| 1 | //===-- llvm/ADT/bit.h - C++20 <bit> ----------------------------*- C++ -*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | /// |
| 9 | /// \file |
| 10 | /// This file implements the C++20 <bit> header. |
| 11 | /// |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #ifndef LLVM_ADT_BIT_H |
| 15 | #define LLVM_ADT_BIT_H |
| 16 | |
| 17 | #include "llvm/Support/Compiler.h" |
| 18 | #include <cstdint> |
| 19 | #include <limits> |
| 20 | #include <type_traits> |
| 21 | |
| 22 | #if !__has_builtin(__builtin_bit_cast) |
| 23 | #include <cstring> |
| 24 | #endif |
| 25 | |
| 26 | #if defined(_MSC_VER) && !defined(_DEBUG) |
| 27 | #include <cstdlib> // for _byteswap_{ushort,ulong,uint64} |
| 28 | #endif |
| 29 | |
| 30 | #if defined(__linux__) || defined(__GNU__) || defined(__HAIKU__) || \ |
| 31 | defined(__Fuchsia__) || defined(__EMSCRIPTEN__) || defined(__NetBSD__) || \ |
| 32 | defined(__OpenBSD__) || defined(__DragonFly__) |
| 33 | #include <endian.h> |
| 34 | #elif defined(_AIX) |
| 35 | #include <sys/machine.h> |
| 36 | #elif defined(__sun) |
| 37 | /* Solaris provides _BIG_ENDIAN/_LITTLE_ENDIAN selector in sys/types.h */ |
| 38 | #include <sys/types.h> |
| 39 | #define BIG_ENDIAN 4321 |
| 40 | #define LITTLE_ENDIAN 1234 |
| 41 | #if defined(_BIG_ENDIAN) |
| 42 | #define BYTE_ORDER BIG_ENDIAN |
| 43 | #else |
| 44 | #define BYTE_ORDER LITTLE_ENDIAN |
| 45 | #endif |
| 46 | #elif defined(__MVS__) |
| 47 | #define BIG_ENDIAN 4321 |
| 48 | #define LITTLE_ENDIAN 1234 |
| 49 | #define BYTE_ORDER BIG_ENDIAN |
| 50 | #else |
| 51 | #if !defined(BYTE_ORDER) && !defined(_WIN32) |
| 52 | #include <machine/endian.h> |
| 53 | #endif |
| 54 | #endif |
| 55 | |
| 56 | #ifdef _MSC_VER |
| 57 | // Declare these intrinsics manually rather including intrin.h. It's very |
| 58 | // expensive, and bit.h is popular via MathExtras.h. |
| 59 | // #include <intrin.h> |
| 60 | extern "C" { |
| 61 | unsigned char _BitScanForward(unsigned long *_Index, unsigned long _Mask); |
| 62 | unsigned char _BitScanForward64(unsigned long *_Index, unsigned __int64 _Mask); |
| 63 | unsigned char _BitScanReverse(unsigned long *_Index, unsigned long _Mask); |
| 64 | unsigned char _BitScanReverse64(unsigned long *_Index, unsigned __int64 _Mask); |
| 65 | } |
| 66 | #endif |
| 67 | |
| 68 | namespace llvm { |
| 69 | |
| 70 | enum class endianness { |
| 71 | big, |
| 72 | little, |
| 73 | #if defined(BYTE_ORDER) && defined(BIG_ENDIAN) && BYTE_ORDER == BIG_ENDIAN |
| 74 | native = big |
| 75 | #else |
| 76 | native = little |
| 77 | #endif |
| 78 | }; |
| 79 | |
| 80 | // This implementation of bit_cast is different from the C++20 one in two ways: |
| 81 | // - It isn't constexpr because that requires compiler support. |
| 82 | // - It requires trivially-constructible To, to avoid UB in the implementation. |
| 83 | template < |
| 84 | typename To, typename From, |
| 85 | typename = std::enable_if_t<sizeof(To) == sizeof(From)>, |
| 86 | typename = std::enable_if_t<std::is_trivially_constructible<To>::value>, |
| 87 | typename = std::enable_if_t<std::is_trivially_copyable<To>::value>, |
