[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 1.21.2.17 - (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 :     #include "../quant/quant_mpeg4.h"
43 :     #include "../quant/quant_h263.h"
44 :     #include "../encoder.h"
45 :    
46 : edgomez 1.21 #include "../image/reduced.h"
47 : Isibaar 1.1
48 : edgomez 1.21 MBFIELDTEST_PTR MBFieldTest;
49 : Isibaar 1.1
50 : edgomez 1.21.2.6 /*
51 :     * Skip blocks having a coefficient sum below this value. This value will be
52 :     * corrected according to the MB quantizer to avoid artifacts for quant==1
53 :     */
54 :     #define PVOP_TOOSMALL_LIMIT 1
55 :     #define BVOP_TOOSMALL_LIMIT 3
56 :    
57 :     /*****************************************************************************
58 :     * Local functions
59 :     ****************************************************************************/
60 : Isibaar 1.1
61 : edgomez 1.21.2.6 /* permute block and return field dct choice */
62 :     static __inline uint32_t
63 :     MBDecideFieldDCT(int16_t data[6 * 64])
64 : Isibaar 1.1 {
65 : edgomez 1.21.2.6 uint32_t field = MBFieldTest(data);
66 : edgomez 1.3
67 : edgomez 1.21.2.6 if (field)
68 :     MBFrameToField(data);
69 : edgomez 1.21
70 : edgomez 1.21.2.6 return field;
71 :     }
72 : h 1.2
73 : edgomez 1.21.2.6 /* Performs Forward DCT on all blocks */
74 :     static __inline void
75 : syskin 1.21.2.8 MBfDCT(const MBParam * const pParam,
76 :     const FRAMEINFO * const frame,
77 :     MACROBLOCK * const pMB,
78 : edgomez 1.21.2.6 uint32_t x_pos,
79 :     uint32_t y_pos,
80 :     int16_t data[6 * 64])
81 : syskin 1.21.2.8 {
82 : edgomez 1.21.2.6 /* Handles interlacing */
83 : h 1.2 start_timer();
84 :     pMB->field_dct = 0;
85 : edgomez 1.21.2.5 if ((frame->vol_flags & XVID_VOL_INTERLACING) &&
86 : h 1.11 (x_pos>0) && (x_pos<pParam->mb_width-1) &&
87 :     (y_pos>0) && (y_pos<pParam->mb_height-1)) {
88 : h 1.2 pMB->field_dct = MBDecideFieldDCT(data);
89 :     }
90 :     stop_interlacing_timer();
91 :    
92 : edgomez 1.21.2.6 /* Perform DCT */
93 :     start_timer();
94 :     fdct(&data[0 * 64]);
95 :     fdct(&data[1 * 64]);
96 :     fdct(&data[2 * 64]);
97 :     fdct(&data[3 * 64]);
98 :     fdct(&data[4 * 64]);
99 :     fdct(&data[5 * 64]);
100 :     stop_dct_timer();
101 :     }
102 :    
103 :     /* Performs Inverse DCT on all blocks */
104 :     static __inline void
105 :     MBiDCT(int16_t data[6 * 64],
106 :     const uint8_t cbp)
107 :     {
108 :     start_timer();
109 :     if(cbp & (1 << (5 - 0))) idct(&data[0 * 64]);
110 :     if(cbp & (1 << (5 - 1))) idct(&data[1 * 64]);
111 :     if(cbp & (1 << (5 - 2))) idct(&data[2 * 64]);
112 :     if(cbp & (1 << (5 - 3))) idct(&data[3 * 64]);
113 :     if(cbp & (1 << (5 - 4))) idct(&data[4 * 64]);
114 :     if(cbp & (1 << (5 - 5))) idct(&data[5 * 64]);
115 :     stop_idct_timer();
116 :     }
117 :    
118 :     /* Quantize all blocks -- Intra mode */
119 :     static __inline void
120 :     MBQuantIntra(const MBParam * pParam,
121 : chl 1.21.2.9 const FRAMEINFO * const frame,
122 : edgomez 1.21.2.6 const MACROBLOCK * pMB,
123 : syskin 1.21.2.8 int16_t qcoeff[6 * 64],
124 : edgomez 1.21.2.6 int16_t data[6*64])
125 :     {
126 : edgomez 1.21.2.16 int mpeg;
127 :     int scaler_lum, scaler_chr;
128 : edgomez 1.21.2.6
129 : edgomez 1.21.2.16 quanth263_intraFuncPtr const quant[2] =
130 :     {
131 :     (quanth263_intraFuncPtr)quant_intra,
132 :     (quanth263_intraFuncPtr)quant4_intra
133 :     };
134 :    
135 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
136 :     scaler_lum = get_dc_scaler(pMB->quant, 1);
137 :     scaler_chr = get_dc_scaler(pMB->quant, 0);
138 : edgomez 1.21.2.6
139 : edgomez 1.21.2.16 /* Quantize the block */
140 :     start_timer();
141 :     quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum);
142 :     quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum);
143 :     quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum);
144 :     quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum);
145 :     quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr);
146 :     quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr);
147 :     stop_quant_timer();
148 : edgomez 1.21.2.6 }
149 :    
150 :     /* DeQuantize all blocks -- Intra mode */
151 :     static __inline void
152 :     MBDeQuantIntra(const MBParam * pParam,
153 :     const int iQuant,
154 :     int16_t qcoeff[6 * 64],
155 :     int16_t data[6*64])
156 :     {
157 : edgomez 1.21.2.16 int mpeg;
158 :     int scaler_lum, scaler_chr;
159 : edgomez 1.21.2.6
160 : edgomez 1.21.2.16 quanth263_intraFuncPtr const dequant[2] =
161 :     {
162 :     (quanth263_intraFuncPtr)dequant_intra,
163 :     (quanth263_intraFuncPtr)dequant4_intra
164 :     };
165 :    
166 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
167 :     scaler_lum = get_dc_scaler(iQuant, 1);
168 :     scaler_chr = get_dc_scaler(iQuant, 0);
169 : Isibaar 1.1
170 : edgomez 1.21.2.16 start_timer();
171 :     dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum);
172 :     dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum);
173 :     dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum);
174 :     dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum);
175 :     dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr);
