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

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

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.21.2.12, Mon May 12 12:33:16 2003 UTC revision 1.21.2.18, Tue Oct 7 13:02:35 2003 UTC
# Line 39  Line 39 
39  #include "../bitstream/zigzag.h"  #include "../bitstream/zigzag.h"
40  #include "../dct/fdct.h"  #include "../dct/fdct.h"
41  #include "../dct/idct.h"  #include "../dct/idct.h"
42  #include "../quant/quant_mpeg4.h"  #include "../quant/quant.h"
 #include "../quant/quant_h263.h"  
43  #include "../encoder.h"  #include "../encoder.h"
44    
45  #include "../image/reduced.h"  #include "../image/reduced.h"
# Line 123  Line 122 
122                           int16_t qcoeff[6 * 64],                           int16_t qcoeff[6 * 64],
123                           int16_t data[6*64])                           int16_t data[6*64])
124  {  {
125          int i;          int mpeg;
126            int scaler_lum, scaler_chr;
127    
128          for (i = 0; i < 6; i++) {          quant_intraFuncPtr const quant[2] =
129                  uint32_t iDcScaler = get_dc_scaler(pMB->quant, i < 4);                  {
130                            quant_h263_intra,
131                            quant_mpeg_intra
132                    };
133    
134            mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
135            scaler_lum = get_dc_scaler(pMB->quant, 1);
136            scaler_chr = get_dc_scaler(pMB->quant, 0);
137    
138                  /* Quantize the block */                  /* Quantize the block */
139                  start_timer();                  start_timer();
140                  if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) {          quant[mpeg](&data[0 * 64], &qcoeff[0 * 64], pMB->quant, scaler_lum);
141                          quant_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler);          quant[mpeg](&data[1 * 64], &qcoeff[1 * 64], pMB->quant, scaler_lum);
142                  } else {          quant[mpeg](&data[2 * 64], &qcoeff[2 * 64], pMB->quant, scaler_lum);
143                          quant4_intra(&data[i * 64], &qcoeff[i * 64], pMB->quant, iDcScaler);          quant[mpeg](&data[3 * 64], &qcoeff[3 * 64], pMB->quant, scaler_lum);
144                  }          quant[mpeg](&data[4 * 64], &qcoeff[4 * 64], pMB->quant, scaler_chr);
145            quant[mpeg](&data[5 * 64], &qcoeff[5 * 64], pMB->quant, scaler_chr);
146                  stop_quant_timer();                  stop_quant_timer();
147          }          }
 }  
148    
149  /* DeQuantize all blocks -- Intra mode */  /* DeQuantize all blocks -- Intra mode */
150  static __inline void  static __inline void
# Line 146  Line 153 
153                             int16_t qcoeff[6 * 64],                             int16_t qcoeff[6 * 64],
154                             int16_t data[6*64])                             int16_t data[6*64])
155  {  {
156          int i;          int mpeg;
157            int scaler_lum, scaler_chr;
158    
159          for (i = 0; i < 6; i++) {          quant_intraFuncPtr const dequant[2] =
160                  uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4);                  {
161                            dequant_h263_intra,
162                            dequant_mpeg_intra
163                    };
164    
165            mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
166            scaler_lum = get_dc_scaler(iQuant, 1);
167            scaler_chr = get_dc_scaler(iQuant, 0);
168    
169                  start_timer();                  start_timer();
170                  if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT))          dequant[mpeg](&qcoeff[0 * 64], &data[0 * 64], iQuant, scaler_lum);
171                          dequant_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler);          dequant[mpeg](&qcoeff[1 * 64], &data[1 * 64], iQuant, scaler_lum);
172                  else          dequant[mpeg](&qcoeff[2 * 64], &data[2 * 64], iQuant, scaler_lum);
173                          dequant4_intra(&qcoeff[i * 64], &data[i * 64], iQuant, iDcScaler);          dequant[mpeg](&qcoeff[3 * 64], &data[3 * 64], iQuant, scaler_lum);
174            dequant[mpeg](&qcoeff[4 * 64], &data[4 * 64], iQuant, scaler_chr);
175            dequant[mpeg](&qcoeff[5 * 64], &data[5 * 64], iQuant, scaler_chr);
176                  stop_iquant_timer();                  stop_iquant_timer();
177          }          }
 }  
178    
179    
180  static int  typedef int (*trellis_func_ptr_t)(int16_t *const Out,
181  dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero);                                                                    const int16_t *const In,
182                                                                      int Q,
183                                                                      const uint16_t * const Zigzag,
184                                                                      int Non_Zero);
185    
186  static int  static int
187  dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero);  dct_quantize_trellis_h263_c(int16_t *const Out,
188                                                            const int16_t *const In,
189                                                            int Q,
190                                                            const uint16_t * const Zigzag,
191                                                            int Non_Zero);
192    
193    static int
194    dct_quantize_trellis_mpeg_c(int16_t *const Out,
