• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2010 Google Inc. All Rights Reserved.
2 //
3 // This code is licensed under the same terms as WebM:
4 //  Software License Agreement:  http://www.webmproject.org/license/software/
5 //  Additional IP Rights Grant:  http://www.webmproject.org/license/additional/
6 // -----------------------------------------------------------------------------
7 //
8 // Boolean decoder
9 //
10 // Author: Skal (pascal.massimino@gmail.com)
11 //         Vikas Arora (vikaas.arora@gmail.com)
12 
13 #ifndef WEBP_UTILS_BIT_READER_H_
14 #define WEBP_UTILS_BIT_READER_H_
15 
16 #include <assert.h>
17 #ifdef _MSC_VER
18 #include <stdlib.h>  // _byteswap_ulong
19 #endif
20 #include <string.h>  // For memcpy
21 #include "webp/types.h"
22 
23 #if defined(__cplusplus) || defined(c_plusplus)
24 extern "C" {
25 #endif
26 
27 // The Boolean decoder needs to maintain infinite precision on the value_ field.
28 // However, since range_ is only 8bit, we only need an active window of 8 bits
29 // for value_. Left bits (MSB) gets zeroed and shifted away when value_ falls
30 // below 128, range_ is updated, and fresh bits read from the bitstream are
31 // brought in as LSB.
32 // To avoid reading the fresh bits one by one (slow), we cache a few of them
33 // ahead (actually, we cache BITS of them ahead. See below). There's two
34 // strategies regarding how to shift these looked-ahead fresh bits into the
35 // 8bit window of value_: either we shift them in, while keeping the position of
36 // the window fixed. Or we slide the window to the right while keeping the cache
37 // bits at a fixed, right-justified, position.
38 //
39 //  Example, for BITS=16: here is the content of value_ for both strategies:
40 //
41 //          !USE_RIGHT_JUSTIFY            ||        USE_RIGHT_JUSTIFY
42 //                                        ||
43 //   <- 8b -><- 8b -><- BITS bits  ->     ||  <- 8b+3b -><- 8b -><- 13 bits ->
44 //   [unused][value_][cached bits][0]     ||  [unused...][value_][cached bits]
45 //  [........00vvvvvvBBBBBBBBBBBBB000]LSB || [...........00vvvvvvBBBBBBBBBBBBB]
46 //                                        ||
47 // After calling VP8Shift(), where we need to shift away two zeros:
48 //  [........vvvvvvvvBBBBBBBBBBB00000]LSB || [.............vvvvvvvvBBBBBBBBBBB]
49 //                                        ||
50 // Just before we need to call VP8LoadNewBytes(), the situation is:
51 //  [........vvvvvv000000000000000000]LSB || [..........................vvvvvv]
52 //                                        ||
53 // And just after calling VP8LoadNewBytes():
54 //  [........vvvvvvvvBBBBBBBBBBBBBBBB]LSB || [........vvvvvvvvBBBBBBBBBBBBBBBB]
55 //
56 // -> we're back to height active 'value_' bits (marked 'v') and BITS cached
57 // bits (marked 'B')
58 //
59 // The right-justify strategy tends to use less shifts and is often faster.
60 
61 //------------------------------------------------------------------------------
62 // BITS can be either 32, 24, 16 or 8.
63 // Pick values that fit natural register size.
64 
65 #if !defined(WEBP_REFERENCE_IMPLEMENTATION)
66 
67 #define USE_RIGHT_JUSTIFY
68 
69 #if defined(__i386__) || defined(_M_IX86)      // x86 32bit
70 #define BITS 16
71 #elif defined(__arm__) || defined(_M_ARM)     // ARM
72 #define BITS 24
73 #else                      // reasonable default
74 #define BITS 24
75 #endif
76 
77 #else     // reference choices
78 
79 #define USE_RIGHT_JUSTIFY
80 #define BITS 8
81 
82 #endif
83 
84 //------------------------------------------------------------------------------
85 // Derived types and constants
86 
87 #if (BITS == 32)
88 typedef uint64_t bit_t;   // natural register type
89 typedef uint32_t lbit_t;  // natural type for memory I/O
90 #elif (BITS == 24)
91 typedef uint32_t bit_t;
92 typedef uint32_t lbit_t;
93 #elif (BITS == 16)
94 typedef uint32_t bit_t;
95 typedef uint16_t lbit_t;
96 #else
97 typedef uint32_t bit_t;
98 typedef uint8_t lbit_t;
99 #endif
100 
101 #ifndef USE_RIGHT_JUSTIFY
102 typedef bit_t range_t;     // type for storing range_
103 #define MASK ((((bit_t)1) << (BITS)) - 1)
104 #else
105 typedef uint32_t range_t;  // range_ only uses 8bits here. No need for bit_t.