176 :     dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr);
177 :     stop_iquant_timer();
178 : edgomez 1.21.2.6 }
179 : Isibaar 1.1
180 : chl 1.21.2.10
181 : edgomez 1.21.2.16 typedef int (*trellis_func_ptr_t)(int16_t *const Out,
182 :     const int16_t *const In,
183 :     int Q,
184 :     const uint16_t * const Zigzag,
185 :     int Non_Zero);
186 :    
187 : edgomez 1.21.2.17 static int
188 : edgomez 1.21.2.13 dct_quantize_trellis_h263_c(int16_t *const Out,
189 :     const int16_t *const In,
190 :     int Q,
191 :     const uint16_t * const Zigzag,
192 :     int Non_Zero);
193 : chl 1.21.2.10
194 : edgomez 1.21.2.17 static int
195 : edgomez 1.21.2.13 dct_quantize_trellis_mpeg_c(int16_t *const Out,
196 :     const int16_t *const In,
197 :     int Q,
198 :     const uint16_t * const Zigzag,
199 :     int Non_Zero);
200 : chl 1.21.2.10
201 : edgomez 1.21.2.6 /* Quantize all blocks -- Inter mode */
202 :     static __inline uint8_t
203 :     MBQuantInter(const MBParam * pParam,
204 : chl 1.21.2.9 const FRAMEINFO * const frame,
205 : edgomez 1.21.2.6 const MACROBLOCK * pMB,
206 :     int16_t data[6 * 64],
207 :     int16_t qcoeff[6 * 64],
208 :     int bvop,
209 :     int limit)
210 :     {
211 :    
212 :     int i;
213 :     uint8_t cbp = 0;
214 :     int sum;
215 : edgomez 1.21.2.16 int code_block, mpeg;
216 :    
217 :     quanth263_interFuncPtr const quant[2] =
218 :     {
219 :     (quanth263_interFuncPtr)quant_inter,
220 :     (quanth263_interFuncPtr)quant4_inter
221 :     };
222 :    
223 :     trellis_func_ptr_t const trellis[2] =
224 :     {
225 :     (trellis_func_ptr_t)dct_quantize_trellis_h263_c,
226 :     (trellis_func_ptr_t)dct_quantize_trellis_mpeg_c
227 :     };
228 :    
229 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
230 : edgomez 1.21.2.6
231 :     for (i = 0; i < 6; i++) {
232 : syskin 1.21.2.8
233 : edgomez 1.21.2.6 /* Quantize the block */
234 :     start_timer();
235 : edgomez 1.21.2.16
236 :     sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant);
237 :    
238 :     if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) {
239 :     sum = trellis[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63);
240 : chl 1.21.2.9 }
241 : edgomez 1.21.2.6 stop_quant_timer();
242 :    
243 :     /*
244 :     * We code the block if the sum is higher than the limit and if the first
245 :     * two AC coefficients in zig zag order are not zero.
246 :     */
247 :     code_block = 0;
248 :     if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) {
249 :     code_block = 1;
250 : edgomez 1.7 } else {
251 : edgomez 1.21
252 : edgomez 1.21.2.6 if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) {
253 :     /* dark blocks prevention for direct mode */
254 :     if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0))
255 :     code_block = 1;
256 : edgomez 1.21 } else {
257 : edgomez 1.21.2.6 /* not direct mode */
258 :     if (qcoeff[i*64] != 0)
259 :     code_block = 1;
260 : edgomez 1.21 }
261 :     }
262 : edgomez 1.21.2.6
263 :     /* Set the corresponding cbp bit */
264 :     cbp |= code_block << (5 - i);
265 : edgomez 1.21 }
266 :    
267 : edgomez 1.21.2.6 return(cbp);
268 :     }
269 : Isibaar 1.1
270 : edgomez 1.21.2.6 /* DeQuantize all blocks -- Inter mode */
271 : syskin 1.21.2.8 static __inline void
272 : edgomez 1.21.2.6 MBDeQuantInter(const MBParam * pParam,
273 :     const int iQuant,
274 :     int16_t data[6 * 64],
275 :     int16_t qcoeff[6 * 64],
276 :     const uint8_t cbp)
277 :     {
278 : edgomez 1.21.2.16 int mpeg;
279 : edgomez 1.21.2.6
280 : edgomez 1.21.2.16 quanth263_interFuncPtr const dequant[2] =
281 :     {
282 :     (quanth263_interFuncPtr)dequant_inter,
283 :     (quanth263_interFuncPtr)dequant4_inter
284 :     };
285 :    
286 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
287 :    
288 :     start_timer();
289 :     if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant);
290 :     if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant);
291 :     if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant);
292 :     if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant);
293 :     if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant);
294 :     if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant);
295 :     stop_iquant_timer();
296 : Isibaar 1.1 }
297 :    
298 : edgomez 1.21.2.6 typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);
299 :     typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);
300 : Isibaar 1.1
301 : edgomez 1.3
302 : edgomez 1.21.2.6 static __inline void
303 : syskin 1.21.2.8 MBTrans8to16(const MBParam * const pParam,
304 :     const FRAMEINFO * const frame,
305 :     const MACROBLOCK * const pMB,
306 : edgomez 1.21.2.6 const uint32_t x_pos,
307 :     const uint32_t y_pos,
308 :     int16_t data[6 * 64])
309 :     {
310 : h 1.4 uint32_t stride = pParam->edged_width;
311 :     uint32_t stride2 = stride / 2;
312 : edgomez 1.21.2.6 uint32_t next_block = stride * 8;
313 : syskin 1.21.2.8 int32_t cst;
314 : edgomez 1.21.2.16 int vop_reduced;
315 : Isibaar 1.1 uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
316 : syskin 1.21.2.8 const IMAGE * const pCurrent = &frame->image;
317 : edgomez 1.21.2.16 transfer_operation_8to16_t * const functions[2] =
318 :     {
319 :     (transfer_operation_8to16_t *)transfer_8to16copy,
