| 1 | // Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file |
| 2 | // for details. All rights reserved. Use of this source code is governed by a |
| 3 | // BSD-style license that can be found in the LICENSE file. |
| 4 | |
| 5 | #ifndef RUNTIME_PLATFORM_UTILS_H_ |
| 6 | #define RUNTIME_PLATFORM_UTILS_H_ |
| 7 | |
| 8 | #include <limits> |
| 9 | #include <memory> |
| 10 | #include <type_traits> |
| 11 | |
| 12 | #include "platform/assert.h" |
| 13 | #include "platform/globals.h" |
| 14 | |
| 15 | namespace dart { |
| 16 | |
| 17 | class Utils { |
| 18 | public: |
| 19 | template <typename T> |
| 20 | static inline T Minimum(T x, T y) { |
| 21 | return x < y ? x : y; |
| 22 | } |
| 23 | |
| 24 | template <typename T> |
| 25 | static constexpr inline T Maximum(T x, T y) { |
| 26 | return x > y ? x : y; |
| 27 | } |
| 28 | |
| 29 | // Calculates absolute value of a given signed integer. |
| 30 | // `x` must not be equal to minimum value representable by `T` |
| 31 | // as its absolute value is out of range. |
| 32 | template <typename T> |
| 33 | static inline T Abs(T x) { |
| 34 | // Note: as a general rule, it is not OK to use STL in Dart VM. |
| 35 | // However, std::numeric_limits<T>::min() and max() are harmless |
| 36 | // and worthwhile exception from this rule. |
| 37 | ASSERT(x != std::numeric_limits<T>::min()); |
| 38 | if (x < 0) return -x; |
| 39 | return x; |
| 40 | } |
| 41 | |
| 42 | // Calculates absolute value of a given signed integer with saturation. |
| 43 | // If `x` equals to minimum value representable by `T`, then |
| 44 | // absolute value is saturated to the maximum value representable by `T`. |
| 45 | template <typename T> |
| 46 | static inline T AbsWithSaturation(T x) { |
| 47 | if (x < 0) { |
| 48 | // Note: as a general rule, it is not OK to use STL in Dart VM. |
| 49 | // However, std::numeric_limits<T>::min() and max() are harmless |
| 50 | // and worthwhile exception from this rule. |
| 51 | if (x == std::numeric_limits<T>::min()) { |
| 52 | return std::numeric_limits<T>::max(); |
| 53 | } |
| 54 | return -x; |
| 55 | } |
| 56 | return x; |
| 57 | } |
| 58 | |
| 59 | template <typename T> |
| 60 | static constexpr bool IsPowerOfTwo(T x) { |
| 61 | return ((x & (x - 1)) == 0) && (x != 0); |
| 62 | } |
| 63 | |
| 64 | template <typename T> |
| 65 | static inline int ShiftForPowerOfTwo(T x) { |
| 66 | ASSERT(IsPowerOfTwo(x)); |
| 67 | int num_shifts = 0; |
| 68 | while (x > 1) { |
| 69 | num_shifts++; |
| 70 | x = x >> 1; |
| 71 | } |
| 72 | return num_shifts; |
| 73 | } |
| 74 | |
| 75 | template <typename T> |
| 76 | static constexpr bool IsAligned(T x, |
| 77 | uintptr_t alignment, |
| 78 | uintptr_t offset = 0) { |
| 79 | ASSERT(IsPowerOfTwo(alignment)); |
| 80 | ASSERT(offset < alignment); |
| 81 | return (x & (alignment - 1)) == offset; |
| 82 | } |
| 83 | |
| 84 | template <typename T> |
| 85 | static constexpr bool IsAligned(T* x, |
| 86 | uintptr_t alignment, |
| 87 | uintptr_t offset = 0) { |
| 88 | return IsAligned(x: reinterpret_cast<uword>(x), alignment, offset); |
| 89 | } |
| 90 | |
| 91 | template <typename T> |
