[svn] / branches / release-1_3-branch / xvidcore / src / utils / mbtransquant.c Repository:
ViewVC logotype

Annotation of /branches/release-1_3-branch/xvidcore/src/utils/mbtransquant.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1985 - (view) (download)

1 : edgomez 1382 /*****************************************************************************
2 :     *
3 :     * XVID MPEG-4 VIDEO CODEC
4 : Isibaar 1909 * - MB Transfer/Quantization functions -
5 : edgomez 1382 *
6 : Isibaar 1909 * Copyright(C) 2001-2010 Peter Ross <pross@xvid.org>
7 :     * 2001-2010 Michael Militzer <michael@xvid.org>
8 : edgomez 1382 * 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 : Isibaar 1985 * $Id$
25 : edgomez 1382 *
26 :     ****************************************************************************/
27 : Isibaar 3
28 : edgomez 1382 #include <stdio.h>
29 :     #include <stdlib.h>
30 : edgomez 78 #include <string.h>
31 :    
32 : Isibaar 3 #include "../portab.h"
33 :     #include "mbfunctions.h"
34 :    
35 :     #include "../global.h"
36 :     #include "mem_transfer.h"
37 :     #include "timer.h"
38 : edgomez 1382 #include "../bitstream/mbcoding.h"
39 :     #include "../bitstream/zigzag.h"
40 : Isibaar 3 #include "../dct/fdct.h"
41 :     #include "../dct/idct.h"
42 : edgomez 1382 #include "../quant/quant.h"
43 : Isibaar 1909 #include "../motion/sad.h"
44 : Isibaar 3 #include "../encoder.h"
45 :    
46 : edgomez 1382 #include "../quant/quant_matrix.h"
47 : Isibaar 3
48 : edgomez 851 MBFIELDTEST_PTR MBFieldTest;
49 : Isibaar 3
50 : edgomez 1382 /*
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 : Isibaar 3
57 : edgomez 1382 /*****************************************************************************
58 :     * Local functions
59 :     ****************************************************************************/
60 :    
61 :     /* permute block and return field dct choice */
62 :     static __inline uint32_t
63 :     MBDecideFieldDCT(int16_t data[6 * 64])
64 :     {
65 :     uint32_t field = MBFieldTest(data);
66 :    
67 :     if (field)
68 :     MBFrameToField(data);
69 :    
70 :     return field;
71 :     }
72 :    
73 :     /* Performs Forward DCT on all blocks */
74 : syskin 937 static __inline void
75 : edgomez 1382 MBfDCT(const MBParam * const pParam,
76 :     const FRAMEINFO * const frame,
77 :     MACROBLOCK * const pMB,
78 :     uint32_t x_pos,
79 :     uint32_t y_pos,
80 :     int16_t data[6 * 64])
81 : syskin 937 {
82 : edgomez 1382 /* Handles interlacing */
83 : syskin 937 start_timer();
84 : edgomez 1382 pMB->field_dct = 0;
85 :     if ((frame->vol_flags & XVID_VOL_INTERLACING) &&
86 :     (x_pos>0) && (x_pos<pParam->mb_width-1) &&
87 :     (y_pos>0) && (y_pos<pParam->mb_height-1)) {
88 :     pMB->field_dct = MBDecideFieldDCT(data);
89 :     }
90 :     stop_interlacing_timer();
91 :    
92 :     /* Perform DCT */
93 :     start_timer();
94 : suxen_drol 1653 fdct((short * const)&data[0 * 64]);
95 :     fdct((short * const)&data[1 * 64]);
96 :     fdct((short * const)&data[2 * 64]);
97 :     fdct((short * const)&data[3 * 64]);
98 :     fdct((short * const)&data[4 * 64]);
99 :     fdct((short * const)&data[5 * 64]);
100 : syskin 937 stop_dct_timer();
101 :     }
102 :    
103 : edgomez 1382 /* 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 : suxen_drol 1653 if(cbp & (1 << (5 - 0))) idct((short * const)&data[0 * 64]);
110 :     if(cbp & (1 << (5 - 1))) idct((short * const)&data[1 * 64]);
111 :     if(cbp & (1 << (5 - 2))) idct((short * const)&data[2 * 64]);
112 :     if(cbp & (1 << (5 - 3))) idct((short * const)&data[3 * 64]);
113 :     if(cbp & (1 << (5 - 4))) idct((short * const)&data[4 * 64]);
114 :     if(cbp & (1 << (5 - 5))) idct((short * const)&data[5 * 64]);
115 : edgomez 1382 stop_idct_timer();
116 :     }
117 : syskin 937
118 : edgomez 1382 /* Quantize all blocks -- Intra mode */
119 :     static __inline void
120 :     MBQuantIntra(const MBParam * pParam,
121 :     const FRAMEINFO * const frame,
122 :     const MACROBLOCK * pMB,
123 :     int16_t qcoeff[6 * 64],
124 :     int16_t data[6*64])
125 : syskin 937 {
126 : edgomez 1382 int scaler_lum, scaler_chr;
127 : syskin 1713 quant_intraFuncPtr quant;
128 : syskin 937
129 : syskin 1713 /* check if quant matrices need to be re-initialized with new quant */
130 :     if (pParam->vol_flags & XVID_VOL_MPEGQUANT) {
131 :     if (pParam->last_quant_initialized_intra != pMB->quant) {
132 :     init_intra_matrix(pParam->mpeg_quant_matrices, pMB->quant);
133 :     }
134 :     quant = quant_mpeg_intra;
135 :     } else {
136 :     quant = quant_h263_intra;
137 :     }
138 : edgomez 1382
139 :     scaler_lum = get_dc_scaler(pMB->quant, 1);
140 :     scaler_chr = get_dc_scaler(pMB->quant, 0);
141 :    
142 :     /* Quantize the block */
143 : syskin 937 start_timer();
144 : syskin 1713 quant(&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices);
145 :     quant(&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices);
146 :     quant(&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices);
147 :     quant(&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum, pParam->mpeg_quant_matrices);
148 :     quant(&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices);
149 :     quant(&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr, pParam->mpeg_quant_matrices);
150 : syskin 937 stop_quant_timer();
151 :     }
152 :    
153 : edgomez 1382 /* DeQuantize all blocks -- Intra mode */
154 :     static __inline void
155 :     MBDeQuantIntra(const MBParam * pParam,
156 :     const int iQuant,
157 :     int16_t qcoeff[6 * 64],
158 :     int16_t data[6*64])
159 : Isibaar 3 {
160 : edgomez 1382 int mpeg;
161 :     int scaler_lum, scaler_chr;
162 : edgomez 78
163 : edgomez 1382 quant_intraFuncPtr const dequant[2] =
164 :     {
165 :     dequant_h263_intra,
166 :     dequant_mpeg_intra
167 :     };
168 : Isibaar 3
169 : edgomez 1382 mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
170 :     scaler_lum = get_dc_scaler(iQuant, 1);
