[svn] / branches / dev-api-4 / xvidcore / src / utils / mbtransquant.c Repository:
ViewVC logotype

Annotation of /branches/dev-api-4/xvidcore/src/utils/mbtransquant.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1011 - (view) (download)

1 : edgomez 965 /*****************************************************************************
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 : chl 1011 * $Id: mbtransquant.c,v 1.21.2.10 2003-05-09 22:03:13 chl Exp $
25 : edgomez 965 *
26 :     ****************************************************************************/
27 : Isibaar 3
28 : edgomez 78 #include <string.h>
29 : edgomez 982 #include <stdlib.h>
30 : edgomez 78
31 : Isibaar 3 #include "../portab.h"
32 :     #include "mbfunctions.h"
33 :    
34 :     #include "../global.h"
35 :     #include "mem_transfer.h"
36 :     #include "timer.h"
37 : chl 995 #include "../bitstream/mbcoding.h"
38 : chl 1011 #include "../bitstream/zigzag.h"
39 : Isibaar 3 #include "../dct/fdct.h"
40 :     #include "../dct/idct.h"
41 :     #include "../quant/quant_mpeg4.h"
42 :     #include "../quant/quant_h263.h"
43 :     #include "../encoder.h"
44 :    
45 : edgomez 851 #include "../image/reduced.h"
46 : Isibaar 3
47 : edgomez 851 MBFIELDTEST_PTR MBFieldTest;
48 : Isibaar 3
49 : edgomez 965 /*
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 : Isibaar 3
56 : edgomez 965 /*****************************************************************************
57 :     * Local functions
58 :     ****************************************************************************/
59 :    
60 :     /* permute block and return field dct choice */
61 :     static __inline uint32_t
62 :     MBDecideFieldDCT(int16_t data[6 * 64])
63 : Isibaar 3 {
64 : edgomez 965 uint32_t field = MBFieldTest(data);
65 : edgomez 78
66 : edgomez 965 if (field)
67 :     MBFrameToField(data);
68 : Isibaar 3
69 : edgomez 965 return field;
70 :     }
71 : h 69
72 : edgomez 965 /* Performs Forward DCT on all blocks */
73 :     static __inline void
74 : syskin 984 MBfDCT(const MBParam * const pParam,
75 :     const FRAMEINFO * const frame,
76 :     MACROBLOCK * const pMB,
77 : edgomez 965 uint32_t x_pos,
78 :     uint32_t y_pos,
79 :     int16_t data[6 * 64])
80 : syskin 984 {
81 : edgomez 965 /* Handles interlacing */
82 : h 69 start_timer();
83 :     pMB->field_dct = 0;
84 : edgomez 949 if ((frame->vol_flags & XVID_VOL_INTERLACING) &&
85 : h 390 (x_pos>0) && (x_pos<pParam->mb_width-1) &&
86 :     (y_pos>0) && (y_pos<pParam->mb_height-1)) {
87 : h 69 pMB->field_dct = MBDecideFieldDCT(data);
88 :     }
89 :     stop_interlacing_timer();
90 :    
91 : edgomez 965 /* 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 995 const FRAMEINFO * const frame,
121 : edgomez 965 const MACROBLOCK * pMB,
122 : syskin 984 int16_t qcoeff[6 * 64],
123 : edgomez 965 int16_t data[6*64])
124 :     {
125 :     int i;
126 :    
127 : edgomez 195 for (i = 0; i < 6; i++) {
128 : edgomez 965 uint32_t iDcScaler = get_dc_scaler(pMB->quant, i < 4);
129 :    
130 :     /* Quantize the block */
131 :     start_timer();
132 : chl 995 if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) {
133 : edgomez 965 quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler);
134 : chl 995 } else {
135 : edgomez 965 quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler);
136 : chl 995 }
137 : edgomez 965 stop_quant_timer();
138 :     }
139 :     }
140 :    
141 :     /* DeQuantize all blocks -- Intra mode */
142 :     static __inline void
143 :     MBDeQuantIntra(const MBParam * pParam,
144 :     const int iQuant,
145 :     int16_t qcoeff[6 * 64],
146 :     int16_t data[6*64])
147 :     {
148 :     int i;
149 :    
150 :     for (i = 0; i < 6; i++) {
151 : Isibaar 3 uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4);
152 :    
153 :     start_timer();
154 : edgomez 965 if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT))
155 :     dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler);
156 :     else
157 :     dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler);
158 :     stop_iquant_timer();
159 :     }
160 :     }
161 : Isibaar 3
162 : chl 1011
163 :     static int
164 :     dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero);
165 :    
166 :     static int
167 :     dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero);
168 :    