| 92 | static constexpr inline T RoundDown(T x, intptr_t alignment) { |
| 93 | ASSERT(IsPowerOfTwo(alignment)); |
| 94 | return (x & -alignment); |
| 95 | } |
| 96 | |
| 97 | template <typename T> |
| 98 | static inline T* RoundDown(T* x, intptr_t alignment) { |
| 99 | return reinterpret_cast<T*>( |
| 100 | RoundDown(x: reinterpret_cast<uword>(x), alignment)); |
| 101 | } |
| 102 | |
| 103 | template <typename T> |
| 104 | static constexpr inline T RoundUp(T x, |
| 105 | uintptr_t alignment, |
| 106 | uintptr_t offset = 0) { |
| 107 | ASSERT(offset < alignment); |
| 108 | return RoundDown(x + alignment - 1 + offset, alignment) - offset; |
| 109 | } |
| 110 | |
| 111 | template <typename T> |
| 112 | static inline T* RoundUp(T* x, uintptr_t alignment, uintptr_t offset = 0) { |
| 113 | return reinterpret_cast<T*>( |
| 114 | RoundUp(x: reinterpret_cast<uword>(x), alignment, offset)); |
| 115 | } |
| 116 | |
| 117 | // Implementation is from "Hacker's Delight" by Henry S. Warren, Jr., |
| 118 | // figure 3-3, page 48, where the function is called clp2. |
| 119 | static constexpr uintptr_t RoundUpToPowerOfTwo(uintptr_t x) { |
| 120 | x = x - 1; |
| 121 | x = x | (x >> 1); |
| 122 | x = x | (x >> 2); |
| 123 | x = x | (x >> 4); |
| 124 | x = x | (x >> 8); |
| 125 | x = x | (x >> 16); |
| 126 | #if defined(ARCH_IS_64_BIT) |
| 127 | x = x | (x >> 32); |
| 128 | #endif // defined(ARCH_IS_64_BIT) |
| 129 | return x + 1; |
| 130 | } |
| 131 | |
| 132 | static constexpr int CountOneBits64(uint64_t x) { |
| 133 | // Apparently there are x64 chips without popcount. |
| 134 | #if __GNUC__ && !defined(HOST_ARCH_IA32) && !defined(HOST_ARCH_X64) |
| 135 | return __builtin_popcountll(x); |
| 136 | #else |
| 137 | x = x - ((x >> 1) & 0x5555555555555555); |
| 138 | x = (x & 0x3333333333333333) + ((x >> 2) & 0x3333333333333333); |
| 139 | x = (((x + (x >> 4)) & 0x0f0f0f0f0f0f0f0f) * 0x0101010101010101) >> 56; |
| 140 | return x; |
| 141 | #endif |
| 142 | } |
| 143 | |
| 144 | static constexpr int CountOneBits32(uint32_t x) { |
| 145 | // Apparently there are x64 chips without popcount. |
| 146 | #if __GNUC__ && !defined(HOST_ARCH_IA32) && !defined(HOST_ARCH_X64) |
| 147 | return __builtin_popcount(x); |
| 148 | #else |
| 149 | // Implementation is from "Hacker's Delight" by Henry S. Warren, Jr., |
| 150 | // figure 5-2, page 66, where the function is called pop. |
| 151 | x = x - ((x >> 1) & 0x55555555); |
| 152 | x = (x & 0x33333333) + ((x >> 2) & 0x33333333); |
| 153 | x = (x + (x >> 4)) & 0x0F0F0F0F; |
| 154 | x = x + (x >> 8); |
| 155 | x = x + (x >> 16); |
| 156 | return static_cast<int>(x & 0x0000003F); |
| 157 | #endif |
| 158 | } |
| 159 | |
| 160 | static constexpr int CountOneBitsWord(uword x) { |
| 161 | #ifdef ARCH_IS_64_BIT |
| 162 | return CountOneBits64(x); |
| 163 | #else |
| 164 | return CountOneBits32(x); |
| 165 | #endif |
| 166 | } |
| 167 | |
| 168 | // TODO(koda): Compare to flsll call/intrinsic. |
| 169 | static constexpr size_t HighestBit(int64_t v) { |
| 170 | uint64_t x = static_cast<uint64_t>((v > 0) ? v : -v); |
| 171 | uint64_t t = 0; |
| 172 | size_t r = 0; |
| 173 | if ((t = x >> 32) != 0) { |
| 174 | x = t; |
| 175 | r += 32; |
