diff options
author | anastasy888 <anastasy888@yandex-team.ru> | 2022-02-10 16:45:54 +0300 |
---|---|---|
committer | Daniil Cherednik <dcherednik@yandex-team.ru> | 2022-02-10 16:45:54 +0300 |
commit | 49f765d71da452ea93138a25559dfa68dd76c7f3 (patch) | |
tree | 1016041feb637349e401dcc0fa85217dd2c2c639 /contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal | |
parent | 7353a3fdea9c67c256980c00a2b3b67f09b23a27 (diff) | |
download | ydb-49f765d71da452ea93138a25559dfa68dd76c7f3.tar.gz |
Restoring authorship annotation for <anastasy888@yandex-team.ru>. Commit 1 of 2.
Diffstat (limited to 'contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal')
5 files changed, 1473 insertions, 1473 deletions
diff --git a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.cc b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.cc index 5f1b655e7e..563cfb3f6f 100644 --- a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.cc +++ b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.cc @@ -1,349 +1,349 @@ -// Copyright 2018 The Abseil Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// https://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. -// -// This file provides CityHash64() and related functions. -// -// It's probably possible to create even faster hash functions by -// writing a program that systematically explores some of the space of -// possible hash functions, by using SIMD instructions, or by -// compromising on hash quality. - +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file provides CityHash64() and related functions. +// +// It's probably possible to create even faster hash functions by +// writing a program that systematically explores some of the space of +// possible hash functions, by using SIMD instructions, or by +// compromising on hash quality. + #include "y_absl/hash/internal/city.h" - -#include <string.h> // for memcpy and memset -#include <algorithm> - + +#include <string.h> // for memcpy and memset +#include <algorithm> + #include "y_absl/base/config.h" #include "y_absl/base/internal/endian.h" #include "y_absl/base/internal/unaligned_access.h" #include "y_absl/base/optimization.h" - + namespace y_absl { ABSL_NAMESPACE_BEGIN -namespace hash_internal { - -#ifdef ABSL_IS_BIG_ENDIAN +namespace hash_internal { + +#ifdef ABSL_IS_BIG_ENDIAN #define uint32_in_expected_order(x) (y_absl::gbswap_32(x)) #define uint64_in_expected_order(x) (y_absl::gbswap_64(x)) -#else -#define uint32_in_expected_order(x) (x) -#define uint64_in_expected_order(x) (x) -#endif - -static uint64_t Fetch64(const char *p) { - return uint64_in_expected_order(ABSL_INTERNAL_UNALIGNED_LOAD64(p)); -} - -static uint32_t Fetch32(const char *p) { - return uint32_in_expected_order(ABSL_INTERNAL_UNALIGNED_LOAD32(p)); -} - -// Some primes between 2^63 and 2^64 for various uses. -static const uint64_t k0 = 0xc3a5c85c97cb3127ULL; -static const uint64_t k1 = 0xb492b66fbe98f273ULL; -static const uint64_t k2 = 0x9ae16a3b2f90404fULL; - -// Magic numbers for 32-bit hashing. Copied from Murmur3. -static const uint32_t c1 = 0xcc9e2d51; -static const uint32_t c2 = 0x1b873593; - -// A 32-bit to 32-bit integer hash copied from Murmur3. -static uint32_t fmix(uint32_t h) { - h ^= h >> 16; - h *= 0x85ebca6b; - h ^= h >> 13; - h *= 0xc2b2ae35; - h ^= h >> 16; - return h; -} - -static uint32_t Rotate32(uint32_t val, int shift) { - // Avoid shifting by 32: doing so yields an undefined result. - return shift == 0 ? val : ((val >> shift) | (val << (32 - shift))); -} - -#undef PERMUTE3 -#define PERMUTE3(a, b, c) \ - do { \ - std::swap(a, b); \ - std::swap(a, c); \ - } while (0) - -static uint32_t Mur(uint32_t a, uint32_t h) { - // Helper from Murmur3 for combining two 32-bit values. - a *= c1; - a = Rotate32(a, 17); - a *= c2; - h ^= a; - h = Rotate32(h, 19); - return h * 5 + 0xe6546b64; -} - -static uint32_t Hash32Len13to24(const char *s, size_t len) { - uint32_t a = Fetch32(s - 4 + (len >> 1)); - uint32_t b = Fetch32(s + 4); - uint32_t c = Fetch32(s + len - 8); - uint32_t d = Fetch32(s + (len >> 1)); - uint32_t e = Fetch32(s); - uint32_t f = Fetch32(s + len - 4); - uint32_t h = len; - - return fmix(Mur(f, Mur(e, Mur(d, Mur(c, Mur(b, Mur(a, h))))))); -} - -static uint32_t Hash32Len0to4(const char *s, size_t len) { - uint32_t b = 0; - uint32_t c = 9; - for (size_t i = 0; i < len; i++) { - signed char v = s[i]; - b = b * c1 + v; - c ^= b; - } - return fmix(Mur(b, Mur(len, c))); -} - -static uint32_t Hash32Len5to12(const char *s, size_t len) { - uint32_t a = len, b = len * 5, c = 9, d = b; - a += Fetch32(s); - b += Fetch32(s + len - 4); - c += Fetch32(s + ((len >> 1) & 4)); - return fmix(Mur(c, Mur(b, Mur(a, d)))); -} - -uint32_t CityHash32(const char *s, size_t len) { - if (len <= 24) { - return len <= 12 - ? (len <= 4 ? Hash32Len0to4(s, len) : Hash32Len5to12(s, len)) - : Hash32Len13to24(s, len); - } - - // len > 24 - uint32_t h = len, g = c1 * len, f = g; - - uint32_t a0 = Rotate32(Fetch32(s + len - 4) * c1, 17) * c2; - uint32_t a1 = Rotate32(Fetch32(s + len - 8) * c1, 17) * c2; - uint32_t a2 = Rotate32(Fetch32(s + len - 16) * c1, 17) * c2; - uint32_t a3 = Rotate32(Fetch32(s + len - 12) * c1, 17) * c2; - uint32_t a4 = Rotate32(Fetch32(s + len - 20) * c1, 17) * c2; - h ^= a0; - h = Rotate32(h, 19); - h = h * 5 + 0xe6546b64; - h ^= a2; - h = Rotate32(h, 19); - h = h * 5 + 0xe6546b64; - g ^= a1; - g = Rotate32(g, 19); - g = g * 5 + 0xe6546b64; - g ^= a3; - g = Rotate32(g, 19); - g = g * 5 + 0xe6546b64; - f += a4; - f = Rotate32(f, 19); - f = f * 5 + 0xe6546b64; - size_t iters = (len - 1) / 20; - do { - uint32_t b0 = Rotate32(Fetch32(s) * c1, 17) * c2; - uint32_t b1 = Fetch32(s + 4); - uint32_t b2 = Rotate32(Fetch32(s + 8) * c1, 17) * c2; - uint32_t b3 = Rotate32(Fetch32(s + 12) * c1, 17) * c2; - uint32_t b4 = Fetch32(s + 16); - h ^= b0; - h = Rotate32(h, 18); - h = h * 5 + 0xe6546b64; - f += b1; - f = Rotate32(f, 19); - f = f * c1; - g += b2; - g = Rotate32(g, 18); - g = g * 5 + 0xe6546b64; - h ^= b3 + b1; - h = Rotate32(h, 19); - h = h * 5 + 0xe6546b64; - g ^= b4; +#else +#define uint32_in_expected_order(x) (x) +#define uint64_in_expected_order(x) (x) +#endif + +static uint64_t Fetch64(const char *p) { + return uint64_in_expected_order(ABSL_INTERNAL_UNALIGNED_LOAD64(p)); +} + +static uint32_t Fetch32(const char *p) { + return uint32_in_expected_order(ABSL_INTERNAL_UNALIGNED_LOAD32(p)); +} + +// Some primes between 2^63 and 2^64 for various uses. +static const uint64_t k0 = 0xc3a5c85c97cb3127ULL; +static const uint64_t k1 = 0xb492b66fbe98f273ULL; +static const uint64_t k2 = 0x9ae16a3b2f90404fULL; + +// Magic numbers for 32-bit hashing. Copied from Murmur3. +static const uint32_t c1 = 0xcc9e2d51; +static const uint32_t c2 = 0x1b873593; + +// A 32-bit to 32-bit integer hash copied from Murmur3. +static uint32_t fmix(uint32_t h) { + h ^= h >> 16; + h *= 0x85ebca6b; + h ^= h >> 13; + h *= 0xc2b2ae35; + h ^= h >> 16; + return h; +} + +static uint32_t Rotate32(uint32_t val, int shift) { + // Avoid shifting by 32: doing so yields an undefined result. + return shift == 0 ? val : ((val >> shift) | (val << (32 - shift))); +} + +#undef PERMUTE3 +#define PERMUTE3(a, b, c) \ + do { \ + std::swap(a, b); \ + std::swap(a, c); \ + } while (0) + +static uint32_t Mur(uint32_t a, uint32_t h) { + // Helper from Murmur3 for combining two 32-bit values. + a *= c1; + a = Rotate32(a, 17); + a *= c2; + h ^= a; + h = Rotate32(h, 19); + return h * 5 + 0xe6546b64; +} + +static uint32_t Hash32Len13to24(const char *s, size_t len) { + uint32_t a = Fetch32(s - 4 + (len >> 1)); + uint32_t b = Fetch32(s + 4); + uint32_t c = Fetch32(s + len - 8); + uint32_t d = Fetch32(s + (len >> 1)); + uint32_t e = Fetch32(s); + uint32_t f = Fetch32(s + len - 4); + uint32_t h = len; + + return fmix(Mur(f, Mur(e, Mur(d, Mur(c, Mur(b, Mur(a, h))))))); +} + +static uint32_t Hash32Len0to4(const char *s, size_t len) { + uint32_t b = 0; + uint32_t c = 9; + for (size_t i = 0; i < len; i++) { + signed char v = s[i]; + b = b * c1 + v; + c ^= b; + } + return fmix(Mur(b, Mur(len, c))); +} + +static uint32_t Hash32Len5to12(const char *s, size_t len) { + uint32_t a = len, b = len * 5, c = 9, d = b; + a += Fetch32(s); + b += Fetch32(s + len - 4); + c += Fetch32(s + ((len >> 1) & 4)); + return fmix(Mur(c, Mur(b, Mur(a, d)))); +} + +uint32_t CityHash32(const char *s, size_t len) { + if (len <= 24) { + return len <= 12 + ? (len <= 4 ? Hash32Len0to4(s, len) : Hash32Len5to12(s, len)) + : Hash32Len13to24(s, len); + } + + // len > 24 + uint32_t h = len, g = c1 * len, f = g; + + uint32_t a0 = Rotate32(Fetch32(s + len - 4) * c1, 17) * c2; + uint32_t a1 = Rotate32(Fetch32(s + len - 8) * c1, 17) * c2; + uint32_t a2 = Rotate32(Fetch32(s + len - 16) * c1, 17) * c2; + uint32_t a3 = Rotate32(Fetch32(s + len - 12) * c1, 17) * c2; + uint32_t a4 = Rotate32(Fetch32(s + len - 20) * c1, 17) * c2; + h ^= a0; + h = Rotate32(h, 19); + h = h * 5 + 0xe6546b64; + h ^= a2; + h = Rotate32(h, 19); + h = h * 5 + 0xe6546b64; + g ^= a1; + g = Rotate32(g, 19); + g = g * 5 + 0xe6546b64; + g ^= a3; + g = Rotate32(g, 19); + g = g * 5 + 0xe6546b64; + f += a4; + f = Rotate32(f, 19); + f = f * 5 + 0xe6546b64; + size_t iters = (len - 1) / 20; + do { + uint32_t b0 = Rotate32(Fetch32(s) * c1, 17) * c2; + uint32_t b1 = Fetch32(s + 4); + uint32_t b2 = Rotate32(Fetch32(s + 8) * c1, 17) * c2; + uint32_t b3 = Rotate32(Fetch32(s + 12) * c1, 17) * c2; + uint32_t b4 = Fetch32(s + 16); + h ^= b0; + h = Rotate32(h, 18); + h = h * 5 + 0xe6546b64; + f += b1; + f = Rotate32(f, 19); + f = f * c1; + g += b2; + g = Rotate32(g, 18); + g = g * 5 + 0xe6546b64; + h ^= b3 + b1; + h = Rotate32(h, 19); + h = h * 5 + 0xe6546b64; + g ^= b4; g = y_absl::gbswap_32(g) * 5; - h += b4 * 5; + h += b4 * 5; h = y_absl::gbswap_32(h); - f += b0; - PERMUTE3(f, h, g); - s += 20; - } while (--iters != 0); - g = Rotate32(g, 11) * c1; - g = Rotate32(g, 17) * c1; - f = Rotate32(f, 11) * c1; - f = Rotate32(f, 17) * c1; - h = Rotate32(h + g, 19); - h = h * 5 + 0xe6546b64; - h = Rotate32(h, 17) * c1; - h = Rotate32(h + f, 19); - h = h * 5 + 0xe6546b64; - h = Rotate32(h, 17) * c1; - return h; -} - -// Bitwise right rotate. Normally this will compile to a single -// instruction, especially if the shift is a manifest constant. -static uint64_t Rotate(uint64_t val, int shift) { - // Avoid shifting by 64: doing so yields an undefined result. - return shift == 0 ? val : ((val >> shift) | (val << (64 - shift))); -} - -static uint64_t ShiftMix(uint64_t val) { return val ^ (val >> 47); } - -static uint64_t HashLen16(uint64_t u, uint64_t v, uint64_t mul) { - // Murmur-inspired hashing. - uint64_t a = (u ^ v) * mul; - a ^= (a >> 47); - uint64_t b = (v ^ a) * mul; - b ^= (b >> 47); - b *= mul; - return b; -} - + f += b0; + PERMUTE3(f, h, g); + s += 20; + } while (--iters != 0); + g = Rotate32(g, 11) * c1; + g = Rotate32(g, 17) * c1; + f = Rotate32(f, 11) * c1; + f = Rotate32(f, 17) * c1; + h = Rotate32(h + g, 19); + h = h * 5 + 0xe6546b64; + h = Rotate32(h, 17) * c1; + h = Rotate32(h + f, 19); + h = h * 5 + 0xe6546b64; + h = Rotate32(h, 17) * c1; + return h; +} + +// Bitwise right rotate. Normally this will compile to a single +// instruction, especially if the shift is a manifest constant. +static uint64_t Rotate(uint64_t val, int shift) { + // Avoid shifting by 64: doing so yields an undefined result. + return shift == 0 ? val : ((val >> shift) | (val << (64 - shift))); +} + +static uint64_t ShiftMix(uint64_t val) { return val ^ (val >> 47); } + +static uint64_t HashLen16(uint64_t u, uint64_t v, uint64_t mul) { + // Murmur-inspired hashing. + uint64_t a = (u ^ v) * mul; + a ^= (a >> 47); + uint64_t b = (v ^ a) * mul; + b ^= (b >> 47); + b *= mul; + return b; +} + static uint64_t HashLen16(uint64_t u, uint64_t v) { const uint64_t kMul = 0x9ddfea08eb382d69ULL; return HashLen16(u, v, kMul); } -static uint64_t HashLen0to16(const char *s, size_t len) { - if (len >= 8) { - uint64_t mul = k2 + len * 2; - uint64_t a = Fetch64(s) + k2; - uint64_t b = Fetch64(s + len - 8); - uint64_t c = Rotate(b, 37) * mul + a; - uint64_t d = (Rotate(a, 25) + b) * mul; - return HashLen16(c, d, mul); - } - if (len >= 4) { - uint64_t mul = k2 + len * 2; - uint64_t a = Fetch32(s); - return HashLen16(len + (a << 3), Fetch32(s + len - 4), mul); - } - if (len > 0) { - uint8_t a = s[0]; - uint8_t b = s[len >> 1]; - uint8_t c = s[len - 1]; - uint32_t y = static_cast<uint32_t>(a) + (static_cast<uint32_t>(b) << 8); - uint32_t z = len + (static_cast<uint32_t>(c) << 2); - return ShiftMix(y * k2 ^ z * k0) * k2; - } - return k2; -} - -// This probably works well for 16-byte strings as well, but it may be overkill -// in that case. -static uint64_t HashLen17to32(const char *s, size_t len) { - uint64_t mul = k2 + len * 2; - uint64_t a = Fetch64(s) * k1; - uint64_t b = Fetch64(s + 8); - uint64_t c = Fetch64(s + len - 8) * mul; - uint64_t d = Fetch64(s + len - 16) * k2; - return HashLen16(Rotate(a + b, 43) + Rotate(c, 30) + d, - a + Rotate(b + k2, 18) + c, mul); -} - -// Return a 16-byte hash for 48 bytes. Quick and dirty. -// Callers do best to use "random-looking" values for a and b. +static uint64_t HashLen0to16(const char *s, size_t len) { + if (len >= 