Tue, 06 Jan 2015 21:39:09 +0100
Conditionally force memory storage according to privacy.thirdparty.isolate;
This solves Tor bug #9701, complying with disk avoidance documented in
https://www.torproject.org/projects/torbrowser/design/#disk-avoidance.
michael@0 | 1 | // Copyright 2011 Google Inc. All Rights Reserved. |
michael@0 | 2 | // |
michael@0 | 3 | // Redistribution and use in source and binary forms, with or without |
michael@0 | 4 | // modification, are permitted provided that the following conditions are |
michael@0 | 5 | // met: |
michael@0 | 6 | // |
michael@0 | 7 | // * Redistributions of source code must retain the above copyright |
michael@0 | 8 | // notice, this list of conditions and the following disclaimer. |
michael@0 | 9 | // * Redistributions in binary form must reproduce the above |
michael@0 | 10 | // copyright notice, this list of conditions and the following disclaimer |
michael@0 | 11 | // in the documentation and/or other materials provided with the |
michael@0 | 12 | // distribution. |
michael@0 | 13 | // * Neither the name of Google Inc. nor the names of its |
michael@0 | 14 | // contributors may be used to endorse or promote products derived from |
michael@0 | 15 | // this software without specific prior written permission. |
michael@0 | 16 | // |
michael@0 | 17 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
michael@0 | 18 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
michael@0 | 19 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
michael@0 | 20 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
michael@0 | 21 | // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
michael@0 | 22 | // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
michael@0 | 23 | // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
michael@0 | 24 | // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
michael@0 | 25 | // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
michael@0 | 26 | // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
michael@0 | 27 | // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
michael@0 | 28 | // |
michael@0 | 29 | // Various stubs for the open-source version of Snappy. |
michael@0 | 30 | |
michael@0 | 31 | #ifndef UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ |
michael@0 | 32 | #define UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ |
michael@0 | 33 | |
michael@0 | 34 | #ifdef HAVE_CONFIG_H |
michael@0 | 35 | #include "config.h" |
michael@0 | 36 | #endif |
michael@0 | 37 | |
michael@0 | 38 | #include <iostream> |
michael@0 | 39 | #include <string> |
michael@0 | 40 | |
michael@0 | 41 | #include <assert.h> |
michael@0 | 42 | #include <stdlib.h> |
michael@0 | 43 | #include <string.h> |
michael@0 | 44 | |
michael@0 | 45 | #ifdef HAVE_SYS_MMAN_H |
michael@0 | 46 | #include <sys/mman.h> |
michael@0 | 47 | #endif |
michael@0 | 48 | |
michael@0 | 49 | #include "snappy-stubs-public.h" |
michael@0 | 50 | |
michael@0 | 51 | #if defined(__x86_64__) |
michael@0 | 52 | |
michael@0 | 53 | // Enable 64-bit optimized versions of some routines. |
michael@0 | 54 | #define ARCH_K8 1 |
