1 | //
|
---|
2 | // Cforall Version 1.0.0 Copyright (C) 2022 University of Waterloo
|
---|
3 | //
|
---|
4 | // The contents of this file are covered under the licence agreement in the
|
---|
5 | // file "LICENCE" distributed with Cforall.
|
---|
6 | //
|
---|
7 | // random.hfa --
|
---|
8 | //
|
---|
9 | // Author : Peter A. Buhr
|
---|
10 | // Created On : Fri Jan 14 07:18:11 2022
|
---|
11 | // Last Modified By : Peter A. Buhr
|
---|
12 | // Last Modified On : Thu Dec 1 11:52:08 2022
|
---|
13 | // Update Count : 115
|
---|
14 | //
|
---|
15 |
|
---|
16 | #pragma once
|
---|
17 |
|
---|
18 | #include <stdint.h>
|
---|
19 |
|
---|
20 | #define GLUE2( x, y ) x##y
|
---|
21 | #define GLUE( x, y ) GLUE2( x, y )
|
---|
22 |
|
---|
23 | // Set default PRNG for architecture size.
|
---|
24 | #ifdef __x86_64__ // 64-bit architecture
|
---|
25 | #define LEHMER64
|
---|
26 | #define XORSHIFT_6_21_7
|
---|
27 | //#define XOSHIRO256PP
|
---|
28 | //#define XOSHIRO128PP
|
---|
29 | #else // 32-bit architecture
|
---|
30 | #define XORSHIFT_13_7_17
|
---|
31 | #define XORSHIFT_6_21_7
|
---|
32 | //#define XOSHIRO256PP
|
---|
33 | //#define XOSHIRO128PP
|
---|
34 | #endif // __x86_64__
|
---|
35 |
|
---|
36 | // C/CFA PRNG name and random-state.
|
---|
37 |
|
---|
38 | #ifdef LEHMER64
|
---|
39 | #define PRNG_NAME_64 lehmer64
|
---|
40 | #define PRNG_STATE_64_T __uint128_t
|
---|
41 | #endif // LEHMER64
|
---|
42 |
|
---|
43 | #ifdef XORSHIFT_13_7_17
|
---|
44 | #define PRNG_NAME_64 xorshift_13_7_17
|
---|
45 | #define PRNG_STATE_64_T uint64_t
|
---|
46 | #endif // XORSHIFT_13_7_17
|
---|
47 |
|
---|
48 | #ifdef XORSHIFT_6_21_7
|
---|
49 | #define PRNG_NAME_32 xorshift_6_21_7
|
---|
50 | #define PRNG_STATE_32_T uint32_t
|
---|
51 | #endif // XORSHIFT_6_21_7
|
---|
52 |
|
---|
53 | #ifdef XOSHIRO256PP
|
---|
54 | #define PRNG_NAME_64 xoshiro256pp
|
---|
55 | #define PRNG_STATE_64_T struct GLUE(PRNG_NAME_64,_t)
|
---|
56 | PRNG_STATE_64_T { uint64_t s[4]; };
|
---|
57 | #endif // XOSHIRO256PP
|
---|
58 |
|
---|
59 | #ifdef XOSHIRO128PP
|
---|
60 | #define PRNG_NAME_32 xoshiro128pp
|
---|
61 | #define PRNG_STATE_32_T struct GLUE(PRNG_NAME_32,_t)
|
---|
62 | PRNG_STATE_32_T { uint32_t s[4]; };
|
---|
63 | #endif // XOSHIRO128PP
|
---|
64 |
|
---|
65 | #define PRNG_SET_SEED_64 GLUE(PRNG_NAME_64,_set_seed)
|
---|
66 | #define PRNG_SET_SEED_32 GLUE(PRNG_NAME_32,_set_seed)
|
---|
67 |
|
---|
68 |
|
---|
69 | // Default PRNG used by runtime.
|
---|
70 | #ifdef __x86_64__ // 64-bit architecture
|
---|
71 | #define PRNG_NAME PRNG_NAME_64
|
---|
72 | #define PRNG_STATE_T PRNG_STATE_64_T
|
---|
73 | #else // 32-bit architecture
|
---|
74 | #define PRNG_NAME PRNG_NAME_32
|
---|
75 | #define PRNG_STATE_T PRNG_STATE_32_T
|
---|
76 | #endif // __x86_64__
|
---|
77 |
|
---|
78 | #define PRNG_SET_SEED GLUE(PRNG_NAME,_set_seed)
|
---|
79 |
|
---|
80 |
|
---|
81 | #ifdef __cforall // don't include in C code (invoke.h)
|
---|
82 |
|
---|
83 | // https://prng.di.unimi.it/xoshiro128plusplus.c
|
---|
84 | //
|
---|
85 | // This is xoshiro128++ 1.0, one of our 32-bit all-purpose, rock-solid generators. It has excellent speed, a state size
|
---|
86 | // (128 bits) that is large enough for mild parallelism, and it passes all tests we are aware of.
|
---|
87 | //
|
---|
88 | // For generating just single-precision (i.e., 32-bit) floating-point numbers, xoshiro128+ is even faster.
|
---|
89 | //
|
---|
90 | // The state must be seeded so that it is not everywhere zero.
|
---|
91 |
|
---|
92 | #ifndef XOSHIRO128PP
|
---|
93 | struct xoshiro128pp_t { uint32_t s[4]; };
|
---|
94 | #endif // ! XOSHIRO128PP
|
---|
95 |
|
---|
96 | static inline uint32_t xoshiro128pp( xoshiro128pp_t & rs ) with(rs) {
|
---|
97 | inline uint32_t rotl( const uint32_t x, int k ) {
|
---|
98 | return (x << k) | (x >> (32 - k));
|
---|
99 | }
|
---|
100 |
|
---|
101 | const uint32_t result = rotl( s[0] + s[3], 7 ) + s[0];
|
---|
102 | const uint32_t t = s[1] << 9;
|
---|
103 |
|
---|
104 | s[2] ^= s[0];
|
---|
105 | s[3] ^= s[1];
|
---|
106 | s[1] ^= s[2];
|
---|
107 | s[0] ^= s[3];
|
---|
108 | s[2] ^= t;
|
---|
109 | s[3] = rotl( s[3], 11 );
|
---|
110 | return result;
|
---|
111 | }
|
---|
112 |
|
---|
113 | static inline void xoshiro128pp_set_seed( xoshiro128pp_t & state, uint32_t seed ) {
|
---|
114 | state = (xoshiro128pp_t){ {seed, seed, seed, seed} };
|
---|
115 | } // xoshiro128pp_set_seed
|
---|
116 |
|
---|
117 | // This is xoshiro256++ 1.0, one of our all-purpose, rock-solid generators. It has excellent (sub-ns) speed, a state
|
---|
118 | // (256 bits) that is large enough for any parallel application, and it passes all tests we are aware of.
|
---|
119 | //
|
---|
120 | // For generating just floating-point numbers, xoshiro256+ is even faster.
|
---|
121 | //
|
---|
122 | // The state must be seeded so that it is not everywhere zero. If you have a 64-bit seed, we suggest to seed a
|
---|
123 | // splitmix64 generator and use its output to fill s.
|
---|
124 |
|
---|
125 | #ifndef XOSHIRO256PP
|
---|
126 | struct xoshiro256pp_t { uint64_t s[4]; };
|
---|
127 | #endif // ! XOSHIRO256PP
|
---|
128 |
|
---|
129 | static inline uint64_t xoshiro256pp( xoshiro256pp_t & rs ) with(rs) {
|
---|
130 | inline uint64_t rotl(const uint64_t x, int k) {
|
---|
131 | return (x << k) | (x >> (64 - k));
|
---|
132 | }
|
---|
133 |
|
---|
134 | const uint64_t result = rotl( s[0] + s[3], 23 ) + s[0];
|
---|
135 | const uint64_t t = s[1] << 17;
|
---|
136 |
|
---|
137 | s[2] ^= s[0];
|
---|
138 | s[3] ^= s[1];
|
---|
139 | s[1] ^= s[2];
|
---|
140 | s[0] ^= s[3];
|
---|
141 | s[2] ^= t;
|
---|
142 | s[3] = rotl( s[3], 45 );
|
---|
143 | return result;
|
---|
144 | }
|
---|
145 |
|
---|
146 | static inline void xoshiro256pp_set_seed( xoshiro256pp_t & state, uint64_t seed ) {
|
---|
147 | state = (xoshiro256pp_t){ {seed, seed, seed, seed} };
|
---|
148 | } // xoshiro256pp_set_seed
|
---|
149 |
|
---|
150 | #ifdef __SIZEOF_INT128__
|
---|
151 | // Pipelined to allow out-of-order overlap with reduced dependencies. Critically, the current random state is
|
---|
152 | // returned (copied), and then compute and store the next random value.