195                                                            const int16_t *const In,
196                                                            int Q,
197                                                            const uint16_t * const Zigzag,
198                                                            int Non_Zero);
199    
200  /* Quantize all blocks -- Inter mode */  /* Quantize all blocks -- Inter mode */
201  static __inline uint8_t  static __inline uint8_t
# Line 182  Line 211 
211          int i;          int i;
212          uint8_t cbp = 0;          uint8_t cbp = 0;
213          int sum;          int sum;
214          int code_block;          int code_block, mpeg;
215    
216            quant_interFuncPtr const quant[2] =
217                    {
218                            quant_h263_inter,
219                            quant_mpeg_inter
220                    };
221    
222            trellis_func_ptr_t const trellis[2] =
223                    {
224                            dct_quantize_trellis_h263_c,
225                            dct_quantize_trellis_mpeg_c
226                    };
227    
228            mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
229    
230          for (i = 0; i < 6; i++) {          for (i = 0; i < 6; i++) {
231    
232                  /* Quantize the block */                  /* Quantize the block */
233                  start_timer();                  start_timer();
234                  if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT)) {  
235                          sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant);                  sum = quant[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant);
236                          if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) {  
237                                  sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1;                  if(sum && (frame->vop_flags & XVID_VOP_TRELLISQUANT)) {
238                                  limit = 1;                          sum = trellis[mpeg](&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63);
                         }  
                 } else {  
                         sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant);  
 //                      if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) )  
 //                              sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1;  
239                  }                  }
240                  stop_quant_timer();                  stop_quant_timer();
241    
# Line 236  Line 274 
274                             int16_t qcoeff[6 * 64],                             int16_t qcoeff[6 * 64],
275                             const uint8_t cbp)                             const uint8_t cbp)
276  {  {
277          int i;          int mpeg;
278    
279            quant_interFuncPtr const dequant[2] =
280                    {
281                            dequant_h263_inter,
282                            dequant_mpeg_inter
283                    };
284    
285            mpeg = !!(pParam->vol_flags & XVID_VOL_MPEGQUANT);
286    
         for (i = 0; i < 6; i++) {  
                 if (cbp & (1 << (5 - i))) {  
287                          start_timer();                          start_timer();
288                          if (!(pParam->vol_flags & XVID_VOL_MPEGQUANT))          if(cbp & (1 << (5 - 0))) dequant[mpeg](&data[0 * 64], &qcoeff[0 * 64], iQuant);
289                                  dequant_inter(&data[i * 64], &qcoeff[i * 64], iQuant);          if(cbp & (1 << (5 - 1))) dequant[mpeg](&data[1 * 64], &qcoeff[1 * 64], iQuant);
290                          else          if(cbp & (1 << (5 - 2))) dequant[mpeg](&data[2 * 64], &qcoeff[2 * 64], iQuant);
291                                  dequant4_inter(&data[i * 64], &qcoeff[i * 64], iQuant);          if(cbp & (1 << (5 - 3))) dequant[mpeg](&data[3 * 64], &qcoeff[3 * 64], iQuant);
292            if(cbp & (1 << (5 - 4))) dequant[mpeg](&data[4 * 64], &qcoeff[4 * 64], iQuant);
293            if(cbp & (1 << (5 - 5))) dequant[mpeg](&data[5 * 64], &qcoeff[5 * 64], iQuant);
294                          stop_iquant_timer();                          stop_iquant_timer();
295                  }                  }
         }  
 }  
296    
297  typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);  typedef void (transfer_operation_8to16_t) (int16_t *Dst, const uint8_t *Src, int BpS);
298  typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);  typedef void (transfer_operation_16to8_t) (uint8_t *Dst, const int16_t *Src, int BpS);
# Line 266  Line 310 
310          uint32_t stride2 = stride / 2;          uint32_t stride2 = stride / 2;
311          uint32_t next_block = stride * 8;          uint32_t next_block = stride * 8;
312          int32_t cst;          int32_t cst;
313            int vop_reduced;
314          uint8_t *pY_Cur, *pU_Cur, *pV_Cur;          uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
315          const IMAGE * const pCurrent = &frame->image;          const IMAGE * const pCurrent = &frame->image;
316            transfer_operation_8to16_t * const functions[2] =
317                    {
318                            (transfer_operation_8to16_t *)transfer_8to16copy,
319                            (transfer_operation_8to16_t *)filter_18x18_to_8x8
320                    };
321          transfer_operation_8to16_t *transfer_op = NULL;          transfer_operation_8to16_t *transfer_op = NULL;
322    
323          if ((frame->vop_flags & XVID_VOP_REDUCED)) {          vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED);
   
