[cvs] / xvidcore / src / utils / mbtransquant.c Repository:
ViewVC logotype

Annotation of /xvidcore/src/utils/mbtransquant.c

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1.21.2.18 - (view) (download)

1 : edgomez 1.21.2.6 /*****************************************************************************
2 :     *
3 :     * XVID MPEG-4 VIDEO CODEC
4 :     * - MB Transfert/Quantization functions -
5 :     *
6 :     * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org>
7 :     * 2001-2003 Michael Militzer <isibaar@xvid.org>
8 :     * 2003 Edouard Gomez <ed.gomez@free.fr>
9 :     *
10 :     * 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
12 :     * the Free Software Foundation ; either version 2 of the License, or
13 :     * (at your option) any later version.
14 :     *
15 :     * This program is distributed in the hope that it will be useful,
16 :     * but WITHOUT ANY WARRANTY ; without even the implied warranty of
17 :     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 :     * GNU General Public License for more details.
19 :     *
20 :     * You should have received a copy of the GNU General Public License
21 :     * along with this program ; if not, write to the Free Software
22 :     * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 :     *
24 : edgomez 1.21.2.17 * $Id$
25 : edgomez 1.21.2.6 *
26 :     ****************************************************************************/
27 : Isibaar 1.1
28 : chl 1.21.2.11 #include <stdio.h>
29 : edgomez 1.21.2.7 #include <stdlib.h>
30 : chl 1.21.2.11 #include <string.h>
31 : edgomez 1.3
32 : Isibaar 1.1 #include "../portab.h"
33 :     #include "mbfunctions.h"
34 :    
35 :     #include "../global.h"
36 :     #include "mem_transfer.h"
37 :     #include "timer.h"
38 : chl 1.21.2.9 #include "../bitstream/mbcoding.h"
39 : chl 1.21.2.10 #include "../bitstream/zigzag.h"
40 : Isibaar 1.1 #include "../dct/fdct.h"
41 :     #include "../dct/idct.h"
42 : edgomez 1.21.2.18 #include "../quant/quant.h"
43 : Isibaar 1.1 #include "../encoder.h"
44 :    
45 : edgomez 1.21 #include "../image/reduced.h"
46 : Isibaar 1.1
47 : edgomez 1.21 MBFIELDTEST_PTR MBFieldTest;
48 : Isibaar 1.1
49 : edgomez 1.21.2.6 /*
50 :     * Skip blocks having a coefficient sum below this value. This value will be
51 :     * corrected according to the MB quantizer to avoid artifacts for quant==1
52 :     */
53 :     #define PVOP_TOOSMALL_LIMIT 1
54 :     #define BVOP_TOOSMALL_LIMIT 3
55 :    
56 :     /*****************************************************************************
57 :     * Local functions
58 :     ****************************************************************************/
59 : Isibaar 1.1
60 : edgomez 1.21.2.6 /* permute block and return field dct choice */
61 :     static __inline uint32_t
62 :     MBDecideFieldDCT(int16_t data[6 * 64])
63 : Isibaar 1.1 {
64 : edgomez 1.21.2.6 uint32_t field = MBFieldTest(data);
65 : edgomez 1.3
66 : edgomez 1.21.2.6 if (field)
67 :     MBFrameToField(data);
68 : edgomez 1.21
69 : edgomez 1.21.2.6 return field;
70 :     }
71 : h 1.2
72 : edgomez 1.21.2.6 /* Performs Forward DCT on all blocks */
73 :     static __inline void
74 : syskin 1.21.2.8 MBfDCT(const MBParam * const pParam,
75 :     const FRAMEINFO * const frame,
76 :     MACROBLOCK * const pMB,
77 : edgomez 1.21.2.6 uint32_t x_pos,
78 :     uint32_t y_pos,
79 :     int16_t data[6 * 64])
80 : syskin 1.21.2.8 {
81 : edgomez 1.21.2.6 /* Handles interlacing */
82 : h 1.2 start_timer();
83 :     pMB->field_dct = 0;
84 : edgomez 1.21.2.5 if ((frame->vol_flags & XVID_VOL_INTERLACING) &&
85 : h 1.11 (x_pos>0) && (x_pos<pParam->mb_width-1) &&
86 :     (y_pos>0) && (y_pos<pParam->mb_height-1)) {
87 : h 1.2 pMB->field_dct = MBDecideFieldDCT(data);
88 :     }
89 :     stop_interlacing_timer();
90 :    
91 : edgomez 1.21.2.6 /* Perform DCT */
92 :     start_timer();
93 :     fdct(&data[0 * 64]);
94 :     fdct(&data[1 * 64]);
95 :     fdct(&data[2 * 64]);
96 :     fdct(&data[3 * 64]);
97 :     fdct(&data[4 * 64]);
98 :     fdct(&data[5 * 64]);
99 :     stop_dct_timer();
100 :     }
101 :    
102 :     /* 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();
108 :     if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]);
109 :     if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]);
110 :     if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]);
111 :     if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]);
112 :     if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]);
113 :     if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]);
114 :     stop_idct_timer();
115 :     }
116 :    
117 :     /* Quantize all blocks -- Intra mode */
118 :     static __inline void
119 :     MBQuantIntra(const MBParam * pParam,
120 : chl 1.21.2.9 const FRAMEINFO * const frame,
121 : edgomez 1.21.2.6 const MACROBLOCK * pMB,
122 : syskin 1.21.2.8 int16_t qcoeff[6 * 64],
123 : edgomez 1.21.2.6 int16_t data[6*64])
124 :     {
125 : edgomez 1.21.2.16 int mpeg;
126 :     int scaler_lum, scaler_chr;
127 : edgomez 1.21.2.6
128 : edgomez 1.21.2.18 quant_intraFuncPtr const quant[2] =
129 : edgomez 1.21.2.16 {
130 : edgomez 1.21.2.18 quant_h263_intra,
131 :     quant_mpeg_intra
132 : edgomez 1.21.2.16 };
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 : edgomez 1.21.2.6
138 : edgomez 1.21.2.16 /* Quantize the block */
139 :     start_timer();
140 :     quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum);
141 :     quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum);
142 :     quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum);
143 :     quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum);
144 :     quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr);
145 :     quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr);
146 :     stop_quant_timer();
147 : edgomez 1.21.2.6 }
148 :    
149 :     /* DeQuantize all blocks -- Intra mode */
150 :     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 : edgomez 1.21.2.16 int mpeg;
157 :     int scaler_lum, scaler_chr;
158 : edgomez 1.21.2.6
159 : edgomez 1.21.2.18 quant_intraFuncPtr const dequant[2] =
160 : edgomez 1.21.2.16 {
161 : edgomez 1.21.2.18 dequant_h263_intra,
162 :     dequant_mpeg_intra
163 : edgomez 1.21.2.16 };
