| OLD | NEW |
| 1 // Copyright 2010 Google Inc. All Rights Reserved. | 1 // Copyright 2010 Google Inc. All Rights Reserved. |
| 2 // | 2 // |
| 3 // Use of this source code is governed by a BSD-style license | 3 // Use of this source code is governed by a BSD-style license |
| 4 // that can be found in the COPYING file in the root of the source | 4 // that can be found in the COPYING file in the root of the source |
| 5 // tree. An additional intellectual property rights grant can be found | 5 // tree. An additional intellectual property rights grant can be found |
| 6 // in the file PATENTS. All contributing project authors may | 6 // in the file PATENTS. All contributing project authors may |
| 7 // be found in the AUTHORS file in the root of the source tree. | 7 // be found in the AUTHORS file in the root of the source tree. |
| 8 // ----------------------------------------------------------------------------- | 8 // ----------------------------------------------------------------------------- |
| 9 // | 9 // |
| 10 // Boolean decoder | 10 // Boolean decoder |
| 11 // | 11 // |
| 12 // Author: Skal (pascal.massimino@gmail.com) | 12 // Author: Skal (pascal.massimino@gmail.com) |
| 13 // Vikas Arora (vikaas.arora@gmail.com) | 13 // Vikas Arora (vikaas.arora@gmail.com) |
| 14 | 14 |
| 15 #ifndef WEBP_UTILS_BIT_READER_H_ | 15 #ifndef WEBP_UTILS_BIT_READER_H_ |
| 16 #define WEBP_UTILS_BIT_READER_H_ | 16 #define WEBP_UTILS_BIT_READER_H_ |
| 17 | 17 |
| 18 #include <assert.h> | 18 #include <assert.h> |
| 19 #ifdef _MSC_VER | 19 #ifdef _MSC_VER |
| 20 #include <stdlib.h> // _byteswap_ulong | 20 #include <stdlib.h> // _byteswap_ulong |
| 21 #endif | 21 #endif |
| 22 #include <string.h> // For memcpy | |
| 23 #include "../webp/types.h" | 22 #include "../webp/types.h" |
| 24 | 23 |
| 25 #if defined(__cplusplus) || defined(c_plusplus) | 24 #ifdef __cplusplus |
| 26 extern "C" { | 25 extern "C" { |
| 27 #endif | 26 #endif |
| 28 | 27 |
| 29 // The Boolean decoder needs to maintain infinite precision on the value_ field. | 28 // The Boolean decoder needs to maintain infinite precision on the value_ field. |
| 30 // However, since range_ is only 8bit, we only need an active window of 8 bits | 29 // However, since range_ is only 8bit, we only need an active window of 8 bits |
| 31 // for value_. Left bits (MSB) gets zeroed and shifted away when value_ falls | 30 // for value_. Left bits (MSB) gets zeroed and shifted away when value_ falls |
| 32 // below 128, range_ is updated, and fresh bits read from the bitstream are | 31 // below 128, range_ is updated, and fresh bits read from the bitstream are |
| 33 // brought in as LSB. | 32 // brought in as LSB. |
| 34 // To avoid reading the fresh bits one by one (slow), we cache a few of them | 33 // To avoid reading the fresh bits one by one (slow), we cache a few of them |
| 35 // ahead (actually, we cache BITS of them ahead. See below). There's two | 34 // ahead (actually, we cache BITS of them ahead. See below). There's two |
| (...skipping 12 matching lines...) Expand all Loading... |
| 48 // || | 47 // || |
| 49 // After calling VP8Shift(), where we need to shift away two zeros: | 48 // After calling VP8Shift(), where we need to shift away two zeros: |
| 50 // [........vvvvvvvvBBBBBBBBBBB00000]LSB || [.............vvvvvvvvBBBBBBBBBBB] | 49 // [........vvvvvvvvBBBBBBBBBBB00000]LSB || [.............vvvvvvvvBBBBBBBBBBB] |
| 51 // || | 50 // || |
| 52 // Just before we need to call VP8LoadNewBytes(), the situation is: | 51 // Just before we need to call VP8LoadNewBytes(), the situation is: |
| 53 // [........vvvvvv000000000000000000]LSB || [..........................vvvvvv] | 52 // [........vvvvvv000000000000000000]LSB || [..........................vvvvvv] |
