64]) |
void |
551 |
|
MBFrameToField(int16_t data[6 * 64]) |
552 |
{ |
{ |
553 |
int16_t tmp[8]; |
int16_t tmp[8]; |
554 |
|
|
555 |
/* left blocks */ |
/* left blocks */ |
556 |
|
|
557 |
// 1=2, 2=4, 4=8, 8=1 |
/* 1=2, 2=4, 4=8, 8=1 */ |
558 |
MOVLINE(tmp, LINE(0,1)); |
MOVLINE(tmp, LINE(0,1)); |
559 |
MOVLINE(LINE(0,1), LINE(0,2)); |
MOVLINE(LINE(0,1), LINE(0,2)); |
560 |
MOVLINE(LINE(0,2), LINE(0,4)); |
MOVLINE(LINE(0,2), LINE(0,4)); |
561 |
MOVLINE(LINE(0,4), LINE(2,0)); |
MOVLINE(LINE(0,4), LINE(2,0)); |
562 |
MOVLINE(LINE(2,0), tmp); |
MOVLINE(LINE(2,0), tmp); |
563 |
|
|
564 |
// 3=6, 6=12, 12=9, 9=3 |
/* 3=6, 6=12, 12=9, 9=3 */ |
565 |
MOVLINE(tmp, LINE(0,3)); |
MOVLINE(tmp, LINE(0,3)); |
566 |
MOVLINE(LINE(0,3), LINE(0,6)); |
MOVLINE(LINE(0,3), LINE(0,6)); |
567 |
MOVLINE(LINE(0,6), LINE(2,4)); |
MOVLINE(LINE(0,6), LINE(2,4)); |
568 |
MOVLINE(LINE(2,4), LINE(2,1)); |
MOVLINE(LINE(2,4), LINE(2,1)); |
569 |
MOVLINE(LINE(2,1), tmp); |
MOVLINE(LINE(2,1), tmp); |
570 |
|
|
571 |
// 5=10, 10=5 |
/* 5=10, 10=5 */ |
572 |
MOVLINE(tmp, LINE(0,5)); |
MOVLINE(tmp, LINE(0,5)); |
573 |
MOVLINE(LINE(0,5), LINE(2,2)); |
MOVLINE(LINE(0,5), LINE(2,2)); |
574 |
MOVLINE(LINE(2,2), tmp); |
MOVLINE(LINE(2,2), tmp); |
575 |
|
|
576 |
// 7=14, 14=13, 13=11, 11=7 |
/* 7=14, 14=13, 13=11, 11=7 */ |
577 |
MOVLINE(tmp, LINE(0,7)); |
MOVLINE(tmp, LINE(0,7)); |
578 |
MOVLINE(LINE(0,7), LINE(2,6)); |
MOVLINE(LINE(0,7), LINE(2,6)); |
579 |
MOVLINE(LINE(2,6), LINE(2,5)); |
MOVLINE(LINE(2,6), LINE(2,5)); |
582 |
|
|
583 |
/* right blocks */ |
/* right blocks */ |
584 |
|
|
585 |
// 1=2, 2=4, 4=8, 8=1 |
/* 1=2, 2=4, 4=8, 8=1 */ |
586 |
MOVLINE(tmp, LINE(1,1)); |
MOVLINE(tmp, LINE(1,1)); |
587 |
MOVLINE(LINE(1,1), LINE(1,2)); |
MOVLINE(LINE(1,1), LINE(1,2)); |
588 |
MOVLINE(LINE(1,2), LINE(1,4)); |
MOVLINE(LINE(1,2), LINE(1,4)); |
589 |
MOVLINE(LINE(1,4), LINE(3,0)); |
MOVLINE(LINE(1,4), LINE(3,0)); |
590 |
MOVLINE(LINE(3,0), tmp); |
MOVLINE(LINE(3,0), tmp); |
591 |
|
|
592 |
// 3=6, 6=12, 12=9, 9=3 |
/* 3=6, 6=12, 12=9, 9=3 */ |
593 |
MOVLINE(tmp, LINE(1,3)); |
MOVLINE(tmp, LINE(1,3)); |
594 |
MOVLINE(LINE(1,3), LINE(1,6)); |
MOVLINE(LINE(1,3), LINE(1,6)); |
595 |
MOVLINE(LINE(1,6), LINE(3,4)); |
MOVLINE(LINE(1,6), LINE(3,4)); |
596 |
MOVLINE(LINE(3,4), LINE(3,1)); |
MOVLINE(LINE(3,4), LINE(3,1)); |
597 |
MOVLINE(LINE(3,1), tmp); |
MOVLINE(LINE(3,1), tmp); |
598 |
|
|
599 |
// 5=10, 10=5 |
/* 5=10, 10=5 */ |
600 |
MOVLINE(tmp, LINE(1,5)); |
MOVLINE(tmp, LINE(1,5)); |
601 |
MOVLINE(LINE(1,5), LINE(3,2)); |
MOVLINE(LINE(1,5), LINE(3,2)); |
602 |
MOVLINE(LINE(3,2), tmp); |
MOVLINE(LINE(3,2), tmp); |
603 |
|
|
604 |
// 7=14, 14=13, 13=11, 11=7 |
/* 7=14, 14=13, 13=11, 11=7 */ |
605 |
MOVLINE(tmp, LINE(1,7)); |
MOVLINE(tmp, LINE(1,7)); |
606 |
MOVLINE(LINE(1,7), LINE(3,6)); |
MOVLINE(LINE(1,7), LINE(3,6)); |
607 |
MOVLINE(LINE(3,6), LINE(3,5)); |
MOVLINE(LINE(3,6), LINE(3,5)); |
609 |
MOVLINE(LINE(3,3), tmp); |
MOVLINE(LINE(3,3), tmp); |
610 |
} |
} |
611 |
|
|
612 |
|
/***************************************************************************** |
613 |
|
* Trellis based R-D optimal quantization |
614 |
|
* |
615 |
|
* Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net |
616 |
|
* |
617 |
|
****************************************************************************/ |
618 |
|
|
619 |
|
/*---------------------------------------------------------------------------- |
620 |
|
* |
621 |
|
* Trellis-Based quantization |
622 |
|
* |
623 |
|
* So far I understand this paper: |
624 |
|
* |
625 |
|
* "Trellis-Based R-D Optimal Quantization in H.263+" |
626 |
|
* J.Wen, M.Luttrell, J.Villasenor |
627 |
|
* IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000. |
628 |
|
* |
629 |
|
* we are at stake with a simplified Bellmand-Ford / Dijkstra Single |
630 |
|