164 :    
165 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
166 :     scaler_lum = get_dc_scaler(iQuant, 1);
167 :     scaler_chr = get_dc_scaler(iQuant, 0);
168 : Isibaar 1.1
169 : edgomez 1.21.2.16 start_timer();
170 :     dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum);
171 :     dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum);
172 :     dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum);
173 :     dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum);
174 :     dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr);
175 :     dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr);
176 :     stop_iquant_timer();
177 : edgomez 1.21.2.6 }
178 : Isibaar 1.1
179 : chl 1.21.2.10
180 : edgomez 1.21.2.16 typedef int (*trellis_func_ptr_t)(int16_t *const Out,
181 :     const int16_t *const In,
182 :     int Q,
183 :     const uint16_t * const Zigzag,
184 :     int Non_Zero);
185 :    
186 : edgomez 1.21.2.17 static int
187 : edgomez 1.21.2.13 dct_quantize_trellis_h263_c(int16_t *const Out,
188 :     const int16_t *const In,
189 :     int Q,
190 :     const uint16_t * const Zigzag,
191 :     int Non_Zero);
192 : chl 1.21.2.10
193 : edgomez 1.21.2.17 static int
194 : edgomez 1.21.2.13 dct_quantize_trellis_mpeg_c(int16_t *const Out,
195 :     const int16_t *const In,
196 :     int Q,
197 :     const uint16_t * const Zigzag,
198 :     int Non_Zero);
199 : chl 1.21.2.10
200 : edgomez 1.21.2.6 /* Quantize all blocks -- Inter mode */
201 :     static __inline uint8_t
202 :     MBQuantInter(const MBParam * pParam,
203 : chl 1.21.2.9 const FRAMEINFO * const frame,
204 : edgomez 1.21.2.6 const MACROBLOCK * pMB,
205 :     int16_t data[6 * 64],
206 :     int16_t qcoeff[6 * 64],
207 :     int bvop,
208 :     int limit)
209 :     {
210 :    
211 :     int i;
212 :     uint8_t cbp = 0;
213 :     int sum;
214 : edgomez 1.21.2.16 int code_block, mpeg;
215 :    
216 : edgomez 1.21.2.18 quant_interFuncPtr const quant[2] =
217 : edgomez 1.21.2.16 {
218 : edgomez 1.21.2.18 quant_h263_inter,
219 :     quant_mpeg_inter
220 : edgomez 1.21.2.16 };
221 :    
222 :     trellis_func_ptr_t const trellis[2] =
223 :     {
224 : edgomez 1.21.2.18 dct_quantize_trellis_h263_c,
225 :     dct_quantize_trellis_mpeg_c
226 : edgomez 1.21.2.16 };
227 :    
228 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
229 : edgomez 1.21.2.6
230 :     for (i = 0; i < 6; i++) {
231 : syskin 1.21.2.8
232 : edgomez 1.21.2.6 /* Quantize the block */
233 :     start_timer();
234 : edgomez 1.21.2.16
235 :     sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant);
236 :    
237 :     if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) {
238 :     sum = trellis[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63);
239 : chl 1.21.2.9 }
240 : edgomez 1.21.2.6 stop_quant_timer();
241 :    
242 :     /*
243 :     * We code the block if the sum is higher than the limit and if the first
244 :     * two AC coefficients in zig zag order are not zero.
245 :     */
246 :     code_block = 0;
247 :     if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) {
248 :     code_block = 1;
249 : edgomez 1.7 } else {
250 : edgomez 1.21
251 : edgomez 1.21.2.6 if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) {
252 :     /* dark blocks prevention for direct mode */
253 :     if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0))
254 :     code_block = 1;
255 : edgomez 1.21 } else {
256 : edgomez 1.21.2.6 /* not direct mode */
257 :     if (qcoeff[i*64] != 0)
258 :     code_block = 1;
259 : edgomez 1.21 }
260 :     }
261 : edgomez 1.21.2.6
262 :     /* Set the corresponding cbp bit */
263 :     cbp |= code_block << (5 - i);
264 : edgomez 1.21 }
265 :    
266 : edgomez 1.21.2.6 return(cbp);
267 :     }
268 : Isibaar 1.1
269 : edgomez 1.21.2.6 /* DeQuantize all blocks -- Inter mode */
270 : syskin 1.21.2.8 static __inline void
271 : edgomez 1.21.2.6 MBDeQuantInter(const MBParam * pParam,
272 :     const int iQuant,
273 :     int16_t data[6 * 64],
274 :     int16_t qcoeff[6 * 64],
275 :     const uint8_t cbp)
276 :     {
277 : edgomez 1.21.2.16 int mpeg;
278 : edgomez 1.21.2.6
279 : edgomez 1.21.2.18 quant_interFuncPtr const dequant[2] =
280 : edgomez 1.21.2.16 {
281 : edgomez 1.21.2.18 dequant_h263_inter,
282 :     dequant_mpeg_inter
283 : edgomez 1.21.2.16 };
284 :    
285 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
286 :    
287 :     start_timer();
288 :     if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant);
289 :     if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant);
290 :     if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant);
291 :     if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant);
292 :     if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant);
293 :     if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant);
294 :     stop_iquant_timer();
295 : Isibaar 1.1 }
296 :    
297 : edgomez 1.21.2.6 typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);
298 :     typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);
299 : Isibaar 1.1
300 : edgomez 1.3
301 : edgomez 1.21.2.6 static __inline void
302 : syskin 1.21.2.8 MBTrans8to16(const MBParam * const pParam,
303 :     const FRAMEINFO * const frame,
304 :     const MACROBLOCK * const pMB,
305 : edgomez 1.21.2.6 const uint32_t x_pos,
306 :     const uint32_t y_pos,
307 :     int16_t data[6 * 64])
308 :     {
309 : h 1.4 uint32_t stride = pParam->edged_width;
310 :     uint32_t stride2 = stride / 2;
311 : edgomez 1.21.2.6 uint32_t next_block = stride * 8;
312 : syskin 1.21.2.8 int32_t cst;
313 : edgomez 1.21.2.16 int vop_reduced;
314 : Isibaar 1.1 uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
315 : syskin 1.21.2.8 const IMAGE * const pCurrent = &frame->image;
316 : edgomez 1.21.2.16 transfer_operation_8to16_t * const functions[2] =
317 :     {
318 :     (transfer_operation_8to16_t *)transfer_8to16copy,
319 :     (transfer_operation_8to16_t *)filter_18x18_to_8x8
320 :     };
321 : edgomez 1.21.2.6 transfer_operation_8to16_t *transfer_op = NULL;
322 : edgomez 1.7
323 : edgomez 1.21.2.16 vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED);
324 : edgomez 1.21.2.6
325 : edgomez 1.21.2.16 /* Image pointers */