171 :     scaler_chr = get_dc_scaler(iQuant, 0);
172 :    
173 : edgomez 851 start_timer();
174 : edgomez 1382 dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices);
175 :     dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices);
176 :     dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices);
177 :     dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum, pParam->mpeg_quant_matrices);
178 :     dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices);
179 :     dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr, pParam->mpeg_quant_matrices);
180 :     stop_iquant_timer();
181 :     }
182 : h 69
183 : edgomez 1382 static int
184 :     dct_quantize_trellis_c(int16_t *const Out,
185 :     const int16_t *const In,
186 :     int Q,
187 :     const uint16_t * const Zigzag,
188 :     const uint16_t * const QuantMatrix,
189 : edgomez 1432 int Non_Zero,
190 : syskin 1660 int Sum,
191 : Isibaar 1909 int Lambda_Mod,
192 :     const uint32_t rel_var8,
193 :     const int Metric);
194 : edgomez 851
195 : edgomez 1382 /* Quantize all blocks -- Inter mode */
196 :     static __inline uint8_t
197 :     MBQuantInter(const MBParam * pParam,
198 :     const FRAMEINFO * const frame,
199 :     const MACROBLOCK * pMB,
200 :     int16_t data[6 * 64],
201 :     int16_t qcoeff[6 * 64],
202 :     int bvop,
203 :     int limit)
204 :     {
205 : h 69
206 : edgomez 1382 int i;
207 :     uint8_t cbp = 0;
208 :     int sum;
209 :     int code_block, mpeg;
210 : h 69
211 : edgomez 1382 quant_interFuncPtr const quant[2] =
212 :     {
213 :     quant_h263_inter,
214 :     quant_mpeg_inter
215 :     };
216 : syskin 937
217 : edgomez 1382 mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
218 :    
219 : edgomez 195 for (i = 0; i < 6; i++) {
220 : Isibaar 3
221 : edgomez 1382 /* Quantize the block */
222 : Isibaar 3 start_timer();
223 :    
224 : edgomez 1382 sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, pParam->mpeg_quant_matrices);
225 : edgomez 851
226 : edgomez 1582 if(sum && (pMB->quant > 2) && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) {
227 : edgomez 1382 const uint16_t *matrix;
228 :     const static uint16_t h263matrix[] =
229 :     {
230 :     16, 16, 16, 16, 16, 16, 16, 16,
231 :     16, 16, 16, 16, 16, 16, 16, 16,
232 :     16, 16, 16, 16, 16, 16, 16, 16,
233 :     16, 16, 16, 16, 16, 16, 16, 16,
234 :     16, 16, 16, 16, 16, 16, 16, 16,
235 :     16, 16, 16, 16, 16, 16, 16, 16,
236 :     16, 16, 16, 16, 16, 16, 16, 16,
237 :     16, 16, 16, 16, 16, 16, 16, 16
238 :     };
239 :    
240 :     matrix = (mpeg)?get_inter_matrix(pParam->mpeg_quant_matrices):h263matrix;
241 :     sum = dct_quantize_trellis_c(&qcoeff[i*64], &data[i*64],
242 :     pMB->quant, &scan_tables[0][0],
243 :     matrix,
244 : edgomez 1432 63,
245 : syskin 1660 sum,
246 : Isibaar 1909 pMB->lambda[i],
247 :     pMB->rel_var8[i],
248 :     !!(frame->vop_flags & XVID_VOP_RD_PSNRHVSM));
249 : Isibaar 3 }
250 : edgomez 1382 stop_quant_timer();
251 : Isibaar 3
252 : edgomez 1382 /*
253 :     * We code the block if the sum is higher than the limit and if the first
254 :     * two AC coefficients in zig zag order are not zero.
255 :     */
256 :     code_block = 0;
257 :     if ((sum >= limit) || (qcoeff[i*64+1] != 0) || (qcoeff[i*64+8] != 0)) {
258 :     code_block = 1;
259 :     } else {
260 : edgomez 851
261 : edgomez 1382 if (bvop && (pMB->mode == MODE_DIRECT || pMB->mode == MODE_DIRECT_NO4V)) {
262 :     /* dark blocks prevention for direct mode */
263 :     if ((qcoeff[i*64] < -1) || (qcoeff[i*64] > 0))
264 :     code_block = 1;
265 :     } else {
266 :     /* not direct mode */
267 :     if (qcoeff[i*64] != 0)
268 :     code_block = 1;
269 :     }
270 : edgomez 851 }
271 :    
272 : edgomez 1382 /* Set the corresponding cbp bit */
273 :     cbp |= code_block << (5 - i);
274 : h 69 }
275 :    
276 : edgomez 1382 return(cbp);
277 : Isibaar 3 }
278 :    
279 : edgomez 1382 /* DeQuantize all blocks -- Inter mode */
280 :     static __inline void
281 :     MBDeQuantInter(const MBParam * pParam,
282 :     const int iQuant,
283 :     int16_t data[6 * 64],
284 :     int16_t qcoeff[6 * 64],
285 :     const uint8_t cbp)
286 : Isibaar 3 {
287 : edgomez 1382 int mpeg;
288 :    
289 :     quant_interFuncPtr const dequant[2] =
290 :     {
291 :     dequant_h263_inter,
292 :     dequant_mpeg_inter
293 :     };
294 :    
295 :     mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
296 :    
297 :     start_timer();
298 :     if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant, pParam->mpeg_quant_matrices);
299 :     if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant, pParam->mpeg_quant_matrices);
300 :     if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant, pParam->mpeg_quant_matrices);
301 :     if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant, pParam->mpeg_quant_matrices);
302 :     if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant, pParam->mpeg_quant_matrices);
303 :     if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant, pParam->mpeg_quant_matrices);
304 :     stop_iquant_timer();
305 :     }
306 :    
307 :     typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);
308 :     typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);
309 :    
310 :    
311 :     static __inline void
312 :     MBTrans8to16(const MBParam * const pParam,
313 :     const FRAMEINFO * const frame,
314 :     const MACROBLOCK * const pMB,
315 :     const uint32_t x_pos,
316 :     const uint32_t y_pos,
317 :     int16_t data[6 * 64])
318 :     {
319 : h 82 uint32_t stride = pParam->edged_width;
320 : edgomez 1382 uint32_t stride2 = stride / 2;
321 :     uint32_t next_block = stride * 8;
322 : Isibaar 3 uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
323 : syskin 937 const IMAGE * const pCurrent = &frame->image;
324 : edgomez 195
325 : edgomez 1382 /* Image pointers */
326 : syskin 1566 pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4);