| 176 | } |
| 177 | if ((t = x >> 16) != 0) { |
| 178 | x = t; |
| 179 | r += 16; |
| 180 | } |
| 181 | if ((t = x >> 8) != 0) { |
| 182 | x = t; |
| 183 | r += 8; |
| 184 | } |
| 185 | if ((t = x >> 4) != 0) { |
| 186 | x = t; |
| 187 | r += 4; |
| 188 | } |
| 189 | if ((t = x >> 2) != 0) { |
| 190 | x = t; |
| 191 | r += 2; |
| 192 | } |
| 193 | if (x > 1) r += 1; |
| 194 | return r; |
| 195 | } |
| 196 | |
| 197 | static constexpr size_t BitLength(int64_t value) { |
| 198 | // Flip bits if negative (-1 becomes 0). |
| 199 | value ^= value >> (8 * sizeof(value) - 1); |
| 200 | return (value == 0) ? 0 : (Utils::HighestBit(v: value) + 1); |
| 201 | } |
| 202 | |
| 203 | static int CountLeadingZeros32(uint32_t x) { |
| 204 | #if defined(DART_HOST_OS_WINDOWS) |
| 205 | unsigned long position; // NOLINT |
| 206 | return (_BitScanReverse(&position, x) == 0) |
| 207 | ? 32 |
| 208 | : 31 - static_cast<int>(position); |
| 209 | #else |
| 210 | return x == 0 ? 32 : __builtin_clz(x); |
| 211 | #endif |
| 212 | } |
| 213 | static int CountLeadingZeros64(uint64_t x) { |
| 214 | #if defined(DART_HOST_OS_WINDOWS) |
| 215 | #if defined(ARCH_IS_32_BIT) |
| 216 | const uint32_t x_hi = static_cast<uint32_t>(x >> 32); |
| 217 | if (x_hi != 0) { |
| 218 | return CountLeadingZeros32(x_hi); |
| 219 | } |
| 220 | return 32 + CountLeadingZeros32(static_cast<uint32_t>(x)); |
| 221 | #else |
| 222 | unsigned long position; // NOLINT |
| 223 | return (_BitScanReverse64(&position, x) == 0) |
| 224 | ? 64 |
| 225 | : 63 - static_cast<int>(position); |
| 226 | #endif |
| 227 | #else |
| 228 | return x == 0 ? 64 : __builtin_clzll(x); |
| 229 | #endif |
| 230 | } |
| 231 | static int CountLeadingZerosWord(uword x) { |
| 232 | #ifdef ARCH_IS_64_BIT |
| 233 | return CountLeadingZeros64(x); |
| 234 | #else |
| 235 | return CountLeadingZeros32(x); |
| 236 | #endif |
| 237 | } |
| 238 | |
| 239 | static int CountTrailingZeros32(uint32_t x) { |
| 240 | #if defined(DART_HOST_OS_WINDOWS) |
| 241 | unsigned long position; // NOLINT |
| 242 | return (_BitScanForward(&position, x) == 0) ? 32 |
| 243 | : static_cast<int>(position); |
| 244 | #else |
| 245 | return x == 0 ? 32 : __builtin_ctz(x); |
| 246 | #endif |
| 247 | } |
| 248 | static int CountTrailingZeros64(uint64_t x) { |
| 249 | #if defined(DART_HOST_OS_WINDOWS) |
| 250 | #if defined(ARCH_IS_32_BIT) |
| 251 | const uint32_t x_lo = static_cast<uint32_t>(x); |
| 252 | if (x_lo != 0) { |
| 253 | return CountTrailingZeros32(x_lo); |
| 254 | } |
| 255 | return 32 + CountTrailingZeros32(static_cast<uint32_t>(x >> 32)); |
| 256 | #else |
| 257 | unsigned long position; // NOLINT |
| 258 | return (_BitScanForward64(&position, x) == 0) ? 64 |
| 259 | : static_cast<int>(position); |
| 260 | #endif |
| 261 | #else |
| 262 | return x == 0 ? 64 : __builtin_ctzll(x); |
| 263 | #endif |
| 264 | } |
| 265 | static int CountTrailingZerosWord(uword x) { |
| 266 | #ifdef ARCH_IS_64_BIT |
| 267 | return CountTrailingZeros64(x); |
| 268 | #else |
| 269 | return CountTrailingZeros32(x); |
| 270 | #endif |
| 271 | } |
| 272 | |
| 273 | static uint64_t ReverseBits64(uint64_t x); |