* Source Shortest Path algo. But due to the underlying graph structure |
631 |
|
* ("Trellis"), it can be turned into a dynamic programming algo, |
632 |
|
* partially saving the explicit graph's nodes representation. And |
633 |
|
* without using a heap, since the open frontier of the DAG is always |
634 |
|
* known, and of fixed size. |
635 |
|
*--------------------------------------------------------------------------*/ |
636 |
|
|
637 |
|
|
638 |
|
|
639 |
|
/* Codes lengths for relevant levels. */ |
640 |
|
|
641 |
|
/* let's factorize: */ |
642 |
|
static const uint8_t Code_Len0[64] = { |
643 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
644 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
645 |
|
static const uint8_t Code_Len1[64] = { |
646 |
|
20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
647 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
648 |
|
static const uint8_t Code_Len2[64] = { |
649 |
|
19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
650 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
651 |
|
static const uint8_t Code_Len3[64] = { |
652 |
|
18,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
653 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
654 |
|
static const uint8_t Code_Len4[64] = { |
655 |
|
17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
656 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
657 |
|
static const uint8_t Code_Len5[64] = { |
658 |
|
16,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
659 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
660 |
|
static const uint8_t Code_Len6[64] = { |
661 |
|
15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
662 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
663 |
|
static const uint8_t Code_Len7[64] = { |
664 |
|
13,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
665 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
666 |
|
static const uint8_t Code_Len8[64] = { |
667 |
|
11,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
668 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
669 |
|
static const uint8_t Code_Len9[64] = { |
670 |
|
12,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
671 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
672 |
|
static const uint8_t Code_Len10[64] = { |
673 |
|
12,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
674 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
675 |
|
static const uint8_t Code_Len11[64] = { |
676 |
|
12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
677 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
678 |
|
static const uint8_t Code_Len12[64] = { |
679 |
|
11,17,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
680 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
681 |
|
static const uint8_t Code_Len13[64] = { |
682 |
|
11,15,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
683 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
684 |
|
static const uint8_t Code_Len14[64] = { |
685 |
|
10,12,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
686 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
687 |
|
static const uint8_t Code_Len15[64] = { |
688 |
|
10,13,17,19,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
689 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
690 |
|
static const uint8_t Code_Len16[64] = { |
691 |
|
9,12,13,18,18,19,19,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
692 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; |
693 |
|
static const uint8_t Code_Len17[64] = { |
694 |
|
8,11,13,14,14,14,15,19,19,19,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
695 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
696 |
|
static const uint8_t Code_Len18[64] = { |
697 |
|
7, 9,11,11,13,13,13,15,15,15,16,22,22,22,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
