• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2001-2016, Alliance for Open Media. All rights reserved
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #include <assert.h>
13 #include "aom_dsp/entdec.h"
14 #include "aom_dsp/prob.h"
15 
16 /*A range decoder.
17   This is an entropy decoder based upon \cite{Mar79}, which is itself a
18    rediscovery of the FIFO arithmetic code introduced by \cite{Pas76}.
19   It is very similar to arithmetic encoding, except that encoding is done with
20    digits in any base, instead of with bits, and so it is faster when using
21    larger bases (i.e.: a byte).
22   The author claims an average waste of $\frac{1}{2}\log_b(2b)$ bits, where $b$
23    is the base, longer than the theoretical optimum, but to my knowledge there
24    is no published justification for this claim.
25   This only seems true when using near-infinite precision arithmetic so that
26    the process is carried out with no rounding errors.
27 
28   An excellent description of implementation details is available at
29    http://www.arturocampos.com/ac_range.html
30   A recent work \cite{MNW98} which proposes several changes to arithmetic
31    encoding for efficiency actually re-discovers many of the principles
32    behind range encoding, and presents a good theoretical analysis of them.
33 
34   End of stream is handled by writing out the smallest number of bits that
35    ensures that the stream will be correctly decoded regardless of the value of
36    any subsequent bits.
37   od_ec_dec_tell() can be used to determine how many bits were needed to decode
38    all the symbols thus far; other data can be packed in the remaining bits of
39    the input buffer.
40   @PHDTHESIS{Pas76,
41     author="Richard Clark Pasco",
42     title="Source coding algorithms for fast data compression",
43     school="Dept. of Electrical Engineering, Stanford University",
44     address="Stanford, CA",
45     month=May,
46     year=1976,
47     URL="http://www.richpasco.org/scaffdc.pdf"
48   }
49   @INPROCEEDINGS{Mar79,
50    author="Martin, G.N.N.",
51    title="Range encoding: an algorithm for removing redundancy from a digitised
52     message",
53    booktitle="Video & Data Recording Conference",
54    year=1979,
55    address="Southampton",
56    month=Jul,
57    URL="http://www.compressconsult.com/rangecoder/rngcod.pdf.gz"
58   }
59   @ARTICLE{MNW98,
60    author="Alistair Moffat and Radford Neal and Ian H. Witten",
61    title="Arithmetic Coding Revisited",
62    journal="{ACM} Transactions on Information Systems",
63    year=1998,
64    volume=16,
65    number=3,
66    pages="256--294",
67    month=Jul,
68    URL="http://researchcommons.waikato.ac.nz/bitstream/handle/10289/78/content.pdf"
69   }*/
70 
71 /*This is meant to be a large, positive constant that can still be efficiently
72    loaded as an immediate (on platforms like ARM, for example).
73   Even relatively modest values like 100 would work fine.*/
74 #define OD_EC_LOTS_OF_BITS (0x4000)
75 
76 /*The return value of od_ec_dec_tell does not change across an od_ec_dec_refill
77    call.*/
od_ec_dec_refill(od_ec_dec * dec)78 static void od_ec_dec_refill(od_ec_dec *dec) {
79   int s;
80   od_ec_window dif;
81   int16_t cnt;
82   const unsigned char *bptr;
83   const unsigned char *end;
84   dif = dec->dif;
85   cnt = dec->cnt;
86   bptr = dec->bptr;
87   end = dec->end;
88   s = OD_EC_WINDOW_SIZE - 9 - (cnt + 15);
89   for (; s >= 0 && bptr < end; s -= 8, bptr++) {
90     /*Each time a byte is inserted into the window (dif), bptr advances and cnt
91        is incremented by 8, so the total number of consumed bits (the return
92        value of od_ec_dec_tell) does not change.*/
93     assert(s <= OD_EC_WINDOW_SIZE - 8);
94     dif ^= (od_ec_window)bptr[0] << s;
95     cnt += 8;
96   }
97   if (bptr >= end) {
98     /*We've reached the end of the buffer. It is perfectly valid for us to need
99        to fill the window with additional bits past the end of the buffer (and
100        this happens in normal operation). These bits should all just be taken
101        as zero. But we cannot increment bptr past 'end' (this is undefined
102        behavior), so we start to increment dec->tell_offs. We also don't want
103        to keep testing bptr against 'end', so we set cnt to OD_EC_LOTS_OF_BITS
104        and adjust dec->tell_offs so that the total number of unconsumed bits in
105        the window (dec->cnt - dec->tell_offs) does not change. This effectively
106        puts lots of zero bits into the window, and means we won't try to refill
107        it from the buffer for a very long time (at which point we'll put lots
108        of zero bits into the window again).*/
109     dec->tell_offs += OD_EC_LOTS_OF_BITS - cnt;
110     cnt = OD_EC_LOTS_OF_BITS;
111   }
112   dec->dif = dif;
113   dec->cnt = cnt;
114   dec->bptr = bptr;
115 }
116 
117 /*Takes updated dif and range values, renormalizes them so that
118    32768 <= rng < 65536 (reading more bytes from the stream into dif if
119    necessary), and stores them back in the decoder context.
120   dif: The new value of dif.
121   rng: The new value of the range.
122   ret: The value to return.
123   Return: ret.
124           This allows the compiler to jump to this function via a tail-call.*/
od_ec_dec_normalize(od_ec_dec * dec,od_ec_window dif,unsigned rng,int ret)125 static int od_ec_dec_normalize(od_ec_dec *dec, od_ec_window dif, unsigned rng,
126                                int ret) {
127   int d;
128   assert(rng <= 65535U);
129   /*The number of leading zeros in the 16-bit binary representation of rng.*/
130   d = 16 - OD_ILOG_NZ(rng);
131   /*d bits in dec->dif are consumed.*/
132   dec->cnt -= d;
133   /*This is equivalent to shifting in 1's instead of 0's.*/
134   dec->dif = ((dif + 1) << d) - 1;
135   dec->rng = rng << d;
136   if (dec->cnt < 0) od_ec_dec_refill(dec);
137   return ret;
138 }
139 
140 /*Initializes the decoder.