michael@0 | 55 | |
michael@0 | 56 | #endif |
michael@0 | 57 | |
michael@0 | 58 | // Needed by OS X, among others. |
michael@0 | 59 | #ifndef MAP_ANONYMOUS |
michael@0 | 60 | #define MAP_ANONYMOUS MAP_ANON |
michael@0 | 61 | #endif |
michael@0 | 62 | |
michael@0 | 63 | // Pull in std::min, std::ostream, and the likes. This is safe because this |
michael@0 | 64 | // header file is never used from any public header files. |
michael@0 | 65 | using namespace std; |
michael@0 | 66 | |
michael@0 | 67 | // The size of an array, if known at compile-time. |
michael@0 | 68 | // Will give unexpected results if used on a pointer. |
michael@0 | 69 | // We undefine it first, since some compilers already have a definition. |
michael@0 | 70 | #ifdef ARRAYSIZE |
michael@0 | 71 | #undef ARRAYSIZE |
michael@0 | 72 | #endif |
michael@0 | 73 | #define ARRAYSIZE(a) (sizeof(a) / sizeof(*(a))) |
michael@0 | 74 | |
michael@0 | 75 | // Static prediction hints. |
michael@0 | 76 | #ifdef HAVE_BUILTIN_EXPECT |
michael@0 | 77 | #define PREDICT_FALSE(x) (__builtin_expect(x, 0)) |
michael@0 | 78 | #define PREDICT_TRUE(x) (__builtin_expect(!!(x), 1)) |
michael@0 | 79 | #else |
michael@0 | 80 | #define PREDICT_FALSE(x) x |
michael@0 | 81 | #define PREDICT_TRUE(x) x |
michael@0 | 82 | #endif |
michael@0 | 83 | |
michael@0 | 84 | // This is only used for recomputing the tag byte table used during |
michael@0 | 85 | // decompression; for simplicity we just remove it from the open-source |
michael@0 | 86 | // version (anyone who wants to regenerate it can just do the call |
michael@0 | 87 | // themselves within main()). |
michael@0 | 88 | #define DEFINE_bool(flag_name, default_value, description) \ |
michael@0 | 89 | bool FLAGS_ ## flag_name = default_value |
michael@0 | 90 | #define DECLARE_bool(flag_name) \ |
michael@0 | 91 | extern bool FLAGS_ ## flag_name |
michael@0 | 92 | |
michael@0 | 93 | namespace snappy { |
michael@0 | 94 | |
michael@0 | 95 | static const uint32 kuint32max = static_cast<uint32>(0xFFFFFFFF); |
michael@0 | 96 | static const int64 kint64max = static_cast<int64>(0x7FFFFFFFFFFFFFFFLL); |
michael@0 | 97 | |
michael@0 | 98 | // Logging. |
michael@0 | 99 | |
michael@0 | 100 | #define LOG(level) LogMessage() |
michael@0 | 101 | #define VLOG(level) true ? (void)0 : \ |
michael@0 | 102 | snappy::LogMessageVoidify() & snappy::LogMessage() |
michael@0 | 103 | |
michael@0 | 104 | class LogMessage { |
michael@0 | 105 | public: |
michael@0 | 106 | LogMessage() { } |
michael@0 | 107 | ~LogMessage() { |
michael@0 | 108 | cerr << endl; |
michael@0 | 109 | } |
michael@0 | 110 | |
michael@0 | 111 | LogMessage& operator<<(const std::string& msg) { |
michael@0 | 112 | cerr << msg; |
michael@0 | 113 | return *this; |
michael@0 | 114 | } |
michael@0 | 115 | LogMessage& operator<<(int x) { |
michael@0 | 116 | cerr << x; |
michael@0 | 117 | return *this; |
michael@0 | 118 | } |
michael@0 | 119 | }; |
michael@0 | 120 | |
michael@0 | 121 | // Asserts, both versions activated in debug mode only, |
michael@0 | 122 | // and ones that are always active. |