326 :     pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced));
327 :     pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
328 :     pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
329 : edgomez 1.21.2.6
330 : edgomez 1.21.2.16 /* Block size */
331 :     cst = 8<<vop_reduced;
332 : edgomez 1.21.2.6
333 : edgomez 1.21.2.16 /* Operation function */
334 :     transfer_op = functions[vop_reduced];
335 : Isibaar 1.1
336 : edgomez 1.21.2.6 /* Do the transfer */
337 : h 1.2 start_timer();
338 : edgomez 1.21.2.6 transfer_op(&data[0 * 64], pY_Cur, stride);
339 :     transfer_op(&data[1 * 64], pY_Cur + cst, stride);
340 :     transfer_op(&data[2 * 64], pY_Cur + next_block, stride);
341 :     transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride);
342 :     transfer_op(&data[4 * 64], pU_Cur, stride2);
343 :     transfer_op(&data[5 * 64], pV_Cur, stride2);
344 :     stop_transfer_timer();
345 : syskin 1.21.2.8 }
346 : h 1.2
347 : edgomez 1.21.2.6 static __inline void
348 : syskin 1.21.2.8 MBTrans16to8(const MBParam * const pParam,
349 :     const FRAMEINFO * const frame,
350 :     const MACROBLOCK * const pMB,
351 : edgomez 1.21.2.6 const uint32_t x_pos,
352 :     const uint32_t y_pos,
353 :     int16_t data[6 * 64],
354 : edgomez 1.21.2.16 const uint32_t add, /* Must be 1 or 0 */
355 : edgomez 1.21.2.6 const uint8_t cbp)
356 :     {
357 :     uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
358 :     uint32_t stride = pParam->edged_width;
359 :     uint32_t stride2 = stride / 2;
360 :     uint32_t next_block = stride * 8;
361 : syskin 1.21.2.8 uint32_t cst;
362 : edgomez 1.21.2.16 int vop_reduced;
363 : syskin 1.21.2.8 const IMAGE * const pCurrent = &frame->image;
364 : edgomez 1.21.2.18
365 : edgomez 1.21.2.16 /* Array of function pointers, indexed by [vop_reduced<<1+add] */
366 :     transfer_operation_16to8_t * const functions[4] =
367 :     {
368 :     (transfer_operation_16to8_t*)transfer_16to8copy,
369 :     (transfer_operation_16to8_t*)transfer_16to8add,
370 :     (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8,
371 :     (transfer_operation_16to8_t*)add_upsampled_8x8_16to8
372 :     };
373 : edgomez 1.21.2.17
374 : edgomez 1.21.2.6 transfer_operation_16to8_t *transfer_op = NULL;
375 : edgomez 1.21
376 : edgomez 1.21.2.6 if (pMB->field_dct) {
377 :     next_block = stride;
378 :     stride *= 2;
379 :     }
380 : Isibaar 1.1
381 : edgomez 1.21.2.16 /* Makes this vars booleans */
382 :     vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED);
383 : edgomez 1.21.2.6
384 : edgomez 1.21.2.16 /* Image pointers */
385 :     pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced));
386 :     pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
387 :     pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
388 : Isibaar 1.1
389 : edgomez 1.21.2.16 /* Block size */
390 :     cst = 8<<vop_reduced;
391 :    
392 :     /* Operation function */
393 :     transfer_op = functions[(vop_reduced<<1) + add];
394 : h 1.2
395 : edgomez 1.21.2.6 /* Do the operation */
396 : h 1.2 start_timer();
397 : edgomez 1.21.2.16 if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride);
398 :     if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride);
399 :     if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride);
400 : edgomez 1.21.2.6 if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride);
401 : edgomez 1.21.2.16 if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2);
402 :     if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2);
403 : h 1.2 stop_transfer_timer();
404 :     }
405 :    
406 : edgomez 1.21.2.6 /*****************************************************************************
407 :     * Module functions
408 :     ****************************************************************************/
409 :    
410 : syskin 1.21.2.8 void
411 :     MBTransQuantIntra(const MBParam * const pParam,
412 :     const FRAMEINFO * const frame,
413 :     MACROBLOCK * const pMB,
414 : chl 1.8 const uint32_t x_pos,
415 :     const uint32_t y_pos,
416 :     int16_t data[6 * 64],
417 :     int16_t qcoeff[6 * 64])
418 :     {
419 :    
420 : edgomez 1.21.2.6 /* Transfer data */
421 :     MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data);
422 : chl 1.8
423 : edgomez 1.21.2.6 /* Perform DCT (and field decision) */
424 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
425 : chl 1.8
426 : edgomez 1.21.2.6 /* Quantize the block */
427 : chl 1.21.2.9 MBQuantIntra(pParam, frame, pMB, data, qcoeff);
428 : edgomez 1.21.2.6
429 :     /* DeQuantize the block */
430 :     MBDeQuantIntra(pParam, pMB->quant, data, qcoeff);
431 :    
432 :     /* Perform inverse DCT*/
433 :     MBiDCT(data, 0x3F);
434 :    
435 :     /* Transfer back the data -- Don't add data */
436 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F);
437 : chl 1.8 }
438 :    
439 : edgomez 1.21.2.6
440 : chl 1.8 uint8_t
441 : syskin 1.21.2.8 MBTransQuantInter(const MBParam * const pParam,
442 :     const FRAMEINFO * const frame,
443 :     MACROBLOCK * const pMB,
444 : edgomez 1.21.2.2 const uint32_t x_pos,
445 :     const uint32_t y_pos,
446 : chl 1.8 int16_t data[6 * 64],
447 :     int16_t qcoeff[6 * 64])
448 :     {
449 :     uint8_t cbp;
450 : edgomez 1.21.2.6 uint32_t limit;
451 : chl 1.8
452 : edgomez 1.21.2.18 /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
453 :     * already */
454 : edgomez 1.21.2.2
455 : edgomez 1.21.2.6 /* Perform DCT (and field decision) */
456 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
457 : chl 1.8
458 : edgomez 1.21.2.6 /* Set the limit threshold */
459 :     limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
460 : chl 1.8
461 : Isibaar 1.21.2.15 if (frame->vop_flags & XVID_VOP_CARTOON)
462 :     limit *= 3;
463 :    
464 : edgomez 1.21.2.6 /* Quantize the block */
465 : chl 1.21.2.9 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
466 : chl 1.8
467 : edgomez 1.21.2.6 /* DeQuantize the block */
468 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
469 : chl 1.8
470 : edgomez 1.21.2.6 /* Perform inverse DCT*/
471 :     MBiDCT(data, cbp);
472 : chl 1.8
473 : edgomez 1.21.2.6 /* Transfer back the data -- Add the data */
474 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