327 :     pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);
328 :     pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);
329 : h 69
330 : edgomez 1382 /* Do the transfer */
331 :     start_timer();
332 : syskin 1566 transfer_8to16copy(&data[0 * 64], pY_Cur, stride);
333 :     transfer_8to16copy(&data[1 * 64], pY_Cur + 8, stride);
334 :     transfer_8to16copy(&data[2 * 64], pY_Cur + next_block, stride);
335 :     transfer_8to16copy(&data[3 * 64], pY_Cur + next_block + 8, stride);
336 :     transfer_8to16copy(&data[4 * 64], pU_Cur, stride2);
337 :     transfer_8to16copy(&data[5 * 64], pV_Cur, stride2);
338 : edgomez 1382 stop_transfer_timer();
339 :     }
340 : Isibaar 3
341 : edgomez 1382 static __inline void
342 :     MBTrans16to8(const MBParam * const pParam,
343 :     const FRAMEINFO * const frame,
344 :     const MACROBLOCK * const pMB,
345 :     const uint32_t x_pos,
346 :     const uint32_t y_pos,
347 :     int16_t data[6 * 64],
348 :     const uint32_t add, /* Must be 1 or 0 */
349 :     const uint8_t cbp)
350 :     {
351 :     uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
352 :     uint32_t stride = pParam->edged_width;
353 :     uint32_t stride2 = stride / 2;
354 :     uint32_t next_block = stride * 8;
355 :     const IMAGE * const pCurrent = &frame->image;
356 : Isibaar 3
357 : syskin 1566 /* Array of function pointers, indexed by [add] */
358 :     transfer_operation_16to8_t * const functions[2] =
359 : edgomez 1382 {
360 :     (transfer_operation_16to8_t*)transfer_16to8copy,
361 :     (transfer_operation_16to8_t*)transfer_16to8add,
362 :     };
363 : Isibaar 3
364 : edgomez 1382 transfer_operation_16to8_t *transfer_op = NULL;
365 : Isibaar 3
366 : edgomez 1382 /* Image pointers */
367 : syskin 1566 pY_Cur = pCurrent->y + (y_pos << 4) * stride + (x_pos << 4);
368 :     pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);
369 :     pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);
370 : edgomez 1382
371 : edgomez 195 if (pMB->field_dct) {
372 : h 82 next_block = stride;
373 :     stride *= 2;
374 : h 69 }
375 :    
376 : edgomez 1382 /* Operation function */
377 : syskin 1566 transfer_op = functions[add];
378 : edgomez 1382
379 :     /* Do the operation */
380 : h 69 start_timer();
381 : syskin 1566 if (cbp&32) transfer_op(pY_Cur, &data[0 * 64], stride);
382 :     if (cbp&16) transfer_op(pY_Cur + 8, &data[1 * 64], stride);
383 :     if (cbp& 8) transfer_op(pY_Cur + next_block, &data[2 * 64], stride);
384 :     if (cbp& 4) transfer_op(pY_Cur + next_block + 8, &data[3 * 64], stride);
385 :     if (cbp& 2) transfer_op(pU_Cur, &data[4 * 64], stride2);
386 :     if (cbp& 1) transfer_op(pV_Cur, &data[5 * 64], stride2);
387 : h 69 stop_transfer_timer();
388 : edgomez 1382 }
389 : h 69
390 : edgomez 1382 /*****************************************************************************
391 :     * Module functions
392 :     ****************************************************************************/
393 :    
394 :     void
395 :     MBTransQuantIntra(const MBParam * const pParam,
396 :     const FRAMEINFO * const frame,
397 :     MACROBLOCK * const pMB,
398 :     const uint32_t x_pos,
399 :     const uint32_t y_pos,
400 :     int16_t data[6 * 64],
401 :     int16_t qcoeff[6 * 64])
402 :     {
403 :    
404 :     /* Transfer data */
405 :     MBTrans8to16(pParam, frame, pMB, x_pos, y_pos, data);
406 :    
407 :     /* Perform DCT (and field decision) */
408 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
409 :    
410 :     /* Quantize the block */
411 :     MBQuantIntra(pParam, frame, pMB, data, qcoeff);
412 :    
413 :     /* DeQuantize the block */
414 :     MBDeQuantIntra(pParam, pMB->quant, data, qcoeff);
415 :    
416 :     /* Perform inverse DCT*/
417 :     MBiDCT(data, 0x3F);
418 :    
419 :     /* Transfer back the data -- Don't add data */
420 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 0, 0x3F);
421 : Isibaar 3 }
422 : h 69
423 : edgomez 1382
424 : chl 368 uint8_t
425 : edgomez 1382 MBTransQuantInter(const MBParam * const pParam,
426 :     const FRAMEINFO * const frame,
427 :     MACROBLOCK * const pMB,
428 :     const uint32_t x_pos,
429 :     const uint32_t y_pos,
430 : chl 368 int16_t data[6 * 64],
431 :     int16_t qcoeff[6 * 64])
432 :     {
433 : edgomez 1382 uint8_t cbp;
434 :     uint32_t limit;
435 : chl 368
436 : edgomez 1382 /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
437 :     * already */
438 : chl 368
439 : edgomez 1382 /* Perform DCT (and field decision) */
440 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
441 : chl 368
442 : edgomez 1382 /* Set the limit threshold */
443 :     limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
444 : chl 368
445 : edgomez 1382 if (frame->vop_flags & XVID_VOP_CARTOON)
446 :     limit *= 3;
447 : chl 368
448 : edgomez 1382 /* Quantize the block */
449 :     cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
450 : chl 368
451 : edgomez 1382 /* DeQuantize the block */
452 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
453 :    
454 :     /* Perform inverse DCT*/
455 :     MBiDCT(data, cbp);
456 :    
457 :     /* Transfer back the data -- Add the data */
458 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
459 :    
460 :     return(cbp);
461 : chl 368 }
462 :    
463 : edgomez 1382 uint8_t
464 :     MBTransQuantInterBVOP(const MBParam * pParam,
465 :     FRAMEINFO * frame,
466 :     MACROBLOCK * pMB,
467 :     const uint32_t x_pos,
468 :     const uint32_t y_pos,
469 :     int16_t data[6 * 64],
470 :     int16_t qcoeff[6 * 64])
471 : h 69 {
472 : edgomez 1382 uint8_t cbp;
473 :     uint32_t limit;
474 : h 69
475 : edgomez 1382 /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
476 :     * already */
477 :    
478 :     /* Perform DCT (and field decision) */
479 :     MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
480 :    
481 :     /* Set the limit threshold */
482 :     limit = BVOP_TOOSMALL_LIMIT;
483 :    
484 :     if (frame->vop_flags & XVID_VOP_CARTOON)