698 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
699 |
|
static const uint8_t Code_Len19[64] = { |
700 |
|
5, 7, 9,10,10,11,11,11,11,11,13,14,16,17,17,18,18,18,18,18,18,18,18,20,20,21,21,30,30,30,30,30, |
701 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 }; |
702 |
|
static const uint8_t Code_Len20[64] = { |
703 |
|
3, 4, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9,10,10,10,10,10,10,10,10,12,12,13,13,12,13,14,15,15, |
704 |
|
15,16,16,16,16,17,17,17,18,18,19,19,19,19,19,19,19,19,21,21,22,22,30,30,30,30,30,30,30,30,30,30 }; |
705 |
|
|
706 |
|
/* a few more table for LAST table: */ |
707 |
|
static const uint8_t Code_Len21[64] = { |
708 |
|
13,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
709 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; |
710 |
|
static const uint8_t Code_Len22[64] = { |
711 |
|
12,15,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30, |
712 |
|
30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; |
713 |
|
static const uint8_t Code_Len23[64] = { |
714 |
|
10,12,15,15,15,16,16,16,16,17,17,17,17,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,20,20,20, |
715 |
|
20,21,21,21,21,21,21,21,21,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30}; |
716 |
|
static const uint8_t Code_Len24[64] = { |
717 |
|
5, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,10,10,10,10,10,10,10,10,11,11,11,11,12,12,12, |
718 |
|
12,13,13,13,13,13,13,13,13,14,16,16,16,16,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,19,19}; |
719 |
|
|
720 |
|
|
721 |
|
static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */ |
722 |
|
Code_Len20,Code_Len19,Code_Len18,Code_Len17, |
723 |
|
Code_Len16,Code_Len15,Code_Len14,Code_Len13, |
724 |
|
Code_Len12,Code_Len11,Code_Len10,Code_Len9, |
725 |
|
Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5, |
726 |
|
Code_Len4, Code_Len3, Code_Len3 ,Code_Len2, |
727 |
|
Code_Len2, Code_Len1, Code_Len1, Code_Len1, |
728 |
|
}; |
729 |
|
|
730 |
|
static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */ |
731 |
|
Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1, |
732 |
|
}; |
733 |
|
|
734 |
|
/* TL_SHIFT controls the precision of the RD optimizations in trellis |
735 |
|
* valid range is [10..16]. The bigger, the more trellis is vulnerable |
736 |
|
* to overflows in cost formulas. |
737 |
|
* - 10 allows ac values up to 2^11 == 2048 |
738 |
|
* - 16 allows ac values up to 2^8 == 256 |
739 |
|
*/ |
740 |
|
#define TL_SHIFT 11 |
741 |
|
#define TL(q) ((0xfe00>>(16-TL_SHIFT))/(q*q)) |
742 |
|
|
743 |
/* interlace Y blocks vertically */ |
static const int Trellis_Lambda_Tabs[31] = { |
744 |
|
TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7), |
745 |
|
TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15), |
746 |
|
TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23), |
747 |
|
TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31) |
748 |
|
}; |
749 |
|
#undef TL |
750 |
|
|
751 |
void MBFieldToFrame(int16_t data[6*64]) |
static int __inline |
752 |
|
Find_Last(const int16_t *C, const uint16_t *Zigzag, int i) |
753 |
|
{ |
754 |
|
while(i>=0) |
755 |
|
if (C[Zigzag[i]]) |
756 |
|
return i; |
757 |
|
else i--; |
758 |
|
return -1; |
759 |
|
} |
760 |
|
|
761 |
|
/* this routine has been strippen of all debug code */ |
762 |
|
static int |
763 |
|
dct_quantize_trellis_c(int16_t *const Out, |
764 |
|
const int16_t *const In, |
765 |
|
int Q, |
766 |
|
const uint16_t * const Zigzag, |
767 |
|
const uint16_t * const QuantMatrix, |
768 |
|
int Non_Zero, |
769 |
|
int Sum, |
770 |
|
int Lambda_Mod) |
771 |
{ |
{ |
|
uint16_t tmp[8]; |
|
772 |
|
|
773 |
/* left blocks */ |
/* Note: We should search last non-zero coeffs on *real* DCT input coeffs |
774 |
|
* (In[]), not quantized one (Out[]). However, it only improves the result |
775 |
|