169 :    
170 : edgomez 965 /* Quantize all blocks -- Inter mode */
171 :     static __inline uint8_t
172 :     MBQuantInter(const MBParam * pParam,
173 : chl 995 const FRAMEINFO * const frame,
174 : edgomez 965 const MACROBLOCK * pMB,
175 :     int16_t data[6 * 64],
176 :     int16_t qcoeff[6 * 64],
177 :     int bvop,
178 :     int limit)
179 :     {
180 :    
181 :     int i;
182 :     uint8_t cbp = 0;
183 :     int sum;
184 :     int code_block;
185 :    
186 :     for (i = 0; i < 6; i++) {
187 : syskin 984
188 : edgomez 965 /* Quantize the block */
189 :     start_timer();
190 : chl 995 if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) {
191 :     sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant);
192 :     if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) {
193 : chl 1011 sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1;
194 : chl 995 limit = 1;
195 :     }
196 :     } else {
197 : edgomez 965 sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant);
198 : chl 995 // if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) )
199 : chl 1011 // sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1;
200 : chl 995 }
201 : edgomez 965 stop_quant_timer();
202 :    
203 :     /*
204 :     * We code the block if the sum is higher than the limit and if the first
205 :     * two AC coefficients in zig zag order are not zero.
206 :     */
207 :     code_block = 0;
208 :     if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) {
209 :     code_block = 1;
210 : edgomez 195 } else {
211 : Isibaar 3
212 : edgomez 965 if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) {
213 :     /* dark blocks prevention for direct mode */
214 :     if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0))
215 :     code_block = 1;
216 : edgomez 851 } else {
217 : edgomez 965 /* not direct mode */
218 :     if (qcoeff[i*64] != 0)
219 :     code_block = 1;
220 : edgomez 851 }
221 : Isibaar 3 }
222 :    
223 : edgomez 965 /* Set the corresponding cbp bit */
224 :     cbp |= code_block << (5 - i);
225 :     }
226 : edgomez 851
227 : edgomez 965 return(cbp);
228 :     }
229 : Isibaar 3
230 : edgomez 965 /* DeQuantize all blocks -- Inter mode */
231 : syskin 984 static __inline void
232 : edgomez 965 MBDeQuantInter(const MBParam * pParam,
233 :     const int iQuant,
234 :     int16_t data[6 * 64],
235 :     int16_t qcoeff[6 * 64],
236 :     const uint8_t cbp)
237 :     {
238 :     int i;
239 :    
240 :     for (i = 0; i < 6; i++) {
241 : syskin 984 if (cbp & (1 << (5 - i))) {
242 : edgomez 965 start_timer();
243 :     if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT))
244 :     dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant);
245 :     else
246 :     dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant);
247 :     stop_iquant_timer();
248 : edgomez 851 }
249 : h 69 }
250 : Isibaar 3 }
251 :    
252 : edgomez 965 typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);
253 :     typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);
254 : Isibaar 3
255 : edgomez 78
256 : edgomez 965 static __inline void
257 : syskin 984 MBTrans8to16(const MBParam * const pParam,
258 :     const FRAMEINFO * const frame,
259 :     const MACROBLOCK * const pMB,
260 : edgomez 965 const uint32_t x_pos,
261 :     const uint32_t y_pos,
262 :     int16_t data[6 * 64])
263 :     {
264 : h 82 uint32_t stride = pParam->edged_width;
265 :     uint32_t stride2 = stride / 2;
266 : edgomez 965 uint32_t next_block = stride * 8;
267 : syskin 984 int32_t cst;
268 : Isibaar 3 uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
269 : syskin 984 const IMAGE * const pCurrent = &frame->image;
270 : edgomez 965 transfer_operation_8to16_t *transfer_op = NULL;
271 : edgomez 195
272 : edgomez 965 if ((frame->vop_flags & XVID_VOP_REDUCED)) {
273 :    
274 :     /* Image pointers */
275 :     pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5);
276 : edgomez 851 pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4);
277 :     pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4);
278 : edgomez 965
279 :     /* Block size */
280 :     cst = 16;
281 :    
282 :     /* Operation function */
283 :     transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8;
284 :     } else {
285 :    
286 :     /* Image pointers */
287 :     pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4);