485 :     limit *= 2;
486 :    
487 :     /* Quantize the block */
488 :     cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
489 :    
490 :     /*
491 :     * History comment:
492 :     * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
493 :     *
494 :     * BUT some plugins require the rebuilt original frame to be passed so we
495 :     * have to take care of that here
496 :     */
497 :     if((pParam->plugin_flags & XVID_REQORIGINAL)) {
498 :    
499 :     /* DeQuantize the block */
500 :     MBDeQuantInter(pParam, pMB->quant, data, qcoeff, cbp);
501 :    
502 :     /* Perform inverse DCT*/
503 :     MBiDCT(data, cbp);
504 :    
505 :     /* Transfer back the data -- Add the data */
506 :     MBTrans16to8(pParam, frame, pMB, x_pos, y_pos, data, 1, cbp);
507 :     }
508 :    
509 :     return(cbp);
510 : edgomez 851 }
511 :    
512 :     /* if sum(diff between field lines) < sum(diff between frame lines), use field dct */
513 :     uint32_t
514 :     MBFieldTest_c(int16_t data[6 * 64])
515 :     {
516 : edgomez 195 const uint8_t blocks[] =
517 :     { 0 * 64, 0 * 64, 0 * 64, 0 * 64, 2 * 64, 2 * 64, 2 * 64, 2 * 64 };
518 :     const uint8_t lines[] = { 0, 16, 32, 48, 0, 16, 32, 48 };
519 : edgomez 78
520 : h 69 int frame = 0, field = 0;
521 :     int i, j;
522 :    
523 : edgomez 195 for (i = 0; i < 7; ++i) {
524 :     for (j = 0; j < 8; ++j) {
525 :     frame +=
526 : edgomez 1382 abs(data[0 * 64 + (i + 1) * 8 + j] - data[0 * 64 + i * 8 + j]);
527 : edgomez 195 frame +=
528 : edgomez 1382 abs(data[1 * 64 + (i + 1) * 8 + j] - data[1 * 64 + i * 8 + j]);
529 : edgomez 195 frame +=
530 : edgomez 1382 abs(data[2 * 64 + (i + 1) * 8 + j] - data[2 * 64 + i * 8 + j]);
531 : edgomez 195 frame +=
532 : edgomez 1382 abs(data[3 * 64 + (i + 1) * 8 + j] - data[3 * 64 + i * 8 + j]);
533 : h 69
534 : edgomez 195 field +=
535 : edgomez 1382 abs(data[blocks[i + 1] + lines[i + 1] + j] -
536 : edgomez 195 data[blocks[i] + lines[i] + j]);
537 :     field +=
538 : edgomez 1382 abs(data[blocks[i + 1] + lines[i + 1] + 8 + j] -
539 : edgomez 195 data[blocks[i] + lines[i] + 8 + j]);
540 :     field +=
541 : edgomez 1382 abs(data[blocks[i + 1] + 64 + lines[i + 1] + j] -
542 : edgomez 195 data[blocks[i] + 64 + lines[i] + j]);
543 :     field +=
544 : edgomez 1382 abs(data[blocks[i + 1] + 64 + lines[i + 1] + 8 + j] -
545 : edgomez 195 data[blocks[i] + 64 + lines[i] + 8 + j]);
546 : h 69 }
547 :     }
548 :    
549 : edgomez 851 return (frame >= (field + 350));
550 : h 69 }
551 :    
552 :    
553 :     /* deinterlace Y blocks vertically */
554 :    
555 :     #define MOVLINE(X,Y) memcpy(X, Y, sizeof(tmp))
556 : edgomez 1382 #define LINE(X,Y) &data[X*64 + Y*8]
557 : h 69
558 : edgomez 195 void
559 :     MBFrameToField(int16_t data[6 * 64])
560 : h 69 {
561 :     int16_t tmp[8];
562 :    
563 :     /* left blocks */
564 :    
565 : edgomez 1382 /* 1=2, 2=4, 4=8, 8=1 */
566 : edgomez 195 MOVLINE(tmp, LINE(0, 1));
567 :     MOVLINE(LINE(0, 1), LINE(0, 2));
568 :     MOVLINE(LINE(0, 2), LINE(0, 4));
569 :     MOVLINE(LINE(0, 4), LINE(2, 0));
570 :     MOVLINE(LINE(2, 0), tmp);
571 : h 69
572 : edgomez 1382 /* 3=6, 6=12, 12=9, 9=3 */
573 : edgomez 195 MOVLINE(tmp, LINE(0, 3));
574 :     MOVLINE(LINE(0, 3), LINE(0, 6));
575 :     MOVLINE(LINE(0, 6), LINE(2, 4));
576 :     MOVLINE(LINE(2, 4), LINE(2, 1));
577 :     MOVLINE(LINE(2, 1), tmp);
578 : h 69
579 : edgomez 1382 /* 5=10, 10=5 */
580 : edgomez 195 MOVLINE(tmp, LINE(0, 5));
581 :     MOVLINE(LINE(0, 5), LINE(2, 2));
582 :     MOVLINE(LINE(2, 2), tmp);
583 : h 69
584 : edgomez 1382 /* 7=14, 14=13, 13=11, 11=7 */
585 : edgomez 195 MOVLINE(tmp, LINE(0, 7));
586 :     MOVLINE(LINE(0, 7), LINE(2, 6));
587 :     MOVLINE(LINE(2, 6), LINE(2, 5));
588 :     MOVLINE(LINE(2, 5), LINE(2, 3));
589 :     MOVLINE(LINE(2, 3), tmp);
590 : h 69
591 :     /* right blocks */
592 :    
593 : edgomez 1382 /* 1=2, 2=4, 4=8, 8=1 */
594 : edgomez 195 MOVLINE(tmp, LINE(1, 1));
595 :     MOVLINE(LINE(1, 1), LINE(1, 2));
596 :     MOVLINE(LINE(1, 2), LINE(1, 4));
597 :     MOVLINE(LINE(1, 4), LINE(3, 0));
598 :     MOVLINE(LINE(3, 0), tmp);
599 : h 69
600 : edgomez 1382 /* 3=6, 6=12, 12=9, 9=3 */
601 : edgomez 195 MOVLINE(tmp, LINE(1, 3));
602 :     MOVLINE(LINE(1, 3), LINE(1, 6));
603 :     MOVLINE(LINE(1, 6), LINE(3, 4));
604 :     MOVLINE(LINE(3, 4), LINE(3, 1));
605 :     MOVLINE(LINE(3, 1), tmp);
606 : h 69
607 : edgomez 1382 /* 5=10, 10=5 */
608 : edgomez 195 MOVLINE(tmp, LINE(1, 5));
609 :     MOVLINE(LINE(1, 5), LINE(3, 2));
610 :     MOVLINE(LINE(3, 2), tmp);
611 : h 69
612 : edgomez 1382 /* 7=14, 14=13, 13=11, 11=7 */
613 : edgomez 195 MOVLINE(tmp, LINE(1, 7));
614 :     MOVLINE(LINE(1, 7), LINE(3, 6));
615 :     MOVLINE(LINE(3, 6), LINE(3, 5));
616 :     MOVLINE(LINE(3, 5), LINE(3, 3));
617 :     MOVLINE(LINE(3, 3), tmp);
618 : h 69 }
619 : edgomez 1382
620 :     /*****************************************************************************
621 :     * Trellis based R-D optimal quantization
622 :     *
623 :     * Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
624 :     *
625 :     ****************************************************************************/
626 :    
627 :     /*----------------------------------------------------------------------------
628 :     *
629 :     * Trellis-Based quantization
630 :     *
631 :     * So far I understand this paper:
632 :     *
633 :     * "Trellis-Based R-D Optimal Quantization in H.263+"
634 :     * J.Wen, M.Luttrell, J.Villasenor
635 :     * IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
636 :     *
637 :     * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
638 :     * Source Shortest Path algo. But due to the underlying graph structure
639 :     * ("Trellis"), it can be turned into a dynamic programming algo,
640 :     * partially saving the explicit graph's nodes representation. And
641 :     * without using a heap, since the open frontier of the DAG is always
642 :     * known, and of fixed size.
