cheaders: update wyhash to version gamma
ka-weihe
GitHub
parent
7f516dbae2
commit
5be892c635
|
|
@ -213,32 +213,11 @@ void _vcleanup();
|
|||
|
||||
// ============== wyhash ==============
|
||||
//Author: Wang Yi
|
||||
#ifndef wyhash_version_4
|
||||
#define wyhash_version_4
|
||||
#ifndef wyhash_version_gamma
|
||||
#define wyhash_version_gamma
|
||||
#include <stdint.h>
|
||||
#include <string.h>
|
||||
#if defined(_MSC_VER) && defined(_M_X64)
|
||||
#include <intrin.h>
|
||||
#pragma intrinsic(_umul128)
|
||||
#endif
|
||||
const uint64_t _wyp0=0xa0761d6478bd642full, _wyp1=0xe7037ed1a0b428dbull, _wyp2=0x8ebc6af09c88c6e3ull, _wyp3=0x589965cc75374cc3ull, _wyp4=0x1d8e4e27c47d124full;
|
||||
static inline uint64_t _wyrotr(uint64_t v, unsigned k) { return (v>>k)|(v<<(64-k)); }
|
||||
static inline uint64_t _wymum(uint64_t A, uint64_t B) {
|
||||
#ifdef WYHASH32
|
||||
uint64_t hh=(A>>32)*(B>>32), hl=(A>>32)*(unsigned)B, lh=(unsigned)A*(B>>32), ll=(uint64_t)(unsigned)A*(unsigned)B;
|
||||
return _wyrotr(hl,32)^_wyrotr(lh,32)^hh^ll;
|
||||
#else
|
||||
#ifdef __SIZEOF_INT128__
|
||||
__uint128_t r=A; r*=B; return (r>>64)^r;
|
||||
#elif defined(_MSC_VER) && defined(_M_X64)
|
||||
A=_umul128(A, B, &B); return A^B;
|
||||
#else
|
||||
uint64_t ha=A>>32, hb=B>>32, la=(uint32_t)A, lb=(uint32_t)B, hi, lo;
|
||||
uint64_t rh=ha*hb, rm0=ha*lb, rm1=hb*la, rl=la*lb, t=rl+(rm0<<32), c=t<rl;
|
||||
lo=t+(rm1<<32); c+=lo<t;hi=rh+(rm0>>32)+(rm1>>32)+c; return hi^lo;
|
||||
#endif
|
||||
#endif
|
||||
}
|
||||
const uint64_t _wyp0=0xa0761d6478bd642full, _wyp1=0xe7037ed1a0b428dbull;
|
||||
#ifndef WYHASH_LITTLE_ENDIAN
|
||||
#if defined(_WIN32) || defined(__LITTLE_ENDIAN__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
|
||||
#define WYHASH_LITTLE_ENDIAN 1
|
||||
|
|
@ -246,50 +225,59 @@ static inline uint64_t _wymum(uint64_t A, uint64_t B) {
|
|||
#define WYHASH_LITTLE_ENDIAN 0
|
||||
#endif
|
||||
#endif
|
||||
#if(WYHASH_LITTLE_ENDIAN) || defined(__TINYC__)
|
||||
static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return v; }
|
||||
static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return v; }
|
||||
#if (WYHASH_LITTLE_ENDIAN)
|
||||
static inline uint64_t _wyread64(const uint8_t *p){ uint64_t v; memcpy(&v, p, 8); return v;}
|
||||
#else
|
||||
#if defined(__GNUC__) || defined(__INTEL_COMPILER)
|
||||
static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return __builtin_bswap64(v); }
|
||||
static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return __builtin_bswap32(v); }
|
||||
#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
|
||||
static inline uint64_t _wyread64(const uint8_t *p){ uint64_t v; memcpy(&v, p, 8); return __builtin_bswap64(v);}
|
||||
#elif defined(_MSC_VER)
|
||||
static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return _byteswap_uint64(v);}
|
||||
static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return _byteswap_ulong(v); }
|
||||
#endif
|
||||
#endif
|
||||
static inline uint64_t _wyr3(const uint8_t *p, unsigned k) { return (((uint64_t)p[0])<<16)|(((uint64_t)p[k>>1])<<8)|p[k-1]; }
|
||||
static inline uint64_t wyhash(const void* key, uint64_t len, uint64_t seed) {
|
||||
const uint8_t *p=(const uint8_t*)key; uint64_t i=len&63;
|
||||
#if defined(__GNUC__) || defined(__INTEL_COMPILER)
|
||||
#define _like_(x) __builtin_expect(x,1)
|
||||
#define _unlike_(x) __builtin_expect(x,0)
|
||||
static inline uint64_t _wyread64(const uint8_t *p){ uint64_t v; memcpy(&v, p, 8); return _byteswap_uint64(v);}
|
||||
#else
|
||||
#define _like_(x) (x)
|
||||
#define _unlike_(x) (x)
|
||||
static inline uint64_t _wyread64(const uint8_t *p){ uint64_t v; memcpy(&v, p, 8); return v;}