| 54 // || | 53 // || |
| 55 // And just after calling VP8LoadNewBytes(): | 54 // And just after calling VP8LoadNewBytes(): |
| 56 // [........vvvvvvvvBBBBBBBBBBBBBBBB]LSB || [........vvvvvvvvBBBBBBBBBBBBBBBB] | 55 // [........vvvvvvvvBBBBBBBBBBBBBBBB]LSB || [........vvvvvvvvBBBBBBBBBBBBBBBB] |
| 57 // | 56 // |
| 58 // -> we're back to height active 'value_' bits (marked 'v') and BITS cached | 57 // -> we're back to eight active 'value_' bits (marked 'v') and BITS cached |
| 59 // bits (marked 'B') | 58 // bits (marked 'B') |
| 60 // | 59 // |
| 61 // The right-justify strategy tends to use less shifts and is often faster. | 60 // The right-justify strategy tends to use less shifts and is often faster. |
| 62 | 61 |
| 63 //------------------------------------------------------------------------------ | 62 //------------------------------------------------------------------------------ |
| 64 // BITS can be any multiple of 8 from 8 to 56 (inclusive). | 63 // BITS can be any multiple of 8 from 8 to 56 (inclusive). |
| 65 // Pick values that fit natural register size. | 64 // Pick values that fit natural register size. |
| 66 | 65 |
| 67 #if !defined(WEBP_REFERENCE_IMPLEMENTATION) | 66 #if !defined(WEBP_REFERENCE_IMPLEMENTATION) |
| 68 | 67 |
| (...skipping 79 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 148 extern const range_t kVP8NewRange[128]; | 147 extern const range_t kVP8NewRange[128]; |
| 149 | 148 |
| 150 void VP8LoadFinalBytes(VP8BitReader* const br); // special case for the tail | 149 void VP8LoadFinalBytes(VP8BitReader* const br); // special case for the tail |
| 151 | 150 |
| 152 static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) { | 151 static WEBP_INLINE void VP8LoadNewBytes(VP8BitReader* const br) { |
| 153 assert(br != NULL && br->buf_ != NULL); | 152 assert(br != NULL && br->buf_ != NULL); |
| 154 // Read 'BITS' bits at a time if possible. | 153 // Read 'BITS' bits at a time if possible. |
| 155 if (br->buf_ + sizeof(lbit_t) <= br->buf_end_) { | 154 if (br->buf_ + sizeof(lbit_t) <= br->buf_end_) { |
| 156 // convert memory type to register type (with some zero'ing!) | 155 // convert memory type to register type (with some zero'ing!) |
| 157 bit_t bits; | 156 bit_t bits; |
| 158 lbit_t in_bits = *(lbit_t*)br->buf_; | 157 const lbit_t in_bits = *(const lbit_t*)br->buf_; |
| 159 br->buf_ += (BITS) >> 3; | 158 br->buf_ += (BITS) >> 3; |
| 160 #if !defined(__BIG_ENDIAN__) | 159 #if !defined(__BIG_ENDIAN__) |
| 161 #if (BITS > 32) | 160 #if (BITS > 32) |
| 162 // gcc 4.3 has builtin functions for swap32/swap64 | 161 // gcc 4.3 has builtin functions for swap32/swap64 |
| 163 #if defined(__GNUC__) && \ | 162 #if defined(__GNUC__) && \ |
| 164 (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) | 163 (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) |
| 165 bits = (bit_t)__builtin_bswap64(in_bits); | 164 bits = (bit_t)__builtin_bswap64(in_bits); |
| 166 #elif defined(_MSC_VER) | 165 #elif defined(_MSC_VER) |
| 167 bits = (bit_t)_byteswap_uint64(in_bits); | 166 bits = (bit_t)_byteswap_uint64(in_bits); |
| 168 #elif defined(__x86_64__) | 167 #elif defined(__x86_64__) |
| 169 __asm__ volatile("bswapq %0" : "=r"(bits) : "0"(in_bits)); | 168 __asm__ volatile("bswapq %0" : "=r"(bits) : "0"(in_bits)); |
| 170 #else // generic code for swapping 64-bit values (suggested by bdb@) | 169 #else // generic code for swapping 64-bit values (suggested by bdb@) |
| 171 bits = (bit_t)in_bits; | 170 bits = (bit_t)in_bits; |
| 172 bits = ((bits & 0xffffffff00000000ull) >> 32) | | 171 bits = ((bits & 0xffffffff00000000ull) >> 32) | |