michael@0 | 123 | |
michael@0 | 124 | #define CRASH_UNLESS(condition) \ |
michael@0 | 125 | PREDICT_TRUE(condition) ? (void)0 : \ |
michael@0 | 126 | snappy::LogMessageVoidify() & snappy::LogMessageCrash() |
michael@0 | 127 | |
michael@0 | 128 | class LogMessageCrash : public LogMessage { |
michael@0 | 129 | public: |
michael@0 | 130 | LogMessageCrash() { } |
michael@0 | 131 | ~LogMessageCrash() { |
michael@0 | 132 | cerr << endl; |
michael@0 | 133 | abort(); |
michael@0 | 134 | } |
michael@0 | 135 | }; |
michael@0 | 136 | |
michael@0 | 137 | // This class is used to explicitly ignore values in the conditional |
michael@0 | 138 | // logging macros. This avoids compiler warnings like "value computed |
michael@0 | 139 | // is not used" and "statement has no effect". |
michael@0 | 140 | |
michael@0 | 141 | class LogMessageVoidify { |
michael@0 | 142 | public: |
michael@0 | 143 | LogMessageVoidify() { } |
michael@0 | 144 | // This has to be an operator with a precedence lower than << but |
michael@0 | 145 | // higher than ?: |
michael@0 | 146 | void operator&(const LogMessage&) { } |
michael@0 | 147 | }; |
michael@0 | 148 | |
michael@0 | 149 | #define CHECK(cond) CRASH_UNLESS(cond) |
michael@0 | 150 | #define CHECK_LE(a, b) CRASH_UNLESS((a) <= (b)) |
michael@0 | 151 | #define CHECK_GE(a, b) CRASH_UNLESS((a) >= (b)) |
michael@0 | 152 | #define CHECK_EQ(a, b) CRASH_UNLESS((a) == (b)) |
michael@0 | 153 | #define CHECK_NE(a, b) CRASH_UNLESS((a) != (b)) |
michael@0 | 154 | #define CHECK_LT(a, b) CRASH_UNLESS((a) < (b)) |
michael@0 | 155 | #define CHECK_GT(a, b) CRASH_UNLESS((a) > (b)) |
michael@0 | 156 | |
michael@0 | 157 | #ifdef NDEBUG |
michael@0 | 158 | |
michael@0 | 159 | #define DCHECK(cond) CRASH_UNLESS(true) |
michael@0 | 160 | #define DCHECK_LE(a, b) CRASH_UNLESS(true) |
michael@0 | 161 | #define DCHECK_GE(a, b) CRASH_UNLESS(true) |
michael@0 | 162 | #define DCHECK_EQ(a, b) CRASH_UNLESS(true) |
michael@0 | 163 | #define DCHECK_NE(a, b) CRASH_UNLESS(true) |
michael@0 | 164 | #define DCHECK_LT(a, b) CRASH_UNLESS(true) |
michael@0 | 165 | #define DCHECK_GT(a, b) CRASH_UNLESS(true) |
michael@0 | 166 | |
michael@0 | 167 | #else |
michael@0 | 168 | |
michael@0 | 169 | #define DCHECK(cond) CHECK(cond) |
michael@0 | 170 | #define DCHECK_LE(a, b) CHECK_LE(a, b) |
michael@0 | 171 | #define DCHECK_GE(a, b) CHECK_GE(a, b) |
michael@0 | 172 | #define DCHECK_EQ(a, b) CHECK_EQ(a, b) |
michael@0 | 173 | #define DCHECK_NE(a, b) CHECK_NE(a, b) |
michael@0 | 174 | #define DCHECK_LT(a, b) CHECK_LT(a, b) |
michael@0 | 175 | #define DCHECK_GT(a, b) CHECK_GT(a, b) |
michael@0 | 176 | |
michael@0 | 177 | #endif |
michael@0 | 178 | |
michael@0 | 179 | // Potentially unaligned loads and stores. |
michael@0 | 180 | |
michael@0 | 181 | #if defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) |
michael@0 | 182 | |
michael@0 | 183 | #define UNALIGNED_LOAD16(_p) (*reinterpret_cast<const uint16 *>(_p)) |
michael@0 | 184 | #define UNALIGNED_LOAD32(_p) (*reinterpret_cast<const uint32 *>(_p)) |
michael@0 | 185 | #define UNALIGNED_LOAD64(_p) (*reinterpret_cast<const uint64 *>(_p)) |
