1 /* Copyright 2020 王一 Wang Yi <godspeed_china@yeah.net>
2 This is free and unencumbered software released into the public domain. http://unlicense.org/
3 See github.com/wangyi-fudan/wyhash/ LICENSE
4 */
5 #ifndef wyhash_final_version
6 #define wyhash_final_version
7 //defines that change behavior
8 #ifndef WYHASH_CONDOM
9 #define WYHASH_CONDOM 1 //0: read 8 bytes before and after boundaries, dangerous but fastest. 1: normal valid behavior 2: extra protection against entropy loss (probability=2^-63), aka. "blind multiplication"
10 #endif
11 #define WYHASH_32BIT_MUM 0 //faster on 32 bit system
12 //includes
13 #include <stdint.h>
14 #include <string.h>
15 #if defined(_MSC_VER) && defined(_M_X64)
16 #include <intrin.h>
17 #pragma intrinsic(_umul128)
18 #endif
19 #if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
20 #define _likely_(x) __builtin_expect(x,1)
21 #define _unlikely_(x) __builtin_expect(x,0)
22 #else
23 #define _likely_(x) (x)
24 #define _unlikely_(x) (x)
25 #endif
26 //mum function
_wyrot(uint64_t x)27 static inline uint64_t _wyrot(uint64_t x) { return (x>>32)|(x<<32); }
_wymum(uint64_t * A,uint64_t * B)28 static inline void _wymum(uint64_t *A, uint64_t *B){
29 #if(WYHASH_32BIT_MUM)
30 uint64_t hh=(*A>>32)*(*B>>32), hl=(*A>>32)*(unsigned)*B, lh=(unsigned)*A*(*B>>32), ll=(uint64_t)(unsigned)*A*(unsigned)*B;
31 #if(WYHASH_CONDOM>1)
32 *A^=_wyrot(hl)^hh; *B^=_wyrot(lh)^ll;
33 #else
34 *A=_wyrot(hl)^hh; *B=_wyrot(lh)^ll;
35 #endif
36 #elif defined(__SIZEOF_INT128__)
37 __uint128_t r=*A; r*=*B;
38 #if(WYHASH_CONDOM>1)
39 *A^=(uint64_t)r; *B^=(uint64_t)(r>>64);
40 #else
41 *A=(uint64_t)r; *B=(uint64_t)(r>>64);
42 #endif
43 #elif defined(_MSC_VER) && defined(_M_X64)
44 #if(WYHASH_CONDOM>1)
45 uint64_t a, b;
46 a=_umul128(*A,*B,&b);
47 *A^=a; *B^=b;
48 #else
49 *A=_umul128(*A,*B,B);
50 #endif
51 #else
52 uint64_t ha=*A>>32, hb=*B>>32, la=(uint32_t)*A, lb=(uint32_t)*B, hi, lo;
53 uint64_t rh=ha*hb, rm0=ha*lb, rm1=hb*la, rl=la*lb, t=rl+(rm0<<32), c=t<rl;
54 lo=t+(rm1<<32); c+=lo<t; hi=rh+(rm0>>32)+(rm1>>32)+c;
55 #if(WYHASH_CONDOM>1)
56 *A^=lo; *B^=hi;
57 #else
58 *A=lo; *B=hi;
59 #endif
60 #endif
61 }
_wymix(uint64_t A,uint64_t B)62 static inline uint64_t _wymix(uint64_t A, uint64_t B){ _wymum(&A,&B); return A^B; }
63 //read functions
64 #ifndef WYHASH_LITTLE_ENDIAN
65 #if defined(_WIN32) || defined(__LITTLE_ENDIAN__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
66 #define WYHASH_LITTLE_ENDIAN 1
67 #elif defined(__BIG_ENDIAN__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
68 #define WYHASH_LITTLE_ENDIAN 0
69 #endif
70 #endif
71 #if (WYHASH_LITTLE_ENDIAN)
_wyr8(const uint8_t * p)72 static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return v;}
_wyr4(const uint8_t * p)73 static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return v;}
74 #elif defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
_wyr8(const uint8_t * p)75 static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return __builtin_bswap64(v);}
_wyr4(const uint8_t * p)76 static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return __builtin_bswap32(v);}
77 #elif defined(_MSC_VER)
_wyr8(const uint8_t * p)78 static inline uint64_t _wyr8(const uint8_t *p) { uint64_t v; memcpy(&v, p, 8); return _byteswap_uint64(v);}
_wyr4(const uint8_t * p)79 static inline uint64_t _wyr4(const uint8_t *p) { unsigned v; memcpy(&v, p, 4); return _byteswap_ulong(v);}
80 #endif
_wyr3(const uint8_t * p,unsigned k)81 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];}
82 //wyhash function