288 : edgomez 851 pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);
289 :     pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);
290 : edgomez 965
291 :     /* Block size */
292 :     cst = 8;
293 :    
294 :     /* Operation function */
295 :     transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy;
296 : edgomez 851 }
297 : Isibaar 3
298 : edgomez 965 /* Do the transfer */
299 : h 69 start_timer();
300 : edgomez 965 transfer_op(&data[0 * 64], pY_Cur, stride);
301 :     transfer_op(&data[1 * 64], pY_Cur + cst, stride);
302 :     transfer_op(&data[2 * 64], pY_Cur + next_block, stride);
303 :     transfer_op(&data[3 * 64], pY_Cur + next_block + cst, stride);
304 :     transfer_op(&data[4 * 64], pU_Cur, stride2);
305 :     transfer_op(&data[5 * 64], pV_Cur, stride2);
306 :     stop_transfer_timer();
307 : syskin 984 }
308 : edgomez 965
309 :     static __inline void
310 : syskin 984 MBTrans16to8(const MBParam * const pParam,
311 :     const FRAMEINFO * const frame,
312 :     const MACROBLOCK * const pMB,
313 : edgomez 965 const uint32_t x_pos,
314 :     const uint32_t y_pos,
315 :     int16_t data[6 * 64],
316 :     const uint32_t add,
317 :     const uint8_t cbp)
318 :     {
319 :     uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
320 :     uint32_t stride = pParam->edged_width;
321 :     uint32_t stride2 = stride / 2;
322 :     uint32_t next_block = stride * 8;
323 : syskin 984 uint32_t cst;
324 :     const IMAGE * const pCurrent = &frame->image;
325 : edgomez 965 transfer_operation_16to8_t *transfer_op = NULL;
326 :    
327 :     if (pMB->field_dct) {
328 :     next_block = stride;
329 :     stride *= 2;
330 : h 69 }
331 :    
332 : edgomez 965 if ((frame->vop_flags & XVID_VOP_REDUCED)) {
333 : edgomez 851
334 : edgomez 965 /* Image pointers */
335 :     pY_Cur = pCurrent->y + (y_pos << 5) * stride + (x_pos << 5);
336 :     pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4);
337 :     pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4);
338 : Isibaar 3
339 : edgomez 965 /* Block size */
340 :     cst = 16;
341 : Isibaar 3
342 : edgomez 965 /* Operation function */
343 :     if(add)
344 :     transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8;
345 :     else
346 :     transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8;
347 :     } else {
348 : Isibaar 3
349 : edgomez 965 /* Image pointers */
350 :     pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4);
351 :     pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);
352 :     pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);
353 : Isibaar 3
354 : edgomez 965 /* Block size */
355 :     cst = 8;
356 : Isibaar 3
357 : edgomez 965 /* Operation function */
358 :     if(add)
359 :     transfer_op = (transfer_operation_16to8_t*)transfer_16to8add;
360 :     else
361 :     transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy;
362 : Isibaar 3 }
363 : h 69
364 : edgomez 965 /* Do the operation */
365 : h 69 start_timer();
366 : edgomez 965 if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride);
367 :     if (cbp&16) transfer_op(pY_Cur + cst, &data[1 * 64], stride);
368 :     if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride);
369 :     if (cbp& 4) transfer_op(pY_Cur + next_block + cst, &data[3 * 64], stride);
370 :     if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2);
371 :     if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2);
372 : h 69 stop_transfer_timer();
373 : Isibaar 3 }
374 : h 69
375 : edgomez 965 /*****************************************************************************
376 :     * Module functions
377 :     ****************************************************************************/
378 :    
379 : syskin 984 void
380 :     MBTransQuantIntra(const MBParam * const pParam,
381 :     const FRAMEINFO * const frame,
382 :     MACROBLOCK * const pMB,
383 : chl 368 const uint32_t x_pos,
384 :     const uint32_t y_pos,
385 :     int16_t data[6 * 64],
386 :     int16_t qcoeff[6 * 64])
387 :     {
388 : h 69
389 : edgomez 965 /* Transfer data */
390 :     MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data);
391 : chl 368
392 : edgomez 965 /* Perform DCT (and field decision) */
393 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
394 : chl 368
395 : edgomez 965 /* Quantize the block */
