Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Add cityhash, a good and fast hash.

Browse source
This commit is contained in:
Henrik Rydgard 2013-01-19 12:01:28 +01:00
parent 200d2cc927
commit b5037341aa
8 changed files with 1007 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
Android.mk
View file

@ -16,6 +16,7 @@ LOCAL_SRC_FILES :=\
base/colorutil.cpp \
base/error_context.cpp \
base/stringutil.cpp \
ext/cityhash/city.cpp \
ext/sha1/sha1.cpp \
ext/stb_image/stb_image.c \
ext/stb_vorbis/stb_vorbis.c.arm \
@ -50,10 +51,10 @@ LOCAL_SRC_FILES :=\
gfx/texture_gen.cpp \
image/zim_load.cpp \
image/png_load.cpp \
ui/ui.cpp \
ui/ui_context.cpp \
ui/screen.cpp \
ui/virtual_input.cpp \
ui/ui.cpp \
ui/ui_context.cpp \
ui/screen.cpp \
ui/virtual_input.cpp \
util/random/perlin.cpp \
util/text/utf8.cpp

19
ext/cityhash/COPYING Normal file
View file

@ -0,0 +1,19 @@
// Copyright (c) 2011 Google, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.

200
ext/cityhash/README Normal file
View file

@ -0,0 +1,200 @@
NOTE: This is a cut down version of Google CityHash for easy building.
CityHash, a family of hash functions for strings.
Introduction
============
CityHash provides hash functions for strings. The functions mix the
input bits thoroughly but are not suitable for cryptography. See
"Hash Quality," below, for details on how CityHash was tested and so on.
We provide reference implementations in C++, with a friendly MIT license.
CityHash32() returns a 32-bit hash.
CityHash64() and similar return a 64-bit hash.
CityHash128() and similar return a 128-bit hash and are tuned for
strings of at least a few hundred bytes. Depending on your compiler
and hardware, it's likely faster than CityHash64() on sufficiently long
strings. It's slower than necessary on shorter strings, but we expect
that case to be relatively unimportant.
CityHashCrc128() and similar are variants of CityHash128() that depend
on _mm_crc32_u64(), an intrinsic that compiles to a CRC32 instruction
on some CPUs. However, none of the functions we provide are CRCs.
CityHashCrc256() is a variant of CityHashCrc128() that also depends
on _mm_crc32_u64(). It returns a 256-bit hash.
All members of the CityHash family were designed with heavy reliance
on previous work by Austin Appleby, Bob Jenkins, and others.
For example, CityHash32 has many similarities with Murmur3a.
Performance on long strings: 64-bit CPUs
========================================
We are most excited by the performance of CityHash64() and its variants on
short strings, but long strings are interesting as well.
CityHash is intended to be fast, under the constraint that it hash very
well. For CPUs with the CRC32 instruction, CRC is speedy, but CRC wasn't
designed as a hash function and shouldn't be used as one. CityHashCrc128()
is not a CRC, but it uses the CRC32 machinery.
On a single core of a 2.67GHz Intel Xeon X5550, CityHashCrc256 peaks at about
5 to 5.5 bytes/cycle. The other CityHashCrc functions are wrappers around
CityHashCrc256 and should have similar performance on long strings.
(CityHashCrc256 in v1.0.3 was even faster, but we decided it wasn't as thorough
as it should be.) CityHash128 peaks at about 4.3 bytes/cycle. The fastest
Murmur variant on that hardware, Murmur3F, peaks at about 2.4 bytes/cycle.
We expect the peak speed of CityHash128 to dominate CityHash64, which is
aimed more toward short strings or use in hash tables.
For long strings, a new function by Bob Jenkins, SpookyHash, is just
slightly slower than CityHash128 on Intel x86-64 CPUs, but noticeably
faster on AMD x86-64 CPUs. For hashing long strings on AMD CPUs
and/or CPUs without the CRC instruction, SpookyHash may be just as
good or better than any of the CityHash variants.
Performance on short strings: 64-bit CPUs
=========================================
For short strings, e.g., most hash table keys, CityHash64 is faster than
CityHash128, and probably faster than all the aforementioned functions,
depending on the mix of string lengths. Here are a few results from that
same hardware, where we (unrealistically) tested a single string length over
and over again:
Hash Results
------------------------------------------------------------------------------