                 /* Image pointers */  
                 pY_Cur = pCurrent->y + (y_pos << 5) * stride  + (x_pos << 5);  
                 pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4);  
                 pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4);  
   
                 /* Block size */  
                 cst = 16;  
   
                 /* Operation function */  
                 transfer_op = (transfer_operation_8to16_t*)filter_18x18_to_8x8;  
         } else {  
324    
325                  /* Image pointers */                  /* Image pointers */
326                  pY_Cur = pCurrent->y + (y_pos << 4) * stride  + (x_pos << 4);          pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride  + (x_pos << (4+vop_reduced));
327                  pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);          pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
328                  pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);          pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
329    
330                  /* Block size */                  /* Block size */
331                  cst = 8;          cst = 8<<vop_reduced;
332    
333                  /* Operation function */                  /* Operation function */
334                  transfer_op = (transfer_operation_8to16_t*)transfer_8to16copy;          transfer_op = functions[vop_reduced];
         }  
335    
336          /* Do the transfer */          /* Do the transfer */
337          start_timer();          start_timer();
# Line 314  Line 351 
351                           const uint32_t x_pos,                           const uint32_t x_pos,
352                           const uint32_t y_pos,                           const uint32_t y_pos,
353                           int16_t data[6 * 64],                           int16_t data[6 * 64],
354                           const uint32_t add,                           const uint32_t add, /* Must be 1 or 0 */
355                           const uint8_t cbp)                           const uint8_t cbp)
356  {  {
357          uint8_t *pY_Cur, *pU_Cur, *pV_Cur;          uint8_t *pY_Cur, *pU_Cur, *pV_Cur;
# Line 322  Line 359 
359          uint32_t stride2 = stride / 2;          uint32_t stride2 = stride / 2;
360          uint32_t next_block = stride * 8;          uint32_t next_block = stride * 8;
361          uint32_t cst;          uint32_t cst;
362            int vop_reduced;
363          const IMAGE * const pCurrent = &frame->image;          const IMAGE * const pCurrent = &frame->image;
364    
365            /* Array of function pointers, indexed by [vop_reduced<<1+add] */
366            transfer_operation_16to8_t  * const functions[4] =
367                    {
368                            (transfer_operation_16to8_t*)transfer_16to8copy,
369                            (transfer_operation_16to8_t*)transfer_16to8add,
370                            (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8,
371                            (transfer_operation_16to8_t*)add_upsampled_8x8_16to8
372                    };
373    
374          transfer_operation_16to8_t *transfer_op = NULL;          transfer_operation_16to8_t *transfer_op = NULL;
375    
376          if (pMB->field_dct) {          if (pMB->field_dct) {
# Line 330  Line 378 
378                  stride *= 2;                  stride *= 2;
379          }          }
380    
381          if ((frame->vop_flags & XVID_VOP_REDUCED)) {          /* Makes this vars booleans */
382            vop_reduced = !!(frame->vop_flags & XVID_VOP_REDUCED);
383    
384                  /* Image pointers */                  /* Image pointers */
385                  pY_Cur = pCurrent->y + (y_pos << 5) * stride  + (x_pos << 5);          pY_Cur = pCurrent->y + (y_pos << (4+vop_reduced)) * stride  + (x_pos << (4+vop_reduced));
386                  pU_Cur = pCurrent->u + (y_pos << 4) * stride2 + (x_pos << 4);          pU_Cur = pCurrent->u + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
387                  pV_Cur = pCurrent->v + (y_pos << 4) * stride2 + (x_pos << 4);          pV_Cur = pCurrent->v + (y_pos << (3+vop_reduced)) * stride2 + (x_pos << (3+vop_reduced));
388    
389                  /* Block size */                  /* Block size */
390                  cst = 16;          cst = 8<<vop_reduced;
391    
392                  /* Operation function */                  /* Operation function */
393                  if(add)          transfer_op = functions[(vop_reduced<<1) + add];
                         transfer_op = (transfer_operation_16to8_t*)add_upsampled_8x8_16to8;  
                 else  
                         transfer_op = (transfer_operation_16to8_t*)copy_upsampled_8x8_16to8;  
         } else {  
   
                 /* Image pointers */  
                 pY_Cur = pCurrent->y + (y_pos << 4) * stride  + (x_pos << 4);  
                 pU_Cur = pCurrent->u + (y_pos << 3) * stride2 + (x_pos << 3);  
                 pV_Cur = pCurrent->v + (y_pos << 3) * stride2 + (x_pos << 3);  
   
                 /* Block size */  
                 cst = 8;  
   