396 : chl 995 MBQuantIntra(pParam, frame, pMB, data, qcoeff);
397 : edgomez 965
398 :     /* DeQuantize the block */
399 :     MBDeQuantIntra(pParam, pMB->quant, data, qcoeff);
400 :    
401 :     /* Perform inverse DCT*/
402 :     MBiDCT(data, 0x3F);
403 :    
404 :     /* Transfer back the data -- Don't add data */
405 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F);
406 : chl 368 }
407 :    
408 : edgomez 965
409 : chl 368 uint8_t
410 : syskin 984 MBTransQuantInter(const MBParam * const pParam,
411 :     const FRAMEINFO * const frame,
412 :     MACROBLOCK * const pMB,
413 : edgomez 914 const uint32_t x_pos,
414 :     const uint32_t y_pos,
415 : chl 368 int16_t data[6 * 64],
416 :     int16_t qcoeff[6 * 64])
417 :     {
418 :     uint8_t cbp;
419 : edgomez 965 uint32_t limit;
420 : chl 368
421 : edgomez 914 /*
422 : edgomez 965 * There is no MBTrans8to16 for Inter block, that's done in motion compensation
423 :     * already
424 : edgomez 914 */
425 : chl 368
426 : edgomez 965 /* Perform DCT (and field decision) */
427 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
428 : edgomez 914
429 : edgomez 965 /* Set the limit threshold */
430 :     limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
431 : chl 368
432 : edgomez 965 /* Quantize the block */
433 : chl 995 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
434 : chl 368
435 : edgomez 965 /* DeQuantize the block */
436 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
437 : chl 368
438 : edgomez 965 /* Perform inverse DCT*/
439 :     MBiDCT(data, cbp);
440 : chl 368
441 : edgomez 965 /* Transfer back the data -- Add the data */
442 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
443 : syskin 984
444 : edgomez 965 return(cbp);
445 : chl 368 }
446 :    
447 : edgomez 965 uint8_t
448 :     MBTransQuantInterBVOP(const MBParam * pParam,
449 : chl 368 FRAMEINFO * frame,
450 :     MACROBLOCK * pMB,
451 :     const uint32_t x_pos,
452 :     const uint32_t y_pos,
453 :     int16_t data[6 * 64],
454 : edgomez 965 int16_t qcoeff[6 * 64])
455 : chl 368 {
456 : edgomez 965 uint8_t cbp;
457 :     uint32_t limit;
458 : syskin 984
459 : edgomez 965 /*
460 :     * There is no MBTrans8to16 for Inter block, that's done in motion compensation
461 :     * already
462 :     */
463 : chl 368
464 : edgomez 965 /* Perform DCT (and field decision) */
465 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
466 : chl 368
467 : edgomez 965 /* Set the limit threshold */
468 :     limit = BVOP_TOOSMALL_LIMIT;
469 : chl 368
470 : edgomez 965 /* Quantize the block */
471 : chl 995 cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
472 : chl 368
473 : edgomez 965 /*
474 :     * History comment:
475 :     * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
476 :     *
477 :     * BUT some plugins require the original frame to be passed so we have
478 :     * to take care of that here
479 :     */
480 :     if((pParam->plugin_flags & XVID_REQORIGINAL)) {
481 : chl 368
482 : edgomez 965 /* DeQuantize the block */
483 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
484 : chl 368
485 : edgomez 965 /* Perform inverse DCT*/
486 :     MBiDCT(data, cbp);
487 : h 69
488 : edgomez 965 /* Transfer back the data -- Add the data */
489 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
490 : edgomez 851 }
491 :    
492 : edgomez 965 return(cbp);
493 : edgomez 851 }
494 :    
495 :     /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */
496 :     uint32_t
497 :     MBFieldTest_c(int16_t data[6 * 64])
498 :     {
499 : edgomez 195 const uint8_t blocks[] =
500 :     { 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 };
501 :     const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 };
502 : edgomez 78
503 : h 69 int frame = 0, field = 0;
504 :     int i, j;
505 :    
506 : edgomez 195 for (i = 0; i < 7; ++i) {
507 :     for (j = 0; j < 8; ++j) {
508 :     frame +=
509 : edgomez 982 abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]);
510 : edgomez 195 frame +=
511 : edgomez 982 abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]);
512 : edgomez 195 frame +=
513 : edgomez 982 abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]);
514 : edgomez 195 frame +=
515 : edgomez 982 abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]);
516 : h 69
517 : edgomez 195 field +=