320 :     (transfer_operation_8to16_t *)filter_18x18_to_8x8
321 :     };
322 : edgomez 1.21.2.6 transfer_operation_8to16_t *transfer_op = NULL;
323 : edgomez 1.7
324 : edgomez 1.21.2.16 vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED);
325 : edgomez 1.21.2.6
326 : edgomez 1.21.2.16 /* Image pointers */
327 :     pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride + (x_pos << (4+vop_reduced));
328 :     pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
329 :     pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
330 : edgomez 1.21.2.6
331 : edgomez 1.21.2.16 /* Block size */
332 :     cst = 8<<vop_reduced;
333 : edgomez 1.21.2.6
334 : edgomez 1.21.2.16 /* Operation function */
335 :     transfer_op = functions[vop_reduced];
336 : Isibaar 1.1
337 : edgomez 1.21.2.6 /* Do the transfer */
338 : h 1.2 start_timer();
339 : edgomez 1.21.2.6 transfer_op(&data[0 * 64], pY_Cur, stride);
340 :     transfer_op(&data[1 * 64], pY_Cur + cst, stride);
341 :     transfer_op(&data[2 * 64], pY_Cur + next_block, stride);
342 :     transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride);
343 :     transfer_op(&data[4 * 64], pU_Cur, stride2);
344 :     transfer_op(&data[5 * 64], pV_Cur, stride2);
345 :     stop_transfer_timer();
346 : syskin 1.21.2.8 }
347 : h 1.2
348 : edgomez 1.21.2.6 static __inline void
349 : syskin 1.21.2.8 MBTrans16to8(const MBParam * const pParam,
350 :     const FRAMEINFO * const frame,
351 :     const MACROBLOCK * const pMB,
352 : edgomez 1.21.2.6 const uint32_t x_pos,
353 :     const uint32_t y_pos,
354 :     int16_t data[6 * 64],
355 : edgomez 1.21.2.16 const uint32_t add, /* Must be 1 or 0 */
356 : edgomez 1.21.2.6 const uint8_t cbp)
357 :     {
358 :     uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
359 :     uint32_t stride = pParam->edged_width;
360 :     uint32_t stride2 = stride / 2;
361 :     uint32_t next_block = stride * 8;
362 : syskin 1.21.2.8 uint32_t cst;
363 : edgomez 1.21.2.16 int vop_reduced;
364 : syskin 1.21.2.8 const IMAGE * const pCurrent = &frame->image;
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 1.21.2.2 /*
453 : edgomez 1.21.2.6 * There is no MBTrans8to16 for Inter block, that's done in motion compensation
454 :     * already
455 : edgomez 1.21.2.2 */
456 :    
457 : edgomez 1.21.2.6 /* Perform DCT (and field decision) */
458 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
459 : chl 1.8
460 : edgomez 1.21.2.6 /* Set the limit threshold */
461 :     limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
462 : chl 1.8
463 : Isibaar 1.21.2.15 if (frame->vop_flags & XVID_VOP_CARTOON)
464 :     limit *= 3;
465 :    
466 : edgomez 1.21.2.6 /* Quantize the block */
467 : chl 1.21.2.9 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
468 : chl 1.8
469 : edgomez 1.21.2.6 /* DeQuantize the block */
470 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
471 : chl 1.8
472 : edgomez 1.21.2.6 /* Perform inverse DCT*/
473 :     MBiDCT(data, cbp);
474 : chl 1.8
475 : edgomez 1.21.2.6 /* Transfer back the data -- Add the data */
476 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
477 : syskin 1.21.2.8
478 : edgomez 1.21.2.6 return(cbp);
479 : chl 1.8 }
480 :    
481 : edgomez 1.21.2.6 uint8_t
482 :     MBTransQuantInterBVOP(const MBParam * pParam,
483 : edgomez 1.21.2.16 FRAMEINFO * frame,
484 :     MACROBLOCK * pMB,
485 :     const uint32_t x_pos,
486 :     const uint32_t y_pos,
487 :     int16_t data[6 * 64],
488 :     int16_t qcoeff[6 * 64])
489 : chl 1.8 {
490 : edgomez 1.21.2.6 uint8_t cbp;
491 :     uint32_t limit;
492 : syskin 1.21.2.8
493 : edgomez 1.21.2.6 /*
494 :     * There is no MBTrans8to16 for Inter block, that's done in motion compensation
495 :     * already
496 :     */
497 : chl 1.8
498 : edgomez 1.21.2.6 /* Perform DCT (and field decision) */
499 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
500 : chl 1.8
501 : edgomez 1.21.2.6 /* Set the limit threshold */
502 :     limit = BVOP_TOOSMALL_LIMIT;
503 : Isibaar 1.21.2.15
504 :     if (frame->vop_flags & XVID_VOP_CARTOON)
505 :     limit *= 2;
506 : chl 1.8
507 : edgomez 1.21.2.6 /* Quantize the block */
508 : chl 1.21.2.9 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
509 : h 1.2
510 : edgomez 1.21.2.6 /*
511 :     * History comment:
512 :     * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
513 :     *
514 :     * BUT some plugins require the original frame to be passed so we have
515 :     * to take care of that here
516 :     */
517 :     if((pParam->plugin_flags & XVID_REQORIGINAL)) {
518 : h 1.2
519 : edgomez 1.21.2.6 /* DeQuantize the block */
520 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
521 : Isibaar 1.1
522 : edgomez 1.21.2.6 /* Perform inverse DCT*/
523 :     MBiDCT(data, cbp);
524 : edgomez 1.21
525 : edgomez 1.21.2.6 /* Transfer back the data -- Add the data */
526 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
527 : edgomez 1.21 }
528 :    
529 : edgomez 1.21.2.6 return(cbp);
530 : edgomez 1.21 }
531 :    
532 :     /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */
533 :     uint32_t
534 :     MBFieldTest_c(int16_t data[6 * 64])
535 :     {
536 : edgomez 1.7 const uint8_t blocks[] =
537 :     { 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 };
538 :     const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 };