                 /* Operation function */  
                 if(add)  
                         transfer_op = (transfer_operation_16to8_t*)transfer_16to8add;  
                 else  
                         transfer_op = (transfer_operation_16to8_t*)transfer_16to8copy;  
         }  
394    
395          /* Do the operation */          /* Do the operation */
396          start_timer();          start_timer();
# Line 419  Line 449 
449          uint8_t cbp;          uint8_t cbp;
450          uint32_t limit;          uint32_t limit;
451    
452          /*          /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
453           * There is no MBTrans8to16 for Inter block, that's done in motion compensation           * already */
          * already  
          */  
454    
455          /* Perform DCT (and field decision) */          /* Perform DCT (and field decision) */
456          MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);          MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
# Line 430  Line 458 
458          /* Set the limit threshold */          /* Set the limit threshold */
459          limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);          limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
460    
461            if (frame->vop_flags & XVID_VOP_CARTOON)
462                    limit *= 3;
463    
464          /* Quantize the block */          /* Quantize the block */
465          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
466    
# Line 457  Line 488 
488          uint8_t cbp;          uint8_t cbp;
489          uint32_t limit;          uint32_t limit;
490    
491          /*          /* There is no MBTrans8to16 for Inter block, that's done in motion compensation
492           * There is no MBTrans8to16 for Inter block, that's done in motion compensation           * already */
          * already  
          */  
493    
494          /* Perform DCT (and field decision) */          /* Perform DCT (and field decision) */
495          MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);          MBfDCT(pParam, frame, pMB, x_pos, y_pos, data);
# Line 468  Line 497 
497          /* Set the limit threshold */          /* Set the limit threshold */
498          limit = BVOP_TOOSMALL_LIMIT;          limit = BVOP_TOOSMALL_LIMIT;
499    
500            if (frame->vop_flags & XVID_VOP_CARTOON)
501                    limit *= 2;
502    
503          /* Quantize the block */          /* Quantize the block */
504          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
505    
# Line 475  Line 507 
507           * History comment:           * History comment:
508           * We don't have to DeQuant, iDCT and Transfer back data for B-frames.           * We don't have to DeQuant, iDCT and Transfer back data for B-frames.
509           *           *
510           * BUT some plugins require the original frame to be passed so we have           * BUT some plugins require the rebuilt original frame to be passed so we
511           * to take care of that here           * have to take care of that here
512           */           */
513          if((pParam->plugin_flags & XVID_REQORIGINAL)) {          if((pParam->plugin_flags & XVID_REQORIGINAL)) {
514    
# Line 546  Line 578 
578    
579          /* left blocks */          /* left blocks */
580    
581          // 1=2, 2=4, 4=8, 8=1          /* 1=2, 2=4, 4=8, 8=1 */
582          MOVLINE(tmp, LINE(0, 1));          MOVLINE(tmp, LINE(0, 1));
583          MOVLINE(LINE(0, 1), LINE(0, 2));          MOVLINE(LINE(0, 1), LINE(0, 2));
584          MOVLINE(LINE(0, 2), LINE(0, 4));          MOVLINE(LINE(0, 2), LINE(0, 4));
585          MOVLINE(LINE(0, 4), LINE(2, 0));          MOVLINE(LINE(0, 4), LINE(2, 0));
586          MOVLINE(LINE(2, 0), tmp);          MOVLINE(LINE(2, 0), tmp);
587    
588          // 3=6, 6=12, 12=9, 9=3          /* 3=6, 6=12, 12=9, 9=3 */
589          MOVLINE(tmp, LINE(0, 3));          MOVLINE(tmp, LINE(0, 3));
590          MOVLINE(LINE(0, 3), LINE(0, 6));          MOVLINE(LINE(0, 3), LINE(0, 6));
591          MOVLINE(LINE(0, 6), LINE(2, 4));          MOVLINE(LINE(0, 6), LINE(2, 4));
592          MOVLINE(LINE(2, 4), LINE(2, 1));          MOVLINE(LINE(2, 4), LINE(2, 1));