| 274 | static uint32_t ReverseBits32(uint32_t x); |
| 275 | |
| 276 | static uword ReverseBitsWord(uword x) { |
| 277 | #ifdef ARCH_IS_64_BIT |
| 278 | return ReverseBits64(x); |
| 279 | #else |
| 280 | return ReverseBits32(x); |
| 281 | #endif |
| 282 | } |
| 283 | |
| 284 | // Computes magic numbers to implement DIV or MOD operator. |
| 285 | static void CalculateMagicAndShiftForDivRem(int64_t divisor, |
| 286 | int64_t* magic, |
| 287 | int64_t* shift); |
| 288 | |
| 289 | // Computes a hash value for the given series of bytes. |
| 290 | static uint32_t StringHash(const void* data, int length); |
| 291 | |
| 292 | // Computes a hash value for the given word. |
| 293 | static uint32_t WordHash(intptr_t key); |
| 294 | |
| 295 | // Check whether an N-bit two's-complement representation can hold value. |
| 296 | template <typename T> |
| 297 | static inline bool IsInt(intptr_t N, T value) { |
| 298 | ASSERT(N >= 1); |
| 299 | constexpr intptr_t value_size_in_bits = kBitsPerByte * sizeof(T); |
| 300 | if constexpr (std::is_signed<T>::value) { |
| 301 | if (N >= value_size_in_bits) return true; // Trivially fits. |
| 302 | } else { |
| 303 | if (N > value_size_in_bits) return true; // Trivially fits. |
| 304 | if (N == value_size_in_bits) { |
| 305 | return static_cast<typename std::make_signed<T>::type>(value) >= 0; |
| 306 | } |
| 307 | } |
| 308 | const T limit = static_cast<T>(1) << (N - 1); |
| 309 | return (-limit <= value) && (value < limit); |
| 310 | } |
| 311 | |
| 312 | template <typename T> |
| 313 | static inline bool IsUint(intptr_t N, T value) { |
| 314 | ASSERT(N >= 1); |
| 315 | constexpr intptr_t value_size_in_bits = kBitsPerByte * sizeof(T); |
| 316 | if constexpr (std::is_signed<T>::value) { |
| 317 | if (value < 0) return false; // Not an unsigned value. |
| 318 | if (N >= value_size_in_bits - 1) { |
| 319 | return true; // N can fit the magnitude bits. |
| 320 | } |
| 321 | } else { |
| 322 | if (N >= value_size_in_bits) return true; // Trivially fits. |
| 323 | } |
| 324 | const T limit = (static_cast<T>(1) << N) - 1; |
| 325 | return value <= limit; |
| 326 | } |
| 327 | |
| 328 | // Check whether the magnitude of value fits in N bits. This differs from |
| 329 | // IsInt(N + 1, value) only in that this returns false for the minimum value |
| 330 | // of a N+1 bit two's complement value. |
| 331 | // |
| 332 | // Primarily used for testing whether a two's complement value can be used in |
| 333 | // a place where the sign is replaced with a marker that says whether the |
| 334 | // magnitude is added or subtracted, e.g., the U bit (bit 23) in some ARM7 |
| 335 | // instructions. |
| 336 | template <typename T> |
| 337 | static inline bool MagnitudeIsUint(intptr_t N, T value) { |
| 338 | ASSERT(N >= 1); |
| 339 | if constexpr (std::is_signed<T>::value) { |
| 340 | using Unsigned = typename std::make_unsigned<T>::type; |
| 341 | if (value < 0) return IsUint<Unsigned>(N, -value); |
| 342 | } |
| 343 | return IsUint(N, value); |
| 344 | } |
| 345 | |
| 346 | static inline int32_t Low16Bits(int32_t value) { |
| 347 | return static_cast<int32_t>(value & 0xffff); |
| 348 | } |
| 349 | |