518 : edgomez 982 abs(data[blocks[i + 1] + lines[i + 1] + j] -
519 : edgomez 195 data[blocks[i] + lines[i] + j]);
520 :     field +=
521 : edgomez 982 abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] -
522 : edgomez 195 data[blocks[i] + lines[i] + 8 + j]);
523 :     field +=
524 : edgomez 982 abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] -
525 : edgomez 195 data[blocks[i] + 64 + lines[i] + j]);
526 :     field +=
527 : edgomez 982 abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] -
528 : edgomez 195 data[blocks[i] + 64 + lines[i] + 8 + j]);
529 : h 69 }
530 :     }
531 :    
532 : edgomez 851 return (frame >= (field + 350));
533 : h 69 }
534 :    
535 :    
536 :     /* deinterlace Y blocks vertically */
537 :    
538 :     #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp))
539 : syskin 984 #define LINE(X,Y) &data[X*64 + Y*8]
540 : h 69
541 : edgomez 195 void
542 :     MBFrameToField(int16_t data[6 * 64])
543 : h 69 {
544 :     int16_t tmp[8];
545 :    
546 :     /* left blocks */
547 :    
548 : edgomez 851 // 1=2, 2=4, 4=8, 8=1
549 : edgomez 195 MOVLINE(tmp, LINE(0, 1));
550 :     MOVLINE(LINE(0, 1), LINE(0, 2));
551 :     MOVLINE(LINE(0, 2), LINE(0, 4));
552 :     MOVLINE(LINE(0, 4), LINE(2, 0));
553 :     MOVLINE(LINE(2, 0), tmp);
554 : h 69
555 : edgomez 851 // 3=6, 6=12, 12=9, 9=3
556 : edgomez 195 MOVLINE(tmp, LINE(0, 3));
557 :     MOVLINE(LINE(0, 3), LINE(0, 6));
558 :     MOVLINE(LINE(0, 6), LINE(2, 4));
559 :     MOVLINE(LINE(2, 4), LINE(2, 1));
560 :     MOVLINE(LINE(2, 1), tmp);
561 : h 69
562 : edgomez 851 // 5=10, 10=5
563 : edgomez 195 MOVLINE(tmp, LINE(0, 5));
564 :     MOVLINE(LINE(0, 5), LINE(2, 2));
565 :     MOVLINE(LINE(2, 2), tmp);
566 : h 69
567 : edgomez 851 // 7=14, 14=13, 13=11, 11=7
568 : edgomez 195 MOVLINE(tmp, LINE(0, 7));
569 :     MOVLINE(LINE(0, 7), LINE(2, 6));
570 :     MOVLINE(LINE(2, 6), LINE(2, 5));
571 :     MOVLINE(LINE(2, 5), LINE(2, 3));
572 :     MOVLINE(LINE(2, 3), tmp);
573 : h 69
574 :     /* right blocks */
575 :    
576 : edgomez 851 // 1=2, 2=4, 4=8, 8=1
577 : edgomez 195 MOVLINE(tmp, LINE(1, 1));
578 :     MOVLINE(LINE(1, 1), LINE(1, 2));
579 :     MOVLINE(LINE(1, 2), LINE(1, 4));
580 :     MOVLINE(LINE(1, 4), LINE(3, 0));
581 :     MOVLINE(LINE(3, 0), tmp);
582 : h 69
583 : edgomez 851 // 3=6, 6=12, 12=9, 9=3
584 : edgomez 195 MOVLINE(tmp, LINE(1, 3));
585 :     MOVLINE(LINE(1, 3), LINE(1, 6));
586 :     MOVLINE(LINE(1, 6), LINE(3, 4));
587 :     MOVLINE(LINE(3, 4), LINE(3, 1));
588 :     MOVLINE(LINE(3, 1), tmp);
589 : h 69
590 : edgomez 851 // 5=10, 10=5
591 : edgomez 195 MOVLINE(tmp, LINE(1, 5));
592 :     MOVLINE(LINE(1, 5), LINE(3, 2));
593 :     MOVLINE(LINE(3, 2), tmp);
594 : h 69
595 : edgomez 851 // 7=14, 14=13, 13=11, 11=7
596 : edgomez 195 MOVLINE(tmp, LINE(1, 7));
597 :     MOVLINE(LINE(1, 7), LINE(3, 6));
598 :     MOVLINE(LINE(3, 6), LINE(3, 5));
599 :     MOVLINE(LINE(3, 5), LINE(3, 3));
600 :     MOVLINE(LINE(3, 3), tmp);
601 : h 69 }
602 : chl 1011
603 :    
604 :    
605 :    
606 :    
607 :     /************************************************************************
608 :     * Trellis based R-D optimal quantization *
609 :     * *
610 :     * Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net *
611 :     * *
612 :     ************************************************************************/
613 :    
614 :    
615 :     static int
616 :     dct_quantize_trellis_inter_mpeg_c (int16_t *qcoeff, const int16_t *data, int quant)
617 :     { return 63; }
618 :    
619 :    
620 :     //////////////////////////////////////////////////////////
621 :     //
622 :     // Trellis-Based quantization
623 :     //
624 :     // So far I understand this paper:
625 :     //
626 :     // "Trellis-Based R-D Optimal Quantization in H.263+"
627 :     // J.Wen, M.Luttrell, J.Villasenor
628 :     // IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
629 :     //
630 :     // we are at stake with a simplified Bellmand-Ford / Dijkstra Single
631 :     // Source Shorted Path algo. But due to the underlying graph structure
632 :     // ("Trellis"), it can be turned into a dynamic programming algo,
633 :     // partially saving the explicit graph's nodes representation. And
634 :     // without using a heap, since the open frontier of the DAG is always
635 :     // known, and of fixed sized.