593          MOVLINE(LINE(2, 1), tmp);          MOVLINE(LINE(2, 1), tmp);
594    
595          // 5=10, 10=5          /* 5=10, 10=5 */
596          MOVLINE(tmp, LINE(0, 5));          MOVLINE(tmp, LINE(0, 5));
597          MOVLINE(LINE(0, 5), LINE(2, 2));          MOVLINE(LINE(0, 5), LINE(2, 2));
598          MOVLINE(LINE(2, 2), tmp);          MOVLINE(LINE(2, 2), tmp);
599    
600          // 7=14, 14=13, 13=11, 11=7          /* 7=14, 14=13, 13=11, 11=7 */
601          MOVLINE(tmp, LINE(0, 7));          MOVLINE(tmp, LINE(0, 7));
602          MOVLINE(LINE(0, 7), LINE(2, 6));          MOVLINE(LINE(0, 7), LINE(2, 6));
603          MOVLINE(LINE(2, 6), LINE(2, 5));          MOVLINE(LINE(2, 6), LINE(2, 5));
# Line 574  Line 606 
606    
607          /* right blocks */          /* right blocks */
608    
609          // 1=2, 2=4, 4=8, 8=1          /* 1=2, 2=4, 4=8, 8=1 */
610          MOVLINE(tmp, LINE(1, 1));          MOVLINE(tmp, LINE(1, 1));
611          MOVLINE(LINE(1, 1), LINE(1, 2));          MOVLINE(LINE(1, 1), LINE(1, 2));
612          MOVLINE(LINE(1, 2), LINE(1, 4));          MOVLINE(LINE(1, 2), LINE(1, 4));
613          MOVLINE(LINE(1, 4), LINE(3, 0));          MOVLINE(LINE(1, 4), LINE(3, 0));
614          MOVLINE(LINE(3, 0), tmp);          MOVLINE(LINE(3, 0), tmp);
615    
616          // 3=6, 6=12, 12=9, 9=3          /* 3=6, 6=12, 12=9, 9=3 */
617          MOVLINE(tmp, LINE(1, 3));          MOVLINE(tmp, LINE(1, 3));
618          MOVLINE(LINE(1, 3), LINE(1, 6));          MOVLINE(LINE(1, 3), LINE(1, 6));
619          MOVLINE(LINE(1, 6), LINE(3, 4));          MOVLINE(LINE(1, 6), LINE(3, 4));
620          MOVLINE(LINE(3, 4), LINE(3, 1));          MOVLINE(LINE(3, 4), LINE(3, 1));
621          MOVLINE(LINE(3, 1), tmp);          MOVLINE(LINE(3, 1), tmp);
622    
623          // 5=10, 10=5          /* 5=10, 10=5 */
624          MOVLINE(tmp, LINE(1, 5));          MOVLINE(tmp, LINE(1, 5));
625          MOVLINE(LINE(1, 5), LINE(3, 2));          MOVLINE(LINE(1, 5), LINE(3, 2));
626          MOVLINE(LINE(3, 2), tmp);          MOVLINE(LINE(3, 2), tmp);
627    
628          // 7=14, 14=13, 13=11, 11=7          /* 7=14, 14=13, 13=11, 11=7 */
629          MOVLINE(tmp, LINE(1, 7));          MOVLINE(tmp, LINE(1, 7));
630          MOVLINE(LINE(1, 7), LINE(3, 6));          MOVLINE(LINE(1, 7), LINE(3, 6));
631          MOVLINE(LINE(3, 6), LINE(3, 5));          MOVLINE(LINE(3, 6), LINE(3, 5));
# Line 601  Line 633 
633          MOVLINE(LINE(3, 3), tmp);          MOVLINE(LINE(3, 3), tmp);
634  }  }
635    
636    /*****************************************************************************
637     *               Trellis based R-D optimal quantization
638     *
639     *   Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
640     *
641     ****************************************************************************/
642    
643    /*----------------------------------------------------------------------------
644     *
645     *        Trellis-Based quantization
646     *
647     * So far I understand this paper:
648     *
649     *  "Trellis-Based R-D Optimal Quantization in H.263+"
650     *    J.Wen, M.Luttrell, J.Villasenor
651     *    IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
652     *
653     * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
654     * Source Shorted Path algo. But due to the underlying graph structure
655     * ("Trellis"), it can be turned into a dynamic programming algo,
656     * partially saving the explicit graph's nodes representation. And
657     * without using a heap, since the open frontier of the DAG is always
658     * known, and of fixed sized.
659     *--------------------------------------------------------------------------*/
660    
661    
662    
663  /************************************************************************  /* Codes lengths for relevant levels. */
  *               Trellis based R-D optimal quantization                 *  
  *                                                                      *  
  *   Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net  *  
  *                                                                      *  
  ************************************************************************/  
   