643 :     *--------------------------------------------------------------------------*/
644 :    
645 :    
646 :    
647 :     /* Codes lengths for relevant levels. */
648 :    
649 :     /* let's factorize: */
650 :     static const uint8_t Code_Len0[64] = {
651 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
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_Len1[64] = {
654 :     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,
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_Len2[64] = {
657 :     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,
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_Len3[64] = {
660 :     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,
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_Len4[64] = {
663 :     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,
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_Len5[64] = {
666 :     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,
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_Len6[64] = {
669 :     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,
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_Len7[64] = {
672 :     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,
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_Len8[64] = {
675 :     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,
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_Len9[64] = {
678 :     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,
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_Len10[64] = {
681 :     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,
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_Len11[64] = {
684 :     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,
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_Len12[64] = {
687 :     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,
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_Len13[64] = {
690 :     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,
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_Len14[64] = {
693 :     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,
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_Len15[64] = {
696 :     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,
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_Len16[64] = {
699 :     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,
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_Len17[64] = {
702 :     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,
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_Len18[64] = {
705 :     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,
706 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
707 :     static const uint8_t Code_Len19[64] = {
708 :     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,
709 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
710 :     static const uint8_t Code_Len20[64] = {
711 :     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,
712 :     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 };
713 :    
714 :     /* a few more table for LAST table: */
715 :     static const uint8_t Code_Len21[64] = {
716 :     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,
717 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
718 :     static const uint8_t Code_Len22[64] = {
719 :     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,
720 :     30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
721 :     static const uint8_t Code_Len23[64] = {
722 :     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,
723 :     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};
724 :     static const uint8_t Code_Len24[64] = {
725 :     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,
726 :     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};
727 :    
728 :    
729 :     static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */
730 :     Code_Len20,Code_Len19,Code_Len18,Code_Len17,
731 :     Code_Len16,Code_Len15,Code_Len14,Code_Len13,
732 :     Code_Len12,Code_Len11,Code_Len10,Code_Len9,
733 :     Code_Len8, Code_Len7 ,Code_Len6 ,Code_Len5,
734 :     Code_Len4, Code_Len3, Code_Len3 ,Code_Len2,
735 :     Code_Len2, Code_Len1, Code_Len1, Code_Len1,
736 :     };
737 :    
738 :     static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */
739 :     Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
740 :     };
741 :    
742 :     /* TL_SHIFT controls the precision of the RD optimizations in trellis
743 :     * valid range is [10..16]. The bigger, the more trellis is vulnerable
744 :     * to overflows in cost formulas.
745 :     * - 10 allows ac values up to 2^11 == 2048
746 :     * - 16 allows ac values up to 2^8 == 256
747 :     */
748 :     #define TL_SHIFT 11
749 :     #define TL(q) ((0xfe00>>(16-TL_SHIFT))/(q*q))
750 :    
751 :     static const int Trellis_Lambda_Tabs[31] = {
752 :     TL( 1),TL( 2),TL( 3),TL( 4),TL( 5),TL( 6), TL( 7),
753 :     TL( 8),TL( 9),TL(10),TL(11),TL(12),TL(13),TL(14), TL(15),
754 :     TL(16),TL(17),TL(18),TL(19),TL(20),TL(21),TL(22), TL(23),
755 :     TL(24),TL(25),TL(26),TL(27),TL(28),TL(29),TL(30), TL(31)
756 :     };
757 :     #undef TL
758 :    
759 :     static int __inline
760 :     Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
761 :     {
762 :     while(i>=0)
763 :     if (C[Zigzag[i]])
764 :     return i;
765 :     else i--;
766 :     return -1;
767 :     }
768 :    
769 : syskin 1663 #define TRELLIS_MIN_EFFORT 3
770 :    
771 : Isibaar 1909 static __inline uint32_t calc_mseh(int16_t dQ, uint16_t mask,
772 :     const int index, const int Lambda)
773 :     {
774 :     uint32_t t = (mask * Inv_iMask_Coeff[index] + 32) >> 7;
775 :     uint16_t u = abs(dQ) << 4;
776 :     uint16_t thresh = (t < 65536) ? t : 65535;
777 :    
778 :     if (u <= thresh)
779 :     u = 0; /* The error is not perceivable */
780 :     else
781 :     u -= thresh;
782 :    
783 :     u = ((u + iCSF_Round[index]) * iCSF_Coeff[index]) >> 16;
784 :    
785 :     return (((Lambda*u*u)>>4) + 4*Lambda*dQ*dQ) / 5;
786 :     }
787 :    
788 : edgomez 1382 /* this routine has been strippen of all debug code */
789 :     static int
790 :     dct_quantize_trellis_c(int16_t *const Out,
791 :     const int16_t *const In,
792 :     int Q,