106 #endif
107 
108 //------------------------------------------------------------------------------
109 // Bitreader
110 
111 typedef struct VP8BitReader VP8BitReader;
112 struct VP8BitReader {
113   const uint8_t* buf_;        // next byte to be read
114   const uint8_t* buf_end_;    // end of read buffer
115   int eof_;                   // true if input is exhausted
116 
117   // boolean decoder
118   range_t range_;            // current range minus 1. In [127, 254] interval.
119   bit_t value_;              // current value
120   int bits_;                 // number of valid bits left
121 };
122 
123 // Initialize the bit reader and the boolean decoder.
124 void VP8InitBitReader(VP8BitReader* const br,
125                       const uint8_t* const start, const uint8_t* const end);
126 
127 // return the next value made of 'num_bits' bits
128 uint32_t VP8GetValue(VP8BitReader* const br, int num_bits);
VP8Get(VP8BitReader * const br)129 static WEBP_INLINE uint32_t VP8Get(VP8BitReader* const br) {
130   return VP8GetValue(br, 1);
131 }
132 
133 // return the next value with sign-extension.
134 int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits);
135 
136 // Read a bit with proba 'prob'. Speed-critical function!
137 extern const uint8_t kVP8Log2Range[128];
138 extern const range_t kVP8NewRange[128];
139 
140 void VP8LoadFinalBytes(VP8BitReader* const br);    // special case for the tail
141 
VP8LoadNewBytes(VP8BitReader * const br)142 static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) {
143   assert(br != NULL && br->buf_ != NULL);
144   // Read 'BITS' bits at a time if possible.
145   if (br->buf_ + sizeof(lbit_t) <= br->buf_end_) {
146     // convert memory type to register type (with some zero'ing!)
147     bit_t bits;
148     lbit_t in_bits = *(lbit_t*)br->buf_;
149     br->buf_ += (BITS) >> 3;
150 #if !defined(__BIG_ENDIAN__)
151 #if (BITS == 32) || (BITS == 24)
152 #if defined(__i386__) || defined(__x86_64__)
153     __asm__ volatile("bswap %k0" : "=r"(in_bits) : "0"(in_bits));
154     bits = (bit_t)in_bits;   // 24b/32b -> 32b/64b zero-extension
155 #elif defined(_MSC_VER)
156     bits = _byteswap_ulong(in_bits);
157 #else
158     bits = (bit_t)(in_bits >> 24) | ((in_bits >> 8) & 0xff00)
159          | ((in_bits << 8) & 0xff0000)  | (in_bits << 24);
160 #endif  // x86
161 #if (BITS == 24)
162     bits >>= 8;
163 #endif
164 #elif (BITS == 16)
165     // gcc will recognize a 'rorw $8, ...' here:
166     bits = (bit_t)(in_bits >> 8) | ((in_bits & 0xff) << 8);
167 #else   // BITS == 8
168     bits = (bit_t)in_bits;
169 #endif
170 #else    // BIG_ENDIAN
171     bits = (bit_t)in_bits;
172 #endif
173 #ifndef USE_RIGHT_JUSTIFY
174     br->value_ |= bits << (-br->bits_);
175 #else
176     br->value_ = bits | (br->value_ << (BITS));
177 #endif
178     br->bits_ += (BITS);
179   } else {
180     VP8LoadFinalBytes(br);    // no need to be inlined
181   }
182 }
183 
VP8BitUpdate(VP8BitReader * const br,range_t split)184 static WEBP_INLINE int VP8BitUpdate(VP8BitReader* const br, range_t split) {
185   if (br->bits_ < 0) {  // Make sure we have a least BITS bits in 'value_'
186     VP8LoadNewBytes(br);
187   }
188 #ifndef USE_RIGHT_JUSTIFY
189   split |= (MASK);
190   if (br->value_ > split) {
191     br->range_ -= split + 1;
192     br->value_ -= split + 1;
193     return 1;
194   } else {
195     br->range_ = split;
196     return 0;
197   }
198 #else
199   {
200     const int pos = br->bits_;
201     const range_t value = (range_t)(br->value_ >> pos);
202     if (value > split) {
203       br->range_ -= split + 1;
204       br->value_ -= (bit_t)(split + 1) << pos;
205       return 1;
206     } else {
207       br->range_ = split;
208       return 0;
209     }
210   }
211 #endif
212 }
213 
VP8Shift(VP8BitReader * const br)214 static WEBP_INLINE void VP8Shift(VP8BitReader* const br) {
215 #ifndef USE_RIGHT_JUSTIFY
216   // range_ is in [0..127] interval here.
