common/cityhash: Use common types

Allow sharing return types with the rest of the code base. For example,
we use 'u128 = std::array<u64, 2>', meanwhile Google's code uses
'uint128 = std::pair<u64, u64>'.

While we are at it, use size_t instead of std::size_t.
This commit is contained in:
ReinUsesLisp 2021-02-18 00:23:53 -03:00
parent 9ca5e52f07
commit f3805376f7
3 changed files with 100 additions and 116 deletions

View File

@ -28,8 +28,10 @@
// compromising on hash quality.
#include <algorithm>
#include <string.h> // for memcpy and memset
#include "cityhash.h"
#include <cstring>
#include <utility>
#include "common/cityhash.h"
#include "common/swap.h"
// #include "config.h"
@ -42,21 +44,17 @@
using namespace std;
typedef uint8_t uint8;
typedef uint32_t uint32;
typedef uint64_t uint64;
namespace Common {
static uint64 UNALIGNED_LOAD64(const char* p) {
uint64 result;
memcpy(&result, p, sizeof(result));
static u64 unaligned_load64(const char* p) {
u64 result;
std::memcpy(&result, p, sizeof(result));
return result;
}
static uint32 UNALIGNED_LOAD32(const char* p) {
uint32 result;
memcpy(&result, p, sizeof(result));
static u32 unaligned_load32(const char* p) {
u32 result;
std::memcpy(&result, p, sizeof(result));
return result;
}
@ -76,64 +74,64 @@ static uint32 UNALIGNED_LOAD32(const char* p) {
#endif
#endif
static uint64 Fetch64(const char* p) {
return uint64_in_expected_order(UNALIGNED_LOAD64(p));
static u64 Fetch64(const char* p) {
return uint64_in_expected_order(unaligned_load64(p));
}
static uint32 Fetch32(const char* p) {
return uint32_in_expected_order(UNALIGNED_LOAD32(p));
static u32 Fetch32(const char* p) {
return uint32_in_expected_order(unaligned_load32(p));
}
// Some primes between 2^63 and 2^64 for various uses.
static const uint64 k0 = 0xc3a5c85c97cb3127ULL;
static const uint64 k1 = 0xb492b66fbe98f273ULL;
static const uint64 k2 = 0x9ae16a3b2f90404fULL;
static constexpr u64 k0 = 0xc3a5c85c97cb3127ULL;
static constexpr u64 k1 = 0xb492b66fbe98f273ULL;
static constexpr u64 k2 = 0x9ae16a3b2f90404fULL;
// Bitwise right rotate. Normally this will compile to a single
// instruction, especially if the shift is a manifest constant.
static uint64 Rotate(uint64 val, int shift) {
static u64 Rotate(u64 val, int shift) {
// Avoid shifting by 64: doing so yields an undefined result.
return shift == 0 ? val : ((val >> shift) | (val << (64 - shift)));
}
static uint64 ShiftMix(uint64 val) {
static u64 ShiftMix(u64 val) {
return val ^ (val >> 47);
}
static uint64 HashLen16(uint64 u, uint64 v) {
return Hash128to64(uint128(u, v));
static u64 HashLen16(u64 u, u64 v) {
return Hash128to64(u128{u, v});
}
static uint64 HashLen16(uint64 u, uint64 v, uint64 mul) {
static u64 HashLen16(u64 u, u64 v, u64 mul) {
// Murmur-inspired hashing.
uint64 a = (u ^ v) * mul;
u64 a = (u ^ v) * mul;
a ^= (a >> 47);
uint64 b = (v ^ a) * mul;
u64 b = (v ^ a) * mul;
b ^= (b >> 47);
b *= mul;
return b;
}
static uint64 HashLen0to16(const char* s, std::size_t len) {
static u64 HashLen0to16(const char* s, size_t len) {
if (len >= 8) {
uint64 mul = k2 + len * 2;
uint64 a = Fetch64(s) + k2;
uint64 b = Fetch64(s + len - 8);
uint64 c = Rotate(b, 37) * mul + a;
uint64 d = (Rotate(a, 25) + b) * mul;
u64 mul = k2 + len * 2;
u64 a = Fetch64(s) + k2;
u64 b = Fetch64(s + len - 8);
u64 c = Rotate(b, 37) * mul + a;
u64 d = (Rotate(a, 25) + b) * mul;
return HashLen16(c, d, mul);
}
if (len >= 4) {
uint64 mul = k2 + len * 2;
uint64 a = Fetch32(s);
u64 mul = k2 + len * 2;
u64 a = Fetch32(s);