michael@0 | 186 | |
michael@0 | 187 | #define UNALIGNED_STORE16(_p, _val) (*reinterpret_cast<uint16 *>(_p) = (_val)) |
michael@0 | 188 | #define UNALIGNED_STORE32(_p, _val) (*reinterpret_cast<uint32 *>(_p) = (_val)) |
michael@0 | 189 | #define UNALIGNED_STORE64(_p, _val) (*reinterpret_cast<uint64 *>(_p) = (_val)) |
michael@0 | 190 | |
michael@0 | 191 | #else |
michael@0 | 192 | |
michael@0 | 193 | // These functions are provided for architectures that don't support |
michael@0 | 194 | // unaligned loads and stores. |
michael@0 | 195 | |
michael@0 | 196 | inline uint16 UNALIGNED_LOAD16(const void *p) { |
michael@0 | 197 | uint16 t; |
michael@0 | 198 | memcpy(&t, p, sizeof t); |
michael@0 | 199 | return t; |
michael@0 | 200 | } |
michael@0 | 201 | |
michael@0 | 202 | inline uint32 UNALIGNED_LOAD32(const void *p) { |
michael@0 | 203 | uint32 t; |
michael@0 | 204 | memcpy(&t, p, sizeof t); |
michael@0 | 205 | return t; |
michael@0 | 206 | } |
michael@0 | 207 | |
michael@0 | 208 | inline uint64 UNALIGNED_LOAD64(const void *p) { |
michael@0 | 209 | uint64 t; |
michael@0 | 210 | memcpy(&t, p, sizeof t); |
michael@0 | 211 | return t; |
michael@0 | 212 | } |
michael@0 | 213 | |
michael@0 | 214 | inline void UNALIGNED_STORE16(void *p, uint16 v) { |
michael@0 | 215 | memcpy(p, &v, sizeof v); |
michael@0 | 216 | } |
michael@0 | 217 | |
michael@0 | 218 | inline void UNALIGNED_STORE32(void *p, uint32 v) { |
michael@0 | 219 | memcpy(p, &v, sizeof v); |
michael@0 | 220 | } |
michael@0 | 221 | |
michael@0 | 222 | inline void UNALIGNED_STORE64(void *p, uint64 v) { |
michael@0 | 223 | memcpy(p, &v, sizeof v); |
michael@0 | 224 | } |
michael@0 | 225 | |
michael@0 | 226 | #endif |
michael@0 | 227 | |
michael@0 | 228 | // The following guarantees declaration of the byte swap functions. |
michael@0 | 229 | #ifdef WORDS_BIGENDIAN |
michael@0 | 230 | |
michael@0 | 231 | #ifdef HAVE_SYS_BYTEORDER_H |
michael@0 | 232 | #include <sys/byteorder.h> |
michael@0 | 233 | #endif |
michael@0 | 234 | |
michael@0 | 235 | #ifdef HAVE_SYS_ENDIAN_H |
michael@0 | 236 | #include <sys/endian.h> |
michael@0 | 237 | #endif |
michael@0 | 238 | |
michael@0 | 239 | #ifdef _MSC_VER |
michael@0 | 240 | #include <stdlib.h> |
michael@0 | 241 | #define bswap_16(x) _byteswap_ushort(x) |
michael@0 | 242 | #define bswap_32(x) _byteswap_ulong(x) |
michael@0 | 243 | #define bswap_64(x) _byteswap_uint64(x) |
michael@0 | 244 | |
michael@0 | 245 | #elif defined(__APPLE__) |
michael@0 | 246 | // Mac OS X / Darwin features |
michael@0 | 247 | #include <libkern/OSByteOrder.h> |
michael@0 | 248 | #define bswap_16(x) OSSwapInt16(x) |
michael@0 | 249 | #define bswap_32(x) OSSwapInt32(x) |
michael@0 | 250 | #define bswap_64(x) OSSwapInt64(x) |
michael@0 | 251 | |
michael@0 | 252 | #elif defined(HAVE_BYTESWAP_H) |
michael@0 | 253 | #include <byteswap.h> |
michael@0 | 254 | |
michael@0 | 255 | #elif defined(bswap32) |
michael@0 | 256 | // FreeBSD defines bswap{16,32,64} in <sys/endian.h> (already #included). |
michael@0 | 257 | #define bswap_16(x) bswap16(x) |
michael@0 | 258 | #define bswap_32(x) bswap32(x) |
michael@0 | 259 | #define bswap_64(x) bswap64(x) |
michael@0 | 260 | |
michael@0 | 261 | #elif defined(BSWAP_64) |
michael@0 | 262 | // Solaris 10 defines BSWAP_{16,32,64} in <sys/byteorder.h> (already #included). |
michael@0 | 263 | #define bswap_16(x) BSWAP_16(x) |
michael@0 | 264 | #define bswap_32(x) BSWAP_32(x) |
michael@0 | 265 | #define bswap_64(x) BSWAP_64(x) |
michael@0 | 266 | |
michael@0 | 267 | #else |
michael@0 | 268 | |
michael@0 | 269 | inline uint16 bswap_16(uint16 x) { |
michael@0 | 270 | return (x << 8) | (x >> 8); |
michael@0 | 271 | } |
michael@0 | 272 | |
michael@0 | 273 | inline uint32 bswap_32(uint32 x) { |
michael@0 | 274 | x = ((x & 0xff00ff00UL) >> 8) | ((x & 0x00ff00ffUL) << 8); |
michael@0 | 275 | return (x >> 16) | (x << 16); |
michael@0 | 276 | } |
michael@0 | 277 | |
michael@0 | 278 | inline uint64 bswap_64(uint64 x) { |
michael@0 | 279 | x = ((x & 0xff00ff00ff00ff00ULL) >> 8) | ((x & 0x00ff00ff00ff00ffULL) << 8); |
michael@0 | 280 | x = ((x & 0xffff0000ffff0000ULL) >> 16) | ((x & 0x0000ffff0000ffffULL) << 16); |
michael@0 | 281 | return (x >> 32) | (x << 32); |
michael@0 | 282 | } |
michael@0 | 283 | |
michael@0 | 284 | #endif |
michael@0 | 285 | |
michael@0 | 286 | #endif // WORDS_BIGENDIAN |
michael@0 | 287 | |
michael@0 | 288 | // Convert to little-endian storage, opposite of network format. |
michael@0 | 289 | // Convert x from host to little endian: x = LittleEndian.FromHost(x); |
michael@0 | 290 | // convert x from little endian to host: x = LittleEndian.ToHost(x); |
michael@0 | 291 | // |
michael@0 | 292 | // Store values into unaligned memory converting to little endian order: |
michael@0 | 293 | // LittleEndian.Store16(p, x); |
michael@0 | 294 | // |
michael@0 | 295 | // Load unaligned values stored in little endian converting to host order: |
michael@0 | 296 | // x = LittleEndian.Load16(p); |
michael@0 | 297 | class LittleEndian { |
michael@0 | 298 | public: |
michael@0 | 299 | // Conversion functions. |
michael@0 | 300 | #ifdef WORDS_BIGENDIAN |
michael@0 | 301 | |
michael@0 | 302 | static uint16 FromHost16(uint16 x) { return bswap_16(x); } |
michael@0 | 303 | static uint16 ToHost16(uint16 x) { return bswap_16(x); } |
michael@0 | 304 | |
michael@0 | 305 | static uint32 FromHost32(uint32 x) { return bswap_32(x); } |
michael@0 | 306 | static uint32 ToHost32(uint32 x) { return bswap_32(x); } |
michael@0 | 307 | |
michael@0 | 308 | static bool IsLittleEndian() { return false; } |
michael@0 | 309 | |
michael@0 | 310 | #else // !defined(WORDS_BIGENDIAN) |
michael@0 | 311 | |
michael@0 | 312 | static uint16 FromHost16(uint16 x) { return x; } |
michael@0 | 313 | static uint16 ToHost16(uint16 x) { return x; } |
michael@0 | 314 | |
michael@0 | 315 | static uint32 FromHost32(uint32 x) { return x; } |
michael@0 | 316 | static uint32 ToHost32(uint32 x) { return x; } |
michael@0 | 317 | |
michael@0 | 318 | static bool IsLittleEndian() { return true; } |
michael@0 | 319 | |
michael@0 | 320 | #endif // !defined(WORDS_BIGENDIAN) |
michael@0 | 321 | |
michael@0 | 322 | // Functions to do unaligned loads and stores in little-endian order. |
michael@0 | 323 | static uint16 Load16(const void *p) { |
michael@0 | 324 | return ToHost16(UNALIGNED_LOAD16(p)); |
michael@0 | 325 | } |
michael@0 | 326 | |
michael@0 | 327 | static void Store16(void *p, uint16 v) { |
michael@0 | 328 | UNALIGNED_STORE16(p, FromHost16(v)); |
michael@0 | 329 | } |
michael@0 | 330 | |
michael@0 | 331 | static uint32 Load32(const void *p) { |
michael@0 | 332 | return ToHost32(UNALIGNED_LOAD32(p)); |
michael@0 | 333 | } |
michael@0 | 334 | |
michael@0 | 335 | static void Store32(void *p, uint32 v) { |
michael@0 | 336 | UNALIGNED_STORE32(p, FromHost32(v)); |
michael@0 | 337 | } |
michael@0 | 338 | }; |
michael@0 | 339 | |
michael@0 | 340 | // Some bit-manipulation functions. |
michael@0 | 341 | class Bits { |
michael@0 | 342 | public: |
michael@0 | 343 | // Return floor(log2(n)) for positive integer n. Returns -1 iff n == 0. |
michael@0 | 344 | static int Log2Floor(uint32 n); |
michael@0 | 345 | |
michael@0 | 346 | // Return the first set least / most significant bit, 0-indexed. Returns an |
michael@0 | 347 | // undefined value if n == 0. FindLSBSetNonZero() is similar to ffs() except |
michael@0 | 348 | // that it's 0-indexed. |
michael@0 | 349 | static int FindLSBSetNonZero(uint32 n); |
michael@0 | 350 | static int FindLSBSetNonZero64(uint64 n); |
michael@0 | 351 | |
michael@0 | 352 | private: |
michael@0 | 353 | DISALLOW_COPY_AND_ASSIGN(Bits); |
michael@0 | 354 | }; |
michael@0 | 355 | |
michael@0 | 356 | #ifdef HAVE_BUILTIN_CTZ |
michael@0 | 357 | |
michael@0 | 358 | inline int Bits::Log2Floor(uint32 n) { |
michael@0 | 359 | return n == 0 ? -1 : 31 ^ __builtin_clz(n); |
michael@0 | 360 | } |
michael@0 | 361 | |
michael@0 | 362 | inline int Bits::FindLSBSetNonZero(uint32 n) { |
michael@0 | 363 | return __builtin_ctz(n); |
michael@0 | 364 | } |
michael@0 | 365 | |
michael@0 | 366 | inline int Bits::FindLSBSetNonZero64(uint64 n) { |
michael@0 | 367 | return __builtin_ctzll(n); |
michael@0 | 368 | } |
michael@0 | 369 | |
michael@0 | 370 | #else // Portable versions. |
michael@0 | 371 | |
michael@0 | 372 | inline int Bits::Log2Floor(uint32 n) { |
michael@0 | 373 | if (n == 0) |
michael@0 | 374 | return -1; |
michael@0 | 375 | int log = 0; |
michael@0 | 376 | uint32 value = n; |
michael@0 | 377 | for (int i = 4; i >= 0; --i) { |
michael@0 | 378 | int shift = (1 << i); |
michael@0 | 379 | uint32 x = value >> shift; |
michael@0 | 380 | if (x != 0) { |
michael@0 | 381 | value = x; |
michael@0 | 382 | log += shift; |
michael@0 | 383 | } |
michael@0 | 384 | } |
michael@0 | 385 | assert(value == 1); |
michael@0 | 386 | return log; |
michael@0 | 387 | } |
michael@0 | 388 | |
michael@0 | 389 | inline int Bits::FindLSBSetNonZero(uint32 n) { |
michael@0 | 390 | int rc = 31; |
michael@0 | 391 | for (int i = 4, shift = 1 << 4; i >= 0; --i) { |
michael@0 | 392 | const uint32 x = n << shift; |
michael@0 | 393 | if (x != 0) { |
michael@0 | 394 | n = x; |
michael@0 | 395 | rc -= shift; |
michael@0 | 396 | } |
michael@0 | 397 | shift >>= 1; |
michael@0 | 398 | } |
michael@0 | 399 | return rc; |