475 : syskin 1.21.2.8
476 : edgomez 1.21.2.6 return(cbp);
477 : chl 1.8 }
478 :    
479 : edgomez 1.21.2.6 uint8_t
480 :     MBTransQuantInterBVOP(const MBParam * pParam,
481 : edgomez 1.21.2.16 FRAMEINFO * frame,
482 :     MACROBLOCK * pMB,
483 :     const uint32_t x_pos,
484 :     const uint32_t y_pos,
485 :     int16_t data[6 * 64],
486 :     int16_t qcoeff[6 * 64])
487 : chl 1.8 {
488 : edgomez 1.21.2.6 uint8_t cbp;
489 :     uint32_t limit;
490 : syskin 1.21.2.8
491 : edgomez 1.21.2.18 /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
492 :     * already */
493 : chl 1.8
494 : edgomez 1.21.2.6 /* Perform DCT (and field decision) */
495 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
496 : chl 1.8
497 : edgomez 1.21.2.6 /* Set the limit threshold */
498 :     limit = BVOP_TOOSMALL_LIMIT;
499 : Isibaar 1.21.2.15
500 :     if (frame->vop_flags & XVID_VOP_CARTOON)
501 :     limit *= 2;
502 : chl 1.8
503 : edgomez 1.21.2.6 /* Quantize the block */
504 : chl 1.21.2.9 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
505 : h 1.2
506 : edgomez 1.21.2.6 /*
507 :     * History comment:
508 :     * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
509 :     *
510 : edgomez 1.21.2.18 * BUT some plugins require the rebuilt original frame to be passed so we
511 :     * have to take care of that here
512 : edgomez 1.21.2.6 */
513 :     if((pParam->plugin_flags & XVID_REQORIGINAL)) {
514 : h 1.2
515 : edgomez 1.21.2.6 /* DeQuantize the block */
516 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
517 : Isibaar 1.1
518 : edgomez 1.21.2.6 /* Perform inverse DCT*/
519 :     MBiDCT(data, cbp);
520 : edgomez 1.21
521 : edgomez 1.21.2.6 /* Transfer back the data -- Add the data */
522 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
523 : edgomez 1.21 }
524 :    
525 : edgomez 1.21.2.6 return(cbp);
526 : edgomez 1.21 }
527 :    
528 :     /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */
529 :     uint32_t
530 :     MBFieldTest_c(int16_t data[6 * 64])
531 :     {
532 : edgomez 1.7 const uint8_t blocks[] =
533 :     { 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 };
534 :     const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 };
535 : h 1.2
536 :     int frame = 0, field = 0;
537 :     int i, j;
538 :    
539 : edgomez 1.7 for (i = 0; i < 7; ++i) {
540 :     for (j = 0; j < 8; ++j) {
541 :     frame +=
542 : edgomez 1.21.2.7 abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]);
543 : edgomez 1.7 frame +=
544 : edgomez 1.21.2.7 abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]);
545 : edgomez 1.7 frame +=
546 : edgomez 1.21.2.7 abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]);
547 : edgomez 1.7 frame +=
548 : edgomez 1.21.2.7 abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]);
549 : edgomez 1.7
550 :     field +=
551 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + lines[i + 1] + j] -
552 : edgomez 1.7 data[blocks[i] + lines[i] + j]);
553 :     field +=
554 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] -
555 : edgomez 1.7 data[blocks[i] + lines[i] + 8 + j]);
556 :     field +=
557 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] -
558 : edgomez 1.7 data[blocks[i] + 64 + lines[i] + j]);
559 :     field +=
560 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] -
561 : edgomez 1.7 data[blocks[i] + 64 + lines[i] + 8 + j]);
562 : Isibaar 1.1 }
563 :     }
564 : h 1.2
565 : edgomez 1.21 return (frame >= (field + 350));
566 : h 1.2 }
567 :    
568 :    
569 :     /* deinterlace Y blocks vertically */
570 :    
571 :     #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp))
572 : syskin 1.21.2.8 #define LINE(X,Y) &data[X*64 + Y*8]
573 : h 1.2
574 : edgomez 1.7 void
575 :     MBFrameToField(int16_t data[6 * 64])
576 : h 1.2 {
577 :     int16_t tmp[8];
578 :    
579 :     /* left blocks */
580 :    
581 : edgomez 1.21.2.13 /* 1=2, 2=4, 4=8, 8=1 */
582 : edgomez 1.7 MOVLINE(tmp, LINE(0, 1));
583 :     MOVLINE(LINE(0, 1), LINE(0, 2));
584 :     MOVLINE(LINE(0, 2), LINE(0, 4));
585 :     MOVLINE(LINE(0, 4), LINE(2, 0));
586 :     MOVLINE(LINE(2, 0), tmp);
587 : h 1.2
588 : edgomez 1.21.2.13 /* 3=6, 6=12, 12=9, 9=3 */
589 : edgomez 1.7 MOVLINE(tmp, LINE(0, 3));
590 :     MOVLINE(LINE(0, 3), LINE(0, 6));
591 :     MOVLINE(LINE(0, 6), LINE(2, 4));
592 :     MOVLINE(LINE(2, 4), LINE(2, 1));
593 :     MOVLINE(LINE(2, 1), tmp);
594 : h 1.2
595 : edgomez 1.21.2.13 /* 5=10, 10=5 */
596 : edgomez 1.7 MOVLINE(tmp, LINE(0, 5));
597 :     MOVLINE(LINE(0, 5), LINE(2, 2));
598 :     MOVLINE(LINE(2, 2), tmp);
599 : h 1.2
600 : edgomez 1.21.2.13 /* 7=14, 14=13, 13=11, 11=7 */
601 : edgomez 1.7 MOVLINE(tmp, LINE(0, 7));
602 :     MOVLINE(LINE(0, 7), LINE(2, 6));
603 :     MOVLINE(LINE(2, 6), LINE(2, 5));
604 :     MOVLINE(LINE(2, 5), LINE(2, 3));
605 :     MOVLINE(LINE(2, 3), tmp);
606 : h 1.2
607 :     /* right blocks */
608 :    
609 : edgomez 1.21.2.13 /* 1=2, 2=4, 4=8, 8=1 */
610 : edgomez 1.7 MOVLINE(tmp, LINE(1, 1));
611 :     MOVLINE(LINE(1, 1), LINE(1, 2));
612 :     MOVLINE(LINE(1, 2), LINE(1, 4));
613 :     MOVLINE(LINE(1, 4), LINE(3, 0));
614 :     MOVLINE(LINE(3, 0), tmp);
615 : h 1.2
616 : edgomez 1.21.2.13 /* 3=6, 6=12, 12=9, 9=3 */
617 : edgomez 1.7 MOVLINE(tmp, LINE(1, 3));
618 :     MOVLINE(LINE(1, 3), LINE(1, 6));
619 :     MOVLINE(LINE(1, 6), LINE(3, 4));
620 :     MOVLINE(LINE(3, 4), LINE(3, 1));
621 :     MOVLINE(LINE(3, 1), tmp);
622 : h 1.2
623 : edgomez 1.21.2.13 /* 5=10, 10=5 */
624 : edgomez 1.7 MOVLINE(tmp, LINE(1, 5));
625 :     MOVLINE(LINE(1, 5), LINE(3, 2));
626 :     MOVLINE(LINE(3, 2), tmp);
627 : h 1.2
628 : edgomez 1.21.2.13 /* 7=14, 14=13, 13=11, 11=7 */
629 : edgomez 1.7 MOVLINE(tmp, LINE(1, 7));
630 :     MOVLINE(LINE(1, 7), LINE(3, 6));
631 :     MOVLINE(LINE(3, 6), LINE(3, 5));
632 :     MOVLINE(LINE(3, 5), LINE(3, 3));
633 :     MOVLINE(LINE(3, 3), tmp);
634 : Isibaar 1.1 }
635 : chl 1.21.2.10
636 : edgomez 1.21.2.13 /*****************************************************************************
637 :     * Trellis based R-D optimal quantization
638 :     *
639 :     * Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
640 :     *
641 :     ****************************************************************************/
642 :    