8) { + uint64_t mul = k2 + len * 2; + uint64_t a = Fetch64(s) + k2; + uint64_t b = Fetch64(s + len - 8); + uint64_t c = Rotate(b, 37) * mul + a; + uint64_t d = (Rotate(a, 25) + b) * mul; + return HashLen16(c, d, mul); + } + if (len >= 4) { + uint64_t mul = k2 + len * 2; + uint64_t a = Fetch32(s); + return HashLen16(len + (a << 3), Fetch32(s + len - 4), mul); + } + if (len > 0) { + uint8_t a = s[0]; + uint8_t b = s[len >> 1]; + uint8_t c = s[len - 1]; + uint32_t y = static_cast<uint32_t>(a) + (static_cast<uint32_t>(b) << 8); + uint32_t z = len + (static_cast<uint32_t>(c) << 2); + return ShiftMix(y * k2 ^ z * k0) * k2; + } + return k2; +} + +// This probably works well for 16-byte strings as well, but it may be overkill +// in that case. +static uint64_t HashLen17to32(const char *s, size_t len) { + uint64_t mul = k2 + len * 2; + uint64_t a = Fetch64(s) * k1; + uint64_t b = Fetch64(s + 8); + uint64_t c = Fetch64(s + len - 8) * mul; + uint64_t d = Fetch64(s + len - 16) * k2; + return HashLen16(Rotate(a + b, 43) + Rotate(c, 30) + d, + a + Rotate(b + k2, 18) + c, mul); +} + +// Return a 16-byte hash for 48 bytes. Quick and dirty. +// Callers do best to use "random-looking" values for a and b. static std::pair<uint64_t, uint64_t> WeakHashLen32WithSeeds( uint64_t w, uint64_t x, uint64_t y, uint64_t z, uint64_t a, uint64_t b) { - a += w; - b = Rotate(b + a + z, 21); - uint64_t c = a; - a += x; - a += y; - b += Rotate(a, 44); - return std::make_pair(a + z, b + c); -} - -// Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty. + a += w; + b = Rotate(b + a + z, 21); + uint64_t c = a; + a += x; + a += y; + b += Rotate(a, 44); + return std::make_pair(a + z, b + c); +} + +// Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty. static std::pair<uint64_t, uint64_t> WeakHashLen32WithSeeds(const char *s, uint64_t a, uint64_t b) { - return WeakHashLen32WithSeeds(Fetch64(s), Fetch64(s + 8), Fetch64(s + 16), - Fetch64(s + 24), a, b); -} - -// Return an 8-byte hash for 33 to 64 bytes. -static uint64_t HashLen33to64(const char *s, size_t len) { - uint64_t mul = k2 + len * 2; - uint64_t a = Fetch64(s) * k2; - uint64_t b = Fetch64(s + 8); - uint64_t c = Fetch64(s + len - 24); - uint64_t d = Fetch64(s + len - 32); - uint64_t e = Fetch64(s + 16) * k2; - uint64_t f = Fetch64(s + 24) * 9; - uint64_t g = Fetch64(s + len - 8); - uint64_t h = Fetch64(s + len - 16) * mul; - uint64_t u = Rotate(a + g, 43) + (Rotate(b, 30) + c) * 9; - uint64_t v = ((a + g) ^ d) + f + 1; + return WeakHashLen32WithSeeds(Fetch64(s), Fetch64(s + 8), Fetch64(s + 16), + Fetch64(s + 24), a, b); +} + +// Return an 8-byte hash for 33 to 64 bytes. +static uint64_t HashLen33to64(const char *s, size_t len) { + uint64_t mul = k2 + len * 2; + uint64_t a = Fetch64(s) * k2; + uint64_t b = Fetch64(s + 8); + uint64_t c = Fetch64(s + len - 24); + uint64_t d = Fetch64(s + len - 32); + uint64_t e = Fetch64(s + 16) * k2; + uint64_t f = Fetch64(s + 24) * 9; + uint64_t g = Fetch64(s + len - 8); + uint64_t h = Fetch64(s + len - 16) * mul; + uint64_t u = Rotate(a + g, 43) + (Rotate(b, 30) + c) * 9; + uint64_t v = ((a + g) ^ d) + f + 1; uint64_t w = y_absl::gbswap_64((u + v) * mul) + h; - uint64_t x = Rotate(e + f, 42) + c; + uint64_t x = Rotate(e + f, 42) + c; uint64_t y = (y_absl::gbswap_64((v + w) * mul) + g) * mul; - uint64_t z = e + f + c; + uint64_t z = e + f + c; a = y_absl::gbswap_64((x + z) * mul + y) + b; - b = ShiftMix((z + a) * mul + d + h) * mul; - return b + x; -} - -uint64_t CityHash64(const char *s, size_t len) { - if (len <= 32) { - if (len <= 16) { - return HashLen0to16(s, len); - } else { - return HashLen17to32(s, len); - } - } else if (len <= 64) { - return HashLen33to64(s, len); - } - - // For strings over 64 bytes we hash the end first, and then as we - // loop we keep 56 bytes of state: v, w, x, y, and z. - uint64_t x = Fetch64(s + len - 40); - uint64_t y = Fetch64(s + len - 16) + Fetch64(s + len - 56); - uint64_t z = HashLen16(Fetch64(s + len - 48) + len, Fetch64(s + len - 24)); + b = ShiftMix((z + a) * mul + d + h) * mul; + return b + x; +} + +uint64_t CityHash64(const char *s, size_t len) { + if (len <= 32) { + if (len <= 16) { + return HashLen0to16(s, len); + } else { + return HashLen17to32(s, len); + } + } else if (len <= 64) { + return HashLen33to64(s, len); + } + + // For strings over 64 bytes we hash the end first, and then as we + // loop we keep 56 bytes of state: v, w, x, y, and z. + uint64_t x = Fetch64(s + len - 40); + uint64_t y = Fetch64(s + len - 16) + Fetch64(s + len - 56); + uint64_t z = HashLen16(Fetch64(s + len - 48) + len, Fetch64(s + len - 24)); std::pair<uint64_t, uint64_t> v = WeakHashLen32WithSeeds(s + len - 64, len, z); std::pair<uint64_t, uint64_t> w = WeakHashLen32WithSeeds(s + len - 32, y + k1, x); - x = x * k1 + Fetch64(s); - - // Decrease len to the nearest multiple of 64, and operate on 64-byte chunks. - len = (len - 1) & ~static_cast<size_t>(63); - do { - x = Rotate(x + y + v.first + Fetch64(s + 8), 37) * k1; - y = Rotate(y + v.second + Fetch64(s + 48), 42) * k1; - x ^= w.second; - y += v.first + Fetch64(s + 40); - z = Rotate(z + w.first, 33) * k1; - v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); - w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + Fetch64(s + 16)); - std::swap(z, x); - s += 64; - len -= 64; - } while (len != 0); - return HashLen16(HashLen16(v.first, w.first) + ShiftMix(y) * k1 + z, - HashLen16(v.second, w.second) + x); -} - -uint64_t CityHash64WithSeed(const char *s, size_t len, uint64_t seed) { - return CityHash64WithSeeds(s, len, k2, seed); -} - -uint64_t CityHash64WithSeeds(const char *s, size_t len, uint64_t seed0, + x = x * k1 + Fetch64(s); + + // Decrease len to the nearest multiple of 64, and operate on 64-byte chunks. + len = (len - 1) & ~static_cast<size_t>(63); + do { + x = Rotate(x + y + v.first + Fetch64(s + 8), 37) * k1; + y = Rotate(y + v.second + Fetch64(s + 48), 42) * k1; + x ^= w.second; + y += v.first + Fetch64(s + 40); + z = Rotate(z + w.first, 33) * k1; + v = WeakHashLen32WithSeeds(s, v.second * k1, x + w.first); + w = WeakHashLen32WithSeeds(s + 32, z + w.second, y + Fetch64(s + 16)); + std::swap(z, x); + s += 64; + len -= 64; + } while (len != 0); + return HashLen16(HashLen16(v.first, w.first) + ShiftMix(y) * k1 + z, + HashLen16(v.second, w.second) + x); +} + +uint64_t CityHash64WithSeed(const char *s, size_t len, uint64_t seed) { + return CityHash64WithSeeds(s, len, k2, seed); +} + +uint64_t CityHash64WithSeeds(const char *s, size_t len, uint64_t seed0, uint64_t seed1) { - return HashLen16(CityHash64(s, len) - seed0, seed1); -} - -} // namespace hash_internal + return HashLen16(CityHash64(s, len) - seed0, seed1); +} + +} // namespace hash_internal ABSL_NAMESPACE_END } // namespace y_absl diff --git a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.h b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.h index d2b32f0068..6a229db1a5 100644 --- a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.h +++ b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/city.h @@ -1,78 +1,78 @@ -// Copyright 2018 The Abseil Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// https://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. -// -// https://code.google.com/p/cityhash/ -// -// This file provides a few functions for hashing strings. All of them are -// high-quality functions in the sense that they pass standard tests such -// as Austin Appleby's SMHasher. They are also fast. -// -// For 64-bit x86 code, on short strings, we don't know of anything faster than -// CityHash64 that is of comparable quality. We believe our nearest competitor -// is Murmur3. For 64-bit x86 code, CityHash64 is an excellent choice for hash -// tables and most other hashing (excluding cryptography). -// -// For 32-bit x86 code, we don't know of anything faster than CityHash32 that -// is of comparable quality. We believe our nearest competitor is Murmur3A. -// (On 64-bit CPUs, it is typically faster to use the other CityHash variants.) -// -// Functions in the CityHash family are not suitable for cryptography. -// -// Please see CityHash's README file for more details on our performance -// measurements and so on. -// -// WARNING: This code has been only lightly tested on big-endian platforms! -// It is known to work well on little-endian platforms that have a small penalty -// for unaligned reads, such as current Intel and AMD moderate-to-high-end CPUs. -// It should work on all 32-bit and 64-bit platforms that allow unaligned reads; -// bug reports are welcome. -// -// By the way, for some hash functions, given strings a and b, the hash -// of a+b is easily derived from the hashes of a and b. This property -// doesn't hold for any hash functions in this file. - -#ifndef ABSL_HASH_INTERNAL_CITY_H_ -#define ABSL_HASH_INTERNAL_CITY_H_ - -#include <stdint.h> -#include <stdlib.h> // for size_t. - -#include <utility> +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// https://code.google.com/p/cityhash/ +// +// This file provides a few functions for hashing strings. All of them are +// high-quality functions in the sense that they pass standard tests such +// as Austin Appleby's SMHasher. They are also fast. +// +// For 64-bit x86 code, on short strings, we don't know of anything faster than +// CityHash64 that is of comparable quality. We believe our nearest competitor +// is Murmur3. For 64-bit x86 code, CityHash64 is an excellent choice for hash +// tables and most other hashing (excluding cryptography). +// +// For 32-bit x86 code, we don't know of anything faster than CityHash32 that +// is of comparable quality. We believe our nearest competitor is Murmur3A. +// (On 64-bit CPUs, it is typically faster to use the other CityHash variants.) +// +// Functions in the CityHash family are not suitable for cryptography. +// +// Please see CityHash's README file for more details on our performance +// measurements and so on. +// +// WARNING: This code has been only lightly tested on big-endian platforms! +// It is known to work well on little-endian platforms that have a small penalty +// for unaligned reads, such as current Intel and AMD moderate-to-high-end CPUs. +// It should work on all 32-bit and 64-bit platforms that allow unaligned reads; +// bug reports are welcome. +// +// By the way, for some hash functions, given strings a and b, the hash +// of a+b is easily derived from the hashes of a and b. This property +// doesn't hold for any hash functions in this file. + +#ifndef ABSL_HASH_INTERNAL_CITY_H_ +#define ABSL_HASH_INTERNAL_CITY_H_ + +#include <stdint.h> +#include <stdlib.h> // for size_t. +#include <utility> + #include "y_absl/base/config.h" namespace y_absl { ABSL_NAMESPACE_BEGIN -namespace hash_internal { - -// Hash function for a byte array. -uint64_t CityHash64(const char *s, size_t len); - -// Hash function for a byte array. For convenience, a 64-bit seed is also -// hashed into the result. -uint64_t CityHash64WithSeed(const char *s, size_t len, uint64_t seed); - -// Hash function for a byte array. For convenience, two seeds are also -// hashed into the result. -uint64_t CityHash64WithSeeds(const char *s, size_t len, uint64_t seed0, +namespace hash_internal { + +// Hash function for a byte array. +uint64_t CityHash64(const char *s, size_t len); + +// Hash function for a byte array. For convenience, a 64-bit seed is also +// hashed into the result. +uint64_t CityHash64WithSeed(const char *s, size_t len, uint64_t seed); + +// Hash function for a byte array. For convenience, two seeds are also +// hashed into the result. +uint64_t CityHash64WithSeeds(const char *s, size_t len, uint64_t seed0, uint64_t seed1); - -// Hash function for a byte array. Most useful in 32-bit binaries. -uint32_t CityHash32(const char *s, size_t len); - -} // namespace hash_internal + +// Hash function for a byte array. Most useful in 32-bit binaries. +uint32_t CityHash32(const char *s, size_t len); + +} // namespace hash_internal ABSL_NAMESPACE_END } // namespace y_absl - -#endif // ABSL_HASH_INTERNAL_CITY_H_ + +#endif // ABSL_HASH_INTERNAL_CITY_H_ diff --git a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.cc b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.cc index fe075de43a..a8a6252323 100644 --- a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.cc +++ b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.cc @@ -1,51 +1,51 @@ -// Copyright 2018 The Abseil Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// https://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include "y_absl/hash/internal/hash.h" - + namespace y_absl { ABSL_NAMESPACE_BEGIN -namespace hash_internal { - +namespace hash_internal { + uint64_t MixingHashState::CombineLargeContiguousImpl32( uint64_t state, const unsigned char* first, size_t len) { - while (len >= PiecewiseChunkSize()) { + while (len >= PiecewiseChunkSize()) { state = Mix(state, hash_internal::CityHash32(reinterpret_cast<const char*>(first), PiecewiseChunkSize())); - len -= PiecewiseChunkSize(); - first += PiecewiseChunkSize(); - } - // Handle the remainder. - return CombineContiguousImpl(state, first, len, - std::integral_constant<int, 4>{}); -} - + len -= PiecewiseChunkSize(); + first += PiecewiseChunkSize(); + } + // Handle the remainder. + return CombineContiguousImpl(state, first, len, + std::integral_constant<int, 4>{}); +} + uint64_t MixingHashState::CombineLargeContiguousImpl64( uint64_t state, const unsigned char* first, size_t len) { - while (len >= PiecewiseChunkSize()) { + while (len >= PiecewiseChunkSize()) { state = Mix(state, Hash64(first, PiecewiseChunkSize())); - len -= PiecewiseChunkSize(); - first += PiecewiseChunkSize(); - } - // Handle the remainder. - return CombineContiguousImpl(state, first, len, - std::integral_constant<int, 8>{}); -} - + len -= PiecewiseChunkSize(); + first += PiecewiseChunkSize(); + } + // Handle the remainder. + return CombineContiguousImpl(state, first, len, + std::integral_constant<int, 8>{}); +} + ABSL_CONST_INIT const void* const MixingHashState::kSeed = &kSeed; - + // The salt array used by LowLevelHash. This array is NOT the mechanism used to // make y_absl::Hash non-deterministic between program invocations. See `Seed()` // for that mechanism. @@ -64,6 +64,6 @@ uint64_t MixingHashState::LowLevelHashImpl(const unsigned char* data, return LowLevelHash(data, len, Seed(), kHashSalt); } -} // namespace hash_internal +} // namespace hash_internal ABSL_NAMESPACE_END } // namespace y_absl diff --git a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.h b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.h index fcbe43accd..d44d650c21 100644 --- a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.h +++ b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/hash.h @@ -1,44 +1,44 @@ -// Copyright 2018 The Abseil Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// https://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. -// -// ----------------------------------------------------------------------------- -// File: hash.h -// ----------------------------------------------------------------------------- -// -#ifndef ABSL_HASH_INTERNAL_HASH_H_ -#define ABSL_HASH_INTERNAL_HASH_H_ - -#include <algorithm> -#include <array> +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: hash.h +// ----------------------------------------------------------------------------- +// +#ifndef ABSL_HASH_INTERNAL_HASH_H_ +#define ABSL_HASH_INTERNAL_HASH_H_ + +#include <algorithm> +#include <array> #include <bitset> -#include <cmath> -#include <cstring> -#include <deque> -#include <forward_list> -#include <functional> -#include <iterator> -#include <limits> -#include <list> -#include <map> -#include <memory> -#include <set> +#include <cmath> +#include <cstring> +#include <deque> +#include <forward_list> +#include <functional> +#include <iterator> +#include <limits> +#include <list> +#include <map> +#include <memory> +#include <set> #include <util/generic/string.h> -#include <tuple> -#include <type_traits> -#include <utility> -#include <vector> - +#include <tuple> +#include <type_traits> +#include <utility> +#include <vector> + #include "y_absl/base/config.h" #include "y_absl/base/internal/unaligned_access.h" #include "y_absl/base/port.h" @@ -51,15 +51,15 @@ #include "y_absl/types/optional.h" #include "y_absl/types/variant.h" #include "y_absl/utility/utility.h" - + namespace y_absl { ABSL_NAMESPACE_BEGIN -namespace hash_internal { - -// Internal detail: Large buffers are hashed in smaller chunks. This function -// returns the size of these chunks. +namespace hash_internal { + +// Internal detail: Large buffers are hashed in smaller chunks. This function +// returns the size of these chunks. constexpr size_t PiecewiseChunkSize() { return 1024; } - + // PiecewiseCombiner // // PiecewiseCombiner is an internal-only helper class for hashing a piecewise @@ -115,393 +115,393 @@ class PiecewiseCombiner { size_t position_; }; -// HashStateBase -// -// A hash state object represents an intermediate state in the computation -// of an unspecified hash algorithm. `HashStateBase` provides a CRTP style -// base class for hash state implementations. Developers adding type support +// HashStateBase +// +// A hash state object represents an intermediate state in the computation +// of an unspecified hash algorithm. `HashStateBase` provides a CRTP style +// base class for hash state implementations. Developers adding type support // for `y_absl::Hash` should not rely on any parts of the state object other than -// the following member functions: -// -// * HashStateBase::combine() -// * HashStateBase::combine_contiguous() -// -// A derived hash state class of type `H` must provide a static member function -// with a signature similar to the following: -// -// `static H combine_contiguous(H state, const unsigned char*, size_t)`. -// -// `HashStateBase` will provide a complete implementation for a hash state -// object in terms of this method. -// -// Example: -// -// // Use CRTP to define your derived class. -// struct MyHashState : HashStateBase<MyHashState> { -// static H combine_contiguous(H state, const unsigned char*, size_t); -// using MyHashState::HashStateBase::combine; -// using MyHashState::HashStateBase::combine_contiguous; -// }; -template <typename H> -class HashStateBase { - public: - // HashStateBase::combine() - // - // Combines an arbitrary number of values into a hash state, returning the - // updated state. - // - // Each of the value types `T` must be separately hashable by the Abseil - // hashing framework. - // - // NOTE: - // - // state = H::combine(std::move(state), value1, value2, value3); - // - // is guaranteed to produce the same hash expansion as: - // - // state = H::combine(std::move(state), value1); - // state = H::combine(std::move(state), value2); - // state = H::combine(std::move(state), value3); - template <typename T, typename... Ts> - static H combine(H state, const T& value, const Ts&... values); - static H combine(H state) { return state; } - - // HashStateBase::combine_contiguous() - // - // Combines a contiguous array of `size` elements into a hash state, returning - // the updated state. - // - // NOTE: - // - // state = H::combine_contiguous(std::move(state), data, size); - // - // is NOT guaranteed to produce the same hash expansion as a for-loop (it may - // perform internal optimizations). If you need this guarantee, use the - // for-loop instead. - template <typename T> - static H combine_contiguous(H state, const T* data, size_t size); - +// the following member functions: +// +// * HashStateBase::combine() +// * HashStateBase::combine_contiguous() +// +// A derived hash state class of type `H` must provide a static member function +// with a signature similar to the following: +// +// `static H combine_contiguous(H state, const unsigned char*, size_t)`. +// +// `HashStateBase` will provide a complete implementation for a hash state +// object in terms of this method. +// +// Example: +// +// // Use CRTP to define your derived class. +// struct MyHashState : HashStateBase<MyHashState> { +// static H combine_contiguous(H state, const unsigned char*, size_t); +// using MyHashState::HashStateBase::combine; +// using MyHashState::HashStateBase::combine_contiguous; +// }; +template <typename H> +class HashStateBase { + public: + // HashStateBase::combine() + // + // Combines an arbitrary number of values into a hash state, returning the + // updated state. + // + // Each of the value types `T` must be separately hashable by the Abseil + // hashing framework. + // + // NOTE: + // + // state = H::combine(std::move(state), value1, value2, value3); + // + // is guaranteed to produce the same hash expansion as: + // + // state = H::combine(std::move(state), value1); + // state = H::combine(std::move(state), value2); + // state = H::combine(std::move(state), value3); + template <typename T, typename... Ts> + static H combine(H state, const T& value, const Ts&... values); + static H combine(H state) { return state; } + + // HashStateBase::combine_contiguous() + // + // Combines a contiguous array of `size` elements into a hash state, returning + // the updated state. + // + // NOTE: + // + // state = H::combine_contiguous(std::move(state), data, size); + // + // is NOT guaranteed to produce the same hash expansion as a for-loop (it may + // perform internal optimizations). If you need this guarantee, use the + // for-loop instead. + template <typename T> + static H combine_contiguous(H state, const T* data, size_t size); + using AbslInternalPiecewiseCombiner = PiecewiseCombiner; -}; - -// is_uniquely_represented -// -// `is_uniquely_represented<T>` is a trait class that indicates whether `T` -// is uniquely represented. -// -// A type is "uniquely represented" if two equal values of that type are -// guaranteed to have the same bytes in their underlying storage. In other -// words, if `a == b`, then `memcmp(&a, &b, sizeof(T))` is guaranteed to be -// zero. This property cannot be detected automatically, so this trait is false -// by default, but can be specialized by types that wish to assert that they are -// uniquely represented. This makes them eligible for certain optimizations. -// -// If you have any doubt whatsoever, do not specialize this template. -// The default is completely safe, and merely disables some optimizations -// that will not matter for most types. Specializing this template, -// on the other hand, can be very hazardous. -// -// To be uniquely represented, a type must not have multiple ways of -// representing the same value; for example, float and double are not -// uniquely represented, because they have distinct representations for -// +0 and -0. Furthermore, the type's byte representation must consist -// solely of user-controlled data, with no padding bits and no compiler- -// controlled data such as vptrs or sanitizer metadata. This is usually -// very difficult to guarantee, because in most cases the compiler can -// insert data and padding bits at its own discretion. -// -// If you specialize this template for a type `T`, you must do so in the file -// that defines that type (or in this file). If you define that specialization -// anywhere else, `is_uniquely_represented<T>` could have different meanings -// in different places. -// -// The Enable parameter is meaningless; it is provided as a convenience, -// to support certain SFINAE techniques when defining specializations. -template <typename T, typename Enable = void> -struct is_uniquely_represented : std::false_type {}; - -// is_uniquely_represented<unsigned char> -// -// unsigned char is a synonym for "byte", so it is guaranteed to be -// uniquely represented. -template <> -struct is_uniquely_represented<unsigned char> : std::true_type {}; - -// is_uniquely_represented for non-standard integral types -// -// Integral types other than bool should be uniquely represented on any -// platform that this will plausibly be ported to. -template <typename Integral> -struct is_uniquely_represented< - Integral, typename std::enable_if<std::is_integral<Integral>::value>::type> - : std::true_type {}; - -// is_uniquely_represented<bool> -// -// -template <> -struct is_uniquely_represented<bool> : std::false_type {}; - -// hash_bytes() -// -// Convenience function that combines `hash_state` with the byte representation -// of `value`. -template <typename H, typename T> -H hash_bytes(H hash_state, const T& value) { - const unsigned char* start = reinterpret_cast<const unsigned char*>(&value); - return H::combine_contiguous(std::move(hash_state), start, sizeof(value)); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Basic Types -// ----------------------------------------------------------------------------- - -// Note: Default `AbslHashValue` implementations live in `hash_internal`. This -// allows us to block lexical scope lookup when doing an unqualified call to -// `AbslHashValue` below. User-defined implementations of `AbslHashValue` can -// only be found via ADL. - -// AbslHashValue() for hashing bool values -// -// We use SFINAE to ensure that this overload only accepts bool, not types that -// are convertible to bool. -template <typename H, typename B> -typename std::enable_if<std::is_same<B, bool>::value, H>::type AbslHashValue( - H hash_state, B value) { - return H::combine(std::move(hash_state), - static_cast<unsigned char>(value ? 1 : 0)); -} - -// AbslHashValue() for hashing enum values -template <typename H, typename Enum> -typename std::enable_if<std::is_enum<Enum>::value, H>::type AbslHashValue( - H hash_state, Enum e) { - // In practice, we could almost certainly just invoke hash_bytes directly, - // but it's possible that a sanitizer might one day want to - // store data in the unused bits of an enum. To avoid that risk, we - // convert to the underlying type before hashing. Hopefully this will get - // optimized away; if not, we can reopen discussion with c-toolchain-team. - return H::combine(std::move(hash_state), - static_cast<typename std::underlying_type<Enum>::type>(e)); -} -// AbslHashValue() for hashing floating-point values -template <typename H, typename Float> -typename std::enable_if<std::is_same<Float, float>::value || - std::is_same<Float, double>::value, - H>::type -AbslHashValue(H hash_state, Float value) { - return hash_internal::hash_bytes(std::move(hash_state), - value == 0 ? 