[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 1.21.2.10, Fri May 9 22:03:13 2003 UTC revision 1.21.2.15, Thu Jul 24 13:09:14 2003 UTC
# Line 25  Line 25 
25   *   *
26   ****************************************************************************/   ****************************************************************************/
27    
28  #include <string.h>  #include <stdio.h>
29  #include <stdlib.h>  #include <stdlib.h>
30    #include <string.h>
31    
32  #include "../portab.h"  #include "../portab.h"
33  #include "mbfunctions.h"  #include "mbfunctions.h"
# Line 161  Line 162 
162    
163    
164  static int  static int
165  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,
166                                                            const int16_t *const In,
167                                                            int Q,
168                                                            const uint16_t * const Zigzag,
169                                                            int Non_Zero);
170    
171    #if 0
172  static int  static int
173  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_mpeg_c(int16_t *const Out,
174                                                            const int16_t *const In,
175                                                            int Q,
176                                                            const uint16_t * const Zigzag,
177                                                            int Non_Zero);
178    #endif
179    
180  /* Quantize all blocks -- Inter mode */  /* Quantize all blocks -- Inter mode */
181  static __inline uint8_t  static __inline uint8_t
# Line 191  Line 201 
201                          sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant);                          sum = quant_inter(&qcoeff[i*64], &data[i*64], pMB->quant);
202                          if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) {                          if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) ) {
203                                  sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1;                                  sum = dct_quantize_trellis_h263_c(&qcoeff[i*64], &data[i*64], pMB->quant, &scan_tables[0][0], 63)+1;
204                                  limit = 1;  /*                              limit = 1; // Isibaar: why? deactivated so far - so please complain! ;-) */
205                          }                          }
206                  } else {                  } else {
207                          sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant);                          sum = quant4_inter(&qcoeff[i * 64], &data[i * 64], pMB->quant);
208  //                      if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) )  #if 0
209  //                              sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1;                          if ( (sum) && (frame->vop_flags & XVID_VOP_TRELLISQUANT) )
210                                    sum = dct_quantize_trellis_mpeg_c (&qcoeff[i*64], &data[i*64], pMB->quant)+1;
211    #endif
212                  }                  }
213                  stop_quant_timer();                  stop_quant_timer();
214    
# Line 429  Line 441 
441          /* Set the limit threshold */          /* Set the limit threshold */
442          limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);          limit = PVOP_TOOSMALL_LIMIT + ((pMB->quant == 1)? 1 : 0);
443    
444            if (frame->vop_flags & XVID_VOP_CARTOON)
445                    limit *= 3;
446    
447          /* Quantize the block */          /* Quantize the block */
448          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 0, limit);
449    
# Line 467  Line 482 
482          /* Set the limit threshold */          /* Set the limit threshold */
483          limit = BVOP_TOOSMALL_LIMIT;          limit = BVOP_TOOSMALL_LIMIT;
484    
485            if (frame->vop_flags & XVID_VOP_CARTOON)
486                    limit *= 2;
487    
488          /* Quantize the block */          /* Quantize the block */
489          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);          cbp = MBQuantInter(pParam, frame, pMB, data, qcoeff, 1, limit);
490    
# Line 545  Line 563 
563    
564          /* left blocks */          /* left blocks */
565    
566          // 1=2, 2=4, 4=8, 8=1          /* 1=2, 2=4, 4=8, 8=1 */
567          MOVLINE(tmp, LINE(0, 1));          MOVLINE(tmp, LINE(0, 1));
568          MOVLINE(LINE(0, 1), LINE(0, 2));          MOVLINE(LINE(0, 1), LINE(0, 2));
569          MOVLINE(LINE(0, 2), LINE(0, 4));          MOVLINE(LINE(0, 2), LINE(0, 4));
570          MOVLINE(LINE(0, 4), LINE(2, 0));          MOVLINE(LINE(0, 4), LINE(2, 0));