|
||||
#endif
|
||||
if(_unlike_(!i)) { }
|
||||
else if(_unlike_(i<4)) seed=_wymum(_wyr3(p,i)^seed^_wyp0,seed^_wyp1);
|
||||
else if(_like_(i<=8)) seed=_wymum(_wyr4(p)^seed^_wyp0,_wyr4(p+i-4)^seed^_wyp1);
|
||||
else if(_like_(i<=16)) seed=_wymum(_wyr8(p)^seed^_wyp0,_wyr8(p+i-8)^seed^_wyp1);
|
||||
else if(_like_(i<=24)) seed=_wymum(_wyr8(p)^seed^_wyp0,_wyr8(p+8)^seed^_wyp1)^_wymum(_wyr8(p+i-8)^seed^_wyp2,seed^_wyp3);
|
||||
else if(_like_(i<=32)) seed=_wymum(_wyr8(p)^seed^_wyp0,_wyr8(p+8)^seed^_wyp1)^_wymum(_wyr8(p+16)^seed^_wyp2,_wyr8(p+i-8)^seed^_wyp3);
|
||||
else{ seed=_wymum(_wyr8(p)^seed^_wyp0,_wyr8(p+8)^seed^_wyp1)^_wymum(_wyr8(p+16)^seed^_wyp2,_wyr8(p+24)^seed^_wyp3)^_wymum(_wyr8(p+i-32)^seed^_wyp1,_wyr8(p+i-24)^seed^_wyp2)^_wymum(_wyr8(p+i-16)^seed^_wyp3,_wyr8(p+i-8)^seed^_wyp0); }
|
||||
if(_like_(i==len)) return _wymum(seed,len^_wyp4);
|
||||
uint64_t see1=seed, see2=seed, see3=seed;
|
||||
for(p+=i,i=len-i; _like_(i>=64); i-=64,p+=64) {
|
||||
seed=_wymum(_wyr8(p)^seed^_wyp0,_wyr8(p+8)^seed^_wyp1); see1=_wymum(_wyr8(p+16)^see1^_wyp2,_wyr8(p+24)^see1^_wyp3);
|
||||
see2=_wymum(_wyr8(p+32)^see2^_wyp1,_wyr8(p+40)^see2^_wyp2); see3=_wymum(_wyr8(p+48)^see3^_wyp3,_wyr8(p+56)^see3^_wyp0);
|
||||
}
|
||||
return _wymum(seed^see1^see2,see3^len^_wyp4);
|
||||
}
|
||||
static inline uint64_t wyhash64(uint64_t A, uint64_t B) { return _wymum(_wymum(A^_wyp0, B^_wyp1), _wyp2); }
|
||||
static inline uint64_t wyrand(uint64_t *seed) { *seed+=_wyp0; return _wymum(*seed^_wyp1,*seed); }
|
||||
static inline double wy2u01(uint64_t r) { const double _wynorm=1.0/(1ull<<52); return (r>>11)*_wynorm; }
|
||||
static inline double wy2gau(uint64_t r) { const double _wynorm=1.0/(1ull<<20); return ((r&0x1fffff)+((r>>21)&0x1fffff)+((r>>42)&0x1fffff))*_wynorm-3.0; }
|
||||
static inline uint64_t fastest_hash(const void *key, size_t len, uint64_t seed) {
|
||||
#endif
|
||||
|
||||
static inline uint64_t _wyrot32(uint64_t x){ return (x>>32)|(x<<32); }
|
||||
|
||||
static inline uint64_t _wymix64(uint64_t x){ return x*_wyrot32(x); }
|
||||
|
||||
static inline uint64_t wyhash(const void *key, uint64_t len, uint64_t seed){
|
||||
const uint8_t *p=(const uint8_t *)key;
|
||||
return _like_(len >= 4) ? (_wyr4(p) + _wyr4(p + len - 4)) * (_wyr4(p + (len >> 1) - 2) ^ seed) : (_like_(len) ? _wyr3(p, len) * (_wyp0 ^ seed) : seed);
|
||||
uint64_t see1=seed, len0=len;
|
||||
for(;len>16;len-=16,p+=16){
|
||||
seed=_wymix64(_wyread64(p)^seed^_wyp0);
|
||||
see1=_wymix64(_wyread64(p+8)^see1^_wyp1);
|
||||
}
|
||||
//intended unsafe read, trade for great speed.
|
||||
uint64_t d0=_wyread64(p), d1=_wyread64(p+len-8);
|
||||
len=(len<8)*((8-len)<<3);
|
||||
d0<<=len; d1>>=len;
|
||||
seed=_wymix64(d0^seed^_wyp0);
|
||||
see1=_wymix64(d1^see1^_wyp1);
|
||||
return _wyrot32(_wymix64(len0^seed^see1))
|
||||
^_wymix64(_wyp1^_wyrot32(seed)^see1);
|
||||
}
|
||||
|
||||
static inline unsigned wyhash2(unsigned A, unsigned B){
|
||||
uint64_t c=(((uint64_t)A)<<32)|B;
|
||||
c=_wymix64(_wymix64(c^_wyp0));
|
||||
return (c>>32)^(unsigned)c;
|
||||
}
|
||||
|
||||
static inline unsigned wyrand(uint64_t *seed){
|
||||
*seed+=_wyp0;
|
||||
uint64_t x=_wymix64(*seed^_wyp1);
|
||||
return (x>>32)^(unsigned)x;
|
||||
}
|
||||
|
||||
static inline float wy2u01(unsigned r){
|
||||
const float _wynorm=1.0f/(1ull<<23);
|
||||
return (r>>9)*_wynorm;
|
||||
}
|
||||
|
||||
static inline float wy2gau(unsigned r){
|
||||
const float _wynorm=1.0f/(1ull<<9);
|
||||
return ((r&0x3ff)+((r>>10)&0x3ff)+((r>>20)&0x3ff))*_wynorm-3.0f;
|
||||
}
|
||||
#endif
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue