[e57de69] | 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 |
---|
[9fce2572] | 12 | // Last Modified On : Mon Nov 21 17:50:12 2022 |
---|
| 13 | // Update Count : 15 |
---|
[e57de69] | 14 | // |
---|
| 15 | |
---|
[13c5e19] | 16 | #pragma once |
---|
| 17 | |
---|
| 18 | #include <stdint.h> |
---|
| 19 | |
---|
[9fce2572] | 20 | // Set default PRNG for architecture size. |
---|
[d2ad151] | 21 | #ifdef __x86_64__ // 64-bit architecture |
---|
| 22 | #define LEHMER64 |
---|
| 23 | #else // 32-bit architecture |
---|
| 24 | #define XORSHIFT_6_21_7 |
---|
| 25 | #endif // __x86_64__ |
---|
| 26 | |
---|
[9fce2572] | 27 | // C/CFA PRNG name and random-state. |
---|
| 28 | #ifdef LEHMER64 |
---|
| 29 | #define PRNG_NAME lehmer64 |
---|
| 30 | #define PRNG_ARG_T __uint128_t |
---|
| 31 | #endif // LEHMER64 |
---|
| 32 | |
---|
| 33 | #ifdef XORSHIFT_6_21_7 |
---|
| 34 | #define PRNG_NAME xorshift_6_21_7 |
---|
| 35 | #define PRNG_ARG_T uint32_t |
---|
| 36 | #endif // XORSHIFT_6_21_7 |
---|
| 37 | |
---|
| 38 | #ifdef __cforall // don't include in C code (invoke.h) |
---|
| 39 | |
---|
[e57de69] | 40 | // Pipelined to allow out-of-order overlap with reduced dependencies. Critically, the current random state is returned |
---|
| 41 | // (copied), and then compute and store the next random value. |
---|
[611f29d] | 42 | |
---|
[7812a7b5] | 43 | #if defined(__SIZEOF_INT128__) |
---|
[e57de69] | 44 | //-------------------------------------------------- |
---|
[611f29d] | 45 | static inline uint64_t lehmer64( __uint128_t & state ) { |
---|
| 46 | __uint128_t ret = state; |
---|
[7812a7b5] | 47 | state *= 0xda942042e4dd58b5; |
---|
[611f29d] | 48 | return ret >> 64; |
---|
[7812a7b5] | 49 | } |
---|
[13c5e19] | 50 | |
---|
| 51 | //-------------------------------------------------- |
---|
[611f29d] | 52 | static inline uint64_t wyhash64( uint64_t & state ) { |
---|
[7812a7b5] | 53 | state += 0x60bee2bee120fc15; |
---|
| 54 | __uint128_t tmp; |
---|
| 55 | tmp = (__uint128_t) state * 0xa3b195354a39b70d; |
---|
| 56 | uint64_t m1 = (tmp >> 64) ^ tmp; |
---|
| 57 | tmp = (__uint128_t)m1 * 0x1b03738712fad5c9; |
---|
| 58 | uint64_t m2 = (tmp >> 64) ^ tmp; |
---|
| 59 | return m2; |
---|
| 60 | } |
---|
| 61 | #endif |
---|
[13c5e19] | 62 | |
---|
| 63 | //-------------------------------------------------- |
---|
[611f29d] | 64 | static inline uint64_t xorshift_13_7_17( uint64_t & state ) { |
---|
| 65 | uint64_t ret = state; |
---|
| 66 | state ^= state << 13; |
---|
| 67 | state ^= state >> 7; |
---|
| 68 | state ^= state << 17; |
---|
| 69 | return ret; |
---|
[13c5e19] | 70 | } |
---|
| 71 | |
---|
[611f29d] | 72 | //-------------------------------------------------- |
---|
| 73 | static inline uint32_t xorshift_6_21_7( uint32_t & state ) { |
---|
| 74 | uint32_t ret = state; |
---|
| 75 | state ^= state << 6; |
---|
| 76 | state ^= state >> 21; |
---|
| 77 | state ^= state << 7; |
---|
| 78 | return ret; |
---|
| 79 | } // xorshift_6_21_7 |
---|
| 80 | |
---|
[13c5e19] | 81 | //-------------------------------------------------- |
---|
| 82 | typedef struct { |
---|
[9fce2572] | 83 | uint32_t a, b, c, d; |
---|
| 84 | uint32_t counter; |
---|
[611f29d] | 85 | } xorwow__state_t; |
---|
[13c5e19] | 86 | |
---|
[e57de69] | 87 | // The state array must be initialized to not be all zero in the first four words. |
---|
[611f29d] | 88 | static inline uint32_t xorwow( xorwow__state_t & state ) { |
---|
[e57de69] | 89 | // Algorithm "xorwow" from p. 5 of Marsaglia, "Xorshift RNGs". |
---|
[611f29d] | 90 | uint32_t ret = state.a + state.counter; |
---|
[13c5e19] | 91 | uint32_t t = state.d; |
---|
| 92 | |
---|
| 93 | uint32_t const s = state.a; |
---|
| 94 | state.d = state.c; |
---|
| 95 | state.c = state.b; |
---|
| 96 | state.b = s; |
---|
| 97 | |
---|
| 98 | t ^= t >> 2; |
---|
| 99 | t ^= t << 1; |
---|
| 100 | t ^= s ^ (s << 4); |
---|
| 101 | state.a = t; |
---|
| 102 | |
---|
| 103 | state.counter += 362437; |
---|
[611f29d] | 104 | return ret; |
---|
| 105 | } |
---|
| 106 | |
---|
| 107 | //-------------------------------------------------- |
---|
| 108 | static inline uint32_t LCG( uint32_t & state ) { // linear congruential generator |
---|
| 109 | uint32_t ret = state; |
---|
| 110 | state = 36969 * (state & 65535) + (state >> 16); // 36969 is NOT prime! No not change it! |
---|
| 111 | return ret; |
---|
| 112 | } // LCG |
---|
| 113 | |
---|
| 114 | //-------------------------------------------------- |
---|
| 115 | #define M (1_l64u << 48_l64u) |
---|
| 116 | #define A (25214903917_l64u) |
---|
| 117 | #define AI (18446708753438544741_l64u) |
---|
| 118 | #define C (11_l64u) |
---|
| 119 | #define D (16_l64u) |
---|
| 120 | |
---|
[e57de69] | 121 | // Bi-directional LCG random-number generator |
---|
[611f29d] | 122 | static inline uint32_t LCGBI_fwd( uint64_t & state ) { |
---|
| 123 | state = (A * state + C) & (M - 1); |
---|
| 124 | return state >> D; |
---|
[4177592f] | 125 | } |
---|
[611f29d] | 126 | |
---|
| 127 | static inline uint32_t LCGBI_bck( uint64_t & state ) { |
---|
| 128 | unsigned int r = state >> D; |
---|
| 129 | state = AI * (state - C) & (M - 1); |
---|
| 130 | return r; |
---|
| 131 | } |
---|
| 132 | |
---|
| 133 | #undef M |
---|
| 134 | #undef A |
---|
| 135 | #undef AI |
---|
| 136 | #undef C |
---|
| 137 | #undef D |
---|
[9fce2572] | 138 | |
---|
| 139 | #endif // __cforall |
---|