return HashLen16(len + (a << 3), Fetch32(s + len - 4), mul);
}
if (len > 0) {
uint8 a = s[0];
uint8 b = s[len >> 1];
uint8 c = s[len - 1];
uint32 y = static_cast<uint32>(a) + (static_cast<uint32>(b) << 8);
uint32 z = static_cast<uint32>(len) + (static_cast<uint32>(c) << 2);
u8 a = s[0];
u8 b = s[len >> 1];
u8 c = s[len - 1];
u32 y = static_cast<u32>(a) + (static_cast<u32>(b) << 8);
u32 z = static_cast<u32>(len) + (static_cast<u32>(c) << 2);
return ShiftMix(y * k2 ^ z * k0) * k2;
}
return k2;
@ -141,22 +139,21 @@ static uint64 HashLen0to16(const char* s, std::size_t len) {
// This probably works well for 16-byte strings as well, but it may be overkill
// in that case.
static uint64 HashLen17to32(const char* s, std::size_t len) {
uint64 mul = k2 + len * 2;
uint64 a = Fetch64(s) * k1;
uint64 b = Fetch64(s + 8);
uint64 c = Fetch64(s + len - 8) * mul;
uint64 d = Fetch64(s + len - 16) * k2;
static u64 HashLen17to32(const char* s, size_t len) {
u64 mul = k2 + len * 2;
u64 a = Fetch64(s) * k1;
u64 b = Fetch64(s + 8);
u64 c = Fetch64(s + len - 8) * mul;
u64 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 pair<uint64, uint64> WeakHashLen32WithSeeds(uint64 w, uint64 x, uint64 y, uint64 z, uint64 a,
uint64 b) {
static pair<u64, u64> WeakHashLen32WithSeeds(u64 w, u64 x, u64 y, u64 z, u64 a, u64 b) {
a += w;
b = Rotate(b + a + z, 21);
uint64 c = a;
u64 c = a;
a += x;
a += y;
b += Rotate(a, 44);
@ -164,34 +161,34 @@ static pair<uint64, uint64> WeakHashLen32WithSeeds(uint64 w, uint64 x, uint64 y,
}
// Return a 16-byte hash for s[0] ... s[31], a, and b. Quick and dirty.
static pair<uint64, uint64> WeakHashLen32WithSeeds(const char* s, uint64 a, uint64 b) {
static pair<u64, u64> WeakHashLen32WithSeeds(const char* s, u64 a, u64 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 HashLen33to64(const char* s, std::size_t len) {
uint64 mul = k2 + len * 2;
uint64 a = Fetch64(s) * k2;
uint64 b = Fetch64(s + 8);
uint64 c = Fetch64(s + len - 24);
uint64 d = Fetch64(s + len - 32);
uint64 e = Fetch64(s + 16) * k2;
uint64 f = Fetch64(s + 24) * 9;
uint64 g = Fetch64(s + len - 8);
uint64 h = Fetch64(s + len - 16) * mul;
uint64 u = Rotate(a + g, 43) + (Rotate(b, 30) + c) * 9;
uint64 v = ((a + g) ^ d) + f + 1;
uint64 w = swap64((u + v) * mul) + h;
uint64 x = Rotate(e + f, 42) + c;
uint64 y = (swap64((v + w) * mul) + g) * mul;
uint64 z = e + f + c;
static u64 HashLen33to64(const char* s, size_t len) {
u64 mul = k2 + len * 2;
u64 a = Fetch64(s) * k2;
u64 b = Fetch64(s + 8);
u64 c = Fetch64(s + len - 24);
u64 d = Fetch64(s + len - 32);
u64 e = Fetch64(s + 16) * k2;
u64 f = Fetch64(s + 24) * 9;
u64 g = Fetch64(s + len - 8);
u64 h = Fetch64(s + len - 16) * mul;
u64 u = Rotate(a + g, 43) + (Rotate(b, 30) + c) * 9;
u64 v = ((a + g) ^ d) + f + 1;
u64 w = swap64((u + v) * mul) + h;
u64 x = Rotate(e + f, 42) + c;
u64 y = (swap64((v + w) * mul) + g) * mul;
u64 z = e + f + c;
a = swap64((x + z) * mul + y) + b;
b = ShiftMix((z + a) * mul + d + h) * mul;
return b + x;
}
uint64 CityHash64(const char* s, std::size_t len) {
u64 CityHash64(const char* s, size_t len) {
if (len <= 32) {
if (len <= 16) {
return HashLen0to16(s, len);
@ -204,15 +201,15 @@ uint64 CityHash64(const char* s, std::size_t 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 x = Fetch64(s + len - 40);
uint64 y = Fetch64(s + len - 16) + Fetch64(s + len - 56);
uint64 z = HashLen16(Fetch64(s + len - 48) + len, Fetch64(s + len - 24));
pair<uint64, uint64> v = WeakHashLen32WithSeeds(s + len - 64, len, z);
