1 /* Copyright (c) 2001-2011 Timothy B. Terriberry
2    Copyright (c) 2008-2009 Xiph.Org Foundation */
3 /*
4    Redistribution and use in source and binary forms, with or without
5    modification, are permitted provided that the following conditions
6    are met:
7 
8    - Redistributions of source code must retain the above copyright
9    notice, this list of conditions and the following disclaimer.
10 
11    - Redistributions in binary form must reproduce the above copyright
12    notice, this list of conditions and the following disclaimer in the
13    documentation and/or other materials provided with the distribution.
14 
15    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
19    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
20    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
21    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
22    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
23    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
24    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27 
28 #ifdef HAVE_CONFIG_H
29 #include "config.h"
30 #endif
31 
32 #include <stddef.h>
33 #include "os_support.h"
34 #include "arch.h"
35 #include "entdec.h"
36 #include "mfrngcod.h"
37 
38 /*A range decoder.
39   This is an entropy decoder based upon \cite{Mar79}, which is itself a
40    rediscovery of the FIFO arithmetic code introduced by \cite{Pas76}.
41   It is very similar to arithmetic encoding, except that encoding is done with
42    digits in any base, instead of with bits, and so it is faster when using
43    larger bases (i.e.: a byte).
44   The author claims an average waste of $\frac{1}{2}\log_b(2b)$ bits, where $b$
45    is the base, longer than the theoretical optimum, but to my knowledge there
46    is no published justification for this claim.
47   This only seems true when using near-infinite precision arithmetic so that
48    the process is carried out with no rounding errors.
49 
50   An excellent description of implementation details is available at
51    http://www.arturocampos.com/ac_range.html
52   A recent work \cite{MNW98} which proposes several changes to arithmetic
53    encoding for efficiency actually re-discovers many of the principles
54    behind range encoding, and presents a good theoretical analysis of them.
55 
56   End of stream is handled by writing out the smallest number of bits that
57    ensures that the stream will be correctly decoded regardless of the value of
58    any subsequent bits.
59   ec_tell() can be used to determine how many bits were needed to decode
60    all the symbols thus far; other data can be packed in the remaining bits of
61    the input buffer.
62   @PHDTHESIS{Pas76,
63     author="Richard Clark Pasco",
64     title="Source coding algorithms for fast data compression",
65     school="Dept. of Electrical Engineering, Stanford University",
66     address="Stanford, CA",
67     month=May,
68     year=1976
69   }
70   @INPROCEEDINGS{Mar79,
71    author="Martin, G.N.N.",
72    title="Range encoding: an algorithm for removing redundancy from a digitised
73     message",
74    booktitle="Video & Data Recording Conference",
75    year=1979,
76    address="Southampton",
77    month=Jul
78   }
79   @ARTICLE{MNW98,
80    author="Alistair Moffat and Radford Neal and Ian H. Witten",
81    title="Arithmetic Coding Revisited",
82    journal="{ACM} Transactions on Information Systems",
83    year=1998,
84    volume=16,
85    number=3,
86    pages="256--294",
87    month=Jul,
88    URL="http://www.stanford.edu/class/ee398a/handouts/papers/Moffat98ArithmCoding.pdf"
89   }*/
90 
ec_read_byte(ec_dec * _this)91 static int ec_read_byte(ec_dec *_this){
92   return _this->offs<_this->storage?_this->buf[_this->offs++]:0;
93 }
94 
ec_read_byte_from_end(ec_dec * _this)95 static int ec_read_byte_from_end(ec_dec *_this){
96   return _this->end_offs<_this->storage?
97    _this->buf[_this->storage-++(_this->end_offs)]:0;
98 }
99 
100 /*Normalizes the contents of val and rng so that rng lies entirely in the
101    high-order symbol.*/
ec_dec_normalize(ec_dec * _this)102 static void ec_dec_normalize(ec_dec *_this){
103   /*If the range is too small, rescale it and input some bits.*/
104   while(_this->rng<=EC_CODE_BOT){
105     int sym;
106     _this->nbits_total+=EC_SYM_BITS;
107     _this->rng<<=EC_SYM_BITS;
108     /*Use up the remaining bits from our last symbol.*/
109     sym=_this->rem;
110     /*Read the next value from the input.*/
111     _this->rem=ec_read_byte(_this);
112     /*Take the rest of the bits we need from this new symbol.*/
113     sym=(sym<<EC_SYM_BITS|_this->rem)>>(EC_SYM_BITS-EC_CODE_EXTRA);
114     /*And subtract them from val, capped to be less than EC_CODE_TOP.*/
115     _this->val=((_this->val<<EC_SYM_BITS)+(EC_SYM_MAX&~sym))&(EC_CODE_TOP-1);
116   }
117 }
118 
ec_dec_init(ec_dec * _this,unsigned char * _buf,opus_uint32 _storage)119 void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage){
120   _this->buf=_buf;
121   _this->storage=_storage;
122   _this->end_offs=0;
123   _this->end_window=0;
124   _this->nend_bits=0;
125   /*This is the offset from which ec_tell() will subtract partial bits.