CityHash64 v1.0.3 7ns for 1 byte, or 6ns for 8 bytes, or 9ns for 64 bytes
Murmur2 (64-bit) 6ns for 1 byte, or 6ns for 8 bytes, or 15ns for 64 bytes
Murmur3F 14ns for 1 byte, or 15ns for 8 bytes, or 23ns for 64 bytes
We don't have CityHash64 benchmarks results for v1.1, but we expect the
numbers to be similar.
Performance: 32-bit CPUs
========================
CityHash32 is the newest variant of CityHash. It is intended for
32-bit hardware in general but has been mostly tested on x86. Our benchmarks
suggest that Murmur3 is the nearest competitor to CityHash32 on x86.
We don't know of anything faster that has comparable quality. The speed rankings
in our testing: CityHash32 > Murmur3f > Murmur3a (for long strings), and
CityHash32 > Murmur3a > Murmur3f (for short strings).
Installation
============
We provide reference implementations of several CityHash functions, written
in C++. The build system is based on autoconf. It defaults the C++
compiler flags to "-g -O2", which is probably slower than -O3 if you are
using gcc. YMMV.
On systems with gcc, we generally recommend:
./configure
make all check CXXFLAGS="-g -O3"
sudo make install
Or, if your system has the CRC32 instruction, and you want to build everything:
./configure --enable-sse4.2
make all check CXXFLAGS="-g -O3 -msse4.2"
sudo make install
Note that our build system doesn't try to determine the appropriate compiler
flag for enabling SSE4.2. For gcc it is "-msse4.2". The --enable-sse4.2
flag to the configure script controls whether citycrc.h is installed when
you "make install." In general, picking the right compiler flags can be
tricky, and may depend on your compiler, your hardware, and even how you
plan to use the library.
For generic information about how to configure this software, please try:
./configure --help
Failing that, please work from city.cc and city*.h, as they contain all the
necessary code.
Usage
=====
The above installation instructions will produce a single library. It will
contain CityHash32(), CityHash64(), and CityHash128(), and their variants,
and possibly CityHashCrc128(), CityHashCrc128WithSeed(), and
CityHashCrc256(). The functions with Crc in the name are declared in
citycrc.h; the rest are declared in city.h.
Limitations
===========
1) CityHash32 is intended for little-endian 32-bit code, and everything else in
the current version of CityHash is intended for little-endian 64-bit CPUs.
All functions that don't use the CRC32 instruction should work in
little-endian 32-bit or 64-bit code. CityHash should work on big-endian CPUs
as well, but we haven't tested that very thoroughly yet.
2) CityHash is fairly complex. As a result of its complexity, it may not
perform as expected on some compilers. For example, preliminary reports
suggest that some Microsoft compilers compile CityHash to assembly that's
10-20% slower than it could be.
Hash Quality
============
We like to test hash functions with SMHasher, among other things.
SMHasher isn't perfect, but it seems to find almost any significant flaw.
SMHasher is available at http://code.google.com/p/smhasher/
SMHasher is designed to pass a 32-bit seed to the hash functions it tests.
No CityHash function is designed to work that way, so we adapt as follows:
For our functions that accept a seed, we use the given seed directly (padded
with zeroes); for our functions that don't accept a seed, we hash the
concatenation of the given seed and the input string.
The CityHash functions have the following flaws according to SMHasher:
(1) CityHash64: none
(2) CityHash64WithSeed: none
(3) CityHash64WithSeeds: did not test
(4) CityHash128: none
(5) CityHash128WithSeed: none
(6) CityHashCrc128: none
(7) CityHashCrc128WithSeed: none
(8) CityHashCrc256: none
(9) CityHash32: none
Some minor flaws in 32-bit and 64-bit functions are harmless, as we
expect the primary use of these functions will be in hash tables. We
may have gone slightly overboard in trying to please SMHasher and other
similar tests, but we don't want anyone to choose a different hash function
because of some minor issue reported by a quality test.
For more information
====================
http://code.google.com/p/cityhash/
cityhash-discuss@googlegroups.com
Please feel free to send us comments, questions, bug reports, or patches.