636 :     //
637 :     //////////////////////////////////////////////////////////
638 :    
639 :    
640 :     //////////////////////////////////////////////////////////
641 :     // Codes lengths for relevant levels.
642 :    
643 :     // let's factorize:
644 :     static const uint8_t Code_Len0[64] = {
645 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
646 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
647 :     static const uint8_t Code_Len1[64] = {
648 :     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,
649 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
650 :     static const uint8_t Code_Len2[64] = {
651 :     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,
652 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
653 :     static const uint8_t Code_Len3[64] = {
654 :     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,
655 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
656 :     static const uint8_t Code_Len4[64] = {
657 :     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,
658 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
659 :     static const uint8_t Code_Len5[64] = {
660 :     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,
661 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
662 :     static const uint8_t Code_Len6[64] = {
663 :     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,
664 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
665 :     static const uint8_t Code_Len7[64] = {
666 :     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,
667 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
668 :     static const uint8_t Code_Len8[64] = {
669 :     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,
670 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
671 :     static const uint8_t Code_Len9[64] = {
672 :     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,
673 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
674 :     static const uint8_t Code_Len10[64] = {
675 :     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,
676 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
677 :     static const uint8_t Code_Len11[64] = {
678 :     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,
679 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
680 :     static const uint8_t Code_Len12[64] = {
681 :     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,
682 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
683 :     static const uint8_t Code_Len13[64] = {
684 :     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,
685 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
686 :     static const uint8_t Code_Len14[64] = {
687 :     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,
688 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
689 :     static const uint8_t Code_Len15[64] = {
690 :     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,
691 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
692 :     static const uint8_t Code_Len16[64] = {
693 :     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,
694 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
695 :     static const uint8_t Code_Len17[64] = {
696 :     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,
697 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
698 :     static const uint8_t Code_Len18[64] = {
699 :     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,
700 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
701 :     static const uint8_t Code_Len19[64] = {
702 :     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,
703 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
704 :     static const uint8_t Code_Len20[64] = {
705 :     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,
706 :     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 };
707 :    
708 :     // a few more table for LAST table:
709 :     static const uint8_t Code_Len21[64] = {
710 :     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,
711 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
712 :     static const uint8_t Code_Len22[64] = {
713 :     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,
714 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
715 :     static const uint8_t Code_Len23[64] = {
716 :     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,
717 :     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};
718 :     static const uint8_t Code_Len24[64] = {
719 :     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,
720 :     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};
721 :    
722 :    
723 :     static const uint8_t * const B16_17_Code_Len[24] = { // levels [1..24]
724 :     Code_Len20,Code_Len19,Code_Len18,Code_Len17,
725 :     Code_Len16,Code_Len15,Code_Len14,Code_Len13,
726 :     Code_Len12,Code_Len11,Code_Len10,Code_Len9,
727 :     Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5,
728 :     Code_Len4, Code_Len3, Code_Len3 ,Code_Len2,
729 :     Code_Len2, Code_Len1, Code_Len1, Code_Len1,
730 :     };
731 :    
732 :     static const uint8_t * const B16_17_Code_Len_Last[6] = { // levels [1..6]
733 :     Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
734 :     };
735 :    
736 :     #define TL(q) 0xfe00/(q*q)
737 :    
738 :     static const int Trellis_Lambda_Tabs[31] = {
739 :     TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7),
740 :     TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15),
741 :     TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23),
742 :     TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31)
743 :     };
744 :     #undef TL
745 :    
746 :     static inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
747 :     {
748 :     while(i>=0)
749 :     if (C[Zigzag[i]])
750 :     return i;
751 :     else i--;
752 :     return -1;
753 :     }
754 :    
755 :     //////////////////////////////////////////////////////////
756 :    
757 :     #define DBG 0
758 :    
759 :     static uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,
760 :     const uint16_t * Zigzag, int Max, int Lambda)
761 :     {
762 :     #if (DBG>0)
763 :     const int16_t * const Ref = C + 6*64;
764 :     int Last = Max;
765 :     while(Last>=0 && C[Zigzag[Last]]==0) Last--;
766 :     int Bits = 0;
767 :     if (Last>=0) {
768 :     Bits = 2; // CBP
769 :     int j=0, j0=0;
770 :     int Run, Level;
771 :     while(j<Last) {
772 :     while(!C[Zigzag[j]]) j++;
773 :     if (j==Last) break;
774 :     Level=C[Zigzag[j]];
775 :     Run = j - j0;
776 :     j0 = ++j;
777 :     if (Level>=-24 && Level<=24) Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];
778 :     else Bits += 30;
779 :     }
780 :     Level = C[Zigzag[Last]];
781 :     Run = j - j0;
782 :     if (Level>=-6 && Level<=6) Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];
783 :     else Bits += 30;
784 :     }
785 :    
786 :     int Dist = 0;
787 :     int i;
788 :     for(i=0; i<=Last; ++i) {
789 :     int V = C[Zigzag[i]]*Mult;
790 :     if (V>0) V += Bias;
791 :     else if (V<0) V -= Bias;
792 :     V -= Ref[Zigzag[i]];
793 :     Dist += V*V;
794 :     }
795 :     uint32_t Cost = Lambda*Dist + (Bits<<16);
796 :     if (DBG==1)
797 :     printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );
798 :     return Cost;
799 :    
800 :     #else
801 :     return 0;
802 :     #endif
803 :     }
804 :    
805 :    
806 :     static int
807 :     dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
808 :     {
809 :    
810 :     // Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
811 :     // not quantized one (Out[]). However, it only improves the result *very*
812 :     // slightly (~0.01dB), whereas speed drops to crawling level :)
813 :     // Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps,
814 :    
815 :     typedef struct { int16_t Run, Level; } NODE;
816 :    
817 :     NODE Nodes[65], Last;
818 :     uint32_t Run_Costs0[64+1], * const Run_Costs = Run_Costs0 + 1;
819 :    
820 :     const int Mult = 2*Q;
821 :     const int Bias = (Q-1) | 1;
822 :     const int Lev0 = Mult + Bias;
823 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; // it's 1/lambda, actually
824 :    
825 :     int Run_Start = -1;
826 :     Run_Costs[-1] = 2<<16; // source (w/ CBP penalty)
827 :     uint32_t Min_Cost = 2<<16;
828 :    
829 :     int Last_Node = -1;
830 :     uint32_t Last_Cost = 0;
831 :    
832 :     #if (DBG>0)
833 :     Last.Level = 0; Last.Run = -1; // just initialize to smthg
834 :     #endif
835 :    
836 :     int i, j;
837 :    
838 :     Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
839 :     if (Non_Zero<0)
840 :     return -1;
841 :    
842 :     for(i=0; i<=Non_Zero; i++)
843 :     {
844 :     const int AC = In[Zigzag[i]];
845 :     const int Level1 = Out[Zigzag[i]];
846 :     const int Dist0 = Lambda* AC*AC;
847 :     uint32_t Best_Cost = 0xf0000000;
848 :     Last_Cost += Dist0;
849 :    
850 :     if ((uint32_t)(Level1+1)<3) // very specialized loop for -1,0,+1
851 :     {
852 :     int dQ;
853 :     int Run;
854 :    
855 :     if (AC<0) {
856 :     Nodes[i].Level = -1;
857 :     dQ = Lev0 + AC;
858 :     } else {
859 :     Nodes[i].Level = 1;
860 :     dQ = Lev0 - AC;
861 :     }
862 :     const uint32_t Cost0 = Lambda*dQ*dQ;
863 :    
864 :     Nodes[i].Run = 1;
865 :     Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
866 :     for(Run=i-Run_Start; Run>0; --Run)
867 :     {
868 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
869 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
870 :     // TODO: what about tie-breaks? Should we favor short runs or
871 :     // long runs? Although the error is the same, it would not be
872 :     // spread the same way along high and low frequencies...