_wyfinish16(const uint8_t * p,uint64_t len,uint64_t seed,const uint64_t * secret,uint64_t i)83 static inline uint64_t _wyfinish16(const uint8_t *p, uint64_t len, uint64_t seed, const uint64_t *secret, uint64_t i){
84 #if(WYHASH_CONDOM>0)
85 uint64_t a, b;
86 if(_likely_(i<=8)){
87 if(_likely_(i>=4)){ a=_wyr4(p); b=_wyr4(p+i-4); }
88 else if (_likely_(i)){ a=_wyr3(p,i); b=0; }
89 else a=b=0;
90 }
91 else{ a=_wyr8(p); b=_wyr8(p+i-8); }
92 return _wymix(secret[1]^len,_wymix(a^secret[1], b^seed));
93 #else
94 #define oneshot_shift ((i<8)*((8-i)<<3))
95 return _wymix(secret[1]^len,_wymix((_wyr8(p)<<oneshot_shift)^secret[1],(_wyr8(p+i-8)>>oneshot_shift)^seed));
96 #endif
97 }
98
_wyfinish(const uint8_t * p,uint64_t len,uint64_t seed,const uint64_t * secret,uint64_t i)99 static inline uint64_t _wyfinish(const uint8_t *p, uint64_t len, uint64_t seed, const uint64_t *secret, uint64_t i){
100 if(_likely_(i<=16)) return _wyfinish16(p,len,seed,secret,i);
101 return _wyfinish(p+16,len,_wymix(_wyr8(p)^secret[1],_wyr8(p+8)^seed),secret,i-16);
102 }
103
wyhash(const void * key,uint64_t len,uint64_t seed,const uint64_t * secret)104 static inline uint64_t wyhash(const void *key, uint64_t len, uint64_t seed, const uint64_t *secret){
105 const uint8_t *p=(const uint8_t *)key;
106 uint64_t i=len; seed^=*secret;
107 if(_unlikely_(i>64)){
108 uint64_t see1=seed;
109 do{
110 seed=_wymix(_wyr8(p)^secret[1],_wyr8(p+8)^seed)^_wymix(_wyr8(p+16)^secret[2],_wyr8(p+24)^seed);
111 see1=_wymix(_wyr8(p+32)^secret[3],_wyr8(p+40)^see1)^_wymix(_wyr8(p+48)^secret[4],_wyr8(p+56)^see1);
112 p+=64; i-=64;
113 }while(i>64);
114 seed^=see1;
115 }
116 return _wyfinish(p,len,seed,secret,i);
117 }
118 //utility functions
119 const uint64_t _wyp[5] = {0xa0761d6478bd642full, 0xe7037ed1a0b428dbull, 0x8ebc6af09c88c6e3ull, 0x589965cc75374cc3ull, 0x1d8e4e27c47d124full};
wyhash64(uint64_t A,uint64_t B)120 static inline uint64_t wyhash64(uint64_t A, uint64_t B){ A^=_wyp[0]; B^=_wyp[1]; _wymum(&A,&B); return _wymix(A^_wyp[0],B^_wyp[1]);}
wyrand(uint64_t * seed)121 static inline uint64_t wyrand(uint64_t *seed){ *seed+=_wyp[0]; return _wymix(*seed,*seed^_wyp[1]);}
wy2u01(uint64_t r)122 static inline double wy2u01(uint64_t r){ const double _wynorm=1.0/(1ull<<52); return (r>>12)*_wynorm;}
wy2gau(uint64_t r)123 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;}
wy2u0k(uint64_t r,uint64_t k)124 static inline uint64_t wy2u0k(uint64_t r, uint64_t k){ _wymum(&r,&k); return k; }
125
make_secret(uint64_t seed,uint64_t * secret)126 static inline void make_secret(uint64_t seed, uint64_t *secret){
127 uint8_t c[] = {15, 23, 27, 29, 30, 39, 43, 45, 46, 51, 53, 54, 57, 58, 60, 71, 75, 77, 78, 83, 85, 86, 89, 90, 92, 99, 101, 102, 105, 106, 108, 113, 114, 116, 120, 135, 139, 141, 142, 147, 149, 150, 153, 154, 156, 163, 165, 166, 169, 170, 172, 177, 178, 180, 184, 195, 197, 198, 201, 202, 204, 209, 210, 212, 216, 225, 226, 228, 232, 240 };
128 for(size_t i=0;i<5;i++){
129 uint8_t ok;
130 do{
131 ok=1; secret[i]=0;
132 for(size_t j=0;j<64;j+=8) secret[i]|=((uint64_t)c[wyrand(&seed)%sizeof(c)])<<j;
133 if(secret[i]%2==0){ ok=0; continue; }
134 for(size_t j=0;j<i;j++)
135 #if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
136 if(__builtin_popcountll(secret[j]^secret[i])!=32){ ok=0; break; }
137 #elif defined(_MSC_VER) && defined(_M_X64)
138 if(_mm_popcnt_u64(secret[j]^secret[i])!=32){ ok=0; break; }
139 #endif
140 if(!ok)continue;
141 for(uint64_t j=3;j<0x100000000ull;j+=2) if(secret[i]%j==0){ ok=0; break; }
142 }while(!ok);
143 }
144 }
145 #endif
146