571          MOVLINE(LINE(2, 0), tmp);          MOVLINE(LINE(2, 0), tmp);
572    
573          // 3=6, 6=12, 12=9, 9=3          /* 3=6, 6=12, 12=9, 9=3 */
574          MOVLINE(tmp, LINE(0, 3));          MOVLINE(tmp, LINE(0, 3));
575          MOVLINE(LINE(0, 3), LINE(0, 6));          MOVLINE(LINE(0, 3), LINE(0, 6));
576          MOVLINE(LINE(0, 6), LINE(2, 4));          MOVLINE(LINE(0, 6), LINE(2, 4));
577          MOVLINE(LINE(2, 4), LINE(2, 1));          MOVLINE(LINE(2, 4), LINE(2, 1));
578          MOVLINE(LINE(2, 1), tmp);          MOVLINE(LINE(2, 1), tmp);
579    
580          // 5=10, 10=5          /* 5=10, 10=5 */
581          MOVLINE(tmp, LINE(0, 5));          MOVLINE(tmp, LINE(0, 5));
582          MOVLINE(LINE(0, 5), LINE(2, 2));          MOVLINE(LINE(0, 5), LINE(2, 2));
583          MOVLINE(LINE(2, 2), tmp);          MOVLINE(LINE(2, 2), tmp);
584    
585          // 7=14, 14=13, 13=11, 11=7          /* 7=14, 14=13, 13=11, 11=7 */
586          MOVLINE(tmp, LINE(0, 7));          MOVLINE(tmp, LINE(0, 7));
587          MOVLINE(LINE(0, 7), LINE(2, 6));          MOVLINE(LINE(0, 7), LINE(2, 6));
588          MOVLINE(LINE(2, 6), LINE(2, 5));          MOVLINE(LINE(2, 6), LINE(2, 5));
# Line 573  Line 591 
591    
592          /* right blocks */          /* right blocks */
593    
594          // 1=2, 2=4, 4=8, 8=1          /* 1=2, 2=4, 4=8, 8=1 */
595          MOVLINE(tmp, LINE(1, 1));          MOVLINE(tmp, LINE(1, 1));
596          MOVLINE(LINE(1, 1), LINE(1, 2));          MOVLINE(LINE(1, 1), LINE(1, 2));
597          MOVLINE(LINE(1, 2), LINE(1, 4));          MOVLINE(LINE(1, 2), LINE(1, 4));
598          MOVLINE(LINE(1, 4), LINE(3, 0));          MOVLINE(LINE(1, 4), LINE(3, 0));
599          MOVLINE(LINE(3, 0), tmp);          MOVLINE(LINE(3, 0), tmp);
600    
601          // 3=6, 6=12, 12=9, 9=3          /* 3=6, 6=12, 12=9, 9=3 */
602          MOVLINE(tmp, LINE(1, 3));          MOVLINE(tmp, LINE(1, 3));
603          MOVLINE(LINE(1, 3), LINE(1, 6));          MOVLINE(LINE(1, 3), LINE(1, 6));
604          MOVLINE(LINE(1, 6), LINE(3, 4));          MOVLINE(LINE(1, 6), LINE(3, 4));
605          MOVLINE(LINE(3, 4), LINE(3, 1));          MOVLINE(LINE(3, 4), LINE(3, 1));
606          MOVLINE(LINE(3, 1), tmp);          MOVLINE(LINE(3, 1), tmp);
607    
608          // 5=10, 10=5          /* 5=10, 10=5 */
609          MOVLINE(tmp, LINE(1, 5));          MOVLINE(tmp, LINE(1, 5));
610          MOVLINE(LINE(1, 5), LINE(3, 2));          MOVLINE(LINE(1, 5), LINE(3, 2));
611          MOVLINE(LINE(3, 2), tmp);          MOVLINE(LINE(3, 2), tmp);
612    
613          // 7=14, 14=13, 13=11, 11=7          /* 7=14, 14=13, 13=11, 11=7 */
614          MOVLINE(tmp, LINE(1, 7));          MOVLINE(tmp, LINE(1, 7));
615          MOVLINE(LINE(1, 7), LINE(3, 6));          MOVLINE(LINE(1, 7), LINE(3, 6));
616          MOVLINE(LINE(3, 6), LINE(3, 5));          MOVLINE(LINE(3, 6), LINE(3, 5));
# Line 604  Line 622 
622    
623    
624    
625  /************************************************************************  /*****************************************************************************
626   *               Trellis based R-D optimal quantization                 *   *               Trellis based R-D optimal quantization
627   *                                                                      *   *
628   *   Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net  *   *   Trellis Quant code (C) 2003 Pascal Massimino skal(at)planet-d.net
629   *                                                                      *   *
630   ************************************************************************/   ****************************************************************************/
631    
632    
633    #if 0
634  static int  static int
635  dct_quantize_trellis_inter_mpeg_c (int16_t *qcoeff, const int16_t *data, int quant)  dct_quantize_trellis_mpeg_c(int16_t *const Out,
636  { return 63; }                                                          const int16_t *const In,
637                                                            int Q,
638                                                            const uint16_t * const Zigzag,
639                                                            int Non_Zero)
640    {
641            return 63;
642    }
643    #endif
644    
645    /*----------------------------------------------------------------------------
646     *
647     *        Trellis-Based quantization
648     *
649     * So far I understand this paper:
650     *
651     *  "Trellis-Based R-D Optimal Quantization in H.263+"
652     *    J.Wen, M.Luttrell, J.Villasenor
653     *    IEEE Transactions on Image Processing, Vol.9, No.8, Aug. 2000.