539 : h 1.2
540 :     int frame = 0, field = 0;
541 :     int i, j;
542 :    
543 : edgomez 1.7 for (i = 0; i < 7; ++i) {
544 :     for (j = 0; j < 8; ++j) {
545 :     frame +=
546 : edgomez 1.21.2.7 abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]);
547 : edgomez 1.7 frame +=
548 : edgomez 1.21.2.7 abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]);
549 : edgomez 1.7 frame +=
550 : edgomez 1.21.2.7 abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]);
551 : edgomez 1.7 frame +=
552 : edgomez 1.21.2.7 abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]);
553 : edgomez 1.7
554 :     field +=
555 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + lines[i + 1] + j] -
556 : edgomez 1.7 data[blocks[i] + lines[i] + j]);
557 :     field +=
558 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] -
559 : edgomez 1.7 data[blocks[i] + lines[i] + 8 + j]);
560 :     field +=
561 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] -
562 : edgomez 1.7 data[blocks[i] + 64 + lines[i] + j]);
563 :     field +=
564 : edgomez 1.21.2.7 abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] -
565 : edgomez 1.7 data[blocks[i] + 64 + lines[i] + 8 + j]);
566 : Isibaar 1.1 }
567 :     }
568 : h 1.2
569 : edgomez 1.21 return (frame >= (field + 350));
570 : h 1.2 }
571 :    
572 :    
573 :     /* deinterlace Y blocks vertically */
574 :    
575 :     #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp))
576 : syskin 1.21.2.8 #define LINE(X,Y) &data[X*64 + Y*8]
577 : h 1.2
578 : edgomez 1.7 void
579 :     MBFrameToField(int16_t data[6 * 64])
580 : h 1.2 {
581 :     int16_t tmp[8];
582 :    
583 :     /* left blocks */
584 :    
585 : edgomez 1.21.2.13 /* 1=2, 2=4, 4=8, 8=1 */
586 : edgomez 1.7 MOVLINE(tmp, LINE(0, 1));
587 :     MOVLINE(LINE(0, 1), LINE(0, 2));
588 :     MOVLINE(LINE(0, 2), LINE(0, 4));
589 :     MOVLINE(LINE(0, 4), LINE(2, 0));
590 :     MOVLINE(LINE(2, 0), tmp);
591 : h 1.2
592 : edgomez 1.21.2.13 /* 3=6, 6=12, 12=9, 9=3 */
593 : edgomez 1.7 MOVLINE(tmp, LINE(0, 3));
594 :     MOVLINE(LINE(0, 3), LINE(0, 6));
595 :     MOVLINE(LINE(0, 6), LINE(2, 4));
596 :     MOVLINE(LINE(2, 4), LINE(2, 1));
597 :     MOVLINE(LINE(2, 1), tmp);
598 : h 1.2
599 : edgomez 1.21.2.13 /* 5=10, 10=5 */
600 : edgomez 1.7 MOVLINE(tmp, LINE(0, 5));
601 :     MOVLINE(LINE(0, 5), LINE(2, 2));
602 :     MOVLINE(LINE(2, 2), tmp);
603 : h 1.2
604 : edgomez 1.21.2.13 /* 7=14, 14=13, 13=11, 11=7 */
605 : edgomez 1.7 MOVLINE(tmp, LINE(0, 7));
606 :     MOVLINE(LINE(0, 7), LINE(2, 6));
607 :     MOVLINE(LINE(2, 6), LINE(2, 5));
608 :     MOVLINE(LINE(2, 5), LINE(2, 3));
609 :     MOVLINE(LINE(2, 3), tmp);
610 : h 1.2
611 :     /* right blocks */
612 :    
613 : edgomez 1.21.2.13 /* 1=2, 2=4, 4=8, 8=1 */
614 : edgomez 1.7 MOVLINE(tmp, LINE(1, 1));
615 :     MOVLINE(LINE(1, 1), LINE(1, 2));
616 :     MOVLINE(LINE(1, 2), LINE(1, 4));
617 :     MOVLINE(LINE(1, 4), LINE(3, 0));
618 :     MOVLINE(LINE(3, 0), tmp);
619 : h 1.2
620 : edgomez 1.21.2.13 /* 3=6, 6=12, 12=9, 9=3 */
621 : edgomez 1.7 MOVLINE(tmp, LINE(1, 3));
622 :     MOVLINE(LINE(1, 3), LINE(1, 6));
623 :     MOVLINE(LINE(1, 6), LINE(3, 4));
624 :     MOVLINE(LINE(3, 4), LINE(3, 1));
625 :     MOVLINE(LINE(3, 1), tmp);
626 : h 1.2
627 : edgomez 1.21.2.13 /* 5=10, 10=5 */
628 : edgomez 1.7 MOVLINE(tmp, LINE(1, 5));
629 :     MOVLINE(LINE(1, 5), LINE(3, 2));
630 :     MOVLINE(LINE(3, 2), tmp);
631 : h 1.2
632 : edgomez 1.21.2.13 /* 7=14, 14=13, 13=11, 11=7 */
633 : edgomez 1.7 MOVLINE(tmp, LINE(1, 7));
634 :     MOVLINE(LINE(1, 7), LINE(3, 6));
635 :     MOVLINE(LINE(3, 6), LINE(3, 5));
636 :     MOVLINE(LINE(3, 5), LINE(3, 3));
637 :     MOVLINE(LINE(3, 3), tmp);
638 : Isibaar 1.1 }
639 : chl 1.21.2.10
640 :    
641 :    
642 :    
643 :    
644 : edgomez 1.21.2.13 /*****************************************************************************
645 :     * Trellis based R-D optimal quantization
646 :     *
647 :     * Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
648 :     *
649 :     ****************************************************************************/
650 : chl 1.21.2.10
651 :    
652 : edgomez 1.21.2.13 #if 0
653 : edgomez 1.21.2.17 static int
654 : edgomez 1.21.2.13 dct_quantize_trellis_mpeg_c(int16_t *const Out,
655 :     const int16_t *const In,
656 : edgomez 1.21.2.17 int Q,
657 : edgomez 1.21.2.13 const uint16_t * const Zigzag,
658 :     int Non_Zero)
659 :     {
660 :     return 63;
661 :     }
662 :     #endif
663 :    
664 :     /*----------------------------------------------------------------------------
665 :     *
666 :     * Trellis-Based quantization
667 :     *
668 :     * So far I understand this paper:
669 :     *
670 :     * "Trellis-Based R-D Optimal Quantization in H.263+"
671 :     * J.Wen, M.Luttrell, J.Villasenor
672 :     * IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
673 :     *
674 :     * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
675 :     * Source Shorted Path algo. But due to the underlying graph structure
676 :     * ("Trellis"), it can be turned into a dynamic programming algo,
677 : edgomez 1.21.2.17 * partially saving the explicit graph's nodes representation. And
678 : edgomez 1.21.2.13 * without using a heap, since the open frontier of the DAG is always
679 :     * known, and of fixed sized.