| 173 ((bits & 0x00000000ffffffffull) << 32); | 172 ((bits & 0x00000000ffffffffull) << 32); |
| 174 bits = ((bits & 0xffff0000ffff0000ull) >> 16) | | 173 bits = ((bits & 0xffff0000ffff0000ull) >> 16) | |
| 175 ((bits & 0x0000ffff0000ffffull) << 16); | 174 ((bits & 0x0000ffff0000ffffull) << 16); |
| 176 bits = ((bits & 0xff00ff00ff00ff00ull) >> 8) | | 175 bits = ((bits & 0xff00ff00ff00ff00ull) >> 8) | |
| 177 ((bits & 0x00ff00ff00ff00ffull) << 8); | 176 ((bits & 0x00ff00ff00ff00ffull) << 8); |
| 178 #endif | 177 #endif |
| 179 bits >>= 64 - BITS; | 178 bits >>= 64 - BITS; |
| 180 #elif (BITS >= 24) | 179 #elif (BITS >= 24) |
| 181 #if defined(__i386__) || defined(__x86_64__) | 180 #if defined(__i386__) || defined(__x86_64__) |
| 182 __asm__ volatile("bswap %k0" : "=r"(in_bits) : "0"(in_bits)); | 181 { |
| 183 bits = (bit_t)in_bits; // 24b/32b -> 32b/64b zero-extension | 182 lbit_t swapped_in_bits; |
| 183 __asm__ volatile("bswap %k0" : "=r"(swapped_in_bits) : "0"(in_bits)); |
| 184 bits = (bit_t)swapped_in_bits; // 24b/32b -> 32b/64b zero-extension |
| 185 } |
| 184 #elif defined(_MSC_VER) | 186 #elif defined(_MSC_VER) |
| 185 bits = (bit_t)_byteswap_ulong(in_bits); | 187 bits = (bit_t)_byteswap_ulong(in_bits); |
| 186 #else | 188 #else |
| 187 bits = (bit_t)(in_bits >> 24) | ((in_bits >> 8) & 0xff00) | 189 bits = (bit_t)(in_bits >> 24) | ((in_bits >> 8) & 0xff00) |
| 188 | ((in_bits << 8) & 0xff0000) | (in_bits << 24); | 190 | ((in_bits << 8) & 0xff0000) | (in_bits << 24); |
| 189 #endif // x86 | 191 #endif // x86 |
| 190 bits >>= (32 - BITS); | 192 bits >>= (32 - BITS); |
| 191 #elif (BITS == 16) | 193 #elif (BITS == 16) |
| 192 // gcc will recognize a 'rorw $8, ...' here: | 194 // gcc will recognize a 'rorw $8, ...' here: |
| 193 bits = (bit_t)(in_bits >> 8) | ((in_bits & 0xff) << 8); | 195 bits = (bit_t)(in_bits >> 8) | ((in_bits & 0xff) << 8); |
| (...skipping 53 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 247 br->range_ = kVP8NewRange[idx]; | 249 br->range_ = kVP8NewRange[idx]; |
| 248 br->value_ <<= shift; | 250 br->value_ <<= shift; |
| 249 br->bits_ -= shift; | 251 br->bits_ -= shift; |
| 250 #else | 252 #else |
| 251 const int shift = kVP8Log2Range[br->range_]; | 253 const int shift = kVP8Log2Range[br->range_]; |
| 252 assert(br->range_ < (range_t)128); | 254 assert(br->range_ < (range_t)128); |
| 253 br->range_ = kVP8NewRange[br->range_]; | 255 br->range_ = kVP8NewRange[br->range_]; |
| 254 br->bits_ -= shift; | 256 br->bits_ -= shift; |
| 255 #endif | 257 #endif |
| 256 } | 258 } |
| 259 |
| 257 static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) { | 260 static WEBP_INLINE int VP8GetBit(VP8BitReader* const br, int prob) { |
| 258 #ifndef USE_RIGHT_JUSTIFY | 261 #ifndef USE_RIGHT_JUSTIFY |
| 259 // It's important to avoid generating a 64bit x 64bit multiply here. | 262 // It's important to avoid generating a 64bit x 64bit multiply here. |
| 260 // We just need an 8b x 8b after all. | 263 // We just need an 8b x 8b after all. |
| 261 const range_t split = | 264 const range_t split = |
| 262 (range_t)((uint32_t)(br->range_ >> (BITS)) * prob) << ((BITS) - 8); | 265 (range_t)((uint32_t)(br->range_ >> (BITS)) * prob) << ((BITS) - 8); |
| 263 const int bit = VP8BitUpdate(br, split); | 266 const int bit = VP8BitUpdate(br, split); |
| 264 if (br->range_ <= (((range_t)0x7e << (BITS)) | (MASK))) { | 267 if (br->range_ <= (((range_t)0x7e << (BITS)) | (MASK))) { |
| 265 VP8Shift(br); | 268 VP8Shift(br); |
| 266 } | 269 } |
| 267 return bit; | 270 return bit; |
| 268 #else | 271 #else |
| 269 const range_t split = (br->range_ * prob) >> 8; | 272 const range_t split = (br->range_ * prob) >> 8; |
| 270 const int bit = VP8BitUpdate(br, split); | 273 const int bit = VP8BitUpdate(br, split); |