michael@0 | 400 | } |
michael@0 | 401 | |
michael@0 | 402 | // FindLSBSetNonZero64() is defined in terms of FindLSBSetNonZero(). |
michael@0 | 403 | inline int Bits::FindLSBSetNonZero64(uint64 n) { |
michael@0 | 404 | const uint32 bottombits = static_cast<uint32>(n); |
michael@0 | 405 | if (bottombits == 0) { |
michael@0 | 406 | // Bottom bits are zero, so scan in top bits |
michael@0 | 407 | return 32 + FindLSBSetNonZero(static_cast<uint32>(n >> 32)); |
michael@0 | 408 | } else { |
michael@0 | 409 | return FindLSBSetNonZero(bottombits); |
michael@0 | 410 | } |
michael@0 | 411 | } |
michael@0 | 412 | |
michael@0 | 413 | #endif // End portable versions. |
michael@0 | 414 | |
michael@0 | 415 | // Variable-length integer encoding. |
michael@0 | 416 | class Varint { |
michael@0 | 417 | public: |
michael@0 | 418 | // Maximum lengths of varint encoding of uint32. |
michael@0 | 419 | static const int kMax32 = 5; |
michael@0 | 420 | |
michael@0 | 421 | // Attempts to parse a varint32 from a prefix of the bytes in [ptr,limit-1]. |
michael@0 | 422 | // Never reads a character at or beyond limit. If a valid/terminated varint32 |
michael@0 | 423 | // was found in the range, stores it in *OUTPUT and returns a pointer just |
michael@0 | 424 | // past the last byte of the varint32. Else returns NULL. On success, |
michael@0 | 425 | // "result <= limit". |
michael@0 | 426 | static const char* Parse32WithLimit(const char* ptr, const char* limit, |
michael@0 | 427 | uint32* OUTPUT); |
michael@0 | 428 | |
michael@0 | 429 | // REQUIRES "ptr" points to a buffer of length sufficient to hold "v". |
michael@0 | 430 | // EFFECTS Encodes "v" into "ptr" and returns a pointer to the |
michael@0 | 431 | // byte just past the last encoded byte. |
michael@0 | 432 | static char* Encode32(char* ptr, uint32 v); |
michael@0 | 433 | |
michael@0 | 434 | // EFFECTS Appends the varint representation of "value" to "*s". |
michael@0 | 435 | static void Append32(string* s, uint32 value); |
michael@0 | 436 | }; |
michael@0 | 437 | |
michael@0 | 438 | inline const char* Varint::Parse32WithLimit(const char* p, |
michael@0 | 439 | const char* l, |
michael@0 | 440 | uint32* OUTPUT) { |
michael@0 | 441 | const unsigned char* ptr = reinterpret_cast<const unsigned char*>(p); |
michael@0 | 442 | const unsigned char* limit = reinterpret_cast<const unsigned char*>(l); |
michael@0 | 443 | uint32 b, result; |
michael@0 | 444 | if (ptr >= limit) return NULL; |
michael@0 | 445 | b = *(ptr++); result = b & 127; if (b < 128) goto done; |
michael@0 | 446 | if (ptr >= limit) return NULL; |
michael@0 | 447 | b = *(ptr++); result |= (b & 127) << 7; if (b < 128) goto done; |
michael@0 | 448 | if (ptr >= limit) return NULL; |
michael@0 | 449 | b = *(ptr++); result |= (b & 127) << 14; if (b < 128) goto done; |
michael@0 | 450 | if (ptr >= limit) return NULL; |
michael@0 | 451 | b = *(ptr++); result |= (b & 127) << 21; if (b < 128) goto done; |
michael@0 | 452 | if (ptr >= limit) return NULL; |
michael@0 | 453 | b = *(ptr++); result |= (b & 127) << 28; if (b < 16) goto done; |