* *very* slightly (~0.01dB), whereas speed drops to crawling level :) |
776 |
|
* Well, actually, taking 1 more coeff past Non_Zero into account sometimes |
777 |
|
* helps. */ |
778 |
|
typedef struct { int16_t Run, Level; } NODE; |
779 |
|
|
780 |
|
NODE Nodes[65], Last = { 0, 0}; |
781 |
|
uint32_t Run_Costs0[64+1]; |
782 |
|
uint32_t * const Run_Costs = Run_Costs0 + 1; |
783 |
|
|
784 |
// 1=8, 8=4, 4=2, 2=1 |
/* it's 1/lambda, actually */ |
785 |
MOVLINE(tmp, LINE(0,1)); |
const int Lambda = (Lambda_Mod*Trellis_Lambda_Tabs[Q-1])>>LAMBDA_EXP; |
|
MOVLINE(LINE(0,1), LINE(2,0)); |
|
|
MOVLINE(LINE(2,0), LINE(0,4)); |
|
|
MOVLINE(LINE(0,4), LINE(0,2)); |
|
|
MOVLINE(LINE(0,2), tmp); |
|
786 |
|
|
787 |
// 3=9, 9=12, 12=6, 6=3 |
int Run_Start = -1; |
788 |
MOVLINE(tmp, LINE(0,3)); |
uint32_t Min_Cost = 2<<TL_SHIFT; |
|
MOVLINE(LINE(0,3), LINE(2,1)); |
|
|
MOVLINE(LINE(2,1), LINE(2,4)); |
|
|
MOVLINE(LINE(2,4), LINE(0,6)); |
|
|
MOVLINE(LINE(0,6), tmp); |
|
789 |
|
|
790 |
// 5=10, 10=5 |
int Last_Node = -1; |
791 |
MOVLINE(tmp, LINE(0,5)); |
uint32_t Last_Cost = 0; |
|
MOVLINE(LINE(0,5), LINE(2,2)); |
|
|
MOVLINE(LINE(2,2), tmp); |
|
792 |
|
|
793 |
// 7=11, 11=13, 13=14, 14=7 |
int i, j; |
|
MOVLINE(tmp, LINE(0,7)); |
|
|
MOVLINE(LINE(0,7), LINE(2,3)); |
|
|
MOVLINE(LINE(2,3), LINE(2,5)); |
|
|
MOVLINE(LINE(2,5), LINE(2,6)); |
|
|
MOVLINE(LINE(2,6), tmp); |
|
794 |
|
|
795 |
/* right blocks */ |
/* source (w/ CBP penalty) */ |
796 |
|
Run_Costs[-1] = 2<<TL_SHIFT; |
797 |
|
|
798 |
// 1=8, 8=4, 4=2, 2=1 |
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
799 |
MOVLINE(tmp, LINE(1,1)); |
if (Non_Zero<0) |
800 |
MOVLINE(LINE(1,1), LINE(3,0)); |
return 0; /* Sum is zero if there are only zero coeffs */ |
801 |
MOVLINE(LINE(3,0), LINE(1,4)); |
|
802 |
MOVLINE(LINE(1,4), LINE(1,2)); |
for(i=0; i<=Non_Zero; i++) { |
803 |
MOVLINE(LINE(1,2), tmp); |
const int q = ((Q*QuantMatrix[Zigzag[i]])>>4); |
804 |
|
const int Mult = 2*q; |
805 |
|
const int Bias = (q-1) | 1; |
806 |
|
const int Lev0 = Mult + Bias; |
807 |
|
|
808 |
|
const int AC = In[Zigzag[i]]; |
809 |
|
const int Level1 = Out[Zigzag[i]]; |
810 |
|
const unsigned int Dist0 = Lambda* AC*AC; |
811 |
|
uint32_t Best_Cost = 0xf0000000; |
812 |
|
Last_Cost += Dist0; |
813 |
|
|
814 |
|
/* very specialized loop for -1,0,+1 */ |
815 |
|
if ((uint32_t)(Level1+1)<3) { |
816 |
|
int dQ; |
817 |
|
int Run; |
818 |
|
uint32_t Cost0; |
819 |
|
|
820 |
|
if (AC<0) { |
821 |
|
Nodes[i].Level = -1; |
822 |
|
dQ = Lev0 + AC; |
823 |
|
} else { |
824 |
|
Nodes[i].Level = 1; |
825 |
|
dQ = Lev0 - AC; |
826 |
|
} |
827 |
|
Cost0 = Lambda*dQ*dQ; |
828 |
|
|
829 |
|
Nodes[i].Run = 1; |
830 |
|
Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0; |
831 |
|
for(Run=i-Run_Start; Run>0; --Run) { |
832 |
|
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
833 |
|
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT); |
834 |
|
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT); |
835 |
|
|
836 |
|
/* TODO: what about tie-breaks? Should we favor short runs or |
837 |
|
* long runs? Although the error is the same, it would not be |
838 |
|
* spread the same way along high and low frequencies... */ |
839 |
|
|
840 |
|
/* Gruel: I'd say, favour short runs => hifreq errors (HVS) */ |
841 |
|
|
842 |
|
if (Cost<Best_Cost) { |
843 |
|
Best_Cost = Cost; |
844 |
|
Nodes[i].Run = Run; |
845 |
|
} |
846 |
|
|
847 |
|
if (lCost<Last_Cost) { |
848 |
|
Last_Cost = lCost; |
849 |
|
Last.Run = Run; |
850 |
|
Last_Node = i; |
851 |
|
} |
852 |
|
} |
853 |
|
if (Last_Node==i) |
854 |
|
Last.Level = Nodes[i].Level; |
855 |
|
} else if (51U>(uint32_t)(Level1+25)) { |
856 |
|
/* "big" levels (not less than ESC3, though) */ |
857 |
|
const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last; |
858 |
|
int Level2; |
859 |
|
int dQ1, dQ2; |
860 |
|
int Run; |
861 |
|
uint32_t Dist1,Dist2; |
862 |
|
int dDist21; |
863 |
|
|
864 |
|
if (Level1>1) { |
865 |
|
dQ1 = Level1*Mult-AC + Bias; |
866 |
|
dQ2 = dQ1 - Mult; |
867 |
|
Level2 = Level1-1; |
868 |
|
Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0; |
869 |
|
Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0; |
870 |
|
Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0; |
871 |
|
Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0; |
872 |
|
} else { /* Level1<-1 */ |
873 |
|
dQ1 = Level1*Mult-AC - Bias; |
874 |
|
dQ2 = dQ1 + Mult; |
875 |
|
Level2 = Level1 + 1; |
876 |
|
Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0; |
877 |
|
Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0; |
878 |
|
Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0; |
879 |
|
Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0; |
880 |
|
} |
881 |
|
|
882 |
|
Dist1 = Lambda*dQ1*dQ1; |
883 |
|
Dist2 = Lambda*dQ2*dQ2; |
884 |
|
dDist21 = Dist2-Dist1; |
885 |
|
|
886 |
// 3=9, 9=12, 12=6, 6=3 |
for(Run=i-Run_Start; Run>0; --Run) |
887 |
MOVLINE(tmp, LINE(1,3)); |
{ |
888 |
MOVLINE(LINE(1,3), LINE(3,1)); |
const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run]; |
889 |
MOVLINE(LINE(3,1), LINE(3,4)); |
uint32_t Cost1, Cost2; |
890 |
MOVLINE(LINE(3,4), LINE(1,6)); |
int bLevel; |
|
MOVLINE(LINE(1,6), tmp); |
|
891 |
|
|
892 |
// 5=10, 10=5 |
/* for sub-optimal (but slightly worth it, speed-wise) search, |
893 |
MOVLINE(tmp, LINE(1,5)); |
* uncomment the following: |
894 |
MOVLINE(LINE(1,5), LINE(3,2)); |
* if (Cost_Base>=Best_Cost) continue; |
895 |
MOVLINE(LINE(3,2), tmp); |
* (? doesn't seem to have any effect -- gruel ) */ |
896 |
|
|
897 |
// 7=11, 11=13, 13=14, 14=7 |
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT); |
898 |
MOVLINE(tmp, LINE(1,7)); |
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21; |
899 |
MOVLINE(LINE(1,7), LINE(3,3)); |
|
900 |
MOVLINE(LINE(3,3), LINE(3,5)); |
if (Cost2<Cost1) { |
901 |
MOVLINE(LINE(3,5), LINE(3,6)); |
Cost1 = Cost2; |
902 |
MOVLINE(LINE(3,6), tmp); |
bLevel = Level2; |
903 |
|
} else { |
904 |
|
bLevel = Level1; |
905 |
|
} |
906 |
|
|
907 |
|
if (Cost1<Best_Cost) { |
908 |
|
Best_Cost = Cost1; |
909 |
|
Nodes[i].Run = Run; |
910 |
|
Nodes[i].Level = bLevel; |
911 |
|
} |
912 |
|
|
913 |
|
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT); |
914 |
|
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21; |
915 |
|
|
916 |
|
if (Cost2<Cost1) { |
917 |
|
Cost1 = Cost2; |
918 |
|
bLevel = Level2; |
919 |
|
} else { |
920 |
|
bLevel = Level1; |
921 |
|
} |
922 |
|
|
923 |
|
if (Cost1<Last_Cost) { |
924 |
|
Last_Cost = Cost1; |
925 |
|
Last.Run = Run; |
926 |
|
Last.Level = bLevel; |
927 |
|
Last_Node = i; |
928 |
|
} |
929 |
|
} /* end of "for Run" */ |
930 |
|
} else { |
931 |
|
/* Very very high levels, with no chance of being optimizable |
932 |
|
* => Simply pick best Run. */ |
933 |
|
int Run; |
934 |
|
for(Run=i-Run_Start; Run>0; --Run) { |
935 |
|
/* 30 bits + no distortion */ |
936 |
|
const uint32_t Cost = (30<<TL_SHIFT) + Run_Costs[i-Run]; |
937 |
|
if (Cost<Best_Cost) { |
938 |
|
Best_Cost = Cost; |
939 |
|
Nodes[i].Run = Run; |
940 |
|
Nodes[i].Level = Level1; |
941 |
|
} |
942 |
|
|
943 |
|
if (Cost<Last_Cost) { |
944 |
|
Last_Cost = Cost; |
945 |
|
Last.Run = Run; |
946 |
|
Last.Level = Level1; |
947 |
|
Last_Node = i; |
948 |
|
} |
949 |
|
} |
950 |
|
} |
951 |
|
|
952 |
|
|
953 |
|
Run_Costs[i] = Best_Cost; |
954 |
|
|
955 |
|
if (Best_Cost < Min_Cost + Dist0) { |
956 |
|
Min_Cost = Best_Cost; |
957 |
|
Run_Start = i; |
958 |
|
} else { |
959 |
|
/* as noticed by Michael Niedermayer (michaelni at gmx.at), |
960 |
|
* there's a code shorter by 1 bit for a larger run (!), same |
961 |
|
* level. We give it a chance by not moving the left barrier too |
962 |
|
* much. */ |
963 |
|