126     The final value after the ec_dec_normalize() call will be the same as in
127      the encoder, but we have to compensate for the bits that are added there.*/
128   _this->nbits_total=EC_CODE_BITS+1
129    -((EC_CODE_BITS-EC_CODE_EXTRA)/EC_SYM_BITS)*EC_SYM_BITS;
130   _this->offs=0;
131   _this->rng=1U<<EC_CODE_EXTRA;
132   _this->rem=ec_read_byte(_this);
133   _this->val=_this->rng-1-(_this->rem>>(EC_SYM_BITS-EC_CODE_EXTRA));
134   _this->error=0;
135   /*Normalize the interval.*/
136   ec_dec_normalize(_this);
137 }
138 
ec_decode(ec_dec * _this,unsigned _ft)139 unsigned ec_decode(ec_dec *_this,unsigned _ft){
140   unsigned s;
141   _this->ext=_this->rng/_ft;
142   s=(unsigned)(_this->val/_this->ext);
143   return _ft-EC_MINI(s+1,_ft);
144 }
145 
ec_decode_bin(ec_dec * _this,unsigned _bits)146 unsigned ec_decode_bin(ec_dec *_this,unsigned _bits){
147    unsigned s;
148    _this->ext=_this->rng>>_bits;
149    s=(unsigned)(_this->val/_this->ext);
150    return (1U<<_bits)-EC_MINI(s+1U,1U<<_bits);
151 }
152 
ec_dec_update(ec_dec * _this,unsigned _fl,unsigned _fh,unsigned _ft)153 void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft){
154   opus_uint32 s;
155   s=IMUL32(_this->ext,_ft-_fh);
156   _this->val-=s;
157   _this->rng=_fl>0?IMUL32(_this->ext,_fh-_fl):_this->rng-s;
158   ec_dec_normalize(_this);
159 }
160 
161 /*The probability of having a "one" is 1/(1<<_logp).*/
ec_dec_bit_logp(ec_dec * _this,unsigned _logp)162 int ec_dec_bit_logp(ec_dec *_this,unsigned _logp){
163   opus_uint32 r;
164   opus_uint32 d;
165   opus_uint32 s;
166   int         ret;
167   r=_this->rng;
168   d=_this->val;
169   s=r>>_logp;
170   ret=d<s;
171   if(!ret)_this->val=d-s;
172   _this->rng=ret?s:r-s;
173   ec_dec_normalize(_this);
174   return ret;
175 }
176 
ec_dec_icdf(ec_dec * _this,const unsigned char * _icdf,unsigned _ftb)177 int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb){
178   opus_uint32 r;
179   opus_uint32 d;
180   opus_uint32 s;
181   opus_uint32 t;
182   int         ret;
183   s=_this->rng;
184   d=_this->val;
185   r=s>>_ftb;
186   ret=-1;
187   do{
188     t=s;
189     s=IMUL32(r,_icdf[++ret]);
190   }
191   while(d<s);
192   _this->val=d-s;
193   _this->rng=t-s;
194   ec_dec_normalize(_this);
195   return ret;
196 }
197 
ec_dec_uint(ec_dec * _this,opus_uint32 _ft)198 opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft){
199   unsigned ft;
200   unsigned s;
201   int      ftb;
202   /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/
203   celt_assert(_ft>1);
204   _ft--;
205   ftb=EC_ILOG(_ft);
206   if(ftb>EC_UINT_BITS){
207     opus_uint32 t;
208     ftb-=EC_UINT_BITS;
209     ft=(unsigned)(_ft>>ftb)+1;
210     s=ec_decode(_this,ft);
211     ec_dec_update(_this,s,s+1,ft);
212     t=(opus_uint32)s<<ftb|ec_dec_bits(_this,ftb);
213     if(t<=_ft)return t;
214     _this->error=1;
215     return _ft;
216   }
217   else{
218     _ft++;
219     s=ec_decode(_this,(unsigned)_ft);
220     ec_dec_update(_this,s,s+1,(unsigned)_ft);
221     return s;
222   }
223 }
224 
ec_dec_bits(ec_dec * _this,unsigned _bits)225 opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _bits){
226   ec_window   window;
227   int         available;
228   opus_uint32 ret;
229   window=_this->end_window;
230   available=_this->nend_bits;
231   if((unsigned)available<_bits){
232     do{
233       window|=(ec_window)ec_read_byte_from_end(_this)<<available;
234       available+=EC_SYM_BITS;
235     }
236     while(available<=EC_WINDOW_SIZE-EC_SYM_BITS);
237   }
238   ret=(opus_uint32)window&(((opus_uint32)1<<_bits)-1U);
239   window>>=_bits;
240   available-=_bits;
241   _this->end_window=window;
242   _this->nend_bits=available;
243   _this->nbits_total+=_bits;
244   return ret;
245 }
246