michael@0 | 454 | return NULL; // Value is too long to be a varint32 |
michael@0 | 455 | done: |
michael@0 | 456 | *OUTPUT = result; |
michael@0 | 457 | return reinterpret_cast<const char*>(ptr); |
michael@0 | 458 | } |
michael@0 | 459 | |
michael@0 | 460 | inline char* Varint::Encode32(char* sptr, uint32 v) { |
michael@0 | 461 | // Operate on characters as unsigneds |
michael@0 | 462 | unsigned char* ptr = reinterpret_cast<unsigned char*>(sptr); |
michael@0 | 463 | static const int B = 128; |
michael@0 | 464 | if (v < (1<<7)) { |
michael@0 | 465 | *(ptr++) = v; |
michael@0 | 466 | } else if (v < (1<<14)) { |
michael@0 | 467 | *(ptr++) = v | B; |
michael@0 | 468 | *(ptr++) = v>>7; |
michael@0 | 469 | } else if (v < (1<<21)) { |
michael@0 | 470 | *(ptr++) = v | B; |
michael@0 | 471 | *(ptr++) = (v>>7) | B; |
michael@0 | 472 | *(ptr++) = v>>14; |
michael@0 | 473 | } else if (v < (1<<28)) { |
michael@0 | 474 | *(ptr++) = v | B; |
michael@0 | 475 | *(ptr++) = (v>>7) | B; |
michael@0 | 476 | *(ptr++) = (v>>14) | B; |
michael@0 | 477 | *(ptr++) = v>>21; |
michael@0 | 478 | } else { |
michael@0 | 479 | *(ptr++) = v | B; |
michael@0 | 480 | *(ptr++) = (v>>7) | B; |
michael@0 | 481 | *(ptr++) = (v>>14) | B; |
michael@0 | 482 | *(ptr++) = (v>>21) | B; |
michael@0 | 483 | *(ptr++) = v>>28; |
michael@0 | 484 | } |
michael@0 | 485 | return reinterpret_cast<char*>(ptr); |
michael@0 | 486 | } |
michael@0 | 487 | |
michael@0 | 488 | // If you know the internal layout of the std::string in use, you can |
michael@0 | 489 | // replace this function with one that resizes the string without |
michael@0 | 490 | // filling the new space with zeros (if applicable) -- |
michael@0 | 491 | // it will be non-portable but faster. |
michael@0 | 492 | inline void STLStringResizeUninitialized(string* s, size_t new_size) { |
michael@0 | 493 | s->resize(new_size); |
michael@0 | 494 | } |
michael@0 | 495 | |
michael@0 | 496 | // Return a mutable char* pointing to a string's internal buffer, |
michael@0 | 497 | // which may not be null-terminated. Writing through this pointer will |
michael@0 | 498 | // modify the string. |
michael@0 | 499 | // |
michael@0 | 500 | // string_as_array(&str)[i] is valid for 0 <= i < str.size() until the |
michael@0 | 501 | // next call to a string method that invalidates iterators. |
michael@0 | 502 | // |
michael@0 | 503 | // As of 2006-04, there is no standard-blessed way of getting a |
michael@0 | 504 | // mutable reference to a string's internal buffer. However, issue 530 |
michael@0 | 505 | // (http://www.open-std.org/JTC1/SC22/WG21/docs/lwg-defects.html#530) |
michael@0 | 506 | // proposes this as the method. It will officially be part of the standard |
michael@0 | 507 | // for C++0x. This should already work on all current implementations. |
michael@0 | 508 | inline char* string_as_array(string* str) { |
michael@0 | 509 | return str->empty() ? NULL : &*str->begin(); |
michael@0 | 510 | } |
michael@0 | 511 | |
michael@0 | 512 | } // namespace snappy |
michael@0 | 513 | |
michael@0 | 514 | #endif // UTIL_SNAPPY_OPENSOURCE_SNAPPY_STUBS_INTERNAL_H_ |