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) 24extern "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) 88typedef uint64_t bit_t; // natural register type 89typedef uint32_t lbit_t; // natural type for memory I/O 90#elif (BITS == 24) 91typedef uint32_t bit_t; 92typedef uint32_t lbit_t; 93#elif (BITS == 16) 94typedef uint32_t bit_t; 95typedef uint16_t lbit_t; 96#else 97typedef uint32_t bit_t; 98typedef uint8_t lbit_t; 99#endif 100 101#ifndef USE_RIGHT_JUSTIFY 102typedef bit_t range_t; // type for storing range_ 103#define MASK ((((bit_t)1) << (BITS)) - 1) 104#else 105typedef uint32_t range_t; // range_ only uses 8bits here. No need for bit_t. 106#endif 107 108//------------------------------------------------------------------------------ 109// Bitreader 110 111typedef struct VP8BitReader VP8BitReader; 112struct 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. 124void 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 128uint32_t VP8GetValue(VP8BitReader* const br, int num_bits); 129static WEBP_INLINE uint32_t VP8Get(VP8BitReader* const br) { 130 return VP8GetValue(br, 1); 131} 132 133// return the next value with sign-extension. 134int32_t VP8GetSignedValue(VP8BitReader* const br, int num_bits); 135 136// Read a bit with proba 'prob'. Speed-critical function! 137extern const uint8_t kVP8Log2Range[128]; 138extern const range_t kVP8NewRange[128]; 139 140void VP8LoadFinalBytes(VP8BitReader* const br); // special case for the tail 141 142static 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 184static 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 214static 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} 229static 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 250static 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 261typedef uint64_t vp8l_val_t; // right now, this bit-reader can only use 64bit. 262 263typedef 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 273void VP8LInitBitReader(VP8LBitReader* const br, 274 const uint8_t* const start, 275 size_t length); 276 277// Sets a new data buffer. 278void 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. 284uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits); 285 286// Return the prefetched bits, so they can be looked up. 287static 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. 292static 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. 297void 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