654     *
655     * we are at stake with a simplified Bellmand-Ford / Dijkstra Single
656     * Source Shorted Path algo. But due to the underlying graph structure
657     * ("Trellis"), it can be turned into a dynamic programming algo,
658     * partially saving the explicit graph's nodes representation. And
659     * without using a heap, since the open frontier of the DAG is always
660     * known, and of fixed sized.
661     *--------------------------------------------------------------------------*/
662    
 //////////////////////////////////////////////////////////  
 //  
 //        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.  
 //  
 //////////////////////////////////////////////////////////  
663    
664    
665  //////////////////////////////////////////////////////////  /* Codes lengths for relevant levels. */
 // Codes lengths for relevant levels.  
666    
667    // let's factorize:    /* let's factorize: */
668  static const uint8_t Code_Len0[64] = {  static const uint8_t Code_Len0[64] = {
669    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,
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 };    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30 };
# Line 705  Line 729 
729     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,
730    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 };
731    
732    // a few more table for LAST table:    /* a few more table for LAST table: */
733  static const uint8_t Code_Len21[64] = {  static const uint8_t Code_Len21[64] = {
734    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,
735    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};    30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30,30};
# Line 720  Line 744 
744    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};
745    
746    
747  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] */
748    Code_Len20,Code_Len19,Code_Len18,Code_Len17,    Code_Len20,Code_Len19,Code_Len18,Code_Len17,
749    Code_Len16,Code_Len15,Code_Len14,Code_Len13,    Code_Len16,Code_Len15,Code_Len14,Code_Len13,
750    Code_Len12,Code_Len11,Code_Len10,Code_Len9,    Code_Len12,Code_Len11,Code_Len10,Code_Len9,
# Line 729  Line 753 
753    Code_Len2, Code_Len1, Code_Len1, Code_Len1,    Code_Len2, Code_Len1, Code_Len1, Code_Len1,
754  };  };
755    
756  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] */
757    Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,    Code_Len24,Code_Len23,Code_Len22,Code_Len21, Code_Len3, Code_Len1,
758  };  };
759    
# Line 743  Line 767 
767  };  };
768  #undef TL  #undef TL
769    
770  static inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)  static __inline int Find_Last(const int16_t *C, const uint16_t *Zigzag, int i)
771  {  {
772    while(i>=0)    while(i>=0)
773      if (C[Zigzag[i]])      if (C[Zigzag[i]])
# Line 752  Line 776 
776    return -1;    return -1;
777  }  }
778    
779  //////////////////////////////////////////////////////////  /* this routine has been strippen of all debug code */
780    
781    static int
782    dct_quantize_trellis_h263_c(int16_t *const Out, const int16_t *const In, int Q, const uint16_t * const Zigzag, int Non_Zero)
783    {
784    
785        /*
786             * Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),
787             * not quantized one (Out[]). However, it only improves the result *very*
788             * slightly (~0.01dB), whereas speed drops to crawling level :)
789             * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
790             */
791      typedef struct { int16_t Run, Level; } NODE;
792    
793      NODE Nodes[65], Last;
794      uint32_t Run_Costs0[64+1];
795      uint32_t * const Run_Costs = Run_Costs0 + 1;
796      const int Mult = 2*Q;
797      const int Bias = (Q-1) | 1;
798      const int Lev0 = Mult + Bias;
799      const int Lambda = Trellis_Lambda_Tabs[Q-1];    /* it's 1/lambda, actually */
800    
801      int Run_Start = -1;
802      uint32_t Min_Cost = 2<<16;
803    
804      int Last_Node = -1;
805      uint32_t Last_Cost = 0;