| 350 | static inline int32_t High16Bits(int32_t value) { |
| 351 | return static_cast<int32_t>(value >> 16); |
| 352 | } |
| 353 | |
| 354 | static inline int32_t Low32Bits(int64_t value) { |
| 355 | return static_cast<int32_t>(value); |
| 356 | } |
| 357 | |
| 358 | static inline int32_t High32Bits(int64_t value) { |
| 359 | return static_cast<int32_t>(value >> 32); |
| 360 | } |
| 361 | |
| 362 | static inline int64_t LowHighTo64Bits(uint32_t low, int32_t high) { |
| 363 | return (static_cast<uint64_t>(high) << 32) | (low & 0x0ffffffffLL); |
| 364 | } |
| 365 | |
| 366 | static inline constexpr bool IsAlphaNumeric(uint32_t c) { |
| 367 | return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || |
| 368 | IsDecimalDigit(c); |
| 369 | } |
| 370 | |
| 371 | static inline constexpr bool IsDecimalDigit(uint32_t c) { |
| 372 | return ('0' <= c) && (c <= '9'); |
| 373 | } |
| 374 | |
| 375 | static bool IsHexDigit(char c) { |
| 376 | return IsDecimalDigit(c) || (('A' <= c) && (c <= 'F')) || |
| 377 | (('a' <= c) && (c <= 'f')); |
| 378 | } |
| 379 | |
| 380 | static int HexDigitToInt(char c) { |
| 381 | ASSERT(IsHexDigit(c)); |
| 382 | if (IsDecimalDigit(c)) return c - '0'; |
| 383 | if (('A' <= c) && (c <= 'F')) return 10 + (c - 'A'); |
| 384 | return 10 + (c - 'a'); |
| 385 | } |
| 386 | |
| 387 | static char IntToHexDigit(int i) { |
| 388 | ASSERT(0 <= i && i < 16); |
| 389 | if (i < 10) return static_cast<char>('0' + i); |
| 390 | return static_cast<char>('A' + (i - 10)); |
| 391 | } |
| 392 | |
| 393 | // Perform a range check, checking if |
| 394 | // offset + count <= length |
| 395 | // without the risk of integer overflow. |
| 396 | static inline bool RangeCheck(intptr_t offset, |
| 397 | intptr_t count, |
| 398 | intptr_t length) { |
| 399 | return offset >= 0 && count >= 0 && length >= 0 && |
| 400 | count <= (length - offset); |
| 401 | } |
| 402 | |
| 403 | static inline bool WillAddOverflow(int64_t a, int64_t b) { |
| 404 | return ((b > 0) && (a > (kMaxInt64 - b))) || |
| 405 | ((b < 0) && (a < (kMinInt64 - b))); |
| 406 | } |
| 407 | |
| 408 | static inline bool WillSubOverflow(int64_t a, int64_t b) { |
| 409 | return ((b > 0) && (a < (kMinInt64 + b))) || |
| 410 | ((b < 0) && (a > (kMaxInt64 + b))); |
| 411 | } |
| 412 | |
| 413 | // Adds two int64_t values with wrapping around |
| 414 | // (two's complement arithmetic). |
| 415 | template <typename T = int64_t> |
| 416 | static inline T AddWithWrapAround(T a, T b) { |
| 417 | // Avoid undefined behavior by doing arithmetic in the unsigned type. |
| 418 | using Unsigned = typename std::make_unsigned<T>::type; |
| 419 | return static_cast<T>(static_cast<Unsigned>(a) + static_cast<Unsigned>(b)); |
| 420 | } |
| 421 | |
| 422 | // Subtracts two int64_t values with wrapping around |
| 423 | // (two's complement arithmetic). |
| 424 | template <typename T = int64_t> |
| 425 | static inline T SubWithWrapAround(T a, T b) { |
| 426 | // Avoid undefined behavior by doing arithmetic in the unsigned type. |
| 427 | using Unsigned = typename std::make_unsigned<T>::type; |
| 428 | return static_cast<T>(static_cast<Unsigned>(a) - static_cast<Unsigned>(b)); |
| 429 | } |
| 430 | |
| 431 | // Multiplies two int64_t values with wrapping around |
| 432 | // (two's complement arithmetic). |
| 433 | template <typename T = int64_t> |
| 434 | static inline T MulWithWrapAround(T a, T b) { |
| 435 | // Avoid undefined behavior by doing arithmetic in the unsigned type. |
| 436 | using Unsigned = typename std::make_unsigned<T>::type; |
| 437 | return static_cast<T>(static_cast<Unsigned>(a) * static_cast<Unsigned>(b)); |
| 438 | } |
| 439 | |
| 440 | template <typename T = int64_t> |
| 441 | static inline T NegWithWrapAround(T a) { |
| 442 | // Avoid undefined behavior by doing arithmetic in the unsigned type. |
| 443 | using Unsigned = typename std::make_unsigned<T>::type; |
| 444 | return static_cast<T>(-static_cast<Unsigned>(a)); |
| 445 | } |
| 446 | |
| 447 | // Shifts int64_t value left. Supports any non-negative number of bits and |
| 448 | // silently discards shifted out bits. |
| 449 | static inline int64_t ShiftLeftWithTruncation(int64_t a, int64_t b) { |
| 450 | ASSERT(b >= 0); |
| 451 | if (b >= kBitsPerInt64) { |
| 452 | return 0; |
| 453 | } |
| 454 | // Avoid undefined behavior by doing arithmetic in the unsigned type. |
| 455 | return static_cast<int64_t>(static_cast<uint64_t>(a) << b); |
| 456 | } |
| 457 | |
| 458 | template <typename T> |
| 459 | static inline T RotateLeft(T value, uint8_t rotate) { |
| 460 | const uint8_t width = sizeof(T) * kBitsPerByte; |
| 461 | ASSERT(0 <= rotate); |
| 462 | ASSERT(rotate <= width); |
| 463 | using Unsigned = typename std::make_unsigned<T>::type; |
| 464 | return (static_cast<Unsigned>(value) << rotate) | |
| 465 | (static_cast<T>(value) >> ((width - rotate) & (width - 1))); |
| 466 | } |
| 467 | template <typename T> |
| 468 | static inline T RotateRight(T value, uint8_t rotate) { |
| 469 | const uint8_t width = sizeof(T) * kBitsPerByte; |
| 470 | ASSERT(0 <= rotate); |
| 471 | ASSERT(rotate <= width); |
| 472 | using Unsigned = typename std::make_unsigned<T>::type; |
| 473 | return (static_cast<T>(value) >> rotate) | |
| 474 | (static_cast<Unsigned>(value) << ((width - rotate) & (width - 1))); |
| 475 | } |
| 476 | |
| 477 | // Utility functions for converting values from host endianness to |
| 478 | // big or little endian values. |
| 479 | static uint16_t HostToBigEndian16(uint16_t host_value); |
| 480 | static uint32_t HostToBigEndian32(uint32_t host_value); |
| 481 | static uint64_t HostToBigEndian64(uint64_t host_value); |
| 482 | static uint16_t HostToLittleEndian16(uint16_t host_value); |
| 483 | static uint32_t HostToLittleEndian32(uint32_t host_value); |
| 484 | static uint64_t HostToLittleEndian64(uint64_t host_value); |
| 485 | |
| 486 | // Going between Host <-> LE/BE is the same operation for all practical |
| 487 | // purposes. |
| 488 | static inline uint32_t BigEndianToHost32(uint32_t be_value) { |
| 489 | return HostToBigEndian32(host_value: be_value); |
| 490 | } |
| 491 | static inline uint64_t LittleEndianToHost64(uint64_t le_value) { |
| 492 | return HostToLittleEndian64(host_value: le_value); |
| 493 | } |
| 494 | |
| 495 | static bool DoublesBitEqual(const double a, const double b) { |