793 :     const uint16_t * const Zigzag,
794 :     const uint16_t * const QuantMatrix,
795 : edgomez 1432 int Non_Zero,
796 : syskin 1660 int Sum,
797 : Isibaar 1909 int Lambda_Mod,
798 :     const uint32_t rel_var8,
799 :     const int Metric)
800 : edgomez 1382 {
801 :    
802 :     /* Note: We should search last non-zero coeffs on *real* DCT input coeffs
803 :     * (In[]), not quantized one (Out[]). However, it only improves the result
804 :     * *very* slightly (~0.01dB), whereas speed drops to crawling level :)
805 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes
806 :     * helps. */
807 :     typedef struct { int16_t Run, Level; } NODE;
808 :    
809 : Skal 1617 NODE Nodes[65], Last = { 0, 0};
810 : edgomez 1382 uint32_t Run_Costs0[64+1];
811 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
812 :    
813 :     /* it's 1/lambda, actually */
814 : syskin 1660 const int Lambda = (Lambda_Mod*Trellis_Lambda_Tabs[Q-1])>>LAMBDA_EXP;
815 : edgomez 1382
816 :     int Run_Start = -1;
817 :     uint32_t Min_Cost = 2<<TL_SHIFT;
818 :    
819 :     int Last_Node = -1;
820 :     uint32_t Last_Cost = 0;
821 :    
822 : edgomez 1432 int i, j;
823 : edgomez 1382
824 : Isibaar 1909 uint32_t mask = (Metric) ? ((isqrt(2*coeff8_energy(In)*rel_var8) + 48) >> 6) : 0;
825 :    
826 : edgomez 1382 /* source (w/ CBP penalty) */
827 :     Run_Costs[-1] = 2<<TL_SHIFT;
828 :    
829 :     Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
830 : syskin 1663 if (Non_Zero < TRELLIS_MIN_EFFORT)
831 :     Non_Zero = TRELLIS_MIN_EFFORT;
832 : edgomez 1382
833 :     for(i=0; i<=Non_Zero; i++) {
834 :     const int q = ((Q*QuantMatrix[Zigzag[i]])>>4);
835 :     const int Mult = 2*q;
836 :     const int Bias = (q-1) | 1;
837 :     const int Lev0 = Mult + Bias;
838 :    
839 :     const int AC = In[Zigzag[i]];
840 :     const int Level1 = Out[Zigzag[i]];
841 : Isibaar 1909 const unsigned int Dist0 = (Metric) ? (calc_mseh(AC, mask, Zigzag[i], Lambda)) : (Lambda* AC*AC);
842 : edgomez 1382 uint32_t Best_Cost = 0xf0000000;
843 :     Last_Cost += Dist0;
844 :    
845 :     /* very specialized loop for -1,0,+1 */
846 :     if ((uint32_t)(Level1+1)<3) {
847 :     int dQ;
848 :     int Run;
849 :     uint32_t Cost0;
850 :    
851 :     if (AC<0) {
852 :     Nodes[i].Level = -1;
853 :     dQ = Lev0 + AC;
854 :     } else {
855 :     Nodes[i].Level = 1;
856 :     dQ = Lev0 - AC;
857 :     }
858 : Isibaar 1909 Cost0 = (Metric) ? (calc_mseh(dQ, mask, Zigzag[i], Lambda)) : (Lambda*dQ*dQ);
859 : edgomez 1382
860 :     Nodes[i].Run = 1;
861 :     Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0;
862 :     for(Run=i-Run_Start; Run>0; --Run) {
863 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
864 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT);
865 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT);
866 :    
867 :     /* TODO: what about tie-breaks? Should we favor short runs or
868 :     * long runs? Although the error is the same, it would not be
869 :     * spread the same way along high and low frequencies... */
870 :    
871 :     /* Gruel: I'd say, favour short runs => hifreq errors (HVS) */
872 :    
873 :     if (Cost<Best_Cost) {
874 :     Best_Cost = Cost;
875 :     Nodes[i].Run = Run;
876 :     }
877 :    
878 :     if (lCost<Last_Cost) {
879 :     Last_Cost = lCost;
880 :     Last.Run = Run;
881 :     Last_Node = i;
882 :     }
883 :     }
884 :     if (Last_Node==i)
885 :     Last.Level = Nodes[i].Level;
886 :     } else if (51U>(uint32_t)(Level1+25)) {
887 :     /* "big" levels (not less than ESC3, though) */
888 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
889 :     int Level2;
890 :     int dQ1, dQ2;
891 :     int Run;
892 :     uint32_t Dist1,Dist2;
893 :     int dDist21;
894 :    
895 :     if (Level1>1) {
896 :     dQ1 = Level1*Mult-AC + Bias;
897 :     dQ2 = dQ1 - Mult;
898 :     Level2 = Level1-1;
899 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
900 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
901 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
902 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
903 :     } else { /* Level1<-1 */
904 :     dQ1 = Level1*Mult-AC - Bias;
905 :     dQ2 = dQ1 + Mult;
906 :     Level2 = Level1 + 1;
907 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
908 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
909 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
910 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
911 :     }
912 :    
913 : Isibaar 1909 if (Metric) {
914 :     Dist1 = calc_mseh(dQ1, mask, Zigzag[i], Lambda);
915 :     Dist2 = calc_mseh(dQ2, mask, Zigzag[i], Lambda);
916 :     }
917 :     else {
918 :     Dist1 = Lambda*dQ1*dQ1;
919 :     Dist2 = Lambda*dQ2*dQ2;
920 :     }
921 : edgomez 1382 dDist21 = Dist2-Dist1;
922 :    
923 :     for(Run=i-Run_Start; Run>0; --Run)
924 :     {
925 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
926 :     uint32_t Cost1, Cost2;
927 :     int bLevel;
928 :    
929 :     /* for sub-optimal (but slightly worth it, speed-wise) search,
930 :     * uncomment the following:
931 :     * if (Cost_Base>=Best_Cost) continue;
932 :     * (? doesn't seem to have any effect -- gruel ) */
933 :    
934 :     Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT);
935 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21;
936 :    
937 :     if (Cost2<Cost1) {
938 :     Cost1 = Cost2;
939 :     bLevel = Level2;
940 :     } else {
941 :     bLevel = Level1;
942 :     }
943 :    
944 :     if (Cost1<Best_Cost) {
945 :     Best_Cost = Cost1;
946 :     Nodes[i].Run = Run;
947 :     Nodes[i].Level = bLevel;
948 :     }
949 :    
950 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT);
951 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21;
952 :    
953 :     if (Cost2<Cost1) {
954 :     Cost1 = Cost2;
955 :     bLevel = Level2;
956 :     } else {
957 :     bLevel = Level1;
958 :     }
959 :    
960 :     if (Cost1<Last_Cost) {
961 :     Last_Cost = Cost1;
962 :     Last.Run = Run;
963 :     Last.Level = bLevel;
964 :     Last_Node = i;
965 :     }
966 :     } /* end of "for Run" */
967 :     } else {
968 :     /* Very very high levels, with no chance of being optimizable
969 :     * => Simply pick best Run. */
970 :     int Run;
971 :     for(Run=i-Run_Start; Run>0; --Run) {
972 :     /* 30 bits + no distortion */
973 :     const uint32_t Cost = (30<<TL_SHIFT) + Run_Costs[i-Run];
974 :     if (Cost<Best_Cost) {
975 :     Best_Cost = Cost;
976 :     Nodes[i].Run = Run;
977 :     Nodes[i].Level = Level1;
978 :     }
979 :    
980 :     if (Cost<Last_Cost) {
981 :     Last_Cost = Cost;
982 :     Last.Run = Run;
983 :     Last.Level = Level1;
984 :     Last_Node = i;
985 :     }
986 :     }
987 :     }
988 :    
989 :    
990 :     Run_Costs[i] = Best_Cost;
991 :    
992 :     if (Best_Cost < Min_Cost + Dist0) {
993 :     Min_Cost = Best_Cost;
994 :     Run_Start = i;
995 :     } else {
996 :     /* as noticed by Michael Niedermayer (michaelni at gmx.at),
997 :     * there's a code shorter by 1 bit for a larger run (!), same
998 :     * level. We give it a chance by not moving the left barrier too
999 :     * much. */
1000 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) )
1001 :     Run_Start++;
1002 :    
1003 :     /* spread on preceding coeffs the cost incurred by skipping this
1004 :     * one */
1005 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1006 :     Min_Cost += Dist0;
1007 :     }
1008 :     }
1009 :    
1010 : edgomez 1432 /* It seems trellis doesn't give good results... just leave the block untouched
1011 :     * and return the original sum value */
1012 : edgomez 1382 if (Last_Node<0)
1013 : edgomez 1432 return Sum;
1014 : edgomez 1382
1015 :     /* reconstruct optimal sequence backward with surviving paths */
1016 :     memset(Out, 0x00, 64*sizeof(*Out));
1017 :     Out[Zigzag[Last_Node]] = Last.Level;
1018 :     i = Last_Node - Last.Run;
1019 : syskin 1454 Sum = abs(Last.Level);
1020 : edgomez 1382 while(i>=0) {
1021 :     Out[Zigzag[i]] = Nodes[i].Level;
1022 : edgomez 1432 Sum += abs(Nodes[i].Level);
1023 : edgomez 1382 i -= Nodes[i].Run;
1024 :     }
1025 :    
1026 : edgomez 1432 return Sum;
1027 : edgomez 1382 }
1028 :    
1029 :     /* original version including heavy debugging info */
1030 :    
1031 :     #ifdef DBGTRELL
1032 :    
1033 :     #define DBG 0
1034 :    
1035 :     static __inline uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,
1036 :     const uint16_t * Zigzag, int Max, int Lambda)
1037 :     {
1038 :     #if (DBG>0)
1039 :     const int16_t * const Ref = C + 6*64;
1040 :     int Last = Max;
1041 :     int Bits = 0;
1042 :     int Dist = 0;
1043 :     int i;
1044 :     uint32_t Cost;
1045 :    
1046 :     while(Last>=0 && C[Zigzag[Last]]==0)
1047 :     Last--;
1048 :    
1049 :     if (Last>=0) {
1050 :     int j=0, j0=0;
1051 :     int Run, Level;
1052 :    
1053 :     Bits = 2; /* CBP */
1054 :     while(j<Last) {
1055 :     while(!C[Zigzag[j]])
1056 :     j++;
1057 :     if (j==Last)
1058 :     break;
1059 :     Level=C[Zigzag[j]];
1060 :     Run = j - j0;
1061 :     j0 = ++j;
1062 :     if (Level>=-24 && Level<=24)
1063 :     Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];
1064 :     else
1065 :     Bits += 30;
1066 :     }
1067 :     Level = C[Zigzag[Last]];
1068 :     Run = j - j0;
1069 :     if (Level>=-6 && Level<=6)
1070 :     Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];
1071 :     else
1072 :     Bits += 30;
1073 :     }
1074 :    
1075 :     for(i=0; i<=Last; ++i) {
1076 :     int V = C[Zigzag[i]]*Mult;
1077 :     if (V>0)
1078 :     V += Bias;
1079 :     else
1080 :     if (V<0)
1081 :     V -= Bias;
1082 :     V -= Ref[Zigzag[i]];
1083 :     Dist += V*V;
1084 :     }
1085 :     Cost = Lambda*Dist + (Bits<<TL_SHIFT);
1086 :     if (DBG==1)
1087 :     printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );
1088 :     return Cost;
1089 :    
1090 :     #else
1091 :     return 0;
1092 :     #endif
1093 :     }
1094 :    
1095 :    
1096 :     static int
1097 :     dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
1098 :     {
1099 :    
1100 :     /*
1101 :     * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
1102 :     * not quantized one (Out[]). However, it only improves the result *very*
1103 :     * slightly (~0.01dB), whereas speed drops to crawling level :)
1104 :     * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
1105 :     */
1106 :     typedef struct { int16_t Run, Level; } NODE;
1107 :    
1108 :     NODE Nodes[65], Last;
1109 :     uint32_t Run_Costs0[64+1];
1110 :     uint32_t * const Run_Costs = Run_Costs0 + 1;
1111 :     const int Mult = 2*Q;
1112 :     const int Bias = (Q-1) | 1;
1113 :     const int Lev0 = Mult + Bias;
1114 :     const int Lambda = Trellis_Lambda_Tabs[Q-1]; /* it's 1/lambda, actually */
1115 :    
1116 :     int Run_Start = -1;
1117 :     Run_Costs[-1] = 2<<TL_SHIFT; /* source (w/ CBP penalty) */
1118 :     uint32_t Min_Cost = 2<<TL_SHIFT;
1119 :    
1120 :     int Last_Node = -1;
1121 :     uint32_t Last_Cost = 0;
1122 :    
1123 :     int i, j;
1124 :    
1125 :     #if (DBG>0)
1126 :     Last.Level = 0; Last.Run = -1; /* just initialize to smthg */
1127 :     #endif
1128 :    
1129 :     Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
1130 :     if (Non_Zero<0)
1131 :     return -1;
1132 :    
1133 :     for(i=0; i<=Non_Zero; i++)
1134 :     {
1135 :     const int AC = In[Zigzag[i]];
1136 :     const int Level1 = Out[Zigzag[i]];
1137 :     const int Dist0 = Lambda* AC*AC;
1138 :     uint32_t Best_Cost = 0xf0000000;
1139 :     Last_Cost += Dist0;
1140 :    
1141 :     if ((uint32_t)(Level1+1)<3) /* very specialized loop for -1,0,+1 */
1142 :     {
1143 :     int dQ;
1144 :     int Run;
1145 :     uint32_t Cost0;
1146 :    
1147 :     if (AC<0) {
1148 :     Nodes[i].Level = -1;
1149 :     dQ = Lev0 + AC;
1150 :     } else {
1151 :     Nodes[i].Level = 1;
1152 :     dQ = Lev0 - AC;
1153 :     }
1154 :     Cost0 = Lambda*dQ*dQ;
1155 :    
1156 :     Nodes[i].Run = 1;
1157 :     Best_Cost = (Code_Len20[0]<<TL_SHIFT) + Run_Costs[i-1]+Cost0;
1158 :     for(Run=i-Run_Start; Run>0; --Run)
1159 :     {
1160 :     const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
1161 :     const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<TL_SHIFT);
1162 :     const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<TL_SHIFT);
1163 :    
1164 :     /*
1165 :     * TODO: what about tie-breaks? Should we favor short runs or
1166 :     * long runs? Although the error is the same, it would not be
1167 :     * spread the same way along high and low frequencies...