| 88 | typename = std::enable_if_t<std::is_trivially_copyable<From>::value>> |
| 89 | [[nodiscard]] inline To bit_cast(const From &from) noexcept { |
| 90 | #if __has_builtin(__builtin_bit_cast) |
| 91 | return __builtin_bit_cast(To, from); |
| 92 | #else |
| 93 | To to; |
| 94 | std::memcpy(&to, &from, sizeof(To)); |
| 95 | return to; |
| 96 | #endif |
| 97 | } |
| 98 | |
| 99 | /// Reverses the bytes in the given integer value V. |
| 100 | template <typename T, typename = std::enable_if_t<std::is_integral_v<T>>> |
| 101 | [[nodiscard]] constexpr T byteswap(T V) noexcept { |
| 102 | if constexpr (sizeof(T) == 1) { |
| 103 | return V; |
| 104 | } else if constexpr (sizeof(T) == 2) { |
| 105 | uint16_t UV = V; |
| 106 | #if defined(_MSC_VER) && !defined(_DEBUG) |
| 107 | // The DLL version of the runtime lacks these functions (bug!?), but in a |
| 108 | // release build they're replaced with BSWAP instructions anyway. |
| 109 | return _byteswap_ushort(UV); |
| 110 | #else |
| 111 | uint16_t Hi = UV << 8; |
| 112 | uint16_t Lo = UV >> 8; |
| 113 | return Hi | Lo; |
| 114 | #endif |
| 115 | } else if constexpr (sizeof(T) == 4) { |
| 116 | uint32_t UV = V; |
| 117 | #if __has_builtin(__builtin_bswap32) |
| 118 | return __builtin_bswap32(UV); |
| 119 | #elif defined(_MSC_VER) && !defined(_DEBUG) |
| 120 | return _byteswap_ulong(UV); |
| 121 | #else |
| 122 | uint32_t Byte0 = UV & 0x000000FF; |
| 123 | uint32_t Byte1 = UV & 0x0000FF00; |
| 124 | uint32_t Byte2 = UV & 0x00FF0000; |
| 125 | uint32_t Byte3 = UV & 0xFF000000; |
| 126 | return (Byte0 << 24) | (Byte1 << 8) | (Byte2 >> 8) | (Byte3 >> 24); |
| 127 | #endif |
| 128 | } else if constexpr (sizeof(T) == 8) { |
| 129 | uint64_t UV = V; |
| 130 | #if __has_builtin(__builtin_bswap64) |
| 131 | return __builtin_bswap64(UV); |
| 132 | #elif defined(_MSC_VER) && !defined(_DEBUG) |
| 133 | return _byteswap_uint64(UV); |
| 134 | #else |
| 135 | uint64_t Hi = llvm::byteswap<uint32_t>(UV); |
| 136 | uint32_t Lo = llvm::byteswap<uint32_t>(UV >> 32); |
| 137 | return (Hi << 32) | Lo; |
| 138 | #endif |
| 139 | } else { |
| 140 | static_assert(!sizeof(T *), "Don't know how to handle the given type." ); |
| 141 | return 0; |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | template <typename T, typename = std::enable_if_t<std::is_unsigned_v<T>>> |
| 146 | [[nodiscard]] constexpr inline bool has_single_bit(T Value) noexcept { |
| 147 | return (Value != 0) && ((Value & (Value - 1)) == 0); |
| 148 | } |
| 149 | |
| 150 | namespace detail { |
| 151 | template <typename T, std::size_t SizeOfT> struct TrailingZerosCounter { |
| 152 | static unsigned count(T Val) { |
| 153 | if (!Val) |
| 154 | return std::numeric_limits<T>::digits; |
| 155 | if (Val & 0x1) |
| 156 | return 0; |
| 157 | |
| 158 | // Bisection method. |
| 159 | unsigned ZeroBits = 0; |
| 160 | T Shift = std::numeric_limits<T>::digits >> 1; |
| 161 | T Mask = std::numeric_limits<T>::max() >> Shift; |
| 162 | while (Shift) { |
| 163 | if ((Val & Mask) == 0) { |
| 164 | Val >>= Shift; |