620
ext/cityhash/city.cpp Normal file
View file

@ -0,0 +1,620 @@
// Copyright (c) 2011 Google, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
//
// CityHash, by Geoff Pike and Jyrki Alakuijala
//
// 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 "city.h"
#include <algorithm>
#include <string.h> // for memcpy and memset
typedef unsigned long long uint64;
typedef unsigned int uint32;
using namespace std;
static uint64 UNALIGNED_LOAD64(const char *p) {
uint64 result;
memcpy(&result, p, sizeof(result));
return result;
}
static uint32 UNALIGNED_LOAD32(const char *p) {
uint32 result;
memcpy(&result, p, sizeof(result));
return result;
}
#ifdef _MSC_VER
#include <stdlib.h>
#define bswap_32(x) _byteswap_ulong(x)
#define bswap_64(x) _byteswap_uint64(x)
#elif defined(__APPLE__)
// Mac OS X / Darwin features
#include <libkern/OSByteOrder.h>
#define bswap_32(x) OSSwapInt32(x)
#define bswap_64(x) OSSwapInt64(x)
#else
#include <byteswap.h>
#endif
#ifdef WORDS_BIGENDIAN
#define uint32_in_expected_order(x) (bswap_32(x))
#define uint64_in_expected_order(x) (bswap_64(x))
#else
#define uint32_in_expected_order(x) (x)
#define uint64_in_expected_order(x) (x)
#endif
#if !defined(LIKELY)
#if HAVE_BUILTIN_EXPECT
#define LIKELY(x) (__builtin_expect(!!(x), 1))
#else
#define LIKELY(x) (x)
#endif
#endif
static uint64 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));
}
// 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;
// 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 fmix(uint32 h)
{
h ^= h >> 16;
h *= 0x85ebca6b;
h ^= h >> 13;
h *= 0xc2b2ae35;
h ^= h >> 16;
return h;
}
static uint32 Rotate32(uint32 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 Mur(uint32 a, uint32 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 Hash32Len13to24(const char *s, size_t len) {
uint32 a = Fetch32(s - 4 + (len >> 1));
uint32 b = Fetch32(s + 4);
uint32 c = Fetch32(s + len - 8);
uint32 d = Fetch32(s + (len >> 1));
uint32 e = Fetch32(s);
uint32 f = Fetch32(s + len - 4);
uint32 h = len;
return fmix(Mur(f, Mur(e, Mur(d, Mur(c, Mur(b, Mur(a, h)))))));
}
static uint32 Hash32Len0to4(const char *s, size_t len) {
uint32 b = 0;
uint32 c = 9;
for (size_t i = 0; i < len; i++) {
b = b * c1 + s[i];
c ^= b;
}
return fmix(Mur(b, Mur(len, c)));
}
static uint32 Hash32Len5to12(const char *s, size_t len) {
uint32 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 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 h = len, g = c1 * len, f = g;
uint32 a0 = Rotate32(Fetch32(s + len - 4) * c1, 17) * c2;
uint32 a1 = Rotate32(Fetch32(s + len - 8) * c1, 17) * c2;
uint32 a2 = Rotate32(Fetch32(s + len - 16) * c1, 17) * c2;
uint32 a3 = Rotate32(Fetch32(s + len - 12) * c1, 17) * c2;
uint32 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 a0 = Rotate32(Fetch32(s) * c1, 17) * c2;
uint32 a1 = Fetch32(s + 4);
uint32 a2 = Rotate32(Fetch32(s + 8) * c1, 17) * c2;
uint32 a3 = Rotate32(Fetch32(s + 12) * c1, 17) * c2;
uint32 a4 = Fetch32(s + 16);
h ^= a0;
h = Rotate32(h, 18);
h = h * 5 + 0xe6546b64;
f += a1;
f = Rotate32(f, 19);
f = f * c1;
g += a2;
g = Rotate32(g, 18);
g = g * 5 + 0xe6546b64;
h ^= a3 + a1;
h = Rotate32(h, 19);
h = h * 5 + 0xe6546b64;
g ^= a4;
g = bswap_32(g) * 5;
h += a4 * 5;
h = bswap_32(h);
f += a0;