1168 :     */
1169 :     if (Cost<Best_Cost) {
1170 :     Best_Cost = Cost;
1171 :     Nodes[i].Run = Run;
1172 :     }
1173 :    
1174 :     if (lCost<Last_Cost) {
1175 :     Last_Cost = lCost;
1176 :     Last.Run = Run;
1177 :     Last_Node = i;
1178 :     }
1179 :     }
1180 :     if (Last_Node==i)
1181 :     Last.Level = Nodes[i].Level;
1182 :    
1183 :     if (DBG==1) {
1184 :     Run_Costs[i] = Best_Cost;
1185 :     printf( "Costs #%2d: ", i);
1186 :     for(j=-1;j<=Non_Zero;++j) {
1187 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1188 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1189 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1190 :     else printf( " - |" );
1191 :     }
1192 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1193 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1194 :     printf( " AC:%3.0d Dist0:%3d Dist(%d)=%d", AC, Dist0>>12, Nodes[i].Level, Cost0>>12 );
1195 :     printf( "\n" );
1196 :     }
1197 :     }
1198 :     else /* "big" levels */
1199 :     {
1200 :     const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
1201 :     int Level2;
1202 :     int dQ1, dQ2;
1203 :     int Run;
1204 :     uint32_t Dist1,Dist2;
1205 :     int dDist21;
1206 :    
1207 :     if (Level1>1) {
1208 :     dQ1 = Level1*Mult-AC + Bias;
1209 :     dQ2 = dQ1 - Mult;
1210 :     Level2 = Level1-1;
1211 :     Tbl_L1 = (Level1<=24) ? B16_17_Code_Len[Level1-1] : Code_Len0;
1212 :     Tbl_L2 = (Level2<=24) ? B16_17_Code_Len[Level2-1] : Code_Len0;
1213 :     Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
1214 :     Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
1215 :     } else { /* Level1<-1 */
1216 :     dQ1 = Level1*Mult-AC - Bias;
1217 :     dQ2 = dQ1 + Mult;
1218 :     Level2 = Level1 + 1;
1219 :     Tbl_L1 = (Level1>=-24) ? B16_17_Code_Len[Level1^-1] : Code_Len0;
1220 :     Tbl_L2 = (Level2>=-24) ? B16_17_Code_Len[Level2^-1] : Code_Len0;
1221 :     Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
1222 :     Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
1223 :     }
1224 :     Dist1 = Lambda*dQ1*dQ1;
1225 :     Dist2 = Lambda*dQ2*dQ2;
1226 :     dDist21 = Dist2-Dist1;
1227 :    
1228 :     for(Run=i-Run_Start; Run>0; --Run)
1229 :     {
1230 :     const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
1231 :     uint32_t Cost1, Cost2;
1232 :     int bLevel;
1233 :    
1234 :     /*
1235 :     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
1236 :     * if (Cost_Base>=Best_Cost) continue;
1237 :     */
1238 :     Cost1 = Cost_Base + (Tbl_L1[Run-1]<<TL_SHIFT);
1239 :     Cost2 = Cost_Base + (Tbl_L2[Run-1]<<TL_SHIFT) + dDist21;
1240 :    
1241 :     if (Cost2<Cost1) {
1242 :     Cost1 = Cost2;
1243 :     bLevel = Level2;
1244 :     } else
1245 :     bLevel = Level1;
1246 :    
1247 :     if (Cost1<Best_Cost) {
1248 :     Best_Cost = Cost1;
1249 :     Nodes[i].Run = Run;
1250 :     Nodes[i].Level = bLevel;
1251 :     }
1252 :    
1253 :     Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<TL_SHIFT);
1254 :     Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<TL_SHIFT) + dDist21;
1255 :    
1256 :     if (Cost2<Cost1) {
1257 :     Cost1 = Cost2;
1258 :     bLevel = Level2;
1259 :     } else
1260 :     bLevel = Level1;
1261 :    
1262 :     if (Cost1<Last_Cost) {
1263 :     Last_Cost = Cost1;
1264 :     Last.Run = Run;
1265 :     Last.Level = bLevel;
1266 :     Last_Node = i;
1267 :     }
1268 :     } /* end of "for Run" */
1269 :    
1270 :     if (DBG==1) {
1271 :     Run_Costs[i] = Best_Cost;
1272 :     printf( "Costs #%2d: ", i);
1273 :     for(j=-1;j<=Non_Zero;++j) {
1274 :     if (j==Run_Start) printf( " %3.0d|", Run_Costs[j]>>12 );
1275 :     else if (j>Run_Start && j<i) printf( " %3.0d|", Run_Costs[j]>>12 );
1276 :     else if (j==i) printf( "(%3.0d)", Run_Costs[j]>>12 );
1277 :     else printf( " - |" );
1278 :     }
1279 :     printf( "<%3.0d %2d %d>", Min_Cost>>12, Nodes[i].Level, Nodes[i].Run );
1280 :     printf( " Last:#%2d {%3.0d %2d %d}", Last_Node, Last_Cost>>12, Last.Level, Last.Run );
1281 :     printf( " AC:%3.0d Dist0:%3d Dist(%2d):%3d Dist(%2d):%3d", AC, Dist0>>12, Level1, Dist1>>12, Level2, Dist2>>12 );
1282 :     printf( "\n" );
1283 :     }
1284 :     }
1285 :    
1286 :     Run_Costs[i] = Best_Cost;
1287 :    
1288 :     if (Best_Cost < Min_Cost + Dist0) {
1289 :     Min_Cost = Best_Cost;
1290 :     Run_Start = i;
1291 :     }
1292 :     else
1293 :     {
1294 :     /*
1295 :     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
1296 :     * a code shorter by 1 bit for a larger run (!), same level. We give
1297 :     * it a chance by not moving the left barrier too much.
1298 :     */
1299 :    
1300 :     while( Run_Costs[Run_Start]>Min_Cost+(1<<TL_SHIFT) )
1301 :     Run_Start++;
1302 :    
1303 :     /* spread on preceding coeffs the cost incurred by skipping this one */
1304 :     for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1305 :     Min_Cost += Dist0;
1306 :     }
1307 :     }
1308 :    
1309 :     if (DBG) {
1310 :     Last_Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1311 :     if (DBG==1) {
1312 :     printf( "=> " );
1313 :     for(i=0; i<=Non_Zero; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1314 :     printf( "\n" );
1315 :     }
1316 :     }
1317 :    
1318 :     if (Last_Node<0)
1319 :     return -1;
1320 :    
1321 :     /* reconstruct optimal sequence backward with surviving paths */
1322 :     memset(Out, 0x00, 64*sizeof(*Out));
1323 :     Out[Zigzag[Last_Node]] = Last.Level;
1324 :     i = Last_Node - Last.Run;
1325 :     while(i>=0) {
1326 :     Out[Zigzag[i]] = Nodes[i].Level;
1327 :     i -= Nodes[i].Run;
1328 :     }
1329 :    
1330 :     if (DBG) {
1331 :     uint32_t Cost = Evaluate_Cost(Out,Mult,Bias, Zigzag,Non_Zero, Lambda);
1332 :     if (DBG==1) {
1333 :     printf( "<= " );
1334 :     for(i=0; i<=Last_Node; ++i) printf( "[%3.0d] ", Out[Zigzag[i]] );
1335 :     printf( "\n--------------------------------\n" );
1336 :     }
1337 :     if (Cost>Last_Cost) printf( "!!! %u > %u\n", Cost, Last_Cost );
1338 :     }
1339 :     return Last_Node;
1340 :     }
1341 :    
1342 :     #undef DBG
1343 :    
1344 :     #endif

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