ghost/libentcode/rangeenc.c

   1 #include <stddef.h>
   2 #include "entenc.h"
   3 #include "mfrngcod.h"
   4
   5
   6
   7 /*A range encoder.
   8   See rangedec.c and the references for implementation details
   9    \cite{Mar79,MNW98}.
  10
  11   @INPROCEEDINGS{Mar79,
  12    author="Martin, G.N.N.",
  13    title="Range encoding: an algorithm for removing redundancy from a digitised
  14     message",
  15    booktitle="Video \& Data Recording Conference",
  16    year=1979,
  17    address="Southampton",
  18    month=Jul
  19   }
  20   @ARTICLE{MNW98,
  21    author="Alistair Moffat and Radford Neal and Ian H. Witten",
  22    title="Arithmetic Coding Revisited",
  23    journal="{ACM} Transactions on Information Systems",
  24    year=1998,
  25    volume=16,
  26    number=3,
  27    pages="256--294",
  28    month=Jul,
  29    URL="http://www.stanford.edu/class/ee398/handouts/papers/Moffat98ArithmCoding.pdf"
  30   }*/
  31
  32
  33
  34 /*Outputs a symbol, with a carry bit.
  35   If there is a potential to propagate a carry over several symbols, they are
  36    buffered until it can be determined whether or not an actual carry will
  37    occur.
  38   If the counter for the buffered symbols overflows, then the stream becomes
  39    undecodable.
  40   This gives a theoretical limit of a few billion symbols in a single packet on
  41    32-bit systems.
  42   The alternative is to truncate the range in order to force a carry, but
  43    requires similar carry tracking in the decoder, needlessly slowing it down.*/
  44 static void ec_enc_carry_out(ec_enc *_this,int _c){
  45   if(_c!=EC_SYM_MAX){
  46     /*No further carry propagation possible, flush buffer.*/
  47     int carry;
  48     carry=_c>>EC_SYM_BITS;
  49     /*Don't output a byte on the first write.
  50       This compare should be taken care of by branch-prediction thereafter.*/
  51     if(_this->rem>=0)ec_byte_write1(_this->buf,_this->rem+carry);
  52     if(_this->ext>0){
  53       unsigned sym;
  54       sym=EC_SYM_MAX+carry&EC_SYM_MAX;
  55       do ec_byte_write1(_this->buf,sym);
  56       while(--(_this->ext)>0);
  57     }
  58     _this->rem=_c&EC_SYM_MAX;
  59   }
  60   else _this->ext++;
  61 }
  62
  63 static void ec_enc_normalize(ec_enc *_this){
  64   /*If the range is too small, output some bits and rescale it.*/
  65   while(_this->rng<=EC_CODE_BOT){
  66     ec_enc_carry_out(_this,(int)(_this->low>>EC_CODE_SHIFT));
  67     /*Move the next-to-high-order symbol into the high-order position.*/
  68     _this->low=_this->low<<EC_SYM_BITS&EC_CODE_TOP-1;
  69     _this->rng<<=EC_SYM_BITS;
  70   }
  71 }
  72
  73 void ec_enc_init(ec_enc *_this,ec_byte_buffer *_buf){
  74   _this->buf=_buf;
  75   _this->rem=-1;
  76   _this->ext=0;
  77   _this->low=0;
  78   _this->rng=EC_CODE_TOP;
  79 }
  80
  81 void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft){
  82   ec_uint32 r;
  83   r=_this->rng/_ft;
  84   if(_fl>0){
  85     _this->low+=_this->rng-r*(_ft-_fl);
  86     _this->rng=r*(_fh-_fl);
  87   }
  88   else _this->rng-=r*(_ft-_fh);
  89   ec_enc_normalize(_this);
  90 }
  91
  92 long ec_enc_tell(ec_enc *_this,int _b){
  93   ec_uint32 r;
  94   int       l;
  95   long      nbits;
  96   nbits=ec_byte_bytes(_this->buf)+(_this->rem>=0)+_this->ext<<3;
  97   /*To handle the non-integral number of bits still left in the encoder state,
  98      we compute the number of bits of low that must be encoded to ensure that
  99      the value is inside the range for any possible subsequent bits.
 100     Note that this is subtly different than the actual value we would end the
 101      stream with, which tries to make as many of the trailing bits zeros as
 102      possible.*/
 103   nbits+=EC_CODE_BITS;
 104   nbits<<=_b;
 105   l=EC_ILOG(_this->rng);
 106   r=_this->rng>>l-16;
 107   while(_b-->0){
 108     int b;
 109     r=r*r>>15;
 110     b=(int)(r>>16);
 111     l=l<<1|b;
 112     r>>=b;
 113   }
 114   return nbits-l;
 115 }
 116
 117 void ec_enc_done(ec_enc *_this){
 118   /*We compute the integer in the current interval that has the largest number
 119      of trailing zeros, and write that to the stream.
 120     This is guaranteed to yield the smallest possible encoding.*/
 121   if(_this->low){
 122     unsigned end;
 123     end=EC_CODE_TOP;
 124     /*Ensure that the end value is in the range.*/
 125     if(end-_this->low>=_this->rng){
 126       unsigned msk;
 127       msk=EC_CODE_TOP-1;
 128       do{
 129         msk>>=1;
 130         end=_this->low+msk&~msk|msk+1;
 131       }
 132       while(end-_this->low>=_this->rng);
 133     }
 134     /*The remaining output is the next free end.*/
 135     while(end){
 136       ec_enc_carry_out(_this,end>>EC_CODE_SHIFT);
 137       end=end<<EC_SYM_BITS&EC_CODE_TOP-1;
 138     }
 139   }
 140   /*If we have a buffered byte...*/
 141   if(_this->rem>=0||_this->ext>0){
 142     unsigned char *buf;
 143     long           i;
 144     /*Flush it into the output buffer.*/
 145     ec_enc_carry_out(_this,0);
 146     _this->rem=-1;
 147     /*We may be able to drop some redundant bytes from the end.*/
 148     buf=ec_byte_get_buffer(_this->buf);
 149     i=ec_byte_bytes(_this->buf);
 150     /*Strip trailing zeros.*/
 151     do i--;
 152     while(i>0&&!buf[i]);
 153     /*Strip one trailing EC_FOF_RSV1 byte if the buffer ends in a string of
 154        consecutive EC_FOF_RSV1 bytes preceded by one (or more) zeros.*/
 155     if(i>0&&buf[i]==EC_FOF_RSV1){
 156       long j;
 157       j=i;
 158       do j--;
 159       while(j>0&&buf[j]==EC_FOF_RSV1);
 160       if(!buf[j])i--;
 161     }
 162     ec_byte_writetrunc(_this->buf,i+1);
 163   }
 164 }