   
 static int  
 dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q,  
                 const uint16_t * const Zigzag, int Non_Zero)  
 { return 63; }  
   
   
 //////////////////////////////////////////////////////////  
 //  
 //        Trellis-Based quantization  
 //  
 // So far I understand this paper:  
 //  
 //  "Trellis-Based R-D Optimal Quantization in H.263+"  
 //    J.Wen, M.Luttrell, J.Villasenor  
 //    IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.  
 //  
 // we are at stake with a simplified Bellmand-Ford / Dijkstra Single  
 // Source Shorted Path algo. But due to the underlying graph structure  
 // ("Trellis"), it can be turned into a dynamic programming algo,  
 // partially saving the explicit graph's nodes representation. And  
 // without using a heap, since the open frontier of the DAG is always  
 // known, and of fixed sized.  
 //  
 //////////////////////////////////////////////////////////  
   
   
 //////////////////////////////////////////////////////////  
 // Codes lengths for relevant levels.  
664    
665    // let's factorize:  /* let's factorize: */
666  static const uint8_t Code_Len0[64] = {  static const uint8_t Code_Len0[64] = {
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,    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
668    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
# Line 707  Line 727 
727     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,     3, 4, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 9, 9,10,10,10,10,10,10,10,10,12,12,13,13,12,13,14,15,15,
728    15,16,16,16,16,17,17,17,18,18,19,19,19,19,19,19,19,19,21,21,22,22,30,30,30,30,30,30,30,30,30,30 };    15,16,16,16,16,17,17,17,18,18,19,19,19,19,19,19,19,19,21,21,22,22,30,30,30,30,30,30,30,30,30,30 };
729    
730    // a few more table for LAST table:  /* a few more table for LAST table: */
731  static const uint8_t Code_Len21[64] = {  static const uint8_t Code_Len21[64] = {
732    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,    13,20,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
733    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
# Line 722  Line 742 
742    12,13,13,13,13,13,13,13,13,14,16,16,16,16,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,19,19};    12,13,13,13,13,13,13,13,13,14,16,16,16,16,17,17,17,17,18,18,18,18,18,18,18,18,19,19,19,19,19,19};
743    
744    
745  static const uint8_t * const B16_17_Code_Len[24] = { // levels [1..24]  static const uint8_t * const B16_17_Code_Len[24] = { /* levels [1..24] */
746    Code_Len20,Code_Len19,Code_Len18,Code_Len17,    Code_Len20,Code_Len19,Code_Len18,Code_Len17,
747    Code_Len16,Code_Len15,Code_Len14,Code_Len13,    Code_Len16,Code_Len15,Code_Len14,Code_Len13,
748    Code_Len12,Code_Len11,Code_Len10,Code_Len9,    Code_Len12,Code_Len11,Code_Len10,Code_Len9,
# Line 731  Line 751 
751    Code_Len2, Code_Len1, Code_Len1, Code_Len1,    Code_Len2, Code_Len1, Code_Len1, Code_Len1,
752  };  };
753    
754  static const uint8_t * const B16_17_Code_Len_Last[6] = { // levels [1..6]  static const uint8_t * const B16_17_Code_Len_Last[6] = { /* levels [1..6] */
755    Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,    Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
756  };  };
757    
# Line 745  Line 765 
765  };  };
766  #undef TL  #undef TL
767    
768  static __inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)  static int __inline
769    Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
770  {  {
771    while(i>=0)    while(i>=0)
772      if (C[Zigzag[i]])      if (C[Zigzag[i]])
# Line 754  Line 775 
775    return -1;    return -1;
776  }  }
777    
778  //////////////////////////////////////////////////////////  static int __inline
779  // this routine has been strippen of all debug code  Compute_Sum(const int16_t *C, int last)
780  //////////////////////////////////////////////////////////  {
781            int sum = 0;
782    
783            while(last--)
784                    sum += abs(C[last]);
785    
786            return(sum);
787    }
788    /* this routine has been strippen of all debug code */
789    
790  static int  static int
791  dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)  dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
792  {  {
793    
794      // Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),      /*
795      // not quantized one (Out[]). However, it only improves the result *very*           * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
796      // slightly (~0.01dB), whereas speed drops to crawling level :)           * not quantized one (Out[]). However, it only improves the result *very*
797      // Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps,           * slightly (~0.01dB), whereas speed drops to crawling level :)
798             * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
799             */
800    typedef struct { int16_t Run, Level; } NODE;    typedef struct { int16_t Run, Level; } NODE;
801    
802    NODE Nodes[65], Last;    NODE Nodes[65], Last;
# Line 775  Line 805 
805    const int Mult = 2*Q;    const int Mult = 2*Q;
806    const int Bias = (Q-1) | 1;    const int Bias = (Q-1) | 1;
807    const int Lev0 = Mult + Bias;    const int Lev0 = Mult + Bias;
808    const int Lambda = Trellis_Lambda_Tabs[Q-1];    // it's 1/lambda, actually          const int Lambda = Trellis_Lambda_Tabs[Q-1];    /* it's 1/lambda, actually */
809    
810    int Run_Start = -1;    int Run_Start = -1;
811    uint32_t Min_Cost = 2<<16;    uint32_t Min_Cost = 2<<16;
# Line 783  Line 813 
813    int Last_Node = -1;    int Last_Node = -1;
814    uint32_t Last_Cost = 0;    uint32_t Last_Cost = 0;
815    
816    int i, j;          int i, j, sum;
817    Run_Costs[-1] = 2<<16;                          // source (w/ CBP penalty)          Run_Costs[-1] = 2<<16;                          /* source (w/ CBP penalty) */
818    
819    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
820    if (Non_Zero<0)    if (Non_Zero<0)
821        return -1;                  return 0; /* Sum is zero if there are only zero coeffs */
822    
823    for(i=0; i<=Non_Zero; i++)          for(i=0; i<=Non_Zero; i++) {
   {  
824      const int AC = In[Zigzag[i]];      const int AC = In[Zigzag[i]];
825      const int Level1 = Out[Zigzag[i]];      const int Level1 = Out[Zigzag[i]];
826      const int Dist0 = Lambda* AC*AC;      const int Dist0 = Lambda* AC*AC;
827      uint32_t Best_Cost = 0xf0000000;      uint32_t Best_Cost = 0xf0000000;
828      Last_Cost += Dist0;      Last_Cost += Dist0;
829    
830      if ((uint32_t)(Level1+1)<3)                 // very specialized loop for -1,0,+1                  /* very specialized loop for -1,0,+1 */
831      {                  if ((uint32_t)(Level1+1)<3) {
832          int dQ;          int dQ;
833                  int Run;                  int Run;
834        uint32_t Cost0;        uint32_t Cost0;
# Line 815  Line 844 
844    
845        Nodes[i].Run = 1;        Nodes[i].Run = 1;
846        Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;        Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
847        for(Run=i-Run_Start; Run>0; --Run)                          for(Run=i-Run_Start; Run>0; --Run) {
       {  
848          const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];          const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