0 : value); -} - -// Long double has the property that it might have extra unused bytes in it. -// For example, in x86 sizeof(long double)==16 but it only really uses 80-bits -// of it. This means we can't use hash_bytes on a long double and have to -// convert it to something else first. -template <typename H, typename LongDouble> -typename std::enable_if<std::is_same<LongDouble, long double>::value, H>::type -AbslHashValue(H hash_state, LongDouble value) { - const int category = std::fpclassify(value); - switch (category) { - case FP_INFINITE: - // Add the sign bit to differentiate between +Inf and -Inf - hash_state = H::combine(std::move(hash_state), std::signbit(value)); - break; - - case FP_NAN: - case FP_ZERO: - default: - // Category is enough for these. - break; - - case FP_NORMAL: - case FP_SUBNORMAL: - // We can't convert `value` directly to double because this would have - // undefined behavior if the value is out of range. - // std::frexp gives us a value in the range (-1, -.5] or [.5, 1) that is - // guaranteed to be in range for `double`. The truncation is - // implementation defined, but that works as long as it is deterministic. - int exp; - auto mantissa = static_cast<double>(std::frexp(value, &exp)); - hash_state = H::combine(std::move(hash_state), mantissa, exp); - } - - return H::combine(std::move(hash_state), category); -} - -// AbslHashValue() for hashing pointers -template <typename H, typename T> -H AbslHashValue(H hash_state, T* ptr) { - auto v = reinterpret_cast<uintptr_t>(ptr); - // Due to alignment, pointers tend to have low bits as zero, and the next few - // bits follow a pattern since they are also multiples of some base value. - // Mixing the pointer twice helps prevent stuck low bits for certain alignment - // values. - return H::combine(std::move(hash_state), v, v); -} - -// AbslHashValue() for hashing nullptr_t -template <typename H> -H AbslHashValue(H hash_state, std::nullptr_t) { - return H::combine(std::move(hash_state), static_cast<void*>(nullptr)); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Composite Types -// ----------------------------------------------------------------------------- - -// is_hashable() -// +}; + +// is_uniquely_represented +// +// `is_uniquely_represented<T>` is a trait class that indicates whether `T` +// is uniquely represented. +// +// A type is "uniquely represented" if two equal values of that type are +// guaranteed to have the same bytes in their underlying storage. In other +// words, if `a == b`, then `memcmp(&a, &b, sizeof(T))` is guaranteed to be +// zero. This property cannot be detected automatically, so this trait is false +// by default, but can be specialized by types that wish to assert that they are +// uniquely represented. This makes them eligible for certain optimizations. +// +// If you have any doubt whatsoever, do not specialize this template. +// The default is completely safe, and merely disables some optimizations +// that will not matter for most types. Specializing this template, +// on the other hand, can be very hazardous. +// +// To be uniquely represented, a type must not have multiple ways of +// representing the same value; for example, float and double are not +// uniquely represented, because they have distinct representations for +// +0 and -0. Furthermore, the type's byte representation must consist +// solely of user-controlled data, with no padding bits and no compiler- +// controlled data such as vptrs or sanitizer metadata. This is usually +// very difficult to guarantee, because in most cases the compiler can +// insert data and padding bits at its own discretion. +// +// If you specialize this template for a type `T`, you must do so in the file +// that defines that type (or in this file). If you define that specialization +// anywhere else, `is_uniquely_represented<T>` could have different meanings +// in different places. +// +// The Enable parameter is meaningless; it is provided as a convenience, +// to support certain SFINAE techniques when defining specializations. +template <typename T, typename Enable = void> +struct is_uniquely_represented : std::false_type {}; + +// is_uniquely_represented<unsigned char> +// +// unsigned char is a synonym for "byte", so it is guaranteed to be +// uniquely represented. +template <> +struct is_uniquely_represented<unsigned char> : std::true_type {}; + +// is_uniquely_represented for non-standard integral types +// +// Integral types other than bool should be uniquely represented on any +// platform that this will plausibly be ported to. +template <typename Integral> +struct is_uniquely_represented< + Integral, typename std::enable_if<std::is_integral<Integral>::value>::type> + : std::true_type {}; + +// is_uniquely_represented<bool> +// +// +template <> +struct is_uniquely_represented<bool> : std::false_type {}; + +// hash_bytes() +// +// Convenience function that combines `hash_state` with the byte representation +// of `value`. +template <typename H, typename T> +H hash_bytes(H hash_state, const T& value) { + const unsigned char* start = reinterpret_cast<const unsigned char*>(&value); + return H::combine_contiguous(std::move(hash_state), start, sizeof(value)); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Basic Types +// ----------------------------------------------------------------------------- + +// Note: Default `AbslHashValue` implementations live in `hash_internal`. This +// allows us to block lexical scope lookup when doing an unqualified call to +// `AbslHashValue` below. User-defined implementations of `AbslHashValue` can +// only be found via ADL. + +// AbslHashValue() for hashing bool values +// +// We use SFINAE to ensure that this overload only accepts bool, not types that +// are convertible to bool. +template <typename H, typename B> +typename std::enable_if<std::is_same<B, bool>::value, H>::type AbslHashValue( + H hash_state, B value) { + return H::combine(std::move(hash_state), + static_cast<unsigned char>(value ? 1 : 0)); +} + +// AbslHashValue() for hashing enum values +template <typename H, typename Enum> +typename std::enable_if<std::is_enum<Enum>::value, H>::type AbslHashValue( + H hash_state, Enum e) { + // In practice, we could almost certainly just invoke hash_bytes directly, + // but it's possible that a sanitizer might one day want to + // store data in the unused bits of an enum. To avoid that risk, we + // convert to the underlying type before hashing. Hopefully this will get + // optimized away; if not, we can reopen discussion with c-toolchain-team. + return H::combine(std::move(hash_state), + static_cast<typename std::underlying_type<Enum>::type>(e)); +} +// AbslHashValue() for hashing floating-point values +template <typename H, typename Float> +typename std::enable_if<std::is_same<Float, float>::value || + std::is_same<Float, double>::value, + H>::type +AbslHashValue(H hash_state, Float value) { + return hash_internal::hash_bytes(std::move(hash_state), + value == 0 ? 0 : value); +} + +// Long double has the property that it might have extra unused bytes in it. +// For example, in x86 sizeof(long double)==16 but it only really uses 80-bits +// of it. This means we can't use hash_bytes on a long double and have to +// convert it to something else first. +template <typename H, typename LongDouble> +typename std::enable_if<std::is_same<LongDouble, long double>::value, H>::type +AbslHashValue(H hash_state, LongDouble value) { + const int category = std::fpclassify(value); + switch (category) { + case FP_INFINITE: + // Add the sign bit to differentiate between +Inf and -Inf + hash_state = H::combine(std::move(hash_state), std::signbit(value)); + break; + + case FP_NAN: + case FP_ZERO: + default: + // Category is enough for these. + break; + + case FP_NORMAL: + case FP_SUBNORMAL: + // We can't convert `value` directly to double because this would have + // undefined behavior if the value is out of range. + // std::frexp gives us a value in the range (-1, -.5] or [.5, 1) that is + // guaranteed to be in range for `double`. The truncation is + // implementation defined, but that works as long as it is deterministic. + int exp; + auto mantissa = static_cast<double>(std::frexp(value, &exp)); + hash_state = H::combine(std::move(hash_state), mantissa, exp); + } + + return H::combine(std::move(hash_state), category); +} + +// AbslHashValue() for hashing pointers +template <typename H, typename T> +H AbslHashValue(H hash_state, T* ptr) { + auto v = reinterpret_cast<uintptr_t>(ptr); + // Due to alignment, pointers tend to have low bits as zero, and the next few + // bits follow a pattern since they are also multiples of some base value. + // Mixing the pointer twice helps prevent stuck low bits for certain alignment + // values. + return H::combine(std::move(hash_state), v, v); +} + +// AbslHashValue() for hashing nullptr_t +template <typename H> +H AbslHashValue(H hash_state, std::nullptr_t) { + return H::combine(std::move(hash_state), static_cast<void*>(nullptr)); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Composite Types +// ----------------------------------------------------------------------------- + +// is_hashable() +// // Trait class which returns true if T is hashable by the y_absl::Hash framework. -// Used for the AbslHashValue implementations for composite types below. -template <typename T> -struct is_hashable; - -// AbslHashValue() for hashing pairs -template <typename H, typename T1, typename T2> -typename std::enable_if<is_hashable<T1>::value && is_hashable<T2>::value, - H>::type -AbslHashValue(H hash_state, const std::pair<T1, T2>& p) { - return H::combine(std::move(hash_state), p.first, p.second); -} - -// hash_tuple() -// -// Helper function for hashing a tuple. The third argument should -// be an index_sequence running from 0 to tuple_size<Tuple> - 1. -template <typename H, typename Tuple, size_t... Is> +// Used for the AbslHashValue implementations for composite types below. +template <typename T> +struct is_hashable; + +// AbslHashValue() for hashing pairs +template <typename H, typename T1, typename T2> +typename std::enable_if<is_hashable<T1>::value && is_hashable<T2>::value, + H>::type +AbslHashValue(H hash_state, const std::pair<T1, T2>& p) { + return H::combine(std::move(hash_state), p.first, p.second); +} + +// hash_tuple() +// +// Helper function for hashing a tuple. The third argument should +// be an index_sequence running from 0 to tuple_size<Tuple> - 1. +template <typename H, typename Tuple, size_t... Is> H hash_tuple(H hash_state, const Tuple& t, y_absl::index_sequence<Is...>) { - return H::combine(std::move(hash_state), std::get<Is>(t)...); -} - -// AbslHashValue for hashing tuples -template <typename H, typename... Ts> -#if defined(_MSC_VER) -// This SFINAE gets MSVC confused under some conditions. Let's just disable it -// for now. -H + return H::combine(std::move(hash_state), std::get<Is>(t)...); +} + +// AbslHashValue for hashing tuples +template <typename H, typename... Ts> +#if defined(_MSC_VER) +// This SFINAE gets MSVC confused under some conditions. Let's just disable it +// for now. +H #else // _MSC_VER typename std::enable_if<y_absl::conjunction<is_hashable<Ts>...>::value, H>::type -#endif // _MSC_VER -AbslHashValue(H hash_state, const std::tuple<Ts...>& t) { - return hash_internal::hash_tuple(std::move(hash_state), t, +#endif // _MSC_VER +AbslHashValue(H hash_state, const std::tuple<Ts...>& t) { + return hash_internal::hash_tuple(std::move(hash_state), t, y_absl::make_index_sequence<sizeof...