643 :     /*----------------------------------------------------------------------------
644 :     *
645 :     * Trellis-Based quantization
646 :     *
647 :     * So far I understand this paper:
648 :     *
649 :     * "Trellis-Based R-D Optimal Quantization in H.263+"
650 :     * J.Wen, M.Luttrell, J.Villasenor
651 :     * IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
652 :     *
653 :     * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
654 :     * Source Shorted Path algo. But due to the underlying graph structure
655 :     * ("Trellis"), it can be turned into a dynamic programming algo,
656 : edgomez 1.21.2.17 * partially saving the explicit graph's nodes representation. And
657 : edgomez 1.21.2.13 * without using a heap, since the open frontier of the DAG is always
658 :     * known, and of fixed sized.
659 :     *--------------------------------------------------------------------------*/
660 :    
661 : chl 1.21.2.10
662 :    
663 : edgomez 1.21.2.13 /* Codes lengths for relevant levels. */
664 : chl 1.21.2.10
665 : edgomez 1.21.2.16 /* let's factorize: */
666 : chl 1.21.2.10 static const uint8_t Code_Len0[64] = {
667 : edgomez 1.21.2.16 30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
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 : chl 1.21.2.10 static const uint8_t Code_Len1[64] = {
670 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len2[64] = {
673 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len3[64] = {
676 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len4[64] = {
679 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len5[64] = {
682 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len6[64] = {
685 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len7[64] = {
688 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len8[64] = {
691 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len9[64] = {
694 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len10[64] = {
697 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len11[64] = {
700 : edgomez 1.21.2.16 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,
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 : chl 1.21.2.10 static const uint8_t Code_Len12[64] = {
703 : edgomez 1.21.2.16 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,
704 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
705 : chl 1.21.2.10 static const uint8_t Code_Len13[64] = {
706 : edgomez 1.21.2.16 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,
707 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
708 : chl 1.21.2.10 static const uint8_t Code_Len14[64] = {
709 : edgomez 1.21.2.16 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,
710 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
711 : chl 1.21.2.10 static const uint8_t Code_Len15[64] = {
712 : edgomez 1.21.2.16 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,
713 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
714 : chl 1.21.2.10 static const uint8_t Code_Len16[64] = {
715 : edgomez 1.21.2.16 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,
716 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
717 : chl 1.21.2.10 static const uint8_t Code_Len17[64] = {
718 : edgomez 1.21.2.16 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,
719 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
720 : chl 1.21.2.10 static const uint8_t Code_Len18[64] = {
721 : edgomez 1.21.2.16 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,
722 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
723 : chl 1.21.2.10 static const uint8_t Code_Len19[64] = {
724 : edgomez 1.21.2.16 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,
725 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
726 : chl 1.21.2.10 static const uint8_t Code_Len20[64] = {
727 : edgomez 1.21.2.16 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,
728 :     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 };
729 : chl 1.21.2.10
730 : edgomez 1.21.2.16 /* a few more table for LAST table: */
731 : chl 1.21.2.10 static const uint8_t Code_Len21[64] = {
732 : edgomez 1.21.2.16 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,
733 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
734 : chl 1.21.2.10 static const uint8_t Code_Len22[64] = {
735 : edgomez 1.21.2.16 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,
736 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
737 : chl 1.21.2.10 static const uint8_t Code_Len23[64] = {
738 : edgomez 1.21.2.16 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,
739 :     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};
740 : chl 1.21.2.10 static const uint8_t Code_Len24[64] = {
741 : edgomez 1.21.2.16 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,
742 :     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};
743 : chl 1.21.2.10
744 :    
745 : edgomez 1.21.2.13 static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */
746 : edgomez 1.21.2.16 Code_Len20,Code_Len19,Code_Len18,Code_Len17,
747 :     Code_Len16,Code_Len15,Code_Len14,Code_Len13,
748 :     Code_Len12,Code_Len11,Code_Len10,Code_Len9,
749 :     Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5,
750 :     Code_Len4, Code_Len3, Code_Len3 ,Code_Len2,
751 :     Code_Len2, Code_Len1, Code_Len1, Code_Len1,
752 : chl 1.21.2.10 };
753 :    
754 : edgomez 1.21.2.13 static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */
755 : edgomez 1.21.2.17 Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
756 : chl 1.21.2.10 };
757 :    
758 :     #define TL(q) 0xfe00/(q*q)
759 :    
760 :     static const int Trellis_Lambda_Tabs[31] = {
761 : edgomez 1.21.2.16 TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7),
762 :     TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15),
763 :     TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23),
764 :     TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31)
765 : chl 1.21.2.10 };
766 :     #undef TL
767 :    
768 : edgomez 1.21.2.16 static int __inline
769 :     Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
770 : chl 1.21.2.10 {
771 : edgomez 1.21.2.16 while(i>=0)
772 :     if (C[Zigzag[i]])
773 :     return i;
774 :     else i--;
775 :     return -1;
776 : chl 1.21.2.10 }
777 :    