849          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
850          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
851    
852            // TODO: what about tie-breaks? Should we favor short runs or                                  /*
853            // long runs? Although the error is the same, it would not be                                   * TODO: what about tie-breaks? Should we favor short runs or
854            // spread the same way along high and low frequencies...                                   * long runs? Although the error is the same, it would not be
855                                     * spread the same way along high and low frequencies...
856                                     */
857    
858                          // (I'd say: favour short runs => hifreq errors (HVS) -- gruel )                                  /* (I'd say: favour short runs => hifreq errors (HVS) -- gruel ) */
859    
860          if (Cost<Best_Cost) {          if (Cost<Best_Cost) {
861            Best_Cost    = Cost;            Best_Cost    = Cost;
# Line 840  Line 870 
870        }        }
871        if (Last_Node==i)        if (Last_Node==i)
872                          Last.Level = Nodes[i].Level;                          Last.Level = Nodes[i].Level;
873      }                  } else { /* "big" levels */
     else                      // "big" levels  
     {  
874        const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;        const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
875        int Level2;        int Level2;
876        int dQ1, dQ2;        int dQ1, dQ2;
# Line 858  Line 886 
886          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;
887          Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;          Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
888          Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;          Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
889        } else { // Level1<-1                          } else { /* Level1<-1 */
890          dQ1 = Level1*Mult-AC - Bias;          dQ1 = Level1*Mult-AC - Bias;
891          dQ2 = dQ1 + Mult;          dQ2 = dQ1 + Mult;
892          Level2 = Level1 + 1;          Level2 = Level1 + 1;
# Line 867  Line 895 
895          Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;          Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
896          Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;          Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
897        }        }
898    
899        Dist1 = Lambda*dQ1*dQ1;        Dist1 = Lambda*dQ1*dQ1;
900        Dist2 = Lambda*dQ2*dQ2;        Dist2 = Lambda*dQ2*dQ2;
901        dDist21 = Dist2-Dist1;        dDist21 = Dist2-Dist1;
# Line 877  Line 906 
906          uint32_t Cost1, Cost2;          uint32_t Cost1, Cost2;
907          int bLevel;          int bLevel;
908    
909  // for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:                                  /*
910  //        if (Cost_Base>=Best_Cost) continue;                                   * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
911  // (? doesn't seem to have any effect -- gruel )                                   *      if (Cost_Base>=Best_Cost) continue;
912                                     * (? doesn't seem to have any effect -- gruel )
913                                     */
914    
915          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
916          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
# Line 887  Line 918 
918          if (Cost2<Cost1) {          if (Cost2<Cost1) {
919                           Cost1 = Cost2;                           Cost1 = Cost2;
920                           bLevel = Level2;                           bLevel = Level2;
921                    } else                                  } else {
922                           bLevel = Level1;                           bLevel = Level1;
923                                    }
924    
925          if (Cost1<Best_Cost) {          if (Cost1<Best_Cost) {
926            Best_Cost = Cost1;            Best_Cost = Cost1;
# Line 902  Line 934 
934          if (Cost2<Cost1) {          if (Cost2<Cost1) {
935                           Cost1 = Cost2;                           Cost1 = Cost2;
936                           bLevel = Level2;                           bLevel = Level2;
937                    } else                                  } else {
938                           bLevel = Level1;                           bLevel = Level1;
939                                    }
940    
941          if (Cost1<Last_Cost) {          if (Cost1<Last_Cost) {
942            Last_Cost  = Cost1;            Last_Cost  = Cost1;
# Line 911  Line 944 
944            Last.Level = bLevel;            Last.Level = bLevel;
945            Last_Node  = i;            Last_Node  = i;
946          }          }
947        } //end of "for Run"                          } /* end of "for Run" */
948    
949      }      }
950    
# Line 920  Line 953 
953      if (Best_Cost < Min_Cost + Dist0) {      if (Best_Cost < Min_Cost + Dist0) {
954        Min_Cost = Best_Cost;        Min_Cost = Best_Cost;
955        Run_Start = i;        Run_Start = i;
956      }                  } else {
957      else                          /*
958      {                           * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
959          // as noticed by Michael Niedermayer (michaelni at gmx.at), there's                           * a code shorter by 1 bit for a larger run (!), same level. We give
960          // a code shorter by 1 bit for a larger run (!), same level. We give                           * it a chance by not moving the left barrier too much.
961          // it a chance by not moving the left barrier too much.                           */
962    
963        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
964          Run_Start++;          Run_Start++;
965    
966          // spread on preceding coeffs the cost incurred by skipping this one                          /* spread on preceding coeffs the cost incurred by skipping this one */
967        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
968        Min_Cost += Dist0;        Min_Cost += Dist0;
969      }      }
970    }    }
971    
972            /* It seems trellis doesn't give good results... just compute the Out sum and
973             * quit (even if we did not modify it, upperlayer relies on this data) */
974    if (Last_Node<0)    if (Last_Node<0)
975      return -1;                  return Compute_Sum(Out, Non_Zero);
976    
977         // reconstruct optimal sequence backward with surviving paths          /* reconstruct optimal sequence backward with surviving paths */
978    memset(Out, 0x00, 64*sizeof(*Out));    memset(Out, 0x00, 64*sizeof(*Out));
979    Out[Zigzag[Last_Node]] = Last.Level;    Out[Zigzag[Last_Node]] = Last.Level;
980    i = Last_Node - Last.Run;    i = Last_Node - Last.Run;
981            sum = 0;
982    while(i>=0) {    while(i>=0) {
983      Out[Zigzag[i]] = Nodes[i].Level;      Out[Zigzag[i]] = Nodes[i].Level;
984                    sum += abs(Nodes[i].Level);
985      i -= Nodes[i].Run;      i -= Nodes[i].Run;
986    }    }
   return Last_Node;  
 }  
   