pair<uint64, uint64> w = WeakHashLen32WithSeeds(s + len - 32, y + k1, x);
u64 x = Fetch64(s + len - 40);
u64 y = Fetch64(s + len - 16) + Fetch64(s + len - 56);
u64 z = HashLen16(Fetch64(s + len - 48) + len, Fetch64(s + len - 24));
pair<u64, u64> v = WeakHashLen32WithSeeds(s + len - 64, len, z);
pair<u64, u64> 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<std::size_t>(63);
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;
@ -229,21 +226,21 @@ uint64 CityHash64(const char* s, std::size_t len) {
HashLen16(v.second, w.second) + x);
}
uint64 CityHash64WithSeed(const char* s, std::size_t len, uint64 seed) {
u64 CityHash64WithSeed(const char* s, size_t len, u64 seed) {
return CityHash64WithSeeds(s, len, k2, seed);
}
uint64 CityHash64WithSeeds(const char* s, std::size_t len, uint64 seed0, uint64 seed1) {
u64 CityHash64WithSeeds(const char* s, size_t len, u64 seed0, u64 seed1) {
return HashLen16(CityHash64(s, len) - seed0, seed1);
}
// A subroutine for CityHash128(). Returns a decent 128-bit hash for strings
// of any length representable in signed long. Based on City and Murmur.
static uint128 CityMurmur(const char* s, std::size_t len, uint128 seed) {
uint64 a = Uint128Low64(seed);
uint64 b = Uint128High64(seed);
uint64 c = 0;
uint64 d = 0;
static u128 CityMurmur(const char* s, size_t len, u128 seed) {
u64 a = seed[0];
u64 b = seed[1];
u64 c = 0;
u64 d = 0;
signed long l = static_cast<long>(len) - 16;
if (l <= 0) { // len <= 16
a = ShiftMix(a * k1) * k1;
@ -266,20 +263,20 @@ static uint128 CityMurmur(const char* s, std::size_t len, uint128 seed) {
}
a = HashLen16(a, c);
b = HashLen16(d, b);
return uint128(a ^ b, HashLen16(b, a));
return u128{a ^ b, HashLen16(b, a)};
}
uint128 CityHash128WithSeed(const char* s, std::size_t len, uint128 seed) {
u128 CityHash128WithSeed(const char* s, size_t len, u128 seed) {
if (len < 128) {
return CityMurmur(s, len, seed);
}
// We expect len >= 128 to be the common case. Keep 56 bytes of state:
// v, w, x, y, and z.
pair<uint64, uint64> v, w;
uint64 x = Uint128Low64(seed);
uint64 y = Uint128High64(seed);
uint64 z = len * k1;
pair<u64, u64> v, w;
u64 x = seed[0];
u64 y = seed[1];
u64 z = len * k1;
v.first = Rotate(y ^ k1, 49) * k1 + Fetch64(s);
v.second = Rotate(v.first, 42) * k1 + Fetch64(s + 8);
w.first = Rotate(y + z, 35) * k1 + x;
@ -313,7 +310,7 @@ uint128 CityHash128WithSeed(const char* s, std::size_t len, uint128 seed) {
w.first *= 9;
v.first *= k0;
// If 0 < len < 128, hash up to 4 chunks of 32 bytes each from the end of s.
for (std::size_t tail_done = 0; tail_done < len;) {
for (size_t tail_done = 0; tail_done < len;) {
tail_done += 32;
y = Rotate(x + y, 42) * k0 + v.second;
w.first += Fetch64(s + len - tail_done + 16);
@ -328,13 +325,12 @@ uint128 CityHash128WithSeed(const char* s, std::size_t len, uint128 seed) {
// different 56-byte-to-8-byte hashes to get a 16-byte final result.
x = HashLen16(x, v.first);
y = HashLen16(y + z, w.first);
return uint128(HashLen16(x + v.second, w.second) + y, HashLen16(x + w.second, y + v.second));
return u128{HashLen16(x + v.second, w.second) + y, HashLen16(x + w.second, y + v.second)};
}
uint128 CityHash128(const char* s, std::size_t len) {
return len >= 16
? CityHash128WithSeed(s + 16, len - 16, uint128(Fetch64(s), Fetch64(s + 8) + k0))
: CityHash128WithSeed(s, len, uint128(k0, k1));
u128 CityHash128(const char* s, size_t len) {
return len >= 16 ? CityHash128WithSeed(s + 16, len - 16, u128{Fetch64(s), Fetch64(s + 8) + k0})
: CityHash128WithSeed(s, len, u128{k0, k1});
}
} // namespace Common