806    
807      int i, j;
808      Run_Costs[-1] = 2<<16;                          /* source (w/ CBP penalty) */
809    
810      Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
811      if (Non_Zero<0)
812          return -1;
813    
814      for(i=0; i<=Non_Zero; i++)
815      {
816        const int AC = In[Zigzag[i]];
817        const int Level1 = Out[Zigzag[i]];
818        const int Dist0 = Lambda* AC*AC;
819        uint32_t Best_Cost = 0xf0000000;
820        Last_Cost += Dist0;
821    
822        if ((uint32_t)(Level1+1)<3)                 /* very specialized loop for -1,0,+1 */
823        {
824            int dQ;
825                    int Run;
826          uint32_t Cost0;
827    
828          if (AC<0) {
829            Nodes[i].Level = -1;
830            dQ = Lev0 + AC;
831          } else {
832            Nodes[i].Level = 1;
833            dQ = Lev0 - AC;
834          }
835                    Cost0 = Lambda*dQ*dQ;
836    
837          Nodes[i].Run = 1;
838          Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
839          for(Run=i-Run_Start; Run>0; --Run)
840          {
841            const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
842            const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
843            const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
844    
845              /*
846                       * TODO: what about tie-breaks? Should we favor short runs or
847                       * long runs? Although the error is the same, it would not be
848                       * spread the same way along high and low frequencies...
849                       */
850    
851                            /* (I'd say: favour short runs => hifreq errors (HVS) -- gruel ) */
852    
853            if (Cost<Best_Cost) {
854              Best_Cost    = Cost;
855              Nodes[i].Run = Run;
856            }
857    
858            if (lCost<Last_Cost) {
859              Last_Cost  = lCost;
860              Last.Run   = Run;
861              Last_Node  = i;
862            }
863          }
864          if (Last_Node==i)
865                            Last.Level = Nodes[i].Level;
866        }
867        else                      /* "big" levels */
868        {
869          const uint8_t *Tbl_L1, *Tbl_L2, *Tbl_L1_Last, *Tbl_L2_Last;
870          int Level2;
871          int dQ1, dQ2;
872          int Run;
873                    uint32_t Dist1,Dist2;
874                    int dDist21;
875    
876              if (Level1>1) {
877            dQ1 = Level1*Mult-AC + Bias;
878            dQ2 = dQ1 - Mult;
879            Level2 = Level1-1;
880            Tbl_L1      = (Level1<=24) ? B16_17_Code_Len[Level1-1]     : Code_Len0;
881            Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;
882            Tbl_L1_Last = (Level1<=6) ? B16_17_Code_Len_Last[Level1-1] : Code_Len0;
883            Tbl_L2_Last = (Level2<=6) ? B16_17_Code_Len_Last[Level2-1] : Code_Len0;
884          } else { /* Level1<-1 */
885            dQ1 = Level1*Mult-AC - Bias;
886            dQ2 = dQ1 + Mult;
887            Level2 = Level1 + 1;
888            Tbl_L1      = (Level1>=-24) ? B16_17_Code_Len[Level1^-1]      : Code_Len0;
889            Tbl_L2      = (Level2>=-24) ? B16_17_Code_Len[Level2^-1]      : Code_Len0;
890            Tbl_L1_Last = (Level1>=- 6) ? B16_17_Code_Len_Last[Level1^-1] : Code_Len0;
891            Tbl_L2_Last = (Level2>=- 6) ? B16_17_Code_Len_Last[Level2^-1] : Code_Len0;
892          }
893          Dist1 = Lambda*dQ1*dQ1;
894          Dist2 = Lambda*dQ2*dQ2;
895          dDist21 = Dist2-Dist1;
896    
897          for(Run=i-Run_Start; Run>0; --Run)
898          {
899            const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
900            uint32_t Cost1, Cost2;
901            int bLevel;
902    
903    /*
904     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
905     *      if (Cost_Base>=Best_Cost) continue;
906     * (? doesn't seem to have any effect -- gruel )
907     */
908    
909            Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