while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) ) |
964 |
|
Run_Start++; |
965 |
|
|
966 |
|
/* spread on preceding coeffs the cost incurred by skipping this |
967 |
|
* one */ |
968 |
|
for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0; |
969 |
|
Min_Cost += Dist0; |
970 |
|
} |
971 |
|
} |
972 |
|
|
973 |
|
/* It seems trellis doesn't give good results... just leave the block untouched |
974 |
|
* and return the original sum value */ |
975 |
|
if (Last_Node<0) |
976 |
|
return Sum; |
977 |
|
|
978 |
|
/* reconstruct optimal sequence backward with surviving paths */ |
979 |
|
memset(Out, 0x00, 64*sizeof(*Out)); |
980 |
|
Out[Zigzag[Last_Node]] = Last.Level; |
981 |
|
i = Last_Node - Last.Run; |
982 |
|
Sum = abs(Last.Level); |
983 |
|
while(i>=0) { |
984 |
|
Out[Zigzag[i]] = Nodes[i].Level; |
985 |
|
Sum += abs(Nodes[i].Level); |
986 |
|
i -= Nodes[i].Run; |
987 |
|
} |
988 |
|
|
989 |
|
return Sum; |
990 |
|
} |
991 |
|
|
992 |
|
/* original version including heavy debugging info */ |
993 |
|
|
994 |
|
#ifdef DBGTRELL |
995 |
|
|
996 |
|
#define DBG 0 |
997 |
|
|
998 |
|
static __inline uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias, |
999 |
|
const uint16_t * Zigzag, int Max, int Lambda) |
1000 |
|
{ |
1001 |
|
#if (DBG>0) |
1002 |
|
const int16_t * const Ref = C + 6*64; |
1003 |
|
int Last = Max; |
1004 |
|
int Bits = 0; |
1005 |
|
int Dist = 0; |
1006 |
|
int i; |
1007 |
|
uint32_t Cost; |
1008 |
|
|
1009 |
|
while(Last>=0 && C[Zigzag[Last]]==0) |
1010 |
|
Last--; |
1011 |
|
|
1012 |
|
if (Last>=0) { |
1013 |
|
int j=0, j0=0; |
1014 |
|
int Run, Level; |
1015 |
|
|
1016 |
|
Bits = 2; /* CBP */ |
1017 |
|
while(j<Last) { |
1018 |
|
while(!C[Zigzag[j]]) |
1019 |
|
j++; |
1020 |
|
if (j==Last) |
1021 |
|
break; |
1022 |
|
Level=C[Zigzag[j]]; |
1023 |
|
Run = j - j0; |
1024 |
|
j0 = ++j; |
1025 |
|
if (Level>=-24 && Level<=24) |
1026 |
|
Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run]; |
1027 |
|
else |
1028 |
|
Bits += 30; |
1029 |
|
} |
1030 |
|
Level = C[Zigzag[Last]]; |
1031 |
|
Run = j - j0; |
1032 |
|
if (Level>=-6 && Level<=6) |
1033 |
|
Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run]; |
1034 |
|
else |
1035 |
|
Bits += 30; |
1036 |
|
} |
1037 |
|
|
1038 |
|
for(i=0; i<=Last; ++i) { |
1039 |
|
int V = C[Zigzag[i]]*Mult; |
1040 |
|
if (V>0) |
1041 |
|
V += Bias; |
1042 |
|
else |
1043 |
|
if (V<0) |
1044 |
|
V -= Bias; |
1045 |
|
V -= Ref[Zigzag[i]]; |
1046 |
|
Dist += V*V; |
1047 |
|
} |
1048 |
|
Cost = Lambda*Dist + (Bits<<TL_SHIFT); |
1049 |
|
if (DBG==1) |
1050 |
|
printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 ); |
1051 |
|
return Cost; |
1052 |
|
|
1053 |
|
#else |
1054 |
|
return 0; |
1055 |
|
#endif |
1056 |
|
} |
1057 |
|
|
1058 |
|
|
1059 |
|
static int |
1060 |
|
dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero) |
1061 |
|
{ |
1062 |
|
|
1063 |
|
/* |
1064 |
|
* Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]), |
1065 |
|
* not quantized one (Out[]). However, it only improves the result *very* |
1066 |
|
* slightly (~0.01dB), whereas speed drops to crawling level :) |
1067 |
|
* Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps. |
1068 |
|
*/ |
1069 |
|
typedef struct { int16_t Run, Level; } NODE; |
1070 |
|
|
1071 |
|
NODE Nodes[65], Last; |
1072 |
|
uint32_t Run_Costs0[64+1]; |
1073 |
|
uint32_t * const Run_Costs = Run_Costs0 + 1; |
1074 |
|
const int Mult = 2*Q; |
1075 |
|
const int Bias = (Q-1) | 1; |
1076 |
|
const int Lev0 = Mult + Bias; |
1077 |
|
const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */ |
1078 |
|
|
1079 |
|
int Run_Start = -1; |
1080 |
|
Run_Costs[-1] = 2<<TL_SHIFT; /* source (w/ CBP penalty) */ |
1081 |
|
uint32_t Min_Cost = 2<<TL_SHIFT; |
1082 |
|
|
1083 |
|
int Last_Node = -1; |
1084 |
|
uint32_t Last_Cost = 0; |
1085 |
|
|
1086 |
|