| 165 | ZeroBits |= Shift; |
| 166 | } |
| 167 | Shift >>= 1; |
| 168 | Mask >>= Shift; |
| 169 | } |
| 170 | return ZeroBits; |
| 171 | } |
| 172 | }; |
| 173 | |
| 174 | #if defined(__GNUC__) || defined(_MSC_VER) |
| 175 | template <typename T> struct TrailingZerosCounter<T, 4> { |
| 176 | static unsigned count(T Val) { |
| 177 | if (Val == 0) |
| 178 | return 32; |
| 179 | |
| 180 | #if __has_builtin(__builtin_ctz) || defined(__GNUC__) |
| 181 | return __builtin_ctz(Val); |
| 182 | #elif defined(_MSC_VER) |
| 183 | unsigned long Index; |
| 184 | _BitScanForward(&Index, Val); |
| 185 | return Index; |
| 186 | #endif |
| 187 | } |
| 188 | }; |
| 189 | |
| 190 | #if !defined(_MSC_VER) || defined(_M_X64) |
| 191 | template <typename T> struct TrailingZerosCounter<T, 8> { |
| 192 | static unsigned count(T Val) { |
| 193 | if (Val == 0) |
| 194 | return 64; |
| 195 | |
| 196 | #if __has_builtin(__builtin_ctzll) || defined(__GNUC__) |
| 197 | return __builtin_ctzll(Val); |
| 198 | #elif defined(_MSC_VER) |
| 199 | unsigned long Index; |
| 200 | _BitScanForward64(&Index, Val); |
| 201 | return Index; |
| 202 | #endif |
| 203 | } |
| 204 | }; |
| 205 | #endif |
| 206 | #endif |
| 207 | } // namespace detail |
| 208 | |
| 209 | /// Count number of 0's from the least significant bit to the most |
| 210 | /// stopping at the first 1. |
| 211 | /// |
| 212 | /// Only unsigned integral types are allowed. |
| 213 | /// |
| 214 | /// Returns std::numeric_limits<T>::digits on an input of 0. |
| 215 | template <typename T> [[nodiscard]] int countr_zero(T Val) { |
| 216 | static_assert(std::is_unsigned_v<T>, |
| 217 | "Only unsigned integral types are allowed." ); |
| 218 | return llvm::detail::TrailingZerosCounter<T, sizeof(T)>::count(Val); |
| 219 | } |
| 220 | |
| 221 | namespace detail { |
| 222 | template <typename T, std::size_t SizeOfT> struct LeadingZerosCounter { |
| 223 | static unsigned count(T Val) { |
| 224 | if (!Val) |
| 225 | return std::numeric_limits<T>::digits; |
| 226 | |
| 227 | // Bisection method. |
| 228 | unsigned ZeroBits = 0; |
| 229 | for (T Shift = std::numeric_limits<T>::digits >> 1; Shift; Shift >>= 1) { |
| 230 | T Tmp = Val >> Shift; |
| 231 | if (Tmp) |
| 232 | Val = Tmp; |
| 233 | else |
| 234 | ZeroBits |= Shift; |
| 235 | } |
| 236 | return ZeroBits; |
| 237 | } |
| 238 | }; |
| 239 | |
| 240 | #if defined(__GNUC__) || defined(_MSC_VER) |
| 241 | template <typename T> struct LeadingZerosCounter<T, 4> { |
| 242 | static unsigned count(T Val) { |
| 243 | if (Val == 0) |
| 244 | return 32; |
| 245 | |
| 246 | #if __has_builtin(__builtin_clz) || defined(__GNUC__) |
| 247 | return __builtin_clz(Val); |
| 248 | #elif defined(_MSC_VER) |
| 249 | unsigned long Index; |
| 250 | _BitScanReverse(&Index, Val); |
| 251 | return Index ^ 31; |
| 252 | #endif |
| 253 | } |
| 254 | }; |
| 255 | |
| 256 | #if !defined(_MSC_VER) || defined(_M_X64) |
| 257 | template <typename T> struct LeadingZerosCounter<T, 8> { |
| 258 | static unsigned count(T Val) { |
| 259 | if (Val == 0) |
| 260 | return 64; |
| 261 | |