| 496 | return bit_cast<int64_t, double>(source: a) == bit_cast<int64_t, double>(source: b); |
| 497 | } |
| 498 | |
| 499 | // A double-to-integer conversion that avoids undefined behavior. |
| 500 | // Out of range values and NaNs are converted to minimum value |
| 501 | // for type T. |
| 502 | template <typename T> |
| 503 | static T SafeDoubleToInt(double v) { |
| 504 | const double min = static_cast<double>(std::numeric_limits<T>::min()); |
| 505 | const double max = static_cast<double>(std::numeric_limits<T>::max()); |
| 506 | return (min <= v && v <= max) ? static_cast<T>(v) |
| 507 | : std::numeric_limits<T>::min(); |
| 508 | } |
| 509 | |
| 510 | // dart2js represents integers as double precision floats, which can |
| 511 | // represent anything in the range -2^53 ... 2^53. |
| 512 | static bool IsJavaScriptInt(int64_t value) { |
| 513 | return ((-0x20000000000000LL <= value) && (value <= 0x20000000000000LL)); |
| 514 | } |
| 515 | |
| 516 | // The lowest n bits are 1, the others are 0. |
| 517 | template <typename T = uword> |
| 518 | static constexpr T NBitMask(size_t n) { |
| 519 | using Unsigned = typename std::make_unsigned<T>::type; |
| 520 | constexpr size_t kBitsPerT = sizeof(T) * kBitsPerByte; |
| 521 | assert(n <= sizeof(T) * kBitsPerT); |
| 522 | return static_cast<T>(n == kBitsPerT ? std::numeric_limits<Unsigned>::max() |
| 523 | : (static_cast<Unsigned>(1) << n) - 1); |
| 524 | } |
| 525 | |
| 526 | template <typename T = uword> |
| 527 | static constexpr T Bit(size_t n) { |
| 528 | ASSERT(n < sizeof(T) * kBitsPerByte); |
| 529 | T bit = 1; |
| 530 | return bit << n; |
| 531 | } |
| 532 | |
| 533 | template <typename T> |
| 534 | static constexpr bool TestBit(T mask, size_t position) { |
| 535 | ASSERT(position < sizeof(T) * kBitsPerByte); |
| 536 | return ((mask >> position) & 1) != 0; |
| 537 | } |
| 538 | |
| 539 | static char* StrError(int err, char* buffer, size_t bufsize); |
| 540 | |
| 541 | // Not all platforms support strndup. |
| 542 | static char* StrNDup(const char* s, intptr_t n); |
| 543 | static char* StrDup(const char* s); |
| 544 | static intptr_t StrNLen(const char* s, intptr_t n); |
| 545 | static bool StrStartsWith(const char* s, const char* prefix) { |
| 546 | return strncmp(s1: s, s2: prefix, n: strlen(s: prefix)) == 0; |
| 547 | } |
| 548 | |
| 549 | static int Close(int fildes); |
| 550 | static size_t Read(int filedes, void* buf, size_t nbyte); |
| 551 | static int Unlink(const char* path); |
| 552 | |
| 553 | // Print formatted output info a buffer. |
| 554 | // |
| 555 | // Does not write more than size characters (including the trailing '\0'). |
| 556 | // |
| 557 | // Returns the number of characters (excluding the trailing '\0') |
| 558 | // that would been written if the buffer had been big enough. If |
| 559 | // the return value is greater or equal than the given size then the |
| 560 | // output has been truncated. The return value is never negative. |
| 561 | // |
| 562 | // The buffer will always be terminated by a '\0', unless the buffer |
| 563 | // is of size 0. The buffer might be nullptr if the size is 0. |
| 564 | // |