680 :     *--------------------------------------------------------------------------*/
681 :    
682 : chl 1.21.2.10
683 :    
684 : edgomez 1.21.2.13 /* Codes lengths for relevant levels. */
685 : chl 1.21.2.10
686 : edgomez 1.21.2.16 /* let's factorize: */
687 : chl 1.21.2.10 static const uint8_t Code_Len0[64] = {
688 : 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,
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_Len1[64] = {
691 : 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,
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_Len2[64] = {
694 : 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,
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_Len3[64] = {
697 : 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,
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_Len4[64] = {
700 : 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,
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_Len5[64] = {
703 : 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,
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_Len6[64] = {
706 : 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,
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_Len7[64] = {
709 : 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,
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_Len8[64] = {
712 : 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,
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_Len9[64] = {
715 : 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,
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_Len10[64] = {
718 : 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,
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_Len11[64] = {
721 : 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,
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_Len12[64] = {
724 : 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,
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_Len13[64] = {
727 : 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,
728 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
729 : chl 1.21.2.10 static const uint8_t Code_Len14[64] = {
730 : 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,
731 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
732 : chl 1.21.2.10 static const uint8_t Code_Len15[64] = {
733 : 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,
734 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
735 : chl 1.21.2.10 static const uint8_t Code_Len16[64] = {
736 : 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,
737 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
738 : chl 1.21.2.10 static const uint8_t Code_Len17[64] = {
739 : 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,
740 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
741 : chl 1.21.2.10 static const uint8_t Code_Len18[64] = {
742 : 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,
743 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
744 : chl 1.21.2.10 static const uint8_t Code_Len19[64] = {
745 : 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,
746 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
747 : chl 1.21.2.10 static const uint8_t Code_Len20[64] = {
748 : 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,
749 :     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 };
750 : chl 1.21.2.10
751 : edgomez 1.21.2.16 /* a few more table for LAST table: */
752 : chl 1.21.2.10 static const uint8_t Code_Len21[64] = {
753 : 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,
754 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
755 : chl 1.21.2.10 static const uint8_t Code_Len22[64] = {
756 : 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,
757 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
758 : chl 1.21.2.10 static const uint8_t Code_Len23[64] = {
759 : 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,
760 :     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};
761 : chl 1.21.2.10 static const uint8_t Code_Len24[64] = {
762 : 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,
763 :     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};
764 : chl 1.21.2.10
765 :    
766 : edgomez 1.21.2.13 static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */
767 : edgomez 1.21.2.16 Code_Len20,Code_Len19,Code_Len18,Code_Len17,
768 :     Code_Len16,Code_Len15,Code_Len14,Code_Len13,
769 :     Code_Len12,Code_Len11,Code_Len10,Code_Len9,
770 :     Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5,
771 :     Code_Len4, Code_Len3, Code_Len3 ,Code_Len2,
772 :     Code_Len2, Code_Len1, Code_Len1, Code_Len1,
773 : chl 1.21.2.10 };
774 :    
775 : edgomez 1.21.2.13 static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */
776 : edgomez 1.21.2.17 Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
777 : chl 1.21.2.10 };
778 :    
779 :     #define TL(q) 0xfe00/(q*q)
780 :    
781 :     static const int Trellis_Lambda_Tabs[31] = {
782 : edgomez 1.21.2.16 TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7),
783 :     TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15),
784 :     TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23),
785 :     TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31)
786 : chl 1.21.2.10 };
787 :     #undef TL
788 :    
789 : edgomez 1.21.2.16 static int __inline
790 :     Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
791 : chl 1.21.2.10 {
792 : edgomez 1.21.2.16 while(i>=0)
793 :     if (C[Zigzag[i]])
794 :     return i;
795 :     else i--;
796 :     return -1;
797 : chl 1.21.2.10 }
798 :    
799 : edgomez 1.21.2.16 static int __inline
800 :     Compute_Sum(const int16_t *C, int last)
801 :     {
802 :     int sum = 0;
803 :    
804 :     while(last--)
805 :     sum += abs(C[last]);
806 :    
807 :     return(sum);
808 :     }
809 : edgomez 1.21.2.13 /* this routine has been strippen of all debug code */
810 : chl 1.21.2.11
811 : edgomez 1.21.2.17 static int
812 : 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)
813 :     {
814 :    
815 : edgomez 1.21.2.13 /*
816 :     * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
817 :     * not quantized one (Out[]). However, it only improves the result *very*
818 :     * slightly (~0.01dB), whereas speed drops to crawling level :)
819 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
820 :     */
821 : edgomez 1.21.2.16 typedef struct { int16_t Run, Level; } NODE;
822 : edgomez 1.21.2.17
823 : edgomez 1.21.2.16 NODE Nodes[65], Last;
824 :     uint32_t Run_Costs0[64+1];
825 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
826 :     const int Mult = 2*Q;
827 :     const int Bias = (Q-1) | 1;
828 :     const int Lev0 = Mult + Bias;
829 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */
830 :    
831 :     int Run_Start = -1;
832 :     uint32_t Min_Cost = 2<<16;
833 :    
834 :     int Last_Node = -1;
835 :     uint32_t Last_Cost = 0;
836 :    
837 :     int i, j, sum;
838 :     Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */
839 :    
840 :     Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
841 :     if (Non_Zero<0)
842 :     return 0; /* Sum is zero if there are only zero coeffs */
843 :    
844 :     for(i=0; i<=Non_Zero; i++) {
845 :     const int AC = In[Zigzag[i]];
846 :     const int Level1 = Out[Zigzag[i]];
847 :     const int Dist0 = Lambda* AC*AC;
848 :     uint32_t Best_Cost = 0xf0000000;
849 :     Last_Cost += Dist0;
850 :    
851 :     /* very specialized loop for -1,0,+1 */
852 :     if ((uint32_t)(Level1+1)<3) {
853 :     int dQ;
854 :     int Run;
855 :     uint32_t Cost0;
856 :    
857 :     if (AC<0) {
858 :     Nodes[i].Level = -1;
859 :     dQ = Lev0 + AC;
860 :     } else {
861 :     Nodes[i].Level = 1;
862 :     dQ = Lev0 - AC;
863 :     }
864 :     Cost0 = Lambda*dQ*dQ;
865 : edgomez 1.21.2.17
866 : edgomez 1.21.2.16 Nodes[i].Run = 1;
867 :     Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
868 :     for(Run=i-Run_Start; Run>0; --Run) {
869 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
870 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
871 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
872 :    
873 :     /*
874 :     * TODO: what about tie-breaks? Should we favor short runs or
875 :     * long runs? Although the error is the same, it would not be
876 :     * spread the same way along high and low frequencies...