int i, j; |
1087 |
|
|
1088 |
|
#if (DBG>0) |
1089 |
|
Last.Level = 0; Last.Run = -1; /* just initialize to smthg */ |
1090 |
|
#endif |
1091 |
|
|
1092 |
|
Non_Zero = Find_Last(Out, Zigzag, Non_Zero); |
1093 |
|
if (Non_Zero<0) |
1094 |
|
return -1; |
1095 |
|
|
1096 |
|
for(i=0; i<=Non_Zero; i++) |
1097 |
|
{ |
1098 |
|
const int AC = In[Zigzag[i]]; |
1099 |
|
const int Level1 = Out[Zigzag[i]]; |
1100 |
|
const int Dist0 = Lambda* AC*AC; |
1101 |
|
uint32_t Best_Cost = 0xf0000000; |
1102 |
|
Last_Cost += Dist0; |
1103 |
|
|
1104 |
|
if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */ |
1105 |
|
{ |
1106 |
|
int dQ; |
1107 |
|
int Run; |
1108 |
|
uint32_t Cost0; |
1109 |
|
|
1110 |
|
if (AC<0) { |
1111 |
|
Nodes[i].Level = -1; |
1112 |
|
dQ = Lev0 + AC; |
1113 |
|
} else { |
1114 |
|
Nodes[i].Level = 1; |
1115 |
|
dQ = Lev0 - AC; |
1116 |
|
} |
1117 |
|
Cost0 = Lambda*dQ*dQ; |
1118 |
|
|
1119 |
|
Nodes[i].Run = 1; |
1120 |
|
Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0; |
1121 |
|
for(Run=i-Run_Start; Run>0; --Run) |
1122 |
|
{ |
1123 |
|
const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run]; |
1124 |
|
const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT); |
1125 |
|
const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT); |
1126 |
|
|
1127 |
|
/* |
1128 |
|
* TODO: what about tie-breaks? Should we favor short runs or |
1129 |
|
* long runs? Although the error is the same, it would not be |
1130 |
|
* spread the same way along high and low frequencies... |
1131 |
|
*/ |
1132 |
|
if (Cost<Best_Cost) { |
1133 |
|
Best_Cost = Cost; |
1134 |
|
Nodes[i].Run = Run; |
1135 |
|
} |
1136 |
|
|
1137 |
|
if (lCost<Last_Cost) { |
1138 |
|
Last_Cost = lCost; |
1139 |
|
Last.Run = Run; |
1140 |
|
Last_Node = i; |
1141 |
|
} |
1142 |
|
} |
1143 |
|
if (Last_Node==i) |
1144 |
|
Last.Level = Nodes[i].Level; |
1145 |
|
|
1146 |
|
if (DBG==1) { |
1147 |
|
Run_Costs[i] = Best_Cost; |
1148 |
|
printf( "Costs #%2d: ", i); |
1149 |
|
for(j=-1;j<=Non_Zero;++j) { |
1150 |
|
if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 ); |
1151 |
|
else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 ); |
1152 |
|
else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 ); |
1153 |
|
else printf( " - |" ); |
1154 |
|
} |
1155 |
|
printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run ); |
1156 |
|
printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run ); |
1157 |
|
printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 ); |
1158 |
|
printf( "\n" ); |
1159 |
|
} |
1160 |
|
} |
1161 |
|
else /* "big" levels */ |
1162 |
|
{ |
1163 |
|
const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last; |
1164 |
|
int Level2; |
1165 |
|
int dQ1, dQ2; |
1166 |
|
int Run; |
1167 |
|
uint32_t Dist1,Dist2; |
1168 |
|
int dDist21; |
1169 |
|
|
1170 |
|
if (Level1>1) { |
1171 |
|
dQ1 = Level1*Mult-AC + Bias; |
1172 |
|
dQ2 = dQ1 - Mult; |
1173 |
|
Level2 = Level1-1; |
1174 |
|
Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0; |
1175 |
|
Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0; |
1176 |
|
Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0; |
1177 |
|
Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0; |
1178 |
|
} else { /* Level1<-1 */ |
1179 |
|
dQ1 = Level1*Mult-AC - Bias; |
1180 |
|
dQ2 = dQ1 + Mult; |
1181 |
|
Level2 = Level1 + 1; |
1182 |
|
Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0; |
1183 |
|
Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0; |
1184 |
|
Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0; |
1185 |
|
Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0; |
1186 |
|
} |
1187 |
|
Dist1 = Lambda*dQ1*dQ1; |
1188 |
|
Dist2 = Lambda*dQ2*dQ2; |
1189 |
|
dDist21 = Dist2-Dist1; |
1190 |
|
|
1191 |
|
for(Run=i-Run_Start; Run>0; --Run) |
1192 |
|
{ |
1193 |
|