| 565 | // This specification conforms to C99 standard which is implemented |
| 566 | // by glibc 2.1+ with one exception: the C99 standard allows a |
| 567 | // negative return value. We will terminate the vm rather than let |
| 568 | // that occur. |
| 569 | static int SNPrint(char* str, size_t size, const char* format, ...) |
| 570 | PRINTF_ATTRIBUTE(3, 4); |
| 571 | static int VSNPrint(char* str, size_t size, const char* format, va_list args); |
| 572 | |
| 573 | // Allocate a string and print formatted output into a malloc'd buffer. |
| 574 | static char* SCreate(const char* format, ...) PRINTF_ATTRIBUTE(1, 2); |
| 575 | static char* VSCreate(const char* format, va_list args); |
| 576 | |
| 577 | typedef std::unique_ptr<char, decltype(std::free)*> CStringUniquePtr; |
| 578 | |
| 579 | // Returns str in a unique_ptr with free used as its deleter. |
| 580 | static CStringUniquePtr CreateCStringUniquePtr(char* str); |
| 581 | |
| 582 | // Load dynamic library from the given |library_path| and return the |
| 583 | // library handle. |library_path| can be |nullptr| in which case |
| 584 | // library handle representing the executable is returned. |
| 585 | // If an error occurs returns |nullptr| and populates |
| 586 | // |error| (if provided) with an error message (caller must free this message |
| 587 | // when it is no longer needed). |
| 588 | static void* LoadDynamicLibrary(const char* library_path, |
| 589 | char** error = nullptr); |
| 590 | |
| 591 | // Resolve the given |symbol| within the library referenced by the |
| 592 | // given |library_handle|. |
| 593 | // If an error occurs populates |error| (if provided) with an error message |
| 594 | // (caller must free this message when it is no longer needed). |
| 595 | // Note: on some platforms |nullptr| is a valid value for a symbol, so to |
| 596 | // check if resolution succeeded one must instead provide non-null |error| |
| 597 | // and then check if it was populated with an error message. |
| 598 | static void* ResolveSymbolInDynamicLibrary(void* library_handle, |
| 599 | const char* symbol, |
| 600 | char** error = nullptr); |
| 601 | |
| 602 | // Unload the library referenced by the given |library_handle|. |
| 603 | // If an error occurs returns |nullptr| and populates |
| 604 | // |error| (if provided) with an error message (caller must free this message |
| 605 | // when it is no longer needed). |
| 606 | static void UnloadDynamicLibrary(void* library_handle, |
| 607 | char** error = nullptr); |
| 608 | }; |
| 609 | |
| 610 | } // namespace dart |
| 611 | |
| 612 | #if defined(DART_HOST_OS_ANDROID) |
| 613 | #include "platform/utils_android.h" |
| 614 | #elif defined(DART_HOST_OS_FUCHSIA) |
| 615 | #include "platform/utils_fuchsia.h" |
| 616 | #elif defined(DART_HOST_OS_LINUX) |
| 617 | #include "platform/utils_linux.h" |
| 618 | #elif defined(DART_HOST_OS_MACOS) |
| 619 | #include "platform/utils_macos.h" |
| 620 | #elif defined(DART_HOST_OS_WINDOWS) |
| 621 | #include "platform/utils_win.h" |
| 622 | #else |
| 623 | #error Unknown target os. |
| 624 | #endif |
| 625 | |
| 626 | #endif // RUNTIME_PLATFORM_UTILS_H_ |
| 627 | |