877 :     */
878 :    
879 :     /* (I'd say: favour short runs => hifreq errors (HVS) -- gruel ) */
880 :    
881 :     if (Cost<Best_Cost) {
882 :     Best_Cost = Cost;
883 :     Nodes[i].Run = Run;
884 :     }
885 :    
886 :     if (lCost<Last_Cost) {
887 :     Last_Cost = lCost;
888 :     Last.Run = Run;
889 :     Last_Node = i;
890 :     }
891 :     }
892 : edgomez 1.21.2.17 if (Last_Node==i)
893 : edgomez 1.21.2.16 Last.Level = Nodes[i].Level;
894 :     } else { /* "big" levels */
895 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
896 :     int Level2;
897 :     int dQ1, dQ2;
898 :     int Run;
899 :     uint32_t Dist1,Dist2;
900 :     int dDist21;
901 : edgomez 1.21.2.17
902 : edgomez 1.21.2.16 if (Level1>1) {
903 :     dQ1 = Level1*Mult-AC + Bias;
904 :     dQ2 = dQ1 - Mult;
905 :     Level2 = Level1-1;
906 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
907 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
908 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
909 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
910 :     } else { /* Level1<-1 */
911 :     dQ1 = Level1*Mult-AC - Bias;
912 :     dQ2 = dQ1 + Mult;
913 :     Level2 = Level1 + 1;
914 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
915 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
916 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
917 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
918 :     }
919 : chl 1.21.2.11
920 : edgomez 1.21.2.16 Dist1 = Lambda*dQ1*dQ1;
921 :     Dist2 = Lambda*dQ2*dQ2;
922 :     dDist21 = Dist2-Dist1;
923 :    
924 :     for(Run=i-Run_Start; Run>0; --Run)
925 :     {
926 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
927 :     uint32_t Cost1, Cost2;
928 :     int bLevel;
929 :    
930 :     /*
931 :     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
932 :     * if (Cost_Base>=Best_Cost) continue;
933 :     * (? doesn't seem to have any effect -- gruel )
934 :     */
935 :    
936 :     Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
937 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
938 :    
939 : edgomez 1.21.2.17 if (Cost2<Cost1) {
940 :     Cost1 = Cost2;
941 :     bLevel = Level2;
942 : edgomez 1.21.2.16 } else {
943 :     bLevel = Level1;
944 :     }
945 :    
946 :     if (Cost1<Best_Cost) {
947 :     Best_Cost = Cost1;
948 :     Nodes[i].Run = Run;
949 :     Nodes[i].Level = bLevel;
950 :     }
951 :    
952 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
953 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
954 :    
955 : edgomez 1.21.2.17 if (Cost2<Cost1) {
956 :     Cost1 = Cost2;
957 :     bLevel = Level2;
958 : edgomez 1.21.2.16 } else {
959 :     bLevel = Level1;
960 :     }
961 : edgomez 1.21.2.17
962 : edgomez 1.21.2.16 if (Cost1<Last_Cost) {
963 :     Last_Cost = Cost1;
964 :     Last.Run = Run;
965 :     Last.Level = bLevel;
966 :     Last_Node = i;
967 :     }
968 :     } /* end of "for Run" */
969 : chl 1.21.2.11
970 : edgomez 1.21.2.16 }
971 : chl 1.21.2.11
972 : edgomez 1.21.2.16 Run_Costs[i] = Best_Cost;
973 : chl 1.21.2.11
974 : edgomez 1.21.2.16 if (Best_Cost < Min_Cost + Dist0) {
975 :     Min_Cost = Best_Cost;
976 :     Run_Start = i;
977 :     } else {
978 :     /*
979 :     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
980 :     * a code shorter by 1 bit for a larger run (!), same level. We give
981 :     * it a chance by not moving the left barrier too much.