141   buf: The input buffer to use.
142   storage: The size in bytes of the input buffer.*/
od_ec_dec_init(od_ec_dec * dec,const unsigned char * buf,uint32_t storage)143 void od_ec_dec_init(od_ec_dec *dec, const unsigned char *buf,
144                     uint32_t storage) {
145   dec->buf = buf;
146   dec->tell_offs = 10 - (OD_EC_WINDOW_SIZE - 8);
147   dec->end = buf + storage;
148   dec->bptr = buf;
149   dec->dif = ((od_ec_window)1 << (OD_EC_WINDOW_SIZE - 1)) - 1;
150   dec->rng = 0x8000;
151   dec->cnt = -15;
152   od_ec_dec_refill(dec);
153 }
154 
155 /*Decode a single binary value.
156   f: The probability that the bit is one, scaled by 32768.
157   Return: The value decoded (0 or 1).*/
od_ec_decode_bool_q15(od_ec_dec * dec,unsigned f)158 int od_ec_decode_bool_q15(od_ec_dec *dec, unsigned f) {
159   od_ec_window dif;
160   od_ec_window vw;
161   unsigned r;
162   unsigned r_new;
163   unsigned v;
164   int ret;
165   assert(0 < f);
166   assert(f < 32768U);
167   dif = dec->dif;
168   r = dec->rng;
169   assert(dif >> (OD_EC_WINDOW_SIZE - 16) < r);
170   assert(32768U <= r);
171   v = ((r >> 8) * (uint32_t)(f >> EC_PROB_SHIFT) >> (7 - EC_PROB_SHIFT));
172   v += EC_MIN_PROB;
173   vw = (od_ec_window)v << (OD_EC_WINDOW_SIZE - 16);
174   ret = 1;
175   r_new = v;
176   if (dif >= vw) {
177     r_new = r - v;
178     dif -= vw;
179     ret = 0;
180   }
181   return od_ec_dec_normalize(dec, dif, r_new, ret);
182 }
183 
184 /*Decodes a symbol given an inverse cumulative distribution function (CDF)
185    table in Q15.
186   icdf: CDF_PROB_TOP minus the CDF, such that symbol s falls in the range
187          [s > 0 ? (CDF_PROB_TOP - icdf[s - 1]) : 0, CDF_PROB_TOP - icdf[s]).
188         The values must be monotonically non-increasing, and icdf[nsyms - 1]
189          must be 0.
190   nsyms: The number of symbols in the alphabet.
191          This should be at most 16.
192   Return: The decoded symbol s.*/
od_ec_decode_cdf_q15(od_ec_dec * dec,const uint16_t * icdf,int nsyms)193 int od_ec_decode_cdf_q15(od_ec_dec *dec, const uint16_t *icdf, int nsyms) {
194   od_ec_window dif;
195   unsigned r;
196   unsigned c;
197   unsigned u;
198   unsigned v;
199   int ret;
200   (void)nsyms;
201   dif = dec->dif;
202   r = dec->rng;
203   const int N = nsyms - 1;
204 
205   assert(dif >> (OD_EC_WINDOW_SIZE - 16) < r);
206   assert(icdf[nsyms - 1] == OD_ICDF(CDF_PROB_TOP));
207   assert(32768U <= r);
208   assert(7 - EC_PROB_SHIFT - CDF_SHIFT >= 0);
209   c = (unsigned)(dif >> (OD_EC_WINDOW_SIZE - 16));
210   v = r;
211   ret = -1;
212   do {
213     u = v;
214     v = ((r >> 8) * (uint32_t)(icdf[++ret] >> EC_PROB_SHIFT) >>
215          (7 - EC_PROB_SHIFT - CDF_SHIFT));
216     v += EC_MIN_PROB * (N - ret);
217   } while (c < v);
218   assert(v < u);
219   assert(u <= r);
220   r = u - v;
221   dif -= (od_ec_window)v << (OD_EC_WINDOW_SIZE - 16);
222   return od_ec_dec_normalize(dec, dif, r, ret);
223 }
224 
225 /*Returns the number of bits "used" by the decoded symbols so far.
226   This same number can be computed in either the encoder or the decoder, and is
227    suitable for making coding decisions.
228   Return: The number of bits.
229           This will always be slightly larger than the exact value (e.g., all
230            rounding error is in the positive direction).*/
od_ec_dec_tell(const od_ec_dec * dec)231 int od_ec_dec_tell(const od_ec_dec *dec) {
232   /*There is a window of bits stored in dec->dif. The difference
233      (dec->bptr - dec->buf) tells us how many bytes have been read into this
234      window. The difference (dec->cnt - dec->tell_offs) tells us how many of
235      the bits in that window remain unconsumed.*/
236   return (int)((dec->bptr - dec->buf) * 8 - dec->cnt + dec->tell_offs);
237 }
238 
239 /*Returns the number of bits "used" by the decoded symbols so far.
240   This same number can be computed in either the encoder or the decoder, and is
241    suitable for making coding decisions.
242   Return: The number of bits scaled by 2**OD_BITRES.
243           This will always be slightly larger than the exact value (e.g., all
244            rounding error is in the positive direction).*/
od_ec_dec_tell_frac(const od_ec_dec * dec)245 uint32_t od_ec_dec_tell_frac(const od_ec_dec *dec) {
246   return od_ec_tell_frac(od_ec_dec_tell(dec), dec->rng);
247 }
248