|
---|
153 | //--------------------------------------------------
|
---|
154 | static inline uint64_t lehmer64( __uint128_t & state ) {
|
---|
155 | __uint128_t ret = state;
|
---|
156 | state *= 0xda942042e4dd58b5;
|
---|
157 | return ret >> 64;
|
---|
158 | } // lehmer64
|
---|
159 |
|
---|
160 | static inline void lehmer64_set_seed( __uint128_t & state, uint64_t seed ) {
|
---|
161 | state = seed;
|
---|
162 | } // lehmer64_set_seed
|
---|
163 |
|
---|
164 | //--------------------------------------------------
|
---|
165 | static inline uint64_t wyhash64( uint64_t & state ) {
|
---|
166 | state += 0x60bee2bee120fc15;
|
---|
167 | __uint128_t tmp;
|
---|
168 | tmp = (__uint128_t) state * 0xa3b195354a39b70d;
|
---|
169 | uint64_t m1 = (tmp >> 64) ^ tmp;
|
---|
170 | tmp = (__uint128_t)m1 * 0x1b03738712fad5c9;
|
---|
171 | uint64_t m2 = (tmp >> 64) ^ tmp;
|
---|
172 | return m2;
|
---|
173 | }
|
---|
174 |
|
---|
175 | static inline void wyhash64_set_seed( __uint128_t & state, uint64_t seed ) {
|
---|
176 | state = seed;
|
---|
177 | } // lehmer64_set_seed
|
---|
178 | #endif // __SIZEOF_INT128__
|
---|
179 |
|
---|
180 | //--------------------------------------------------
|
---|
181 | static inline uint64_t xorshift_13_7_17( uint64_t & state ) {
|
---|
182 | uint64_t ret = state;
|
---|
183 | state ^= state << 13;
|
---|
184 | state ^= state >> 7;
|
---|
185 | state ^= state << 17;
|
---|
186 | return ret;
|
---|
187 | }
|
---|
188 |
|
---|
189 | static inline void xorshift_13_7_17_set_seed( uint64_t & state, uint32_t seed ) {
|
---|
190 | state = seed;
|
---|
191 | }
|
---|
192 |
|
---|
193 | //--------------------------------------------------
|
---|
194 | static inline uint32_t xorshift_6_21_7( uint32_t & state ) {
|
---|
195 | uint32_t ret = state;
|
---|
196 | state ^= state << 6;
|
---|
197 | state ^= state >> 21;
|
---|
198 | state ^= state << 7;
|
---|
199 | return ret;
|
---|
200 | } // xorshift_6_21_7
|
---|
201 |
|
---|
202 | static inline void xorshift_6_21_7_set_seed( uint32_t & state, uint32_t seed ) {
|
---|
203 | state = seed;
|
---|
204 | }
|
---|
205 |
|
---|
206 | //--------------------------------------------------
|
---|
207 | typedef struct {
|
---|
208 | uint32_t a, b, c, d;
|
---|
209 | uint32_t counter;
|
---|
210 | } xorwow__state_t;
|
---|
211 |
|
---|
212 | // The state array must be initialized to not be all zero in the first four words.
|
---|
213 | static inline uint32_t xorwow( xorwow__state_t & state ) {
|
---|
214 | // Algorithm "xorwow" from p. 5 of Marsaglia, "Xorshift RNGs".
|
---|
215 | uint32_t ret = state.a + state.counter;
|
---|
216 | uint32_t t = state.d;
|
---|
217 |
|
---|
218 | uint32_t const s = state.a;
|
---|
219 | state.d = state.c;
|
---|
220 | state.c = state.b;
|
---|
221 | state.b = s;
|
---|
222 |
|
---|
223 | t ^= t >> 2;
|
---|
224 | t ^= t << 1;
|
---|
225 | t ^= s ^ (s << 4);
|
---|
226 | state.a = t;
|
---|
227 |
|
---|
228 | state.counter += 362437;
|
---|
229 | return ret;
|
---|
230 | }
|
---|
231 |
|
---|
232 | // Used in __tls_rand_fwd
|
---|
233 | //--------------------------------------------------
|
---|
234 | #define M (1_l64u << 48_l64u)
|
---|
235 | #define A (25214903917_l64u)
|
---|
236 | #define AI (18446708753438544741_l64u)
|
---|
237 | #define C (11_l64u)
|
---|
238 | #define D (16_l64u)
|
---|
239 |
|
---|
240 | // Bi-directional LCG random-number generator
|
---|
241 | static inline uint32_t LCGBI_fwd( uint64_t & state ) {
|
---|
242 | state = (A * state + C) & (M - 1);
|
---|
243 | return state >> D;
|
---|
244 | }
|
---|
245 |
|
---|
246 | static inline uint32_t LCGBI_bck( uint64_t & state ) {
|
---|
247 | unsigned int r = state >> D;
|
---|
248 | state = AI * (state - C) & (M - 1);
|
---|
249 | return r;
|
---|
250 | }
|
---|
251 |
|
---|
252 | #undef M
|
---|
253 | #undef A
|
---|
254 | #undef AI
|
---|
255 | #undef C
|
---|
256 | #undef D
|
---|
257 |
|
---|
258 | #endif // __cforall
|
---|