982 :     */
983 :    
984 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
985 :     Run_Start++;
986 :    
987 :     /* spread on preceding coeffs the cost incurred by skipping this one */
988 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
989 :     Min_Cost += Dist0;
990 :     }
991 :     }
992 : chl 1.21.2.11
993 : edgomez 1.21.2.16 /* It seems trellis doesn't give good results... just compute the Out sum and
994 :     * quit (even if we did not modify it, upperlayer relies on this data) */
995 :     if (Last_Node<0)
996 :     return Compute_Sum(Out, Non_Zero);
997 :    
998 :     /* reconstruct optimal sequence backward with surviving paths */
999 :     memset(Out, 0x00, 64*sizeof(*Out));
1000 :     Out[Zigzag[Last_Node]] = Last.Level;
1001 :     i = Last_Node - Last.Run;
1002 :     sum = 0;
1003 :     while(i>=0) {
1004 :     Out[Zigzag[i]] = Nodes[i].Level;
1005 :     sum += abs(Nodes[i].Level);
1006 :     i -= Nodes[i].Run;
1007 :     }
1008 : chl 1.21.2.11
1009 : edgomez 1.21.2.16 return sum;
1010 :     }
1011 : chl 1.21.2.11
1012 : edgomez 1.21.2.17 static int
1013 : 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)
1014 :     {
1015 :     /* ToDo: Ok ok it's just a place holder for Gruel -- damn write this one :-) */
1016 :     return Compute_Sum(Out, 63);
1017 :     }
1018 : chl 1.21.2.11
1019 : edgomez 1.21.2.13 /* original version including heavy debugging info */
1020 : chl 1.21.2.11
1021 :     #ifdef DBGTRELL
1022 : chl 1.21.2.10
1023 :     #define DBG 0
1024 :    
1025 : suxen_drol 1.21.2.12 static __inline uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,
1026 : edgomez 1.21.2.16 const uint16_t * Zigzag, int Max, int Lambda)
1027 : chl 1.21.2.10 {
1028 :     #if (DBG>0)
1029 : edgomez 1.21.2.16 const int16_t * const Ref = C + 6*64;
1030 :     int Last = Max;
1031 :     int Bits = 0;
1032 :     int Dist = 0;
1033 : edgomez 1.21.2.17 int i;
1034 : edgomez 1.21.2.16 uint32_t Cost;
1035 : edgomez 1.21.2.17
1036 :     while(Last>=0 && C[Zigzag[Last]]==0)
1037 : edgomez 1.21.2.16 Last--;
1038 : edgomez 1.21.2.17
1039 : edgomez 1.21.2.16 if (Last>=0) {
1040 :     int j=0, j0=0;
1041 :     int Run, Level;
1042 :    
1043 :     Bits = 2; /* CBP */
1044 :     while(j<Last) {
1045 : edgomez 1.21.2.17 while(!C[Zigzag[j]])
1046 : edgomez 1.21.2.16 j++;
1047 : edgomez 1.21.2.17 if (j==Last)
1048 : edgomez 1.21.2.16 break;
1049 :     Level=C[Zigzag[j]];
1050 :     Run = j - j0;
1051 :     j0 = ++j;
1052 : edgomez 1.21.2.17 if (Level>=-24 && Level<=24)
1053 : edgomez 1.21.2.16 Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];
1054 : edgomez 1.21.2.17 else
1055 : edgomez 1.21.2.16 Bits += 30;
1056 :     }
1057 :     Level = C[Zigzag[Last]];
1058 :     Run = j - j0;
1059 : edgomez 1.21.2.17 if (Level>=-6 && Level<=6)
1060 : edgomez 1.21.2.16 Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];
1061 : edgomez 1.21.2.17 else
1062 : chl 1.21.2.11 Bits += 30;
1063 : edgomez 1.21.2.16 }
1064 :    
1065 :     for(i=0; i<=Last; ++i) {
1066 :     int V = C[Zigzag[i]]*Mult;
1067 : edgomez 1.21.2.17 if (V>0)
1068 : edgomez 1.21.2.16 V += Bias;
1069 : edgomez 1.21.2.17 else
1070 :     if (V<0)
1071 : edgomez 1.21.2.16 V -= Bias;
1072 :     V -= Ref[Zigzag[i]];
1073 :     Dist += V*V;
1074 :     }
1075 :     Cost = Lambda*Dist + (Bits<<16);
1076 :     if (DBG==1)
1077 :     printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );
1078 :     return Cost;
1079 : chl 1.21.2.10
1080 :     #else
1081 : edgomez 1.21.2.16 return 0;
1082 : chl 1.21.2.10 #endif
1083 :     }
1084 :    
1085 :    
1086 : edgomez 1.21.2.17 static int
1087 : 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)
1088 :     {
1089 :    
1090 : edgomez 1.21.2.13 /*
1091 :     * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
1092 :     * not quantized one (Out[]). However, it only improves the result *very*
1093 :     * slightly (~0.01dB), whereas speed drops to crawling level :)
1094 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
1095 :     */
1096 : edgomez 1.21.2.16 typedef struct { int16_t Run, Level; } NODE;
1097 : edgomez 1.21.2.17
1098 : edgomez 1.21.2.16 NODE Nodes[65], Last;
1099 :     uint32_t Run_Costs0[64+1];
1100 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
1101 :     const int Mult = 2*Q;
1102 :     const int Bias = (Q-1) | 1;
1103 :     const int Lev0 = Mult + Bias;
1104 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */
1105 :    
1106 :     int Run_Start = -1;
1107 :     Run_Costs[-1] = 2<<16; /* source (w/ CBP penalty) */
1108 :     uint32_t Min_Cost = 2<<16;
1109 : chl 1.21.2.10
1110 : edgomez 1.21.2.16 int Last_Node = -1;
1111 :     uint32_t Last_Cost = 0;
1112 : chl 1.21.2.10
1113 : edgomez 1.21.2.16 int i, j;
1114 : chl 1.21.2.11
1115 : chl 1.21.2.10 #if (DBG>0)
1116 : edgomez 1.21.2.16 Last.Level = 0; Last.Run = -1; /* just initialize to smthg */
1117 : chl 1.21.2.10 #endif
1118 :    
1119 : edgomez 1.21.2.16 Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
1120 :     if (Non_Zero<0)
1121 : edgomez 1.21.2.17 return -1;
1122 : edgomez 1.21.2.16
1123 :     for(i=0; i<=Non_Zero; i++)
1124 :     {
1125 :     const int AC = In[Zigzag[i]];
1126 :     const int Level1 = Out[Zigzag[i]];
1127 :     const int Dist0 = Lambda* AC*AC;
1128 :     uint32_t Best_Cost = 0xf0000000;
1129 :     Last_Cost += Dist0;
1130 :    
1131 :     if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */
1132 :     {
1133 :     int dQ;
1134 :     int Run;
1135 :     uint32_t Cost0;
1136 :    
1137 :     if (AC<0) {
1138 :     Nodes[i].Level = -1;
1139 :     dQ = Lev0 + AC;
1140 :     } else {
1141 :     Nodes[i].Level = 1;
1142 :     dQ = Lev0 - AC;
1143 :     }
1144 :     Cost0 = Lambda*dQ*dQ;
1145 : edgomez 1.21.2.17
1146 : edgomez 1.21.2.16 Nodes[i].Run = 1;
1147 :     Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
1148 :     for(Run=i-Run_Start; Run>0; --Run)
1149 :     {
1150 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
1151 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
1152 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
1153 :    
1154 :     /*
1155 :     * TODO: what about tie-breaks? Should we favor short runs or
1156 :     * long runs? Although the error is the same, it would not be
1157 :     * spread the same way along high and low frequencies...