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 Rotate(uint64 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) {
return val ^ (val >> 47);
}
static uint64 HashLen16(uint64 u, uint64 v) {
return Hash128to64(uint128(u, v));
}
static uint64 HashLen16(uint64 u, uint64 v, uint64 mul) {
// Murmur-inspired hashing.
uint64 a = (u ^ v) * mul;
a ^= (a >> 47);
uint64 b = (v ^ a) * mul;
b ^= (b >> 47);
b *= mul;
return b;
}
static uint64 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;
return HashLen16(c, d, mul);
}
if (len >= 4) {
uint64 mul = k2 + len * 2;
uint64 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 = len + (static_cast<uint32>(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 HashLen17to32(const char *s, 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;
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) {
a += w;
b = Rotate(b + a + z, 21);
uint64 c = a;
a += x;
a += y;
b += Rotate(a, 44);
return make_pair(a + z, b + c);
}
// 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) {
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, 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 = bswap_64((u + v) * mul) + h;
uint64 x = Rotate(e + f, 42) + c;
uint64 y = (bswap_64((v + w) * mul) + g) * mul;
uint64 z = e + f + c;
a = bswap_64((x + z) * mul + y) + b;
b = ShiftMix((z + a) * mul + d + h) * mul;
return b + x;
}
uint64 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 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);
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 CityHash64WithSeed(const char *s, size_t len, uint64 seed) {
return CityHash64WithSeeds(s, len, k2, seed);
}
uint64 CityHash64WithSeeds(const char *s, size_t len,
uint64 seed0, uint64 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, size_t len, uint128 seed) {
uint64 a = Uint128Low64(seed);
uint64 b = Uint128High64(seed);
uint64 c = 0;
uint64 d = 0;
signed long l = len - 16;
if (l <= 0) { // len <= 16
a = ShiftMix(a * k1) * k1;
c = b * k1 + HashLen0to16(s, len);
d = ShiftMix(a + (len >= 8 ? Fetch64(s) : c));
} else { // len > 16
c = HashLen16(Fetch64(s + len - 8) + k1, a);
d = HashLen16(b + len, c + Fetch64(s + len - 16));
a += d;
do {
a ^= ShiftMix(Fetch64(s) * k1) * k1;
a *= k1;
b ^= a;
c ^= ShiftMix(Fetch64(s + 8) * k1) * k1;
c *= k1;
d ^= c;
s += 16;
l -= 16;
} while (l > 0);
}
a = HashLen16(a, c);
b = HashLen16(d, b);
return uint128(a ^ b, HashLen16(b, a));
}
uint128 CityHash128WithSeed(const char *s, size_t len, uint128 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;
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;
w.second = Rotate(x + Fetch64(s + 88), 53) * k1;
// This is the same inner loop as CityHash64(), manually unrolled.
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;
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 -= 128;
} while (LIKELY(len >= 128));
x += Rotate(v.first + z, 49) * k0;
y = y * k0 + Rotate(w.second, 37);
z = z * k0 + Rotate(w.first, 27);
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 (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);
x = x * k0 + w.first;
z += w.second + Fetch64(s + len - tail_done);
w.second += v.first;
v = WeakHashLen32WithSeeds(s + len - tail_done, v.first + z, v.second);
v.first *= k0;
}
// At this point our 56 bytes of state should contain more than
// enough information for a strong 128-bit hash. We use two
// 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));
}
uint128 CityHash128(const char *s, size_t len) {
return len >= 16 ?
CityHash128WithSeed(s + 16, len - 16,
uint128(Fetch64(s), Fetch64(s + 8) + k0)) :
CityHash128WithSeed(s, len, uint128(k0, k1));
}
#ifdef __SSE4_2__
#include <citycrc.h>
#include <nmmintrin.h>
// Requires len >= 240.
static void CityHashCrc256Long(const char *s, size_t len,
uint32 seed, uint64 *result) {
uint64 a = Fetch64(s + 56) + k0;
uint64 b = Fetch64(s + 96) + k0;
uint64 c = result[0] = HashLen16(b, len);
uint64 d = result[1] = Fetch64(s + 120) * k0 + len;
uint64 e = Fetch64(s + 184) + seed;
uint64 f = 0;
uint64 g = 0;
uint64 h = c + d;
uint64 x = seed;
uint64 y = 0;
uint64 z = 0;
// 240 bytes of input per iter.
size_t iters = len / 240;
len -= iters * 240;
do {
#undef CHUNK
#define CHUNK(r) \
PERMUTE3(x, z, y); \
b += Fetch64(s); \
c += Fetch64(s + 8); \
d += Fetch64(s + 16); \
e += Fetch64(s + 24); \
f += Fetch64(s + 32); \
a += b; \
h += f; \
b += c; \
f += d; \
g += e; \
e += z; \
g += x; \
z = _mm_crc32_u64(z, b + g); \
y = _mm_crc32_u64(y, e + h); \
x = _mm_crc32_u64(x, f + a); \
e = Rotate(e, r); \
c += e; \
s += 40
CHUNK(0); PERMUTE3(a, h, c);
CHUNK(33); PERMUTE3(a, h, f);
CHUNK(0); PERMUTE3(b, h, f);
CHUNK(42); PERMUTE3(b, h, d);
CHUNK(0); PERMUTE3(b, h, e);
CHUNK(33); PERMUTE3(a, h, e);
} while (--iters > 0);
while (len >= 40) {
CHUNK(29);
e ^= Rotate(a, 20);
h += Rotate(b, 30);
g ^= Rotate(c, 40);
f += Rotate(d, 34);
PERMUTE3(c, h, g);
len -= 40;
}
if (len > 0) {
s = s + len - 40;
CHUNK(33);
e ^= Rotate(a, 43);
h += Rotate(b, 42);
g ^= Rotate(c, 41);
f += Rotate(d, 40);
}
result[0] ^= h;
result[1] ^= g;
g += h;
a = HashLen16(a, g + z);
x += y << 32;
b += x;
c = HashLen16(c, z) + h;
d = HashLen16(d, e + result[0]);
g += e;
h += HashLen16(x, f);
e = HashLen16(a, d) + g;
z = HashLen16(b, c) + a;
y = HashLen16(g, h) + c;
result[0] = e + z + y + x;
a = ShiftMix((a + y) * k0) * k0 + b;
result[1] += a + result[0];
a = ShiftMix(a * k0) * k0 + c;
result[2] = a + result[1];
a = ShiftMix((a + e) * k0) * k0;
result[3] = a + result[2];
}
// Requires len < 240.
static void CityHashCrc256Short(const char *s, size_t len, uint64 *result) {
char buf[240];
memcpy(buf, s, len);
memset(buf + len, 0, 240 - len);
CityHashCrc256Long(buf, 240, ~static_cast<uint32>(len), result);
}
void CityHashCrc256(const char *s, size_t len, uint64 *result) {
if (LIKELY(len >= 240)) {
CityHashCrc256Long(s, len, 0, result);
} else {
CityHashCrc256Short(s, len, result);
}
}
uint128 CityHashCrc128WithSeed(const char *s, size_t len, uint128 seed) {
if (len <= 900) {
return CityHash128WithSeed(s, len, seed);
} else {
uint64 result[4];
CityHashCrc256(s, len, result);
uint64 u = Uint128High64(seed) + result[0];
uint64 v = Uint128Low64(seed) + result[1];
return uint128(HashLen16(u, v + result[2]),
HashLen16(Rotate(v, 32), u * k0 + result[3]));
}
}
uint128 CityHashCrc128(const char *s, size_t len) {
if (len <= 900) {
return CityHash128(s, len);
} else {
uint64 result[4];
CityHashCrc256(s, len, result);
return uint128(result[2], result[3]);
}
}
#endif