217   const bit_t idx = br->range_ >> (BITS);
218   const int shift = kVP8Log2Range[idx];
219   br->range_ = kVP8NewRange[idx];
220   br->value_ <<= shift;
221   br->bits_ -= shift;
222 #else
223   const int shift = kVP8Log2Range[br->range_];
224   assert(br->range_ < (range_t)128);
225   br->range_ = kVP8NewRange[br->range_];
226   br->bits_ -= shift;
227 #endif
228 }
VP8GetBit(VP8BitReader * const br,int prob)229 static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) {
230 #ifndef USE_RIGHT_JUSTIFY
231   // It's important to avoid generating a 64bit x 64bit multiply here.
232   // We just need an 8b x 8b after all.
233   const range_t split =
234       (range_t)((uint32_t)(br->range_ >> (BITS)) * prob) << ((BITS) - 8);
235   const int bit = VP8BitUpdate(br, split);
236   if (br->range_ <= (((range_t)0x7e << (BITS)) | (MASK))) {
237     VP8Shift(br);
238   }
239   return bit;
240 #else
241   const range_t split = (br->range_ * prob) >> 8;
242   const int bit = VP8BitUpdate(br, split);
243   if (br->range_ <= (range_t)0x7e) {
244     VP8Shift(br);
245   }
246   return bit;
247 #endif
248 }
249 
VP8GetSigned(VP8BitReader * const br,int v)250 static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v) {
251   const bit_t split = (br->range_ >> 1);
252   const int bit = VP8BitUpdate(br, split);
253   VP8Shift(br);
254   return bit ? -v : v;
255 }
256 
257 
258 // -----------------------------------------------------------------------------
259 // Bitreader for lossless format
260 
261 typedef uint64_t vp8l_val_t;  // right now, this bit-reader can only use 64bit.
262 
263 typedef struct {
264   vp8l_val_t     val_;        // pre-fetched bits
265   const uint8_t* buf_;        // input byte buffer
266   size_t         len_;        // buffer length
267   size_t         pos_;        // byte position in buf_
268   int            bit_pos_;    // current bit-reading position in val_
269   int            eos_;        // bitstream is finished
270   int            error_;      // an error occurred (buffer overflow attempt...)
271 } VP8LBitReader;
272 
273 void VP8LInitBitReader(VP8LBitReader* const br,
274                        const uint8_t* const start,
275                        size_t length);
276 
277 //  Sets a new data buffer.
278 void VP8LBitReaderSetBuffer(VP8LBitReader* const br,
279                             const uint8_t* const buffer, size_t length);
280 
281 // Reads the specified number of bits from Read Buffer.
282 // Flags an error in case end_of_stream or n_bits is more than allowed limit.
283 // Flags eos if this read attempt is going to cross the read buffer.
284 uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits);
285 
286 // Return the prefetched bits, so they can be looked up.
VP8LPrefetchBits(VP8LBitReader * const br)287 static WEBP_INLINE uint32_t VP8LPrefetchBits(VP8LBitReader* const br) {
288   return (uint32_t)(br->val_ >> br->bit_pos_);
289 }
290 
291 // Discard 'num_bits' bits from the cache.
VP8LDiscardBits(VP8LBitReader * const br,int num_bits)292 static WEBP_INLINE void VP8LDiscardBits(VP8LBitReader* const br, int num_bits) {
293   br->bit_pos_ += num_bits;
294 }
295 
296 // Advances the Read buffer by 4 bytes to make room for reading next 32 bits.
297 void VP8LFillBitWindow(VP8LBitReader* const br);
298 
299 #if defined(__cplusplus) || defined(c_plusplus)
300 }    // extern "C"
301 #endif
302 
303 #endif  /* WEBP_UTILS_BIT_READER_H_ */
304