910            Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
911    
912            if (Cost2<Cost1) {
913                             Cost1 = Cost2;
914                             bLevel = Level2;
915                      } else
916                             bLevel = Level1;
917    
918            if (Cost1<Best_Cost) {
919              Best_Cost = Cost1;
920              Nodes[i].Run   = Run;
921              Nodes[i].Level = bLevel;
922            }
923    
924            Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
925            Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
926    
927            if (Cost2<Cost1) {
928                             Cost1 = Cost2;
929                             bLevel = Level2;
930                      } else
931                             bLevel = Level1;
932    
933            if (Cost1<Last_Cost) {
934              Last_Cost  = Cost1;
935              Last.Run   = Run;
936              Last.Level = bLevel;
937              Last_Node  = i;
938            }
939          } /* end of "for Run" */
940    
941        }
942    
943        Run_Costs[i] = Best_Cost;
944    
945        if (Best_Cost < Min_Cost + Dist0) {
946          Min_Cost = Best_Cost;
947          Run_Start = i;
948        }
949        else
950        {
951            /*
952                     * as noticed by Michael Niedermayer (michaelni at gmx.at), there's
953                     * a code shorter by 1 bit for a larger run (!), same level. We give
954                     * it a chance by not moving the left barrier too much.
955                     */
956    
957          while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
958            Run_Start++;
959    
960            /* spread on preceding coeffs the cost incurred by skipping this one */
961          for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
962          Min_Cost += Dist0;
963        }
964      }
965    
966      if (Last_Node<0)
967        return -1;
968    
969           /* reconstruct optimal sequence backward with surviving paths */
970      memset(Out, 0x00, 64*sizeof(*Out));
971      Out[Zigzag[Last_Node]] = Last.Level;
972      i = Last_Node - Last.Run;
973      while(i>=0) {
974        Out[Zigzag[i]] = Nodes[i].Level;
975        i -= Nodes[i].Run;
976      }
977      return Last_Node;
978    }
979    
980    
981    
982    
983    
984    
985    
986    
987    
988    
989    
990    /* original version including heavy debugging info */
991    
992    #ifdef DBGTRELL
993    
994  #define DBG 0  #define DBG 0
995    
996  static uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,  static __inline uint32_t Evaluate_Cost(const int16_t *C, int Mult, int Bias,
997                                  const uint16_t * Zigzag, int Max, int Lambda)                                  const uint16_t * Zigzag, int Max, int Lambda)
998  {  {
999  #if (DBG>0)  #if (DBG>0)
1000    const int16_t * const Ref = C + 6*64;    const int16_t * const Ref = C + 6*64;
1001    int Last = Max;    int Last = Max;
   while(Last>=0 && C[Zigzag[Last]]==0) Last--;  
1002    int Bits = 0;    int Bits = 0;
1003      int Dist = 0;
1004      int i;
1005      uint32_t Cost;
1006    
1007      while(Last>=0 && C[Zigzag[Last]]==0)
1008            Last--;
1009    
1010    if (Last>=0) {    if (Last>=0) {
     Bits = 2;   // CBP  
1011      int j=0, j0=0;      int j=0, j0=0;
1012      int Run, Level;      int Run, Level;
1013    
1014        Bits = 2;   /* CBP */
1015      while(j<Last) {      while(j<Last) {
1016        while(!C[Zigzag[j]]) j++;        while(!C[Zigzag[j]])
1017        if (j==Last) break;                          j++;
1018          if (j==Last)
1019                            break;
1020        Level=C[Zigzag[j]];        Level=C[Zigzag[j]];
1021        Run = j - j0;        Run = j - j0;
1022        j0 = ++j;        j0 = ++j;
1023        if (Level>=-24 && Level<=24) Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];        if (Level>=-24 && Level<=24)
1024        else Bits += 30;                          Bits += B16_17_Code_Len[(Level<0) ? -Level-1 : Level-1][Run];
1025          else
1026                            Bits += 30;
1027      }      }
1028      Level = C[Zigzag[Last]];      Level = C[Zigzag[Last]];
1029      Run = j - j0;      Run = j - j0;
1030      if (Level>=-6 && Level<=6) Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];      if (Level>=-6 && Level<=6)
1031      else Bits += 30;                  Bits += B16_17_Code_Len_Last[(Level<0) ? -Level-1 : Level-1][Run];
1032        else
1033                    Bits += 30;
1034    }    }
1035    
   int Dist = 0;  
   int i;  
1036    for(i=0; i<=Last; ++i) {    for(i=0; i<=Last; ++i) {
1037      int V = C[Zigzag[i]]*Mult;      int V = C[Zigzag[i]]*Mult;
1038      if      (V>0) V += Bias;      if (V>0)
1039      else if (V<0) V -= Bias;                  V += Bias;
1040        else
1041                    if (V<0)
1042                            V -= Bias;
1043      V -= Ref[Zigzag[i]];      V -= Ref[Zigzag[i]];
1044      Dist += V*V;      Dist += V*V;
1045    }    }
1046    uint32_t Cost = Lambda*Dist + (Bits<<16);    Cost = Lambda*Dist + (Bits<<16);
1047    if (DBG==1)    if (DBG==1)
1048      printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );      printf( " Last:%2d/%2d Cost = [(Bits=%5.0d) + Lambda*(Dist=%6.0d) = %d ] >>12= %d ", Last,Max, Bits, Dist, Cost, Cost>>12 );
1049    return Cost;    return Cost;
# Line 807  Line 1058 
1058  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)
1059  {  {
1060    
1061      // Note: We should search last non-zero coeffs on *real* DCT input coeffs (In[]),      /*
1062      // 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[]),
1063      // slightly (~0.01dB), whereas speed drops to crawling level :)           * not quantized one (Out[]). However, it only improves the result *very*
1064      // Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps,           * slightly (~0.01dB), whereas speed drops to crawling level :)
1065             * Well, actually, taking 1 more coeff past Non_Zero into account sometimes helps.
1066             */
1067    typedef struct { int16_t Run, Level; } NODE;    typedef struct { int16_t Run, Level; } NODE;
1068    
1069    NODE Nodes[65], Last;    NODE Nodes[65], Last;
1070    uint32_t Run_Costs0[64+1], * const Run_Costs = Run_Costs0 + 1;    uint32_t Run_Costs0[64+1];
1071      uint32_t * const Run_Costs = Run_Costs0 + 1;
1072    const int Mult = 2*Q;    const int Mult = 2*Q;
1073    const int Bias = (Q-1) | 1;    const int Bias = (Q-1) | 1;
1074    const int Lev0 = Mult + Bias;    const int Lev0 = Mult + Bias;
1075    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 */
1076    
1077    int Run_Start = -1;    int Run_Start = -1;
1078    Run_Costs[-1] = 2<<16;                          // source (w/ CBP penalty)    Run_Costs[-1] = 2<<16;                          /* source (w/ CBP penalty) */
1079    uint32_t Min_Cost = 2<<16;    uint32_t Min_Cost = 2<<16;
1080    
1081    int Last_Node = -1;    int Last_Node = -1;
1082    uint32_t Last_Cost = 0;    uint32_t Last_Cost = 0;
1083    
1084      int i, j;
1085    
1086  #if (DBG>0)  #if (DBG>0)
1087    Last.Level = 0; Last.Run = -1; // just initialize to smthg    Last.Level = 0; Last.Run = -1; /* just initialize to smthg */
1088  #endif  #endif
1089    
   int i, j;  
   
1090    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);    Non_Zero = Find_Last(Out, Zigzag, Non_Zero);
1091    if (Non_Zero<0)    if (Non_Zero<0)
1092        return -1;        return -1;
# Line 847  Line 1099 
1099      uint32_t Best_Cost = 0xf0000000;      uint32_t Best_Cost = 0xf0000000;
1100      Last_Cost += Dist0;      Last_Cost += Dist0;
1101    
1102      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 */
1103      {      {
1104        int dQ;        int dQ;
1105            int Run;            int Run;
1106          uint32_t Cost0;
1107    
1108        if (AC<0) {        if (AC<0) {
1109          Nodes[i].Level = -1;          Nodes[i].Level = -1;
# Line 859  Line 1112 
1112          Nodes[i].Level = 1;          Nodes[i].Level = 1;
1113          dQ = Lev0 - AC;          dQ = Lev0 - AC;