873 :     if (Cost<Best_Cost)
874 :     {
875 :     Best_Cost = Cost;
876 :     Nodes[i].Run = Run;
877 :     }
878 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
879 :     if (lCost<Last_Cost)
880 :     {
881 :     Last_Cost = lCost;
882 :     Last.Run = Run;
883 :     Last_Node = i;
884 :     }
885 :     }
886 :     if (Last_Node==i) Last.Level = Nodes[i].Level;
887 :    
888 :    
889 :     if (DBG==1) {
890 :     Run_Costs[i] = Best_Cost;
891 :     printf( "Costs #%2d: ", i);
892 :     for(j=-1;j<=Non_Zero;++j) {
893 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
894 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
895 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
896 :     else printf( " - |" );
897 :     }
898 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
899 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
900 :     printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 );
901 :     printf( "\n" );
902 :     }
903 :     }
904 :     else // "big" levels
905 :     {
906 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
907 :     int Level2;
908 :     int dQ1, dQ2;
909 :     int Run;
910 :    
911 :     if (Level1>1) {
912 :     dQ1 = Level1*Mult-AC + Bias;
913 :     dQ2 = dQ1 - Mult;
914 :     Level2 = Level1-1;
915 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
916 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
917 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
918 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
919 :     }
920 :     else { // Level1<-1
921 :     dQ1 = Level1*Mult-AC - Bias;
922 :     dQ2 = dQ1 + Mult;
923 :     Level2 = Level1 + 1;
924 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
925 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
926 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
927 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
928 :     }
929 :     const uint32_t Dist1 = Lambda*dQ1*dQ1;
930 :     const uint32_t Dist2 = Lambda*dQ2*dQ2;
931 :     const int dDist21 = Dist2-Dist1;
932 :    
933 :     for(Run=i-Run_Start; Run>0; --Run)
934 :     {
935 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
936 :    
937 :     // for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
938 :     // if (Cost_Base>=Best_Cost) continue;
939 :    
940 :     uint32_t Cost1, Cost2;
941 :     int bLevel;
942 :    
943 :     Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
944 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
945 :    
946 :     if (Cost2<Cost1) { Cost1 = Cost2; bLevel = Level2; }
947 :     else bLevel = Level1;
948 :    
949 :     if (Cost1<Best_Cost)
950 :     {
951 :     Best_Cost = Cost1;
952 :     Nodes[i].Run = Run;
953 :     Nodes[i].Level = bLevel;
954 :     }
955 :    
956 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
957 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
958 :    
959 :     if (Cost2<Cost1) { Cost1 = Cost2; bLevel = Level2; }
960 :     else bLevel = Level1;
961 :     if (Cost1<Last_Cost)
962 :     {
963 :     Last_Cost = Cost1;
964 :     Last.Run = Run;
965 :     Last.Level = bLevel;
966 :     Last_Node = i;
967 :     }
968 :     }
969 :    
970 :     if (DBG==1) {
971 :     Run_Costs[i] = Best_Cost;
972 :     printf( "Costs #%2d: ", i);
973 :     for(j=-1;j<=Non_Zero;++j) {
974 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
975 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
976 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
977 :     else printf( " - |" );
978 :     }
979 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
980 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
981 :     printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 );
982 :     printf( "\n" );
983 :     }
984 :     }
985 :    
986 :     Run_Costs[i] = Best_Cost;
987 :    
988 :     if (Best_Cost < Min_Cost + Dist0) {
989 :     Min_Cost = Best_Cost;
990 :     Run_Start = i;
991 :     }
992 :     else
993 :     {
994 :     // as noticed by Michael Niedermayer (michaelni at gmx.at), there's
995 :     // a code shorter by 1 bit for a larger run (!), same level. We give
996 :     // it a chance by not moving the left barrier too much.
997 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
998 :     Run_Start++;
999 :    
1000 :     // spread on preceding coeffs the cost incurred by skipping this one
1001 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1002 :     Min_Cost += Dist0;
1003 :     }
1004 :     }
1005 :    
1006 :     if (DBG) {
1007 :     Last_Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1008 :     if (DBG==1) {
1009 :     printf( "=> " );
1010 :     for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1011 :     printf( "\n" );
1012 :     }
1013 :     }
1014 :    
1015 :     if (Last_Node<0)
1016 :     return -1;
1017 :    
1018 :     // reconstruct optimal sequence backward with surviving paths
1019 :     bzero(Out, 64*sizeof(*Out));
1020 :     Out[Zigzag[Last_Node]] = Last.Level;
1021 :     i = Last_Node - Last.Run;
1022 :     while(i>=0) {
1023 :     Out[Zigzag[i]] = Nodes[i].Level;
1024 :     i -= Nodes[i].Run;
1025 :     }
1026 :    
1027 :     if (DBG) {
1028 :     uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1029 :     if (DBG==1) {
1030 :     printf( "<= " );
1031 :     for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1032 :     printf( "\n--------------------------------\n" );
1033 :     }
1034 :     if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost );
1035 :     }
1036 :     return Last_Node;
1037 :     }
1038 :    
1039 :     #undef DBG

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