| 262 | #if __has_builtin(__builtin_clzll) || defined(__GNUC__) |
| 263 | return __builtin_clzll(Val); |
| 264 | #elif defined(_MSC_VER) |
| 265 | unsigned long Index; |
| 266 | _BitScanReverse64(&Index, Val); |
| 267 | return Index ^ 63; |
| 268 | #endif |
| 269 | } |
| 270 | }; |
| 271 | #endif |
| 272 | #endif |
| 273 | } // namespace detail |
| 274 | |
| 275 | /// Count number of 0's from the most significant bit to the least |
| 276 | /// stopping at the first 1. |
| 277 | /// |
| 278 | /// Only unsigned integral types are allowed. |
| 279 | /// |
| 280 | /// Returns std::numeric_limits<T>::digits on an input of 0. |
| 281 | template <typename T> [[nodiscard]] int countl_zero(T Val) { |
| 282 | static_assert(std::is_unsigned_v<T>, |
| 283 | "Only unsigned integral types are allowed." ); |
| 284 | return llvm::detail::LeadingZerosCounter<T, sizeof(T)>::count(Val); |
| 285 | } |
| 286 | |
| 287 | /// Count the number of ones from the most significant bit to the first |
| 288 | /// zero bit. |
| 289 | /// |
| 290 | /// Ex. countl_one(0xFF0FFF00) == 8. |
| 291 | /// Only unsigned integral types are allowed. |
| 292 | /// |
| 293 | /// Returns std::numeric_limits<T>::digits on an input of all ones. |
| 294 | template <typename T> [[nodiscard]] int countl_one(T Value) { |
| 295 | static_assert(std::is_unsigned_v<T>, |
| 296 | "Only unsigned integral types are allowed." ); |
| 297 | return llvm::countl_zero<T>(~Value); |
| 298 | } |
| 299 | |
| 300 | /// Count the number of ones from the least significant bit to the first |
| 301 | /// zero bit. |
| 302 | /// |
| 303 | /// Ex. countr_one(0x00FF00FF) == 8. |
| 304 | /// Only unsigned integral types are allowed. |
| 305 | /// |
| 306 | /// Returns std::numeric_limits<T>::digits on an input of all ones. |
| 307 | template <typename T> [[nodiscard]] int countr_one(T Value) { |
| 308 | static_assert(std::is_unsigned_v<T>, |
| 309 | "Only unsigned integral types are allowed." ); |
| 310 | return llvm::countr_zero<T>(~Value); |
| 311 | } |
| 312 | |
| 313 | /// Returns the number of bits needed to represent Value if Value is nonzero. |
| 314 | /// Returns 0 otherwise. |
| 315 | /// |
| 316 | /// Ex. bit_width(5) == 3. |
| 317 | template <typename T> [[nodiscard]] int bit_width(T Value) { |
| 318 | static_assert(std::is_unsigned_v<T>, |
| 319 | "Only unsigned integral types are allowed." ); |
| 320 | return std::numeric_limits<T>::digits - llvm::countl_zero(Value); |
| 321 | } |
| 322 | |
| 323 | /// Returns the largest integral power of two no greater than Value if Value is |
| 324 | /// nonzero. Returns 0 otherwise. |
| 325 | /// |
| 326 | /// Ex. bit_floor(5) == 4. |
| 327 | template <typename T> [[nodiscard]] T bit_floor(T Value) { |
| 328 | static_assert(std::is_unsigned_v<T>, |
| 329 | "Only unsigned integral types are allowed." ); |
| 330 | if (!Value) |
| 331 | return 0; |
| 332 | return T(1) << (llvm::bit_width(Value) - 1); |
| 333 | } |
| 334 | |
| 335 | /// Returns the smallest integral power of two no smaller than Value if Value is |
| 336 | /// nonzero. Returns 1 otherwise. |
| 337 | /// |