| 271 if (br->range_ <= (range_t)0x7e) { | 274 if (br->range_ <= (range_t)0x7e) { |
| 272 VP8Shift(br); | 275 VP8Shift(br); |
| 273 } | 276 } |
| 274 return bit; | 277 return bit; |
| 275 #endif | 278 #endif |
| 276 } | 279 } |
| 277 | 280 |
| 278 static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v) { | 281 static WEBP_INLINE int VP8GetSigned(VP8BitReader* const br, int v) { |
| 279 const range_t split = (br->range_ >> 1); | 282 const range_t split = (br->range_ >> 1); |
| 280 const int bit = VP8BitUpdate(br, split); | 283 const int bit = VP8BitUpdate(br, split); |
| 281 VP8Shift(br); | 284 VP8Shift(br); |
| 282 return bit ? -v : v; | 285 return bit ? -v : v; |
| 283 } | 286 } |
| 284 | 287 |
| 285 | |
| 286 // ----------------------------------------------------------------------------- | 288 // ----------------------------------------------------------------------------- |
| 287 // Bitreader for lossless format | 289 // Bitreader for lossless format |
| 288 | 290 |
| 289 typedef uint64_t vp8l_val_t; // right now, this bit-reader can only use 64bit. | 291 typedef uint64_t vp8l_val_t; // right now, this bit-reader can only use 64bit. |
| 290 | 292 |
| 291 typedef struct { | 293 typedef struct { |
| 292 vp8l_val_t val_; // pre-fetched bits | 294 vp8l_val_t val_; // pre-fetched bits |
| 293 const uint8_t* buf_; // input byte buffer | 295 const uint8_t* buf_; // input byte buffer |
| 294 size_t len_; // buffer length | 296 size_t len_; // buffer length |
| 295 size_t pos_; // byte position in buf_ | 297 size_t pos_; // byte position in buf_ |
| (...skipping 13 matching lines...) Expand all Loading... |
| 309 // Reads the specified number of bits from Read Buffer. | 311 // Reads the specified number of bits from Read Buffer. |
| 310 // Flags an error in case end_of_stream or n_bits is more than allowed limit. | 312 // Flags an error in case end_of_stream or n_bits is more than allowed limit. |
| 311 // Flags eos if this read attempt is going to cross the read buffer. | 313 // Flags eos if this read attempt is going to cross the read buffer. |
| 312 uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits); | 314 uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits); |
| 313 | 315 |
| 314 // Return the prefetched bits, so they can be looked up. | 316 // Return the prefetched bits, so they can be looked up. |
| 315 static WEBP_INLINE uint32_t VP8LPrefetchBits(VP8LBitReader* const br) { | 317 static WEBP_INLINE uint32_t VP8LPrefetchBits(VP8LBitReader* const br) { |
| 316 return (uint32_t)(br->val_ >> br->bit_pos_); | 318 return (uint32_t)(br->val_ >> br->bit_pos_); |
| 317 } | 319 } |
| 318 | 320 |
| 319 // Discard 'num_bits' bits from the cache. | 321 // For jumping over a number of bits in the bit stream when accessed with |
| 320 static WEBP_INLINE void VP8LDiscardBits(VP8LBitReader* const br, int num_bits) { | 322 // VP8LPrefetchBits and VP8LFillBitWindow. |
| 321 br->bit_pos_ += num_bits; | 323 static WEBP_INLINE void VP8LSetBitPos(VP8LBitReader* const br, int val) { |
| 324 br->bit_pos_ = val; |
| 322 } | 325 } |
| 323 | 326 |
| 324 // Advances the Read buffer by 4 bytes to make room for reading next 32 bits. | 327 // Advances the read buffer by 4 bytes to make room for reading next 32 bits. |
| 325 void VP8LFillBitWindow(VP8LBitReader* const br); | 328 void VP8LFillBitWindow(VP8LBitReader* const br); |
| 326 | 329 |
| 327 #if defined(__cplusplus) || defined(c_plusplus) | 330 #ifdef __cplusplus |
| 328 } // extern "C" | 331 } // extern "C" |
| 329 #endif | 332 #endif |
| 330 | 333 |
| 331 #endif /* WEBP_UTILS_BIT_READER_H_ */ | 334 #endif /* WEBP_UTILS_BIT_READER_H_ */ |
| OLD | NEW |