(Ts)>()); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Pointers -// ----------------------------------------------------------------------------- - -// AbslHashValue for hashing unique_ptr -template <typename H, typename T, typename D> -H AbslHashValue(H hash_state, const std::unique_ptr<T, D>& ptr) { - return H::combine(std::move(hash_state), ptr.get()); -} - -// AbslHashValue for hashing shared_ptr -template <typename H, typename T> -H AbslHashValue(H hash_state, const std::shared_ptr<T>& ptr) { - return H::combine(std::move(hash_state), ptr.get()); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for String-Like Types -// ----------------------------------------------------------------------------- - -// AbslHashValue for hashing strings -// -// All the string-like types supported here provide the same hash expansion for -// the same character sequence. These types are: -// +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Pointers +// ----------------------------------------------------------------------------- + +// AbslHashValue for hashing unique_ptr +template <typename H, typename T, typename D> +H AbslHashValue(H hash_state, const std::unique_ptr<T, D>& ptr) { + return H::combine(std::move(hash_state), ptr.get()); +} + +// AbslHashValue for hashing shared_ptr +template <typename H, typename T> +H AbslHashValue(H hash_state, const std::shared_ptr<T>& ptr) { + return H::combine(std::move(hash_state), ptr.get()); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for String-Like Types +// ----------------------------------------------------------------------------- + +// AbslHashValue for hashing strings +// +// All the string-like types supported here provide the same hash expansion for +// the same character sequence. These types are: +// // - `y_absl::Cord` // - `TString` (and std::basic_string<char, std::char_traits<char>, A> for -// any allocator A) +// any allocator A) // - `y_absl::string_view` and `std::string_view` -// -// For simplicity, we currently support only `char` strings. This support may -// be broadened, if necessary, but with some caution - this overload would -// misbehave in cases where the traits' `eq()` member isn't equivalent to `==` -// on the underlying character type. -template <typename H> +// +// For simplicity, we currently support only `char` strings. This support may +// be broadened, if necessary, but with some caution - this overload would +// misbehave in cases where the traits' `eq()` member isn't equivalent to `==` +// on the underlying character type. +template <typename H> H AbslHashValue(H hash_state, y_absl::string_view str) { - return H::combine( - H::combine_contiguous(std::move(hash_state), str.data(), str.size()), - str.size()); -} - -// Support std::wstring, std::u16string and std::u32string. -template <typename Char, typename Alloc, typename H, + return H::combine( + H::combine_contiguous(std::move(hash_state), str.data(), str.size()), + str.size()); +} + +// Support std::wstring, std::u16string and std::u32string. +template <typename Char, typename Alloc, typename H, typename = y_absl::enable_if_t<std::is_same<Char, wchar_t>::value || - std::is_same<Char, char16_t>::value || - std::is_same<Char, char32_t>::value>> -H AbslHashValue( - H hash_state, - const std::basic_string<Char, std::char_traits<Char>, Alloc>& str) { - return H::combine( - H::combine_contiguous(std::move(hash_state), str.data(), str.size()), - str.size()); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Sequence Containers -// ----------------------------------------------------------------------------- - -// AbslHashValue for hashing std::array -template <typename H, typename T, size_t N> -typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( - H hash_state, const std::array<T, N>& array) { - return H::combine_contiguous(std::move(hash_state), array.data(), - array.size()); -} - -// AbslHashValue for hashing std::deque -template <typename H, typename T, typename Allocator> -typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( - H hash_state, const std::deque<T, Allocator>& deque) { - // TODO(gromer): investigate a more efficient implementation taking - // advantage of the chunk structure. - for (const auto& t : deque) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), deque.size()); -} - -// AbslHashValue for hashing std::forward_list -template <typename H, typename T, typename Allocator> -typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( - H hash_state, const std::forward_list<T, Allocator>& list) { - size_t size = 0; - for (const T& t : list) { - hash_state = H::combine(std::move(hash_state), t); - ++size; - } - return H::combine(std::move(hash_state), size); -} - -// AbslHashValue for hashing std::list -template <typename H, typename T, typename Allocator> -typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( - H hash_state, const std::list<T, Allocator>& list) { - for (const auto& t : list) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), list.size()); -} - -// AbslHashValue for hashing std::vector -// + std::is_same<Char, char16_t>::value || + std::is_same<Char, char32_t>::value>> +H AbslHashValue( + H hash_state, + const std::basic_string<Char, std::char_traits<Char>, Alloc>& str) { + return H::combine( + H::combine_contiguous(std::move(hash_state), str.data(), str.size()), + str.size()); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Sequence Containers +// ----------------------------------------------------------------------------- + +// AbslHashValue for hashing std::array +template <typename H, typename T, size_t N> +typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( + H hash_state, const std::array<T, N>& array) { + return H::combine_contiguous(std::move(hash_state), array.data(), + array.size()); +} + +// AbslHashValue for hashing std::deque +template <typename H, typename T, typename Allocator> +typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( + H hash_state, const std::deque<T, Allocator>& deque) { + // TODO(gromer): investigate a more efficient implementation taking + // advantage of the chunk structure. + for (const auto& t : deque) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), deque.size()); +} + +// AbslHashValue for hashing std::forward_list +template <typename H, typename T, typename Allocator> +typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( + H hash_state, const std::forward_list<T, Allocator>& list) { + size_t size = 0; + for (const T& t : list) { + hash_state = H::combine(std::move(hash_state), t); + ++size; + } + return H::combine(std::move(hash_state), size); +} + +// AbslHashValue for hashing std::list +template <typename H, typename T, typename Allocator> +typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( + H hash_state, const std::list<T, Allocator>& list) { + for (const auto& t : list) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), list.size()); +} + +// AbslHashValue for hashing std::vector +// // Do not use this for vector<bool> on platforms that have a working // implementation of std::hash. It does not have a .data(), and a fallback for // std::hash<> is most likely faster. -template <typename H, typename T, typename Allocator> -typename std::enable_if<is_hashable<T>::value && !std::is_same<T, bool>::value, - H>::type -AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) { - return H::combine(H::combine_contiguous(std::move(hash_state), vector.data(), - vector.size()), - vector.size()); -} - +template <typename H, typename T, typename Allocator> +typename std::enable_if<is_hashable<T>::value && !std::is_same<T, bool>::value, + H>::type +AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) { + return H::combine(H::combine_contiguous(std::move(hash_state), vector.data(), + vector.size()), + vector.size()); +} + #if defined(ABSL_IS_BIG_ENDIAN) && \ (defined(__GLIBCXX__) || defined(__GLIBCPP__)) // AbslHashValue for hashing std::vector<bool> @@ -523,59 +523,59 @@ AbslHashValue(H hash_state, const std::vector<T, Allocator>& vector) { } #endif -// ----------------------------------------------------------------------------- -// AbslHashValue for Ordered Associative Containers -// ----------------------------------------------------------------------------- - -// AbslHashValue for hashing std::map -template <typename H, typename Key, typename T, typename Compare, - typename Allocator> -typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value, - H>::type -AbslHashValue(H hash_state, const std::map<Key, T, Compare, Allocator>& map) { - for (const auto& t : map) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), map.size()); -} - -// AbslHashValue for hashing std::multimap -template <typename H, typename Key, typename T, typename Compare, - typename Allocator> -typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value, - H>::type -AbslHashValue(H hash_state, - const std::multimap<Key, T, Compare, Allocator>& map) { - for (const auto& t : map) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), map.size()); -} - -// AbslHashValue for hashing std::set -template <typename H, typename Key, typename Compare, typename Allocator> -typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue( - H hash_state, const std::set<Key, Compare, Allocator>& set) { - for (const auto& t : set) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), set.size()); -} - -// AbslHashValue for hashing std::multiset -template <typename H, typename Key, typename Compare, typename Allocator> -typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue( - H hash_state, const std::multiset<Key, Compare, Allocator>& set) { - for (const auto& t : set) { - hash_state = H::combine(std::move(hash_state), t); - } - return H::combine(std::move(hash_state), set.size()); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Wrapper Types -// ----------------------------------------------------------------------------- - +// ----------------------------------------------------------------------------- +// AbslHashValue for Ordered Associative Containers +// ----------------------------------------------------------------------------- + +// AbslHashValue for hashing std::map +template <typename H, typename Key, typename T, typename Compare, + typename Allocator> +typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value, + H>::type +AbslHashValue(H hash_state, const std::map<Key, T, Compare, Allocator>& map) { + for (const auto& t : map) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), map.size()); +} + +// AbslHashValue for hashing std::multimap +template <typename H, typename Key, typename T, typename Compare, + typename Allocator> +typename std::enable_if<is_hashable<Key>::value && is_hashable<T>::value, + H>::type +AbslHashValue(H hash_state, + const std::multimap<Key, T, Compare, Allocator>& map) { + for (const auto& t : map) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), map.size()); +} + +// AbslHashValue for hashing std::set +template <typename H, typename Key, typename Compare, typename Allocator> +typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue( + H hash_state, const std::set<Key, Compare, Allocator>& set) { + for (const auto& t : set) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), set.size()); +} + +// AbslHashValue for hashing std::multiset +template <typename H, typename Key, typename Compare, typename Allocator> +typename std::enable_if<is_hashable<Key>::value, H>::type AbslHashValue( + H hash_state, const std::multiset<Key, Compare, Allocator>& set) { + for (const auto& t : set) { + hash_state = H::combine(std::move(hash_state), t); + } + return H::combine(std::move(hash_state), set.size()); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Wrapper Types +// ----------------------------------------------------------------------------- + // AbslHashValue for hashing std::reference_wrapper template <typename H, typename T> typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( @@ -584,42 +584,42 @@ typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( } // AbslHashValue for hashing y_absl::optional -template <typename H, typename T> -typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( +template <typename H, typename T> +typename std::enable_if<is_hashable<T>::value, H>::type AbslHashValue( H hash_state, const y_absl::optional<T>& opt) { - if (opt) hash_state = H::combine(std::move(hash_state), *opt); - return H::combine(std::move(hash_state), opt.has_value()); -} - -// VariantVisitor -template <typename H> -struct VariantVisitor { - H&& hash_state; - template <typename T> - H operator()(const T& t) const { - return H::combine(std::move(hash_state), t); - } -}; - + if (opt) hash_state = H::combine(std::move(hash_state), *opt); + return H::combine(std::move(hash_state), opt.has_value()); +} + +// VariantVisitor +template <typename H> +struct VariantVisitor { + H&& hash_state; + template <typename T> + H operator()(const T& t) const { + return H::combine(std::move(hash_state), t); + } +}; + // AbslHashValue for hashing y_absl::variant -template <typename H, typename... T> -typename std::enable_if<conjunction<is_hashable<T>...>::value, H>::type +template <typename H, typename... T> +typename std::enable_if<conjunction<is_hashable<T>...>::value, H>::type AbslHashValue(H hash_state, const y_absl::variant<T...