1114        }        }
1115        const uint32_t Cost0 = Lambda*dQ*dQ;                  Cost0 = Lambda*dQ*dQ;
1116    
1117        Nodes[i].Run = 1;        Nodes[i].Run = 1;
1118        Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;        Best_Cost = (Code_Len20[0]<<16) + Run_Costs[i-1]+Cost0;
# Line 867  Line 1120 
1120        {        {
1121          const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];          const uint32_t Cost_Base = Cost0 + Run_Costs[i-Run];
1122          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);          const uint32_t Cost = Cost_Base + (Code_Len20[Run-1]<<16);
1123            // TODO: what about tie-breaks? Should we favor short runs or          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);
1124            // long runs? Although the error is the same, it would not be  
1125            // spread the same way along high and low frequencies...            /*
1126          if (Cost<Best_Cost)                     * TODO: what about tie-breaks? Should we favor short runs or
1127          {                     * long runs? Although the error is the same, it would not be
1128                       * spread the same way along high and low frequencies...
1129                       */
1130            if (Cost<Best_Cost) {
1131            Best_Cost    = Cost;            Best_Cost    = Cost;
1132            Nodes[i].Run = Run;            Nodes[i].Run = Run;
1133          }          }
1134          const uint32_t lCost = Cost_Base + (Code_Len24[Run-1]<<16);  
1135          if (lCost<Last_Cost)          if (lCost<Last_Cost) {
         {  
1136            Last_Cost  = lCost;            Last_Cost  = lCost;
1137            Last.Run   = Run;            Last.Run   = Run;
1138            Last_Node  = i;            Last_Node  = i;
1139          }          }
1140        }        }
1141        if (Last_Node==i) Last.Level = Nodes[i].Level;        if (Last_Node==i)
1142                            Last.Level = Nodes[i].Level;
1143    
1144        if (DBG==1) {        if (DBG==1) {
1145          Run_Costs[i] = Best_Cost;          Run_Costs[i] = Best_Cost;
# Line 901  Line 1156 
1156          printf( "\n" );          printf( "\n" );
1157        }        }
1158      }      }
1159      else                      // "big" levels      else                      /* "big" levels */
1160      {      {
1161        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;
1162        int Level2;        int Level2;
1163        int dQ1, dQ2;        int dQ1, dQ2;
1164        int Run;        int Run;
1165                    uint32_t Dist1,Dist2;
1166                    int dDist21;
1167    
1168            if (Level1>1) {            if (Level1>1) {
1169          dQ1 = Level1*Mult-AC + Bias;          dQ1 = Level1*Mult-AC + Bias;
# Line 916  Line 1173 
1173          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;          Tbl_L2      = (Level2<=24) ? B16_17_Code_Len[Level2-1]     : Code_Len0;
1174          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;
1175          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;
1176        }        } else { /* Level1<-1 */
       else { // Level1<-1  
1177          dQ1 = Level1*Mult-AC - Bias;          dQ1 = Level1*Mult-AC - Bias;
1178          dQ2 = dQ1 + Mult;          dQ2 = dQ1 + Mult;
1179          Level2 = Level1 + 1;          Level2 = Level1 + 1;
# Line 926  Line 1182 
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        }        }
1185        const uint32_t Dist1 = Lambda*dQ1*dQ1;        Dist1 = Lambda*dQ1*dQ1;
1186        const uint32_t Dist2 = Lambda*dQ2*dQ2;        Dist2 = Lambda*dQ2*dQ2;
1187        const int dDist21 = Dist2-Dist1;        dDist21 = Dist2-Dist1;
1188    
1189        for(Run=i-Run_Start; Run>0; --Run)        for(Run=i-Run_Start; Run>0; --Run)
1190        {        {
1191          const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];          const uint32_t Cost_Base = Dist1 + Run_Costs[i-Run];
   
 // for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:  