778 : edgomez 1.21.2.16 static int __inline
779 :     Compute_Sum(const int16_t *C, int last)
780 :     {
781 :     int sum = 0;
782 :    
783 :     while(last--)
784 :     sum += abs(C[last]);
785 :    
786 :     return(sum);
787 :     }
788 : edgomez 1.21.2.13 /* this routine has been strippen of all debug code */
789 : chl 1.21.2.11
790 : edgomez 1.21.2.17 static int
791 : chl 1.21.2.11 dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
792 :     {
793 :    
794 : edgomez 1.21.2.13 /*
795 :     * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
796 :     * not quantized one (Out[]). However, it only improves the result *very*
797 :     * slightly (~0.01dB), whereas speed drops to crawling level :)
798 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
799 :     */
800 : edgomez 1.21.2.16 typedef struct { int16_t Run, Level; } NODE;
801 : edgomez 1.21.2.17
802 : edgomez 1.21.2.16 NODE Nodes[65], Last;
803 :     uint32_t Run_Costs0[64+1];
804 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
805 :     const int Mult = 2*Q;
806 :     const int Bias = (Q-1) | 1;
807 :     const int Lev0 = Mult + Bias;
808 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */
809 :    
810 :     int Run_Start = -1;
811 :     uint32_t Min_Cost = 2<<16;
812 :    
813 :     int Last_Node = -1;
814 :     uint32_t Last_Cost = 0;
815 :    
816 :     int i, j, sum;
817 :     Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */
818 :    
819 :     Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
820 :     if (Non_Zero<0)
821 :     return 0; /* Sum is zero if there are only zero coeffs */
822 :    
823 :     for(i=0; i<=Non_Zero; i++) {
824 :     const int AC = In[Zigzag[i]];
825 :     const int Level1 = Out[Zigzag[i]];
826 :     const int Dist0 = Lambda* AC*AC;
827 :     uint32_t Best_Cost = 0xf0000000;
828 :     Last_Cost += Dist0;
829 :    
830 :     /* very specialized loop for -1,0,+1 */
831 :     if ((uint32_t)(Level1+1)<3) {
832 :     int dQ;
833 :     int Run;
834 :     uint32_t Cost0;
835 :    
836 :     if (AC<0) {
837 :     Nodes[i].Level = -1;
838 :     dQ = Lev0 + AC;
839 :     } else {
840 :     Nodes[i].Level = 1;
841 :     dQ = Lev0 - AC;
842 :     }
843 :     Cost0 = Lambda*dQ*dQ;
844 : edgomez 1.21.2.17
845 : edgomez 1.21.2.16 Nodes[i].Run = 1;
846 :     Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
847 :     for(Run=i-Run_Start; Run>0; --Run) {
848 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
849 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
850 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
851 :    
852 :     /*
853 :     * TODO: what about tie-breaks? Should we favor short runs or
854 :     * long runs? Although the error is the same, it would not be
855 :     * spread the same way along high and low frequencies...
856 :     */
857 :    
858 :     /* (I'd say: favour short runs => hifreq errors (HVS) -- gruel ) */
859 :    
860 :     if (Cost<Best_Cost) {
861 :     Best_Cost = Cost;
862 :     Nodes[i].Run = Run;
863 :     }
864 :    
865 :     if (lCost<Last_Cost) {
866 :     Last_Cost = lCost;
867 :     Last.Run = Run;
868 :     Last_Node = i;
869 :     }
870 :     }
871 : edgomez 1.21.2.17 if (Last_Node==i)
872 : edgomez 1.21.2.16 Last.Level = Nodes[i].Level;
873 :     } else { /* "big" levels */
874 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
875 :     int Level2;
876 :     int dQ1, dQ2;
877 :     int Run;
878 :     uint32_t Dist1,Dist2;
879 :     int dDist21;
880 : edgomez 1.21.2.17
881 : edgomez 1.21.2.16 if (Level1>1) {
882 :     dQ1 = Level1*Mult-AC + Bias;
883 :     dQ2 = dQ1 - Mult;
884 :     Level2 = Level1-1;
885 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
886 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
887 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
888 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
889 :     } else { /* Level1<-1 */
890 :     dQ1 = Level1*Mult-AC - Bias;
891 :     dQ2 = dQ1 + Mult;
892 :     Level2 = Level1 + 1;
893 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
894 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
895 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
896 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
897 :     }
898 : chl 1.21.2.11
899 : edgomez 1.21.2.16 Dist1 = Lambda*dQ1*dQ1;
900 :     Dist2 = Lambda*dQ2*dQ2;
901 :     dDist21 = Dist2-Dist1;
902 :    
903 :     for(Run=i-Run_Start; Run>0; --Run)
904 :     {
905 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
906 :     uint32_t Cost1, Cost2;
907 :     int bLevel;
908 :    
909 :     /*
910 :     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
911 :     * if (Cost_Base>=Best_Cost) continue;
912 :     * (? doesn't seem to have any effect -- gruel )
913 :     */
914 :    
915 :     Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
916 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
917 :    
918 : edgomez 1.21.2.17 if (Cost2<Cost1) {
919 :     Cost1 = Cost2;
920 :     bLevel = Level2;
921 : edgomez 1.21.2.16 } else {
922 :     bLevel = Level1;
923 :     }
924 :    
925 :     if (Cost1<Best_Cost) {
926 :     Best_Cost = Cost1;
927 :     Nodes[i].Run = Run;
928 :     Nodes[i].Level = bLevel;
929 :     }
930 :    
931 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
932 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
933 :    
934 : edgomez 1.21.2.17 if (Cost2<Cost1) {
935 :     Cost1 = Cost2;
936 :     bLevel = Level2;
937 : edgomez 1.21.2.16 } else {
938 :     bLevel = Level1;
939 :     }
940 : edgomez 1.21.2.17
941 : edgomez 1.21.2.16 if (Cost1<Last_Cost) {
942 :     Last_Cost = Cost1;
943 :     Last.Run = Run;
944 :     Last.Level = bLevel;
945 :     Last_Node = i;
946 :     }
947 :     } /* end of "for Run" */
948 : chl 1.21.2.11
949 : edgomez 1.21.2.16 }
950 : chl 1.21.2.11
951 : edgomez 1.21.2.16 Run_Costs[i] = Best_Cost;
952 : chl 1.21.2.11
953 : edgomez 1.21.2.16 if (Best_Cost < Min_Cost + Dist0) {
954 :     Min_Cost = Best_Cost;
955 :     Run_Start = i;
956 :     } else {
957 :     /*
958 :     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
959 :     * a code shorter by 1 bit for a larger run (!), same level. We give
960 :     * it a chance by not moving the left barrier too much.