   
   
   
   
   
   
   
   
987    
988            return sum;
989    }
990    
991  //////////////////////////////////////////////////////////  static int
992  // original version including heavy debugging info  dct_quantize_trellis_mpeg_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
993  //////////////////////////////////////////////////////////  {
994            /* ToDo: Ok ok it's just a place holder for Gruel -- damn write this one :-) */
995            return Compute_Sum(Out, 63);
996    }
997    
998    /* original version including heavy debugging info */
999    
1000  #ifdef DBGTRELL  #ifdef DBGTRELL
1001    
# Line 987  Line 1019 
1019      int j=0, j0=0;      int j=0, j0=0;
1020      int Run, Level;      int Run, Level;
1021    
1022      Bits = 2;   // CBP                  Bits = 2;   /* CBP */
1023      while(j<Last) {      while(j<Last) {
1024        while(!C[Zigzag[j]])        while(!C[Zigzag[j]])
1025                          j++;                          j++;
# Line 1034  Line 1066 
1066  dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)  dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
1067  {  {
1068    
1069      // Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),      /*
1070      // not quantized one (Out[]). However, it only improves the result *very*           * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
1071      // slightly (~0.01dB), whereas speed drops to crawling level :)           * not quantized one (Out[]). However, it only improves the result *very*
1072      // Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps,           * slightly (~0.01dB), whereas speed drops to crawling level :)
1073             * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
1074             */
1075    typedef struct { int16_t Run, Level; } NODE;    typedef struct { int16_t Run, Level; } NODE;
1076    
1077    NODE Nodes[65], Last;    NODE Nodes[65], Last;
# Line 1047  Line 1080 
1080    const int Mult = 2*Q;    const int Mult = 2*Q;
1081    const int Bias = (Q-1) | 1;    const int Bias = (Q-1) | 1;
1082    const int Lev0 = Mult + Bias;    const int Lev0 = Mult + Bias;
1083    const int Lambda = Trellis_Lambda_Tabs[Q-1];    // it's 1/lambda, actually          const int Lambda = Trellis_Lambda_Tabs[Q-1];    /* it's 1/lambda, actually */
1084    
1085    int Run_Start = -1;    int Run_Start = -1;
1086    Run_Costs[-1] = 2<<16;                          // source (w/ CBP penalty)          Run_Costs[-1] = 2<<16;                          /* source (w/ CBP penalty) */
1087    uint32_t Min_Cost = 2<<16;    uint32_t Min_Cost = 2<<16;
1088    
1089    int Last_Node = -1;    int Last_Node = -1;
# Line 1059  Line 1092 
1092    int i, j;    int i, j;
1093    
1094  #if (DBG>0)  #if (DBG>0)
1095    Last.Level = 0; Last.Run = -1; // just initialize to smthg          Last.Level = 0; Last.Run = -1; /* just initialize to smthg */
1096  #endif  #endif
1097    
1098    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
# Line 1074  Line 1107 
1107      uint32_t Best_Cost = 0xf0000000;      uint32_t Best_Cost = 0xf0000000;
1108      Last_Cost += Dist0;      Last_Cost += Dist0;
1109    
1110      if ((uint32_t)(Level1+1)<3)                 // very specialized loop for -1,0,+1                  if ((uint32_t)(Level1+1)<3)                 /* very specialized loop for -1,0,+1 */
1111      {      {
1112          int dQ;          int dQ;
1113                  int Run;                  int Run;
# Line 1097  Line 1130 
1130          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
1131          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
1132    
1133            // TODO: what about tie-breaks? Should we favor short runs or                                  /*
1134            // long runs? Although the error is the same, it would not be                                   * TODO: what about tie-breaks? Should we favor short runs or
1135            // spread the same way along high and low frequencies...                                   * long runs? Although the error is the same, it would not be
1136                                     * spread the same way along high and low frequencies...
1137                                     */
1138          if (Cost<Best_Cost) {          if (Cost<Best_Cost) {
1139            Best_Cost    = Cost;            Best_Cost    = Cost;
1140            Nodes[i].Run = Run;            Nodes[i].Run = Run;
# Line 1129  Line 1164 
1164          printf( "\n" );          printf( "\n" );
1165        }        }
1166      }      }
1167      else                      // "big" levels                  else                      /* "big" levels */
1168      {      {
1169        const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;        const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
1170        int Level2;        int Level2;
# Line 1146  Line 1181 
1181          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;
1182          Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;          Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
1183          Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;          Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
1184        } else { // Level1<-1                          } else { /* Level1<-1 */
1185          dQ1 = Level1*Mult-AC - Bias;          dQ1 = Level1*Mult-AC - Bias;
1186          dQ2 = dQ1 + Mult;          dQ2 = dQ1 + Mult;
1187          Level2 = Level1 + 1;          Level2 = Level1 + 1;
# Line 1165  Line 1200 
1200          uint32_t Cost1, Cost2;          uint32_t Cost1, Cost2;
1201          int bLevel;          int bLevel;
1202    
1203  // for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:  /*
1204  //        if (Cost_Base>=Best_Cost) continue;   * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
1205     *        if (Cost_Base>=Best_Cost) continue;
1206     */
1207          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
1208          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
1209    
# Line 1198  Line 1234 
1234            Last.Level = bLevel;            Last.Level = bLevel;
1235            Last_Node  = i;            Last_Node  = i;
1236          }          }
1237        } //end of "for Run"                          } /* end of "for Run" */
1238    
1239        if (DBG==1) {        if (DBG==1) {
1240          Run_Costs[i] = Best_Cost;          Run_Costs[i] = Best_Cost;
# Line 1224  Line 1260 
1260      }      }
1261      else      else
1262      {      {
1263          // as noticed by Michael Niedermayer (michaelni at gmx.at), there's                          /*
1264          // a code shorter by 1 bit for a larger run (!), same level. We give                           * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
1265          // it a chance by not moving the left barrier too much.                           * a code shorter by 1 bit for a larger run (!), same level. We give
1266                             * it a chance by not moving the left barrier too much.
1267                             */
1268    
1269        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
1270          Run_Start++;          Run_Start++;
1271    
1272          // spread on preceding coeffs the cost incurred by skipping this one                          /* spread on preceding coeffs the cost incurred by skipping this one */
1273        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1274        Min_Cost += Dist0;        Min_Cost += Dist0;
1275      }      }
# Line 1249  Line 1287 
1287    if (Last_Node<0)    if (Last_Node<0)
1288      return -1;      return -1;
1289    
1290         // reconstruct optimal sequence backward with surviving paths          /* reconstruct optimal sequence backward with surviving paths */
1291    memset(Out, 0x00, 64*sizeof(*Out));    memset(Out, 0x00, 64*sizeof(*Out));
1292    Out[Zigzag[Last_Node]] = Last.Level;    Out[Zigzag[Last_Node]] = Last.Level;
1293    i = Last_Node - Last.Run;    i = Last_Node - Last.Run;
Legend:
Removed from v.1.21.2.12  
changed lines
  Added in v.1.21.2.18
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4