| 338 | /// Ex. bit_ceil(5) == 8. |
| 339 | /// |
| 340 | /// The return value is undefined if the input is larger than the largest power |
| 341 | /// of two representable in T. |
| 342 | template <typename T> [[nodiscard]] T bit_ceil(T Value) { |
| 343 | static_assert(std::is_unsigned_v<T>, |
| 344 | "Only unsigned integral types are allowed." ); |
| 345 | if (Value < 2) |
| 346 | return 1; |
| 347 | return T(1) << llvm::bit_width<T>(Value - 1u); |
| 348 | } |
| 349 | |
| 350 | namespace detail { |
| 351 | template <typename T, std::size_t SizeOfT> struct PopulationCounter { |
| 352 | static int count(T Value) { |
| 353 | // Generic version, forward to 32 bits. |
| 354 | static_assert(SizeOfT <= 4, "Not implemented!" ); |
| 355 | #if defined(__GNUC__) |
| 356 | return (int)__builtin_popcount(Value); |
| 357 | #else |
| 358 | uint32_t v = Value; |
| 359 | v = v - ((v >> 1) & 0x55555555); |
| 360 | v = (v & 0x33333333) + ((v >> 2) & 0x33333333); |
| 361 | return int(((v + (v >> 4) & 0xF0F0F0F) * 0x1010101) >> 24); |
| 362 | #endif |
| 363 | } |
| 364 | }; |
| 365 | |
| 366 | template <typename T> struct PopulationCounter<T, 8> { |
| 367 | static int count(T Value) { |
| 368 | #if defined(__GNUC__) |
| 369 | return (int)__builtin_popcountll(Value); |
| 370 | #else |
| 371 | uint64_t v = Value; |
| 372 | v = v - ((v >> 1) & 0x5555555555555555ULL); |
| 373 | v = (v & 0x3333333333333333ULL) + ((v >> 2) & 0x3333333333333333ULL); |
| 374 | v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0FULL; |
| 375 | return int((uint64_t)(v * 0x0101010101010101ULL) >> 56); |
| 376 | #endif |
| 377 | } |
| 378 | }; |
| 379 | } // namespace detail |
| 380 | |
| 381 | /// Count the number of set bits in a value. |
| 382 | /// Ex. popcount(0xF000F000) = 8 |
| 383 | /// Returns 0 if the word is zero. |
| 384 | template <typename T, typename = std::enable_if_t<std::is_unsigned_v<T>>> |
| 385 | [[nodiscard]] inline int popcount(T Value) noexcept { |
| 386 | return detail::PopulationCounter<T, sizeof(T)>::count(Value); |
| 387 | } |
| 388 | |
| 389 | // Forward-declare rotr so that rotl can use it. |
| 390 | template <typename T, typename = std::enable_if_t<std::is_unsigned_v<T>>> |
| 391 | [[nodiscard]] constexpr T rotr(T V, int R); |
| 392 | |
| 393 | template <typename T, typename = std::enable_if_t<std::is_unsigned_v<T>>> |
| 394 | [[nodiscard]] constexpr T rotl(T V, int R) { |
| 395 | unsigned N = std::numeric_limits<T>::digits; |
| 396 | |
| 397 | R = R % N; |
| 398 | if (!R) |
| 399 | return V; |
| 400 | |
| 401 | if (R < 0) |
| 402 | return llvm::rotr(V, -R); |
| 403 | |
| 404 | return (V << R) | (V >> (N - R)); |
| 405 | } |
| 406 | |
| 407 | template <typename T, typename> [[nodiscard]] constexpr T rotr(T V, int R) { |
| 408 | unsigned N = std::numeric_limits<T>::digits; |
| 409 | |
| 410 | R = R % N; |
| 411 | if (!R) |
| 412 | return V; |
| 413 | |
| 414 | if (R < 0) |
| 415 | return llvm::rotl(V, -R); |
| 416 | |
| 417 | return (V >> R) | (V << (N - R)); |
| 418 | } |
| 419 | |
| 420 | } // namespace llvm |
| 421 | |
| 422 | #endif |
| 423 | |