961 :     */
962 :    
963 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
964 :     Run_Start++;
965 :    
966 :     /* spread on preceding coeffs the cost incurred by skipping this one */
967 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
968 :     Min_Cost += Dist0;
969 :     }
970 :     }
971 : chl 1.21.2.11
972 : edgomez 1.21.2.16 /* It seems trellis doesn't give good results... just compute the Out sum and
973 :     * quit (even if we did not modify it, upperlayer relies on this data) */
974 :     if (Last_Node<0)
975 :     return Compute_Sum(Out, Non_Zero);
976 :    
977 :     /* reconstruct optimal sequence backward with surviving paths */
978 :     memset(Out, 0x00, 64*sizeof(*Out));
979 :     Out[Zigzag[Last_Node]] = Last.Level;
980 :     i = Last_Node - Last.Run;
981 :     sum = 0;
982 :     while(i>=0) {
983 :     Out[Zigzag[i]] = Nodes[i].Level;
984 :     sum += abs(Nodes[i].Level);
985 :     i -= Nodes[i].Run;
986 :     }
987 : chl 1.21.2.11
988 : edgomez 1.21.2.16 return sum;
989 :     }
990 : chl 1.21.2.11
991 : edgomez 1.21.2.17 static int
992 : edgomez 1.21.2.16 dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
993 :     {
994 :     /* ToDo: Ok ok it's just a place holder for Gruel -- damn write this one :-) */
995 :     return Compute_Sum(Out, 63);
996 :     }
997 : chl 1.21.2.11
998 : edgomez 1.21.2.13 /* original version including heavy debugging info */
999 : chl 1.21.2.11
1000 :     #ifdef DBGTRELL
1001 : chl 1.21.2.10
1002 :     #define DBG 0
1003 :    
1004 : suxen_drol 1.21.2.12 static __inline uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,
1005 : edgomez 1.21.2.16 const uint16_t * Zigzag, int Max, int Lambda)
1006 : chl 1.21.2.10 {
1007 :     #if (DBG>0)
1008 : edgomez 1.21.2.16 const int16_t * const Ref = C + 6*64;
1009 :     int Last = Max;
1010 :     int Bits = 0;
1011 :     int Dist = 0;
1012 : edgomez 1.21.2.17 int i;
1013 : edgomez 1.21.2.16 uint32_t Cost;
1014 : edgomez 1.21.2.17
1015 :     while(Last>=0 && C[Zigzag[Last]]==0)
1016 : edgomez 1.21.2.16 Last--;
1017 : edgomez 1.21.2.17
1018 : edgomez 1.21.2.16 if (Last>=0) {
1019 :     int j=0, j0=0;
1020 :     int Run, Level;
1021 :    
1022 :     Bits = 2; /* CBP */
1023 :     while(j<Last) {
1024 : edgomez 1.21.2.17 while(!C[Zigzag[j]])
1025 : edgomez 1.21.2.16 j++;
1026 : edgomez 1.21.2.17 if (j==Last)
1027 : edgomez 1.21.2.16 break;
1028 :     Level=C[Zigzag[j]];
1029 :     Run = j - j0;
1030 :     j0 = ++j;
1031 : edgomez 1.21.2.17 if (Level>=-24 && Level<=24)
1032 : edgomez 1.21.2.16 Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];
1033 : edgomez 1.21.2.17 else
1034 : edgomez 1.21.2.16 Bits += 30;
1035 :     }
1036 :     Level = C[Zigzag[Last]];
1037 :     Run = j - j0;
1038 : edgomez 1.21.2.17 if (Level>=-6 && Level<=6)
1039 : edgomez 1.21.2.16 Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];
1040 : edgomez 1.21.2.17 else
1041 : chl 1.21.2.11 Bits += 30;
1042 : edgomez 1.21.2.16 }
1043 :    
1044 :     for(i=0; i<=Last; ++i) {
1045 :     int V = C[Zigzag[i]]*Mult;
1046 : edgomez 1.21.2.17 if (V>0)
1047 : edgomez 1.21.2.16 V += Bias;
1048 : edgomez 1.21.2.17 else
1049 :     if (V<0)
1050 : edgomez 1.21.2.16 V -= Bias;
1051 :     V -= Ref[Zigzag[i]];
1052 :     Dist += V*V;
1053 :     }
1054 :     Cost = Lambda*Dist + (Bits<<16);
1055 :     if (DBG==1)
1056 :     printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );
1057 :     return Cost;
1058 : chl 1.21.2.10
1059 :     #else
1060 : edgomez 1.21.2.16 return 0;
1061 : chl 1.21.2.10 #endif
1062 :     }
1063 :    
1064 :    
1065 : edgomez 1.21.2.17 static int
1066 : chl 1.21.2.10 dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
1067 :     {
1068 :    
1069 : edgomez 1.21.2.13 /*
1070 :     * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
1071 :     * not quantized one (Out[]). However, it only improves the result *very*
1072 :     * slightly (~0.01dB), whereas speed drops to crawling level :)
1073 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
1074 :     */
1075 : edgomez 1.21.2.16 typedef struct { int16_t Run, Level; } NODE;
1076 : edgomez 1.21.2.17
1077 : edgomez 1.21.2.16 NODE Nodes[65], Last;
1078 :     uint32_t Run_Costs0[64+1];
1079 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
1080 :     const int Mult = 2*Q;
1081 :     const int Bias = (Q-1) | 1;
1082 :     const int Lev0 = Mult + Bias;
1083 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */
1084 :    
1085 :     int Run_Start = -1;
1086 :     Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */
1087 :     uint32_t Min_Cost = 2<<16;
1088 : chl 1.21.2.10
1089 : edgomez 1.21.2.16 int Last_Node = -1;
1090 :     uint32_t Last_Cost = 0;
1091 : chl 1.21.2.10
1092 : edgomez 1.21.2.16 int i, j;
1093 : chl 1.21.2.11
1094 : chl 1.21.2.10 #if (DBG>0)
1095 : edgomez 1.21.2.16 Last.Level = 0; Last.Run = -1; /* just initialize to smthg */
1096 : chl 1.21.2.10 #endif
1097 :    
1098 : edgomez 1.21.2.16 Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
1099 :     if (Non_Zero<0)
1100 : edgomez 1.21.2.17 return -1;
1101 : edgomez 1.21.2.16
1102 :     for(i=0; i<=Non_Zero; i++)
1103 :     {
1104 :     const int AC = In[Zigzag[i]];
1105 :     const int Level1 = Out[Zigzag[i]];
1106 :     const int Dist0 = Lambda* AC*AC;
1107 :     uint32_t Best_Cost = 0xf0000000;
1108 :     Last_Cost += Dist0;
1109 :    
1110 :     if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */
1111 :     {
1112 :     int dQ;
1113 :     int Run;
1114 :     uint32_t Cost0;
1115 :    
1116 :     if (AC<0) {
1117 :     Nodes[i].Level = -1;
1118 :     dQ = Lev0 + AC;
1119 :     } else {
1120 :     Nodes[i].Level = 1;
1121 :     dQ = Lev0 - AC;
1122 :     }
1123 :     Cost0 = Lambda*dQ*dQ;
1124 : edgomez 1.21.2.17
1125 : edgomez 1.21.2.16 Nodes[i].Run = 1;
1126 :     Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
1127 :     for(Run=i-Run_Start; Run>0; --Run)
1128 :     {
1129 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
1130 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
1131 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
1132 :    
1133 :     /*
1134 :     * TODO: what about tie-breaks? Should we favor short runs or
1135 :     * long runs? Although the error is the same, it would not be
1136 :     * spread the same way along high and low frequencies...