>& v) { - if (!v.valueless_by_exception()) { + if (!v.valueless_by_exception()) { hash_state = y_absl::visit(VariantVisitor<H>{std::move(hash_state)}, v); - } - return H::combine(std::move(hash_state), v.index()); -} - -// ----------------------------------------------------------------------------- -// AbslHashValue for Other Types -// ----------------------------------------------------------------------------- - + } + return H::combine(std::move(hash_state), v.index()); +} + +// ----------------------------------------------------------------------------- +// AbslHashValue for Other Types +// ----------------------------------------------------------------------------- + // AbslHashValue for hashing std::bitset is not defined on Little Endian // platforms, for the same reason as for vector<bool> (see std::vector above): // It does not expose the raw bytes, and a fallback to std::hash<> is most // likely faster. - + #if defined(ABSL_IS_BIG_ENDIAN) && \ (defined(__GLIBCXX__) || defined(__GLIBCPP__)) // AbslHashValue for hashing std::bitset @@ -638,219 +638,219 @@ H AbslHashValue(H hash_state, const std::bitset<N>& set) { } #endif -// ----------------------------------------------------------------------------- - -// hash_range_or_bytes() -// -// Mixes all values in the range [data, data+size) into the hash state. -// This overload accepts only uniquely-represented types, and hashes them by -// hashing the entire range of bytes. -template <typename H, typename T> -typename std::enable_if<is_uniquely_represented<T>::value, H>::type -hash_range_or_bytes(H hash_state, const T* data, size_t size) { - const auto* bytes = reinterpret_cast<const unsigned char*>(data); - return H::combine_contiguous(std::move(hash_state), bytes, sizeof(T) * size); -} - -// hash_range_or_bytes() -template <typename H, typename T> -typename std::enable_if<!is_uniquely_represented<T>::value, H>::type -hash_range_or_bytes(H hash_state, const T* data, size_t size) { - for (const auto end = data + size; data < end; ++data) { - hash_state = H::combine(std::move(hash_state), *data); - } - return hash_state; -} - -#if defined(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE) && \ - ABSL_META_INTERNAL_STD_HASH_SFINAE_FRIENDLY_ -#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 1 -#else -#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 0 -#endif - -// HashSelect -// -// Type trait to select the appropriate hash implementation to use. -// HashSelect::type<T> will give the proper hash implementation, to be invoked -// as: -// HashSelect::type<T>::Invoke(state, value) -// Also, HashSelect::type<T>::value is a boolean equal to `true` if there is a -// valid `Invoke` function. Types that are not hashable will have a ::value of -// `false`. -struct HashSelect { - private: - struct State : HashStateBase<State> { - static State combine_contiguous(State hash_state, const unsigned char*, - size_t); - using State::HashStateBase::combine_contiguous; - }; - - struct UniquelyRepresentedProbe { - template <typename H, typename T> - static auto Invoke(H state, const T& value) +// ----------------------------------------------------------------------------- + +// hash_range_or_bytes() +// +// Mixes all values in the range [data, data+size) into the hash state. +// This overload accepts only uniquely-represented types, and hashes them by +// hashing the entire range of bytes. +template <typename H, typename T> +typename std::enable_if<is_uniquely_represented<T>::value, H>::type +hash_range_or_bytes(H hash_state, const T* data, size_t size) { + const auto* bytes = reinterpret_cast<const unsigned char*>(data); + return H::combine_contiguous(std::move(hash_state), bytes, sizeof(T) * size); +} + +// hash_range_or_bytes() +template <typename H, typename T> +typename std::enable_if<!is_uniquely_represented<T>::value, H>::type +hash_range_or_bytes(H hash_state, const T* data, size_t size) { + for (const auto end = data + size; data < end; ++data) { + hash_state = H::combine(std::move(hash_state), *data); + } + return hash_state; +} + +#if defined(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE) && \ + ABSL_META_INTERNAL_STD_HASH_SFINAE_FRIENDLY_ +#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 1 +#else +#define ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ 0 +#endif + +// HashSelect +// +// Type trait to select the appropriate hash implementation to use. +// HashSelect::type<T> will give the proper hash implementation, to be invoked +// as: +// HashSelect::type<T>::Invoke(state, value) +// Also, HashSelect::type<T>::value is a boolean equal to `true` if there is a +// valid `Invoke` function. Types that are not hashable will have a ::value of +// `false`. +struct HashSelect { + private: + struct State : HashStateBase<State> { + static State combine_contiguous(State hash_state, const unsigned char*, + size_t); + using State::HashStateBase::combine_contiguous; + }; + + struct UniquelyRepresentedProbe { + template <typename H, typename T> + static auto Invoke(H state, const T& value) -> y_absl::enable_if_t<is_uniquely_represented<T>::value, H> { - return hash_internal::hash_bytes(std::move(state), value); - } - }; - - struct HashValueProbe { - template <typename H, typename T> + return hash_internal::hash_bytes(std::move(state), value); + } + }; + + struct HashValueProbe { + template <typename H, typename T> static auto Invoke(H state, const T& value) -> y_absl::enable_if_t< - std::is_same<H, - decltype(AbslHashValue(std::move(state), value))>::value, - H> { - return AbslHashValue(std::move(state), value); - } - }; - - struct LegacyHashProbe { -#if ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ - template <typename H, typename T> + std::is_same<H, + decltype(AbslHashValue(std::move(state), value))>::value, + H> { + return AbslHashValue(std::move(state), value); + } + }; + + struct LegacyHashProbe { +#if ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ + template <typename H, typename T> static auto Invoke(H state, const T& value) -> y_absl::enable_if_t< - std::is_convertible< - decltype(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>()(value)), - size_t>::value, - H> { - return hash_internal::hash_bytes( - std::move(state), - ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>{}(value)); - } -#endif // ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ - }; - - struct StdHashProbe { - template <typename H, typename T> - static auto Invoke(H state, const T& value) + std::is_convertible< + decltype(ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>()(value)), + size_t>::value, + H> { + return hash_internal::hash_bytes( + std::move(state), + ABSL_INTERNAL_LEGACY_HASH_NAMESPACE::hash<T>{}(value)); + } +#endif // ABSL_HASH_INTERNAL_SUPPORT_LEGACY_HASH_ + }; + + struct StdHashProbe { + template <typename H, typename T> + static auto Invoke(H state, const T& value) -> y_absl::enable_if_t<type_traits_internal::IsHashable<T>::value, H> { - return hash_internal::hash_bytes(std::move(state), std::hash<T>{}(value)); - } - }; - - template <typename Hash, typename T> - struct Probe : Hash { - private: - template <typename H, typename = decltype(H::Invoke( - std::declval<State>(), std::declval<const T&>()))> - static std::true_type Test(int); - template <typename U> - static std::false_type Test(char); - - public: - static constexpr bool value = decltype(Test<Hash>(0))::value; - }; - - public: - // Probe each implementation in order. - // disjunction provides short circuiting wrt instantiation. - template <typename T> + return hash_internal::hash_bytes(std::move(state), std::hash<T>{}(value)); + } + }; + + template <typename Hash, typename T> + struct Probe : Hash { + private: + template <typename H, typename = decltype(H::Invoke( + std::declval<State>(), std::declval<const T&>()))> + static std::true_type Test(int); + template <typename U> + static std::false_type Test(char); + + public: + static constexpr bool value = decltype(Test<Hash>(0))::value; + }; + + public: + // Probe each implementation in order. + // disjunction provides short circuiting wrt instantiation. + template <typename T> using Apply = y_absl::disjunction< // - Probe<UniquelyRepresentedProbe, T>, // - Probe<HashValueProbe, T>, // - Probe<LegacyHashProbe, T>, // - Probe<StdHashProbe, T>, // - std::false_type>; -}; - -template <typename T> -struct is_hashable - : std::integral_constant<bool, HashSelect::template Apply<T>::value> {}; - + Probe<UniquelyRepresentedProbe, T>, // + Probe<HashValueProbe, T>, // + Probe<LegacyHashProbe, T>, // + Probe<StdHashProbe, T>, // + std::false_type>; +}; + +template <typename T> +struct is_hashable + : std::integral_constant<bool, HashSelect::template Apply<T>::value> {}; + // MixingHashState class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> { // y_absl::uint128 is not an alias or a thin wrapper around the intrinsic. - // We use the intrinsic when available to improve performance. -#ifdef ABSL_HAVE_INTRINSIC_INT128 - using uint128 = __uint128_t; -#else // ABSL_HAVE_INTRINSIC_INT128 + // We use the intrinsic when available to improve performance. +#ifdef ABSL_HAVE_INTRINSIC_INT128 + using uint128 = __uint128_t; +#else // ABSL_HAVE_INTRINSIC_INT128 using uint128 = y_absl::uint128; -#endif // ABSL_HAVE_INTRINSIC_INT128 - - static constexpr uint64_t kMul = - sizeof(size_t) == 4 ? uint64_t{0xcc9e2d51} - : uint64_t{0x9ddfea08eb382d69}; - - template <typename T> - using IntegralFastPath = - conjunction<std::is_integral<T>, is_uniquely_represented<T>>; - - public: - // Move only +#endif // ABSL_HAVE_INTRINSIC_INT128 + + static constexpr uint64_t kMul = + sizeof(size_t) == 4 ? uint64_t{0xcc9e2d51} + : uint64_t{0x9ddfea08eb382d69}; + + template <typename T> + using IntegralFastPath = + conjunction<std::is_integral<T>, is_uniquely_represented<T>>; + + public: + // Move only MixingHashState(MixingHashState&&) = default; MixingHashState& operator=(MixingHashState&&) = default; - + // MixingHashState::combine_contiguous() - // - // Fundamental base case for hash recursion: mixes the given range of bytes - // into the hash state. + // + // Fundamental base case for hash recursion: mixes the given range of bytes + // into the hash state. static MixingHashState combine_contiguous(MixingHashState hash_state, const unsigned char* first, size_t size) { return MixingHashState( - CombineContiguousImpl(hash_state.state_, first, size, - std::integral_constant<int, sizeof(size_t)>{})); - } + CombineContiguousImpl(hash_state.state_, first, size, + std::integral_constant<int, sizeof(size_t)>{})); + } using MixingHashState::HashStateBase::combine_contiguous; - + // MixingHashState::hash() - // - // For performance reasons in non-opt mode, we specialize this for - // integral types. - // Otherwise we would be instantiating and calling dozens of functions for - // something that is just one multiplication and a couple xor's. - // The result should be the same as running the whole algorithm, but faster. + // + // For performance reasons in non-opt mode, we specialize this for + // integral types. + // Otherwise we would be instantiating and calling dozens of functions for + // something that is just one multiplication and a couple xor's. + // The result should be the same as running the whole algorithm, but faster. template <typename T, y_absl::enable_if_t<IntegralFastPath<T>::value, int> = 0> - static size_t hash(T value) { - return static_cast<size_t>(Mix(Seed(), static_cast<uint64_t>(value))); - } - + static size_t hash(T value) { + return static_cast<size_t>(Mix(Seed(), static_cast<uint64_t>(value))); + } + // Overload of MixingHashState::hash() template <typename T, y_absl::enable_if_t<!IntegralFastPath<T>::value, int> = 0> - static size_t hash(const T& value) { + static size_t hash(const T& value) { return static_cast<size_t>(combine(MixingHashState{}, value).state_); - } - - private: - // Invoked only once for a given argument; that plus the fact that this is - // move-only ensures that there is only one non-moved-from object. + } + + private: + // Invoked only once for a given argument; that plus the fact that this is + // move-only ensures that there is only one non-moved-from object. MixingHashState() : state_(Seed()) {} - - // Workaround for MSVC bug. - // We make the type copyable to fix the calling convention, even though we - // never actually copy it. Keep it private to not affect the public API of the - // type. + + // Workaround for MSVC bug. + // We make the type copyable to fix the calling convention, even though we + // never actually copy it. Keep it private to not affect the public API of the + // type. MixingHashState(const MixingHashState&) = default; - + explicit MixingHashState(uint64_t state) : state_(state) {} - - // Implementation of the base case for combine_contiguous where we actually - // mix the bytes into the state. - // Dispatch to different implementations of the combine_contiguous depending - // on the value of `sizeof(size_t)`. - static uint64_t CombineContiguousImpl(uint64_t state, - const unsigned char* first, size_t len, - std::integral_constant<int, 4> + + // Implementation of the base case for combine_contiguous where we actually + // mix the bytes into the state. + // Dispatch to different implementations of the combine_contiguous depending + // on the value of `sizeof(size_t)`. + static uint64_t CombineContiguousImpl(uint64_t state, + const unsigned char* first, size_t len, + std::integral_constant<int, 4> + /* sizeof_size_t */); + static uint64_t CombineContiguousImpl(uint64_t state, + const unsigned char* first, size_t len, + std::integral_constant<int, 8> /* sizeof_size_t */); - static uint64_t CombineContiguousImpl(uint64_t state, - const unsigned char* first, size_t len, - std::integral_constant<int, 8> - /* sizeof_size_t */); - - // Slow dispatch path for calls to CombineContiguousImpl with a size argument - // larger than PiecewiseChunkSize(). Has the same effect as calling - // CombineContiguousImpl() repeatedly with the chunk stride size. - static uint64_t CombineLargeContiguousImpl32(uint64_t state, - const unsigned char* first, - size_t len); - static uint64_t CombineLargeContiguousImpl64(uint64_t state, - const unsigned char* first, - size_t len); - - // Reads 9 to 16 bytes from p. + + // Slow dispatch path for calls to CombineContiguousImpl with a size argument + // larger than PiecewiseChunkSize(). Has the same effect as calling + // CombineContiguousImpl() repeatedly with the chunk stride size. + static uint64_t CombineLargeContiguousImpl32(uint64_t state, + const unsigned char* first, + size_t len); + static uint64_t CombineLargeContiguousImpl64(uint64_t state, + const unsigned char* first, + size_t len); + + // Reads 9 to 16 bytes from p. // The least significant 8 bytes are in .first, the rest (zero padded) bytes // are in .second. - static std::pair<uint64_t, uint64_t> Read9To16(const unsigned char* p, - size_t len) { + static std::pair<uint64_t, uint64_t> Read9To16(const unsigned char* p, + size_t len) { uint64_t low_mem = y_absl::base_internal::UnalignedLoad64(p); uint64_t high_mem = y_absl::base_internal::UnalignedLoad64(p + len - 8); #ifdef ABSL_IS_LITTLE_ENDIAN @@ -861,10 +861,10 @@ class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> { uint64_t least_significant = high_mem; #endif return {least_significant, most_significant >> (128 - len * 8)}; - } - - // Reads 4 to 8 bytes from p. Zero pads to fill uint64_t. - static uint64_t Read4To8(const unsigned char* p, size_t len) { + } + + // Reads 4 to 8 bytes from p. Zero pads to fill uint64_t. + static uint64_t Read4To8(const unsigned char* p, size_t len) { uint32_t low_mem = y_absl::base_internal::UnalignedLoad32(p); uint32_t high_mem = y_absl::base_internal::UnalignedLoad32(p + len - 4); #ifdef ABSL_IS_LITTLE_ENDIAN @@ -876,10 +876,10 @@ class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> { #endif return (static_cast<uint64_t>(most_significant) << (len - 4) * 8) | least_significant; - } - - // Reads 1 to 3 bytes from p. Zero pads to fill uint32_t. - static uint32_t Read1To3(const unsigned char* p, size_t len) { + } + + // Reads 1 to 3 bytes from p. Zero pads to fill uint32_t. + static uint32_t Read1To3(const unsigned char* p, size_t len) { unsigned char mem0 = p[0]; unsigned char mem1 = p[len / 2]; unsigned char mem2 = p[len - 1]; @@ -895,27 +895,27 @@ class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> { return static_cast<uint32_t>(significant0 | // (significant1 << (len / 2 * 8)) | // (significant2 << ((len - 1) * 8))); - } - - ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Mix(uint64_t state, uint64_t v) { + } + + ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Mix(uint64_t state, uint64_t v) { #if defined(__aarch64__) // On AArch64, calculating a 128-bit product is inefficient, because it // requires a sequence of two instructions to calculate the upper and lower // halves of the result. using MultType = uint64_t; #else - using MultType = + using MultType = y_absl::conditional_t<sizeof(size_t) == 4, uint64_t, uint128>; #endif - // We do the addition in 64-bit space to make sure the 128-bit - // multiplication is fast. If we were to do it as MultType the compiler has - // to assume that the high word is non-zero and needs to perform 2 - // multiplications instead of one. - MultType m = state + v; - m *= kMul; - return static_cast<uint64_t>(m ^ (m >> (sizeof(m) * 8 / 2))); - } - + // We do the addition in 64-bit space to make sure the 128-bit + // multiplication is fast. If we were to do it as MultType the compiler has + // to assume that the high word is non-zero and needs to perform 2 + // multiplications instead of one. + MultType m = state + v; + m *= kMul; + return static_cast<uint64_t>(m ^ (m >> (sizeof(m) * 8 / 2))); + } + // An extern to avoid bloat on a direct call to LowLevelHash() with fixed // values for both the seed and salt parameters. static uint64_t LowLevelHashImpl(const unsigned char* data, size_t len); @@ -929,137 +929,137 @@ class ABSL_DLL MixingHashState : public HashStateBase<MixingHashState> { #endif } - // Seed() - // - // A non-deterministic seed. - // - // The current purpose of this seed is to generate non-deterministic results - // and prevent having users depend on the particular hash values. - // It is not meant as a security feature right now, but it leaves the door - // open to upgrade it to a true per-process random seed. A true random seed - // costs more and we don't need to pay for that right now. - // - // On platforms with ASLR, we take advantage of it to make a per-process - // random value. - // See https://en.wikipedia.org/wiki/Address_space_layout_randomization - // - // On other platforms this is still going to be non-deterministic but most - // probably per-build and not per-process. - ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Seed() { + // Seed() + // + // A non-deterministic seed. + // + // The current purpose of this seed is to generate non-deterministic results + // and prevent having users depend on the particular hash values. + // It is not meant as a security feature right now, but it leaves the door + // open to upgrade it to a true per-process random seed. A true random seed + // costs more and we don't need to pay for that right now. + // + // On platforms with ASLR, we take advantage of it to make a per-process + // random value. + // See https://en.wikipedia.org/wiki/Address_space_layout_randomization + // + // On other platforms this is still going to be non-deterministic but most + // probably per-build and not per-process. + ABSL_ATTRIBUTE_ALWAYS_INLINE static uint64_t Seed() { #if (!defined(__clang__) || __clang_major__ > 11) && \ !defined(__apple_build_version__) return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(&kSeed)); #else // Workaround the absence of // https://github.com/llvm/llvm-project/commit/bc15bf66dcca76cc06fe71fca35b74dc4d521021. - return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(kSeed)); + return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(kSeed)); #endif - } - static const void* const kSeed; - - uint64_t state_; -}; - + } + static const void* const kSeed; + + uint64_t state_; +}; + // MixingHashState::CombineContiguousImpl() inline uint64_t MixingHashState::CombineContiguousImpl( - uint64_t state, const unsigned char* first, size_t len, - std::integral_constant<int, 4> /* sizeof_size_t */) { - // For large values we use CityHash, for small ones we just use a - // multiplicative hash. - uint64_t v; - if (len > 8) { - if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) { - return CombineLargeContiguousImpl32(state, first, len); - } + uint64_t state, const unsigned char* first, size_t len, + std::integral_constant<int, 4> /* sizeof_size_t */) { + // For large values we use CityHash, for small ones we just use a + // multiplicative hash. + uint64_t v; + if (len > 8) { + if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) { + return CombineLargeContiguousImpl32(state, first, len); + } v = hash_internal::CityHash32(reinterpret_cast<const char*>(first), len); - } else if (len >= 4) { - v = Read4To8(first, len); - } else if (len > 0) { - v = Read1To3(first, len); - } else { - // Empty ranges have no effect. - return state; - } - return Mix(state, v); -} - + } else if (len >= 4) { + v = Read4To8(first, len); + } else if (len > 0) { + v = Read1To3(first, len); + } else { + // Empty ranges have no effect. + return state; + } + return Mix(state, v); +} + // Overload of MixingHashState::CombineContiguousImpl() inline uint64_t MixingHashState::CombineContiguousImpl( - uint64_t state, const unsigned char* first, size_t len, - std::integral_constant<int, 8> /* sizeof_size_t */) { + uint64_t state, const unsigned char* first, size_t len, + std::integral_constant<int, 8> /* sizeof_size_t */) { // For large values we use LowLevelHash or CityHash depending on the platform, // for small ones we just use a multiplicative hash. - uint64_t v; - if (len > 16) { - if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) { - return CombineLargeContiguousImpl64(state, first, len); - } + uint64_t v; + if (len > 16) { + if (ABSL_PREDICT_FALSE(len > PiecewiseChunkSize())) { + return CombineLargeContiguousImpl64(state, first, len); + } v = Hash64(first, len); - } else if (len > 8) { - auto p = Read9To16(first, len); - state = Mix(state, p.first); - v = p.second; - } else if (len >= 4) { - v = Read4To8(first, len); - } else if (len > 0) { - v = Read1To3(first, len); - } else { - // Empty ranges have no effect. - return state; - } - return Mix(state, v); -} - -struct AggregateBarrier {}; - -// HashImpl - -// Add a private base class to make sure this type is not an aggregate. -// Aggregates can be aggregate initialized even if the default constructor is -// deleted. -struct PoisonedHash : private AggregateBarrier { - PoisonedHash() = delete; - PoisonedHash(const PoisonedHash&) = delete; - PoisonedHash& operator=(const PoisonedHash&) = delete; -}; - -template <typename T> -struct HashImpl { + } else if (len > 8) { + auto p = Read9To16(first, len); + state = Mix(state, p.first); + v = p.second; + } else if (len >= 4) { + v = Read4To8(first, len); + } else if (len > 0) { + v = Read1To3(first, len); + } else { + // Empty ranges have no effect. + return state; + } + return Mix(state, v); +} + +struct AggregateBarrier {}; + +// HashImpl + +// Add a private base class to make sure this type is not an aggregate. +// Aggregates can be aggregate initialized even if the default constructor is +// deleted. +struct PoisonedHash : private AggregateBarrier { + PoisonedHash() = delete; + PoisonedHash(const PoisonedHash&) = delete; + PoisonedHash& operator=(const PoisonedHash&) = delete; +}; + +template <typename T> +struct HashImpl { size_t operator()(const T& value) const { return MixingHashState::hash(value); } -}; - -template <typename T> -struct Hash +}; + +template <typename T> +struct Hash : y_absl::conditional_t<is_hashable<T>::value, HashImpl<T>, PoisonedHash> {}; - -template <typename H> -template <typename T, typename... Ts> -H HashStateBase<H>::combine(H state, const T& value, const Ts&... values) { - return H::combine(hash_internal::HashSelect::template Apply<T>::Invoke( - std::move(state), value), - values...); -} - -// HashStateBase::combine_contiguous() -template <typename H> -template <typename T> -H HashStateBase<H>::combine_contiguous(H state, const T* data, size_t size) { - return hash_internal::hash_range_or_bytes(std::move(state), data, size); -} - -// HashStateBase::PiecewiseCombiner::add_buffer() -template <typename H> -H PiecewiseCombiner::add_buffer(H state, const unsigned char* data, - size_t size) { - if (position_ + size < PiecewiseChunkSize()) { - // This partial chunk does not fill our existing buffer - memcpy(buf_ + position_, data, size); - position_ += size; + +template <typename H> +template <typename T, typename... Ts> +H HashStateBase<H>::combine(H state, const T& value, const Ts&... values) { + return H::combine(hash_internal::HashSelect::template Apply<T>::Invoke( + std::move(state), value), + values...); +} + +// HashStateBase::combine_contiguous() +template <typename H> +template <typename T> +H HashStateBase<H>::combine_contiguous(H state, const T* data, size_t size) { + return hash_internal::hash_range_or_bytes(std::move(state), data, size); +} + +// HashStateBase::PiecewiseCombiner::add_buffer() +template <typename H> +H PiecewiseCombiner::add_buffer(H state, const unsigned char* data, + size_t size) { + if (position_ + size < PiecewiseChunkSize()) { + // This partial chunk does not fill our existing buffer + memcpy(buf_ + position_, data, size); + position_ += size; return state; - } - + } + // If the buffer is partially filled we need to complete the buffer // and hash it. if (position_ != 0) { @@ -1069,28 +1069,28 @@ H PiecewiseCombiner::add_buffer(H state, const unsigned char* data, data += bytes_needed; size -= bytes_needed; } - - // Hash whatever chunks we can without copying - while (size >= PiecewiseChunkSize()) { - state = H::combine_contiguous(std::move(state), data, PiecewiseChunkSize()); - data += PiecewiseChunkSize(); - size -= PiecewiseChunkSize(); - } - // Fill the buffer with the remainder - memcpy(buf_, data, size); - position_ = size; + + // Hash whatever chunks we can without copying + while (size >= PiecewiseChunkSize()) { + state = H::combine_contiguous(std::move(state), data, PiecewiseChunkSize()); + data += PiecewiseChunkSize(); + size -= PiecewiseChunkSize(); + } + // Fill the buffer with the remainder + memcpy(buf_, data, size); + position_ = size; return state; -} - -// HashStateBase::PiecewiseCombiner::finalize() -template <typename H> -H PiecewiseCombiner::finalize(H state) { - // Hash the remainder left in the buffer, which may be empty - return H::combine_contiguous(std::move(state), buf_, position_); -} - -} // namespace hash_internal +} + +// HashStateBase::PiecewiseCombiner::finalize() +template <typename H> +H PiecewiseCombiner::finalize(H state) { + // Hash the remainder left in the buffer, which may be empty + return H::combine_contiguous(std::move(state), buf_, position_); +} + +} // namespace hash_internal ABSL_NAMESPACE_END } // namespace y_absl - -#endif // ABSL_HASH_INTERNAL_HASH_H_ + +#endif // ABSL_HASH_INTERNAL_HASH_H_ diff --git a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/spy_hash_state.h b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/spy_hash_state.h index 520cbf904f..a3e09aba2e 100644 --- a/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/spy_hash_state.h +++ b/contrib/restricted/abseil-cpp-tstring/y_absl/hash/internal/spy_hash_state.h @@ -1,231 +1,231 @@ -// Copyright 2018 The Abseil Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// https://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -#ifndef ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ -#define ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ - -#include <ostream> +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ +#define ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ + +#include <ostream> #include <util/generic/string.h> -#include <vector> - +#include <vector> + #include "y_absl/hash/hash.h" #include "y_absl/strings/match.h" #include "y_absl/strings/str_format.h" #include "y_absl/strings/str_join.h" - + namespace y_absl { ABSL_NAMESPACE_BEGIN -namespace hash_internal { - -// SpyHashState is an implementation of the HashState API that simply -// accumulates all input bytes in an internal buffer. This makes it useful -// for testing AbslHashValue overloads (so long as they are templated on the -// HashState parameter), since it can report the exact hash representation -// that the AbslHashValue overload produces. -// -// Sample usage: -// EXPECT_EQ(SpyHashState::combine(SpyHashState(), foo), -// SpyHashState::combine(SpyHashState(), bar)); -template <typename T> -class SpyHashStateImpl : public HashStateBase<SpyHashStateImpl<T>> { - public: +namespace hash_internal { + +// SpyHashState is an implementation of the HashState API that simply +// accumulates all input bytes in an internal buffer. This makes it useful +// for testing AbslHashValue overloads (so long as they are templated on the +// HashState parameter), since it can report the exact hash representation +// that the AbslHashValue overload produces. +// +// Sample usage: +// EXPECT_EQ(SpyHashState::combine(SpyHashState(), foo), +// SpyHashState::combine(SpyHashState(), bar)); +template <typename T> +class SpyHashStateImpl : public HashStateBase<SpyHashStateImpl<T>> { + public: SpyHashStateImpl() : error_(std::make_shared<y_absl::optional<TString>>()) { - static_assert(std::is_void<T>::value, ""); - } - - // Move-only - SpyHashStateImpl(const SpyHashStateImpl&) = delete; - SpyHashStateImpl& operator=(const SpyHashStateImpl&) = delete; - - SpyHashStateImpl(SpyHashStateImpl&& other) noexcept { - *this = std::move(other); - } - - SpyHashStateImpl& operator=(SpyHashStateImpl&& other) noexcept { - hash_representation_ = std::move(other.hash_representation_); - error_ = other.error_; - moved_from_ = other.moved_from_; - other.moved_from_ = true; - return *this; - } - - template <typename U> - SpyHashStateImpl(SpyHashStateImpl<U>&& other) { // NOLINT - hash_representation_ = std::move(other.hash_representation_); - error_ = other.error_; - moved_from_ = other.moved_from_; - other.moved_from_ = true; - } - - template <typename A, typename... Args> - static SpyHashStateImpl combine(SpyHashStateImpl s, const A& a, - const Args&... args) { - // Pass an instance of SpyHashStateImpl<A> when trying to combine `A`. This - // allows us to test that the user only uses this instance for combine calls - // and does not call AbslHashValue directly. - // See AbslHashValue implementation at the bottom. - s = SpyHashStateImpl<A>::HashStateBase::combine(std::move(s), a); - return SpyHashStateImpl::combine(std::move(s), args...); - } - static SpyHashStateImpl combine(SpyHashStateImpl s) { - if (direct_absl_hash_value_error_) { - *s.error_ = "AbslHashValue should not be invoked directly."; - } else if (s.moved_from_) { - *s.error_ = "Used moved-from instance of the hash state object."; - } - return s; - } - - static void SetDirectAbslHashValueError() { - direct_absl_hash_value_error_ = true; - } - - // Two SpyHashStateImpl objects are equal if they hold equal hash - // representations. - friend bool operator==(const SpyHashStateImpl& lhs, - const SpyHashStateImpl& rhs) { - return lhs.hash_representation_ == rhs.hash_representation_; - } - - friend bool operator!=(const SpyHashStateImpl& lhs, - const SpyHashStateImpl& rhs) { - return !(lhs == rhs); - } - - enum class CompareResult { - kEqual, - kASuffixB, - kBSuffixA, - kUnequal, - }; - - static CompareResult Compare(const SpyHashStateImpl& a, - const SpyHashStateImpl& b) { + static_assert(std::is_void<T>::value, ""); + } + + // Move-only + SpyHashStateImpl(const SpyHashStateImpl&) = delete; + SpyHashStateImpl& operator=(const SpyHashStateImpl&) = delete; + + SpyHashStateImpl(SpyHashStateImpl&& other) noexcept { + *this = std::move(other); + } + + SpyHashStateImpl& operator=(SpyHashStateImpl&& other) noexcept { + hash_representation_ = std::move(other.hash_representation_); + error_ = other.error_; + moved_from_ = other.moved_from_; + other.moved_from_ = true; + return *this; + } + + template <typename U> + SpyHashStateImpl(SpyHashStateImpl<U>&& other) { // NOLINT + hash_representation_ = std::move(other.hash_representation_); + error_ = other.error_; + moved_from_ = other.moved_from_; + other.moved_from_ = true; + } + + template <typename A, typename... Args> + static SpyHashStateImpl combine(SpyHashStateImpl s, const A& a, + const Args&... args) { + // Pass an instance of SpyHashStateImpl<A> when trying to combine `A`. This + // allows us to test that the user only uses this instance for combine calls + // and does not call AbslHashValue directly. + // See AbslHashValue implementation at the bottom. + s = SpyHashStateImpl<A>::HashStateBase::combine(std::move(s), a); + return SpyHashStateImpl::combine(std::move(s), args...); + } + static SpyHashStateImpl combine(SpyHashStateImpl s) { + if (direct_absl_hash_value_error_) { + *s.error_ = "AbslHashValue should not be invoked directly."; + } else if (s.moved_from_) { + *s.error_ = "Used moved-from instance of the hash state object."; + } + return s; + } + + static void SetDirectAbslHashValueError() { + direct_absl_hash_value_error_ = true; + } + + // Two SpyHashStateImpl objects are equal if they hold equal hash + // representations. + friend bool operator==(const SpyHashStateImpl& lhs, + const SpyHashStateImpl& rhs) { + return lhs.hash_representation_ == rhs.hash_representation_; + } + + friend bool operator!=(const SpyHashStateImpl& lhs, + const SpyHashStateImpl& rhs) { + return !(lhs == rhs); + } + + enum class CompareResult { + kEqual, + kASuffixB, + kBSuffixA, + kUnequal, + }; + + static CompareResult Compare(const SpyHashStateImpl& a, + const SpyHashStateImpl& b) { const TString a_flat = y_absl::StrJoin(a.hash_representation_, ""); const TString b_flat = y_absl::StrJoin(b.hash_representation_, ""); - if (a_flat == b_flat) return CompareResult::kEqual; + if (a_flat == b_flat) return CompareResult::kEqual; if (y_absl::EndsWith(a_flat, b_flat)) return CompareResult::kBSuffixA; if (y_absl::EndsWith(b_flat, a_flat)) return CompareResult::kASuffixB; - return CompareResult::kUnequal; - } - - // operator<< prints the hash representation as a hex and ASCII dump, to - // facilitate debugging. - friend std::ostream& operator<<(std::ostream& out, - const SpyHashStateImpl& hash_state) { - out << "[\n"; - for (auto& s : hash_state.hash_representation_) { - size_t offset = 0; - for (char c : s) { - if (offset % 16 == 0) { + return CompareResult::kUnequal; + } + + // operator<< prints the hash representation as a hex and ASCII dump, to + // facilitate debugging. + friend std::ostream& operator<<(std::ostream& out, + const SpyHashStateImpl& hash_state) { + out << "[\n"; + for (auto& s : hash_state.hash_representation_) { + size_t offset = 0; + for (char c : s) { + if (offset % 16 == 0) { out << y_absl::StreamFormat("\n0x%04x: ", offset); - } - if (offset % 2 == 0) { - out << " "; - } + } + if (offset % 2 == 0) { + out << " "; + } out << y_absl::StreamFormat("%02x", c); - ++offset; - } - out << "\n"; - } - return out << "]"; - } - - // The base case of the combine recursion, which writes raw bytes into the - // internal buffer. - static SpyHashStateImpl combine_contiguous(SpyHashStateImpl hash_state, - const unsigned char* begin, - size_t size) { - const size_t large_chunk_stride = PiecewiseChunkSize(); - if (size > large_chunk_stride) { - // Combining a large contiguous buffer must have the same effect as - // doing it piecewise by the stride length, followed by the (possibly - // empty) remainder. - while (size >= large_chunk_stride) { - hash_state = SpyHashStateImpl::combine_contiguous( - std::move(hash_state), begin, large_chunk_stride); - begin += large_chunk_stride; - size -= large_chunk_stride; - } - } - - hash_state.hash_representation_.emplace_back( - reinterpret_cast<const char*>(begin), size); - return hash_state; - } - - using SpyHashStateImpl::HashStateBase::combine_contiguous; - + ++offset; + } + out << "\n"; + } + return out << "]"; + } + + // The base case of the combine recursion, which writes raw bytes into the + // internal buffer. + static SpyHashStateImpl combine_contiguous(SpyHashStateImpl hash_state, + const unsigned char* begin, + size_t size) { + const size_t large_chunk_stride = PiecewiseChunkSize(); + if (size > large_chunk_stride) { + // Combining a large contiguous buffer must have the same effect as + // doing it piecewise by the stride length, followed by the (possibly + // empty) remainder. + while (size >= large_chunk_stride) { + hash_state = SpyHashStateImpl::combine_contiguous( + std::move(hash_state), begin, large_chunk_stride); + begin += large_chunk_stride; + size -= large_chunk_stride; + } + } + + hash_state.hash_representation_.emplace_back( + reinterpret_cast<const char*>(begin), size); + return hash_state; + } + + using SpyHashStateImpl::HashStateBase::combine_contiguous; + y_absl::optional<TString> error() const { - if (moved_from_) { - return "Returned a moved-from instance of the hash state object."; - } - return *error_; - } - - private: - template <typename U> - friend class SpyHashStateImpl; - - // This is true if SpyHashStateImpl<T> has been passed to a call of - // AbslHashValue with the wrong type. This detects that the user called - // AbslHashValue directly (because the hash state type does not match). - static bool direct_absl_hash_value_error_; - + if (moved_from_) { + return "Returned a moved-from instance of the hash state object."; + } + return *error_; + } + + private: + template <typename U> + friend class SpyHashStateImpl; + + // This is true if SpyHashStateImpl<T> has been passed to a call of + // AbslHashValue with the wrong type. This detects that the user called + // AbslHashValue directly (because the hash state type does not match). + static bool direct_absl_hash_value_error_; + std::vector<TString> hash_representation_; - // This is a shared_ptr because we want all instances of the particular - // SpyHashState run to share the field. This way we can set the error for - // use-after-move and all the copies will see it. + // This is a shared_ptr because we want all instances of the particular + // SpyHashState run to share the field. This way we can set the error for + // use-after-move and all the copies will see it. std::shared_ptr<y_absl::optional<TString>> error_; - bool moved_from_ = false; -}; - -template <typename T> -bool SpyHashStateImpl<T>::direct_absl_hash_value_error_; - -template <bool& B> -struct OdrUse { - constexpr OdrUse() {} - bool& b = B; -}; - -template <void (*)()> -struct RunOnStartup { - static bool run; - static constexpr OdrUse<run> kOdrUse{}; -}; - -template <void (*f)()> -bool RunOnStartup<f>::run = (f(), true); - -template < - typename T, typename U, - // Only trigger for when (T != U), + bool moved_from_ = false; +}; + +template <typename T> +bool SpyHashStateImpl<T>::direct_absl_hash_value_error_; + +template <bool& B> +struct OdrUse { + constexpr OdrUse() {} + bool& b = B; +}; + +template <void (*)()> +struct RunOnStartup { + static bool run; + static constexpr OdrUse<run> kOdrUse{}; +}; + +template <void (*f)()> +bool RunOnStartup<f>::run = (f(), true); + +template < + typename T, typename U, + // Only trigger for when (T != U), typename = y_absl::enable_if_t<!std::is_same<T, U>::value>, - // This statement works in two ways: - // - First, it instantiates RunOnStartup and forces the initialization of - // `run`, which set the global variable. - // - Second, it triggers a SFINAE error disabling the overload to prevent - // compile time errors. If we didn't disable the overload we would get - // ambiguous overload errors, which we don't want. - int = RunOnStartup<SpyHashStateImpl<T>::SetDirectAbslHashValueError>::run> -void AbslHashValue(SpyHashStateImpl<T>, const U&); - -using SpyHashState = SpyHashStateImpl<void>; - -} // namespace hash_internal + // This statement works in two ways: + // - First, it instantiates RunOnStartup and forces the initialization of + // `run`, which set the global variable. + // - Second, it triggers a SFINAE error disabling the overload to prevent + // compile time errors. If we didn't disable the overload we would get + // ambiguous overload errors, which we don't want. + int = RunOnStartup<SpyHashStateImpl<T>::SetDirectAbslHashValueError>::run> +void AbslHashValue(SpyHashStateImpl<T>, const U&); + +using SpyHashState = SpyHashStateImpl<void>; + +} // namespace hash_internal ABSL_NAMESPACE_END } // namespace y_absl - -#endif // ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ + +#endif // ABSL_HASH_INTERNAL_SPY_HASH_STATE_H_ |