108
ext/cityhash/city.h Normal file
View file

@ -0,0 +1,108 @@
// Copyright (c) 2011 Google, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
//
// CityHash, by Geoff Pike and Jyrki Alakuijala
//
// http://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 64-bit x86 code, on long strings, the picture is more complicated.
// On many recent Intel CPUs, such as Nehalem, Westmere, Sandy Bridge, etc.,
// CityHashCrc128 appears to be faster than all competitors of comparable
// quality. CityHash128 is also good but not quite as fast. We believe our
// nearest competitor is Bob Jenkins' Spooky. We don't have great data for
// other 64-bit CPUs, but for long strings we know that Spooky is slightly
// faster than CityHash on some relatively recent AMD x86-64 CPUs, for example.
//
// 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.
//
// 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 CITY_HASH_H_
#define CITY_HASH_H_
#include <stdlib.h> // for size_t.
#include <stdint.h>
#include <utility>
typedef uint8_t uint8;
typedef uint32_t uint32;
typedef uint64_t uint64;
typedef std::pair<uint64, uint64> uint128;
inline uint64 Uint128Low64(const uint128& x) { return x.first; }
inline uint64 Uint128High64(const uint128& x) { return x.second; }
// Hash function for a byte array.
uint64 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.
uint64 CityHash64WithSeed(const char *buf, size_t len, uint64 seed);
// Hash function for a byte array. For convenience, two seeds are also
// hashed into the result.
uint64 CityHash64WithSeeds(const char *buf, size_t len,
uint64 seed0, uint64 seed1);
// Hash function for a byte array.
uint128 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.
uint128 CityHash128WithSeed(const char *s, size_t len, uint128 seed);
// Hash function for a byte array. Most useful in 32-bit binaries.
uint32 CityHash32(const char *buf, size_t len);
// Hash 128 input bits down to 64 bits of output.
// This is intended to be a reasonably good hash function.
inline uint64 Hash128to64(const uint128& x) {
// Murmur-inspired hashing.
const uint64 kMul = 0x9ddfea08eb382d69ULL;
uint64 a = (Uint128Low64(x) ^ Uint128High64(x)) * kMul;
a ^= (a >> 47);
uint64 b = (Uint128High64(x) ^ a) * kMul;
b ^= (b >> 47);
b *= kMul;
return b;
}
#endif // CITY_HASH_H_

43
ext/cityhash/citycrc.h Normal file
View file

@ -0,0 +1,43 @@
// Copyright (c) 2011 Google, Inc.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
//
// CityHash, by Geoff Pike and Jyrki Alakuijala
//
// This file declares the subset of the CityHash functions that require
// _mm_crc32_u64(). See the CityHash README for details.
//
// Functions in the CityHash family are not suitable for cryptography.
#ifndef CITY_HASH_CRC_H_
#define CITY_HASH_CRC_H_
#include <city.h>
// Hash function for a byte array.
uint128 CityHashCrc128(const char *s, size_t len);
// Hash function for a byte array. For convenience, a 128-bit seed is also
// hashed into the result.
uint128 CityHashCrc128WithSeed(const char *s, size_t len, uint128 seed);
// Hash function for a byte array. Sets result[0] ... result[3].
void CityHashCrc256(const char *s, size_t len, uint64 *result);
#endif // CITY_HASH_CRC_H_

3
native.vcxproj
View file

@ -199,6 +199,8 @@
<ClInclude Include="base\threadutil.h" />
<ClInclude Include="base\timeutil.h" />
<ClInclude Include="data\listable.h" />
<ClInclude Include="ext\cityhash\city.h" />
<ClInclude Include="ext\cityhash\citycrc.h" />
<ClInclude Include="ext\etcpack\etcdec.h" />
<ClInclude Include="ext\etcpack\etcpack.h" />
<ClInclude Include="ext\sha1\sha1.h" />
@ -291,6 +293,7 @@
<ClCompile Include="base\stringutil.cpp" />
<ClCompile Include="base\threadutil.cpp" />
<ClCompile Include="base\timeutil.cpp" />
<ClCompile Include="ext\cityhash\city.cpp" />
<ClCompile Include="ext\etcpack\etcdec.cpp" />
<ClCompile Include="ext\etcpack\etcpack.cpp" />
<ClCompile Include="ext\sha1\sha1.cpp" />

9
native.vcxproj.filters
View file

@ -236,6 +236,12 @@
<ClInclude Include="util\text\utf8.h">
<Filter>util</Filter>
</ClInclude>
<ClInclude Include="ext\cityhash\city.h">
<Filter>ext</Filter>
</ClInclude>
<ClInclude Include="ext\cityhash\citycrc.h">
<Filter>ext</Filter>
</ClInclude>
</ItemGroup>
<ItemGroup>
<ClCompile Include="gfx\gl_debug_log.cpp">
@ -420,6 +426,9 @@
<ClCompile Include="util\text\utf8.cpp">
<Filter>util</Filter>
</ClCompile>
<ClCompile Include="ext\cityhash\city.cpp">
<Filter>ext</Filter>
</ClCompile>
</ItemGroup>
<ItemGroup>
<Filter Include="gfx">