const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run]; |
1194 |
|
uint32_t Cost1, Cost2; |
1195 |
|
int bLevel; |
1196 |
|
|
1197 |
|
/* |
1198 |
|
* for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following: |
1199 |
|
* if (Cost_Base>=Best_Cost) continue; |
1200 |
|
*/ |
1201 |
|
Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT); |
1202 |
|
Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21; |
1203 |
|
|
1204 |
|
if (Cost2<Cost1) { |
1205 |
|
Cost1 = Cost2; |
1206 |
|
bLevel = Level2; |
1207 |
|
} else |
1208 |
|
bLevel = Level1; |
1209 |
|
|
1210 |
|
if (Cost1<Best_Cost) { |
1211 |
|
Best_Cost = Cost1; |
1212 |
|
Nodes[i].Run = Run; |
1213 |
|
Nodes[i].Level = bLevel; |
1214 |
|
} |
1215 |
|
|
1216 |
|
Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT); |
1217 |
|
Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21; |
1218 |
|
|
1219 |
|
if (Cost2<Cost1) { |
1220 |
|
Cost1 = Cost2; |
1221 |
|
bLevel = Level2; |
1222 |
|
} else |
1223 |
|
bLevel = Level1; |
1224 |
|
|
1225 |
|
if (Cost1<Last_Cost) { |
1226 |
|
Last_Cost = Cost1; |
1227 |
|
Last.Run = Run; |
1228 |
|
Last.Level = bLevel; |
1229 |
|
Last_Node = i; |
1230 |
|
} |
1231 |
|
} /* end of "for Run" */ |
1232 |
|
|
1233 |
|
if (DBG==1) { |
1234 |
|
Run_Costs[i] = Best_Cost; |
1235 |
|
printf( "Costs #%2d: ", i); |
1236 |
|
for(j=-1;j<=Non_Zero;++j) { |
1237 |
|
if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 ); |
1238 |
|
else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 ); |
1239 |
|
else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 ); |
1240 |
|
else printf( " - |" ); |
1241 |
|
} |
1242 |
|
printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run ); |
1243 |
|
printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run ); |
1244 |
|
printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 ); |
1245 |
|
printf( "\n" ); |
1246 |
|
} |
1247 |
|
} |
1248 |
|
|
1249 |
|
Run_Costs[i] = Best_Cost; |
1250 |
|
|
1251 |
|
if (Best_Cost < Min_Cost + Dist0) { |
1252 |
|
Min_Cost = Best_Cost; |
1253 |
|
Run_Start = i; |
1254 |
|
} |
1255 |
|
else |
1256 |
|
{ |
1257 |
|
/* |
1258 |
|
* as noticed by Michael Niedermayer (michaelni at gmx.at), there's |
1259 |
|
* a code shorter by 1 bit for a larger run (!), same level. We give |
1260 |
|
* it a chance by not moving the left barrier too much. |
1261 |
|
*/ |
1262 |
|
|
1263 |
|
while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) ) |
1264 |
|
Run_Start++; |
1265 |
|
|
1266 |
|
/* spread on preceding coeffs the cost incurred by skipping this one */ |
1267 |
|
for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0; |
1268 |
|
Min_Cost += Dist0; |
1269 |
|
} |
1270 |
} |
} |
1271 |
|
|
1272 |
|
if (DBG) { |
1273 |
|
Last_Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda); |
1274 |
|
if (DBG==1) { |
1275 |
|
printf( "=> " ); |
1276 |
|
for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] ); |
1277 |
|
printf( "\n" ); |
1278 |
|
} |
1279 |
|
} |
1280 |
|
|
1281 |
|
if (Last_Node<0) |
1282 |
|
return -1; |
1283 |
|
|
1284 |
|
/* reconstruct optimal sequence backward with surviving paths */ |
1285 |
|
memset(Out, 0x00, 64*sizeof(*Out)); |
1286 |
|
Out[Zigzag[Last_Node]] = Last.Level; |
1287 |
|
i = Last_Node - Last.Run; |
1288 |
|
while(i>=0) { |
1289 |
|
Out[Zigzag[i]] = Nodes[i].Level; |
1290 |
|
i -= Nodes[i].Run; |
1291 |
|
} |
1292 |
|
|
1293 |
|
if (DBG) { |
1294 |
|
uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda); |
1295 |
|
if (DBG==1) { |
1296 |
|
printf( "<= " ); |
1297 |
|
for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] ); |
1298 |
|
printf( "\n--------------------------------\n" ); |
1299 |
|
} |
1300 |
|
if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost ); |
1301 |
|
} |
1302 |
|
return Last_Node; |
1303 |
|
} |
1304 |
|
|
1305 |
|
#undef DBG |
1306 |
|
|
1307 |
|
#endif |