1158 :     */
1159 :     if (Cost<Best_Cost) {
1160 :     Best_Cost = Cost;
1161 :     Nodes[i].Run = Run;
1162 :     }
1163 :    
1164 :     if (lCost<Last_Cost) {
1165 :     Last_Cost = lCost;
1166 :     Last.Run = Run;
1167 :     Last_Node = i;
1168 :     }
1169 :     }
1170 : edgomez 1.21.2.17 if (Last_Node==i)
1171 : edgomez 1.21.2.16 Last.Level = Nodes[i].Level;
1172 :    
1173 :     if (DBG==1) {
1174 :     Run_Costs[i] = Best_Cost;
1175 :     printf( "Costs #%2d: ", i);
1176 :     for(j=-1;j<=Non_Zero;++j) {
1177 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1178 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1179 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1180 :     else printf( " - |" );
1181 :     }
1182 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1183 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1184 :     printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 );
1185 :     printf( "\n" );
1186 :     }
1187 :     }
1188 :     else /* "big" levels */
1189 :     {
1190 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
1191 :     int Level2;
1192 :     int dQ1, dQ2;
1193 :     int Run;
1194 :     uint32_t Dist1,Dist2;
1195 :     int dDist21;
1196 : edgomez 1.21.2.17
1197 : edgomez 1.21.2.16 if (Level1>1) {
1198 :     dQ1 = Level1*Mult-AC + Bias;
1199 :     dQ2 = dQ1 - Mult;
1200 :     Level2 = Level1-1;
1201 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
1202 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
1203 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
1204 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
1205 :     } else { /* Level1<-1 */
1206 :     dQ1 = Level1*Mult-AC - Bias;
1207 :     dQ2 = dQ1 + Mult;
1208 :     Level2 = Level1 + 1;
1209 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
1210 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
1211 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
1212 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
1213 :     }
1214 :     Dist1 = Lambda*dQ1*dQ1;
1215 :     Dist2 = Lambda*dQ2*dQ2;
1216 :     dDist21 = Dist2-Dist1;
1217 :    
1218 :     for(Run=i-Run_Start; Run>0; --Run)
1219 :     {
1220 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
1221 :     uint32_t Cost1, Cost2;
1222 :     int bLevel;
1223 : chl 1.21.2.10
1224 : edgomez 1.21.2.13 /*
1225 :     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
1226 :     * if (Cost_Base>=Best_Cost) continue;
1227 :     */
1228 : edgomez 1.21.2.16 Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
1229 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
1230 : chl 1.21.2.10
1231 : edgomez 1.21.2.17 if (Cost2<Cost1) {
1232 :     Cost1 = Cost2;
1233 :     bLevel = Level2;
1234 :     } else
1235 : edgomez 1.21.2.16 bLevel = Level1;
1236 :    
1237 :     if (Cost1<Best_Cost) {
1238 :     Best_Cost = Cost1;
1239 :     Nodes[i].Run = Run;
1240 :     Nodes[i].Level = bLevel;
1241 :     }
1242 :    
1243 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
1244 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
1245 :    
1246 : edgomez 1.21.2.17 if (Cost2<Cost1) {
1247 :     Cost1 = Cost2;
1248 :     bLevel = Level2;
1249 :     } else
1250 : edgomez 1.21.2.16 bLevel = Level1;
1251 : edgomez 1.21.2.17
1252 : edgomez 1.21.2.16 if (Cost1<Last_Cost) {
1253 :     Last_Cost = Cost1;
1254 :     Last.Run = Run;
1255 :     Last.Level = bLevel;
1256 :     Last_Node = i;
1257 :     }
1258 :     } /* end of "for Run" */
1259 :    
1260 :     if (DBG==1) {
1261 :     Run_Costs[i] = Best_Cost;
1262 :     printf( "Costs #%2d: ", i);
1263 :     for(j=-1;j<=Non_Zero;++j) {
1264 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1265 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1266 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1267 :     else printf( " - |" );
1268 :     }
1269 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1270 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1271 :     printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 );
1272 :     printf( "\n" );
1273 :     }
1274 :     }
1275 : chl 1.21.2.11
1276 : edgomez 1.21.2.16 Run_Costs[i] = Best_Cost;
1277 : chl 1.21.2.10
1278 : edgomez 1.21.2.16 if (Best_Cost < Min_Cost + Dist0) {
1279 :     Min_Cost = Best_Cost;
1280 :     Run_Start = i;
1281 :     }
1282 :     else
1283 :     {
1284 :     /*
1285 :     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
1286 :     * a code shorter by 1 bit for a larger run (!), same level. We give
1287 :     * it a chance by not moving the left barrier too much.
1288 :     */
1289 :    
1290 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
1291 :     Run_Start++;
1292 :    
1293 :     /* spread on preceding coeffs the cost incurred by skipping this one */
1294 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1295 :     Min_Cost += Dist0;
1296 :     }
1297 :     }
1298 :    
1299 :     if (DBG) {
1300 :     Last_Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1301 :     if (DBG==1) {
1302 :     printf( "=> " );
1303 :     for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1304 :     printf( "\n" );
1305 :     }
1306 :     }
1307 :    
1308 :     if (Last_Node<0)
1309 :     return -1;
1310 :    
1311 :     /* reconstruct optimal sequence backward with surviving paths */
1312 :     memset(Out, 0x00, 64*sizeof(*Out));
1313 :     Out[Zigzag[Last_Node]] = Last.Level;
1314 :     i = Last_Node - Last.Run;
1315 :     while(i>=0) {
1316 :     Out[Zigzag[i]] = Nodes[i].Level;
1317 :     i -= Nodes[i].Run;
1318 :     }
1319 :    
1320 :     if (DBG) {
1321 :     uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1322 :     if (DBG==1) {
1323 : edgomez 1.21.2.17 printf( "<= " );
1324 : edgomez 1.21.2.16 for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1325 :     printf( "\n--------------------------------\n" );
1326 :     }
1327 :     if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost );
1328 :     }
1329 :     return Last_Node;
1330 : chl 1.21.2.10 }
1331 :    
1332 :     #undef DBG
1333 : chl 1.21.2.11
1334 :     #endif

No admin address has been configured
ViewVC Help
Powered by ViewVC 1.0.4