1137 :     */
1138 :     if (Cost<Best_Cost) {
1139 :     Best_Cost = Cost;
1140 :     Nodes[i].Run = Run;
1141 :     }
1142 :    
1143 :     if (lCost<Last_Cost) {
1144 :     Last_Cost = lCost;
1145 :     Last.Run = Run;
1146 :     Last_Node = i;
1147 :     }
1148 :     }
1149 : edgomez 1.21.2.17 if (Last_Node==i)
1150 : edgomez 1.21.2.16 Last.Level = Nodes[i].Level;
1151 :    
1152 :     if (DBG==1) {
1153 :     Run_Costs[i] = Best_Cost;
1154 :     printf( "Costs #%2d: ", i);
1155 :     for(j=-1;j<=Non_Zero;++j) {
1156 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1157 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1158 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1159 :     else printf( " - |" );
1160 :     }
1161 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1162 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1163 :     printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 );
1164 :     printf( "\n" );
1165 :     }
1166 :     }
1167 :     else /* "big" levels */
1168 :     {
1169 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
1170 :     int Level2;
1171 :     int dQ1, dQ2;
1172 :     int Run;
1173 :     uint32_t Dist1,Dist2;
1174 :     int dDist21;
1175 : edgomez 1.21.2.17
1176 : edgomez 1.21.2.16 if (Level1>1) {
1177 :     dQ1 = Level1*Mult-AC + Bias;
1178 :     dQ2 = dQ1 - Mult;
1179 :     Level2 = Level1-1;
1180 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
1181 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
1182 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
1183 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
1184 :     } else { /* Level1<-1 */
1185 :     dQ1 = Level1*Mult-AC - Bias;
1186 :     dQ2 = dQ1 + Mult;
1187 :     Level2 = Level1 + 1;
1188 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
1189 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
1190 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
1191 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
1192 :     }
1193 :     Dist1 = Lambda*dQ1*dQ1;
1194 :     Dist2 = Lambda*dQ2*dQ2;
1195 :     dDist21 = Dist2-Dist1;
1196 :    
1197 :     for(Run=i-Run_Start; Run>0; --Run)
1198 :     {
1199 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
1200 :     uint32_t Cost1, Cost2;
1201 :     int bLevel;
1202 : chl 1.21.2.10
1203 : edgomez 1.21.2.13 /*
1204 :     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
1205 :     * if (Cost_Base>=Best_Cost) continue;
1206 :     */
1207 : edgomez 1.21.2.16 Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
1208 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
1209 : chl 1.21.2.10
1210 : edgomez 1.21.2.17 if (Cost2<Cost1) {
1211 :     Cost1 = Cost2;
1212 :     bLevel = Level2;
1213 :     } else
1214 : edgomez 1.21.2.16 bLevel = Level1;
1215 :    
1216 :     if (Cost1<Best_Cost) {
1217 :     Best_Cost = Cost1;
1218 :     Nodes[i].Run = Run;
1219 :     Nodes[i].Level = bLevel;
1220 :     }
1221 :    
1222 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
1223 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
1224 :    
1225 : edgomez 1.21.2.17 if (Cost2<Cost1) {
1226 :     Cost1 = Cost2;
1227 :     bLevel = Level2;
1228 :     } else
1229 : edgomez 1.21.2.16 bLevel = Level1;
1230 : edgomez 1.21.2.17
1231 : edgomez 1.21.2.16 if (Cost1<Last_Cost) {
1232 :     Last_Cost = Cost1;
1233 :     Last.Run = Run;
1234 :     Last.Level = bLevel;
1235 :     Last_Node = i;
1236 :     }
1237 :     } /* end of "for Run" */
1238 :    
1239 :     if (DBG==1) {
1240 :     Run_Costs[i] = Best_Cost;
1241 :     printf( "Costs #%2d: ", i);
1242 :     for(j=-1;j<=Non_Zero;++j) {
1243 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1244 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1245 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1246 :     else printf( " - |" );
1247 :     }
1248 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1249 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1250 :     printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 );
1251 :     printf( "\n" );
1252 :     }
1253 :     }
1254 : chl 1.21.2.11
1255 : edgomez 1.21.2.16 Run_Costs[i] = Best_Cost;
1256 : chl 1.21.2.10
1257 : edgomez 1.21.2.16 if (Best_Cost < Min_Cost + Dist0) {
1258 :     Min_Cost = Best_Cost;
1259 :     Run_Start = i;
1260 :     }
1261 :     else
1262 :     {
1263 :     /*
1264 :     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
1265 :     * a code shorter by 1 bit for a larger run (!), same level. We give
1266 :     * it a chance by not moving the left barrier too much.
1267 :     */
1268 :    
1269 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
1270 :     Run_Start++;
1271 :    
1272 :     /* spread on preceding coeffs the cost incurred by skipping this one */
1273 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1274 :     Min_Cost += Dist0;
1275 :     }
1276 :     }
1277 :    
1278 :     if (DBG) {
1279 :     Last_Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1280 :     if (DBG==1) {
1281 :     printf( "=> " );
1282 :     for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1283 :     printf( "\n" );
1284 :     }
1285 :     }
1286 :    
1287 :     if (Last_Node<0)
1288 :     return -1;
1289 :    
1290 :     /* reconstruct optimal sequence backward with surviving paths */
1291 :     memset(Out, 0x00, 64*sizeof(*Out));
1292 :     Out[Zigzag[Last_Node]] = Last.Level;
1293 :     i = Last_Node - Last.Run;
1294 :     while(i>=0) {
1295 :     Out[Zigzag[i]] = Nodes[i].Level;
1296 :     i -= Nodes[i].Run;
1297 :     }
1298 :    
1299 :     if (DBG) {
1300 :     uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1301 :     if (DBG==1) {
1302 : edgomez 1.21.2.17 printf( "<= " );
1303 : edgomez 1.21.2.16 for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1304 :     printf( "\n--------------------------------\n" );
1305 :     }
1306 :     if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost );
1307 :     }
1308 :     return Last_Node;
1309 : chl 1.21.2.10 }
1310 :    
1311 :     #undef DBG
1312 : chl 1.21.2.11
1313 :     #endif
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4