Add City64 as local hash, and update RTTI accordingly.

This commit is contained in:
Nikolay Igotti
2016-10-11 18:04:08 +03:00
parent fb61896825
commit ecb5a67db7
10 changed files with 432 additions and 186 deletions
+218
View File
@@ -0,0 +1,218 @@
#include "City.h"
#include <string.h>
#include <algorithm>
namespace {
// 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;
uint64_t UNALIGNED_LOAD64(const char *p) {
uint64_t result;
memcpy(&result, p, sizeof(result));
return result;
}
uint32_t UNALIGNED_LOAD32(const char *p) {
uint32_t result;
memcpy(&result, p, sizeof(result));
return result;
}
#define bswap32(x) __builtin_bswap32(x)
#define bswap64(x) __builtin_bswap64(x)
#ifdef WORDS_BIGENDIAN
#define uint32_in_expected_order(x) (bswap32(x))
#define uint64_in_expected_order(x) (bswap64(x))
#else
#define uint32_in_expected_order(x) (x)
#define uint64_in_expected_order(x) (x)
#endif
uint64_t Fetch64(const char *p) {
return uint64_in_expected_order(UNALIGNED_LOAD64(p));
}
uint32_t Fetch32(const char *p) {
return uint32_in_expected_order(UNALIGNED_LOAD32(p));
}
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)));
}
// Bitwise right rotate. Normally this will compile to a single
// instruction, especially if the shift is a manifest constant.
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)));
}
uint64_t ShiftMix(uint64_t val) {
return val ^ (val >> 47);
}
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;
}
typedef std::pair<uint64_t, uint64_t> uint128_t;
uint64_t Hash128to64(const uint128_t& x) {
// Murmur-inspired hashing.
const uint64_t kMul = 0x9ddfea08eb382d69ULL;
uint64_t a = (x.first ^ x.second) * kMul;
a ^= (a >> 47);
uint64_t b = (x.second ^ a) * kMul;
b ^= (b >> 47);
b *= kMul;
return b;
}
uint64_t HashLen16(uint64_t u, uint64_t v) {
return Hash128to64(uint128_t(u, v));
}
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.
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.
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.
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 = bswap64((u + v) * mul) + h;
uint64_t x = Rotate(e + f, 42) + c;
uint64_t y = (bswap64((v + w) * mul) + g) * mul;
uint64_t z = e + f + c;
a = bswap64((x + z) * mul + y) + b;
b = ShiftMix((z + a) * mul + d + h) * mul;
return b + x;
}
} // namespace
extern "C" {
uint64_t CityHash64(const void* data, size_t len) {
const char* s = reinterpret_cast<const char*>(data);
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);
}
} // extern "C"
+20
View File
@@ -0,0 +1,20 @@
#ifndef RUNTIME_CITY_H
#define RUNTIME_CITY_H
// CityHash, by Geoff Pike and Jyrki Alakuijala.
#include <stddef.h>
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
// Hash function for a byte array.
uint64_t CityHash64(const void* buf, size_t len);
#ifdef __cplusplus
}
#endif
#endif // RUNTIME_CITY_H
+2 -1
View File
@@ -2,6 +2,7 @@
#include "Names.h"
#include "City.h"
#include "Sha1.h"
namespace {
@@ -21,7 +22,7 @@ extern "C" {
// Make local hash out of arbitrary data.
void MakeLocalHash(const void* data, uint32_t size, LocalHash* hash) {
assert(false); // not implemented yet
*hash = CityHash64(data, size);
}
// Make global hash out of arbitrary data.
+2 -2
View File
@@ -6,12 +6,12 @@
// All names in system are stored as hashes (or maybe, for debug builds,
// as pointers to uniqued C strings containing names?).
// There are two types of hashes:
// - local hash, must be unique per class (CitiHash64 is being used)
// - local hash, must be unique per class (CityHash64 is being used)
// - global hash, must be unique globally (SHA1 is being used)
// Generic guideline is that global hash is being used in global persistent context, while local
// hashes are more local in scope.
// Local hash.
typedef int64_t LocalHash;
typedef uint64_t LocalHash;
// Hash of field name.
typedef LocalHash FieldNameHash;
// Hash of open method name.
-1
View File
@@ -1,6 +1,5 @@
#ifndef RUNTIME_SHA1_H
#define RUNTIME_SHA1_H
/* ================ sha1.h ================ */
/*
SHA-1 in C
By Steve Reid <steve@edmweb.com>