 //        if (Cost_Base>=Best_Cost) continue;  
   
1192          uint32_t Cost1, Cost2;          uint32_t Cost1, Cost2;
1193          int bLevel;          int bLevel;
1194    
1195    /*
1196     * for sub-optimal (but slightly worth it, speed-wise) search, uncomment the following:
1197     *        if (Cost_Base>=Best_Cost) continue;
1198     */
1199          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);          Cost1 = Cost_Base + (Tbl_L1[Run-1]<<16);
1200          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;          Cost2 = Cost_Base + (Tbl_L2[Run-1]<<16) + dDist21;
1201    
1202          if (Cost2<Cost1) { Cost1 = Cost2; bLevel = Level2; }          if (Cost2<Cost1) {
1203          else bLevel = Level1;                           Cost1 = Cost2;
1204                             bLevel = Level2;
1205                      } else
1206                             bLevel = Level1;
1207    
1208          if (Cost1<Best_Cost)          if (Cost1<Best_Cost) {
         {  
1209            Best_Cost = Cost1;            Best_Cost = Cost1;
1210            Nodes[i].Run   = Run;            Nodes[i].Run   = Run;
1211            Nodes[i].Level = bLevel;            Nodes[i].Level = bLevel;
# Line 956  Line 1214 
1214          Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);          Cost1 = Cost_Base + (Tbl_L1_Last[Run-1]<<16);
1215          Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;          Cost2 = Cost_Base + (Tbl_L2_Last[Run-1]<<16) + dDist21;
1216    
1217          if (Cost2<Cost1) { Cost1 = Cost2; bLevel = Level2; }          if (Cost2<Cost1) {
1218          else bLevel = Level1;                           Cost1 = Cost2;
1219          if (Cost1<Last_Cost)                           bLevel = Level2;
1220          {                    } else
1221                             bLevel = Level1;
1222    
1223            if (Cost1<Last_Cost) {
1224            Last_Cost  = Cost1;            Last_Cost  = Cost1;
1225            Last.Run   = Run;            Last.Run   = Run;
1226            Last.Level = bLevel;            Last.Level = bLevel;
1227            Last_Node  = i;            Last_Node  = i;
1228          }          }
1229        }        } /* end of "for Run" */
1230    
1231        if (DBG==1) {        if (DBG==1) {
1232          Run_Costs[i] = Best_Cost;          Run_Costs[i] = Best_Cost;
# Line 991  Line 1252 
1252      }      }
1253      else      else
1254      {      {
1255          // as noticed by Michael Niedermayer (michaelni at gmx.at), there's          /*
1256          // 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
1257          // 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
1258                     * it a chance by not moving the left barrier too much.
1259                     */
1260    
1261        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )        while( Run_Costs[Run_Start]>Min_Cost+(1<<16) )
1262          Run_Start++;          Run_Start++;
1263    
1264          // spread on preceding coeffs the cost incurred by skipping this one          /* spread on preceding coeffs the cost incurred by skipping this one */
1265        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;        for(j=Run_Start; j<i; ++j) Run_Costs[j] += Dist0;
1266        Min_Cost += Dist0;        Min_Cost += Dist0;
1267      }      }
# Line 1015  Line 1279 
1279    if (Last_Node<0)    if (Last_Node<0)
1280      return -1;      return -1;
1281    
1282         // reconstruct optimal sequence backward with surviving paths         /* reconstruct optimal sequence backward with surviving paths */
1283    bzero(Out, 64*sizeof(*Out));    memset(Out, 0x00, 64*sizeof(*Out));
1284    Out[Zigzag[Last_Node]] = Last.Level;    Out[Zigzag[Last_Node]] = Last.Level;
1285    i = Last_Node - Last.Run;    i = Last_Node - Last.Run;
1286    while(i>=0) {    while(i>=0) {
# Line 1037  Line 1301 
1301  }  }
1302    
1303  #undef DBG  #undef DBG
1304    
1305    #endif
Legend:
Removed from v.1.21.2.10  
changed lines
  Added in v.1.21.2.15
No admin address has been configured
 ViewVC Help
Powered by ViewVC 1.0.4