View File

@ -61,50 +61,38 @@
#pragma once
#include <cstddef>
#include <cstdint>
#include <utility>
#include "common/common_types.h"
namespace Common {
using uint128 = std::pair<uint64_t, uint64_t>;
[[nodiscard]] inline uint64_t Uint128Low64(const uint128& x) {
return x.first;
}
[[nodiscard]] inline uint64_t Uint128High64(const uint128& x) {
return x.second;
}
// Hash function for a byte array.
[[nodiscard]] uint64_t CityHash64(const char* buf, std::size_t len);
[[nodiscard]] u64 CityHash64(const char* buf, size_t len);
// Hash function for a byte array. For convenience, a 64-bit seed is also
// hashed into the result.
[[nodiscard]] uint64_t CityHash64WithSeed(const char* buf, std::size_t len, uint64_t seed);
[[nodiscard]] u64 CityHash64WithSeed(const char* buf, size_t len, u64 seed);
// Hash function for a byte array. For convenience, two seeds are also
// hashed into the result.
[[nodiscard]] uint64_t CityHash64WithSeeds(const char* buf, std::size_t len, uint64_t seed0,
uint64_t seed1);
[[nodiscard]] u64 CityHash64WithSeeds(const char* buf, size_t len, u64 seed0, u64 seed1);
// Hash function for a byte array.
[[nodiscard]] uint128 CityHash128(const char* s, std::size_t len);
[[nodiscard]] u128 CityHash128(const char* s, size_t len);
// Hash function for a byte array. For convenience, a 128-bit seed is also
// hashed into the result.
[[nodiscard]] uint128 CityHash128WithSeed(const char* s, std::size_t len, uint128 seed);
[[nodiscard]] u128 CityHash128WithSeed(const char* s, size_t len, u128 seed);
// Hash 128 input bits down to 64 bits of output.
// This is intended to be a reasonably good hash function.
[[nodiscard]] inline uint64_t Hash128to64(const uint128& x) {
[[nodiscard]] inline u64 Hash128to64(const u128& x) {
// Murmur-inspired hashing.
const uint64_t kMul = 0x9ddfea08eb382d69ULL;
uint64_t a = (Uint128Low64(x) ^ Uint128High64(x)) * kMul;
const u64 mul = 0x9ddfea08eb382d69ULL;
u64 a = (x[0] ^ x[1]) * mul;
a ^= (a >> 47);
uint64_t b = (Uint128High64(x) ^ a) * kMul;
u64 b = (x[1] ^ a) * mul;
b ^= (b >> 47);
b *= kMul;
b *= mul;
return b;
}

View File

@ -16,7 +16,7 @@ TEST_CASE("CityHash", "[common]") {
REQUIRE(CityHash64(msg, sizeof(msg)) == 0x92d5c2e9cbfbbc01);
REQUIRE(CityHash64WithSeed(msg, sizeof(msg), 0xdead) == 0xbfbe93f21a2820dd);
REQUIRE(CityHash64WithSeeds(msg, sizeof(msg), 0xbeef, 0xcafe) == 0xb343317955fc8a06);
REQUIRE(CityHash128(msg, sizeof(msg)) == uint128{0x98e60d0423747eaa, 0xd8694c5b6fcaede9});
REQUIRE(CityHash128(msg, sizeof(msg)) == u128{0x98e60d0423747eaa, 0xd8694c5b6fcaede9});
REQUIRE(CityHash128WithSeed(msg, sizeof(msg), {0xdead, 0xbeef}) ==
uint128{0xf0307dba81199ebe, 0xd77764e0c4a9eb74});
u128{0xf0307dba81199ebe, 0xd77764e0c4a9eb74});
}