view src/SFMT/SFMT-alti.h @ 135:eb6ba88088d3

Wrote a more efficient input-number-assembly program; 91 oc -> 60 oc.
author Dylan Holmes <ocsenave@gmail.com>
date Sun, 18 Mar 2012 05:13:19 -0500
parents f9f4f1b99eed
children
line wrap: on
line source
1 /**
2 * @file SFMT-alti.h
3 *
4 * @brief SIMD oriented Fast Mersenne Twister(SFMT)
5 * pseudorandom number generator
6 *
7 * @author Mutsuo Saito (Hiroshima University)
8 * @author Makoto Matsumoto (Hiroshima University)
9 *
10 * Copyright (C) 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
11 * University. All rights reserved.
12 *
13 * The new BSD License is applied to this software.
14 * see LICENSE.txt
15 */
17 #ifndef SFMT_ALTI_H
18 #define SFMT_ALTI_H
20 inline static vector unsigned int vec_recursion(vector unsigned int a,
21 vector unsigned int b,
22 vector unsigned int c,
23 vector unsigned int d)
24 ALWAYSINLINE;
26 /**
27 * This function represents the recursion formula in AltiVec and BIG ENDIAN.
28 * @param a a 128-bit part of the interal state array
29 * @param b a 128-bit part of the interal state array
30 * @param c a 128-bit part of the interal state array
31 * @param d a 128-bit part of the interal state array
32 * @return output
33 */
34 inline static vector unsigned int vec_recursion(vector unsigned int a,
35 vector unsigned int b,
36 vector unsigned int c,
37 vector unsigned int d) {
39 const vector unsigned int sl1 = ALTI_SL1;
40 const vector unsigned int sr1 = ALTI_SR1;
41 #ifdef ONLY64
42 const vector unsigned int mask = ALTI_MSK64;
43 const vector unsigned char perm_sl = ALTI_SL2_PERM64;
44 const vector unsigned char perm_sr = ALTI_SR2_PERM64;
45 #else
46 const vector unsigned int mask = ALTI_MSK;
47 const vector unsigned char perm_sl = ALTI_SL2_PERM;
48 const vector unsigned char perm_sr = ALTI_SR2_PERM;
49 #endif
50 vector unsigned int v, w, x, y, z;
51 x = vec_perm(a, (vector unsigned int)perm_sl, perm_sl);
52 v = a;
53 y = vec_sr(b, sr1);
54 z = vec_perm(c, (vector unsigned int)perm_sr, perm_sr);
55 w = vec_sl(d, sl1);
56 z = vec_xor(z, w);
57 y = vec_and(y, mask);
58 v = vec_xor(v, x);
59 z = vec_xor(z, y);
60 z = vec_xor(z, v);
61 return z;
62 }
64 /**
65 * This function fills the internal state array with pseudorandom
66 * integers.
67 */
68 inline static void gen_rand_all(void) {
69 int i;
70 vector unsigned int r, r1, r2;
72 r1 = sfmt[N - 2].s;
73 r2 = sfmt[N - 1].s;
74 for (i = 0; i < N - POS1; i++) {
75 r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
76 sfmt[i].s = r;
77 r1 = r2;
78 r2 = r;
79 }
80 for (; i < N; i++) {
81 r = vec_recursion(sfmt[i].s, sfmt[i + POS1 - N].s, r1, r2);
82 sfmt[i].s = r;
83 r1 = r2;
84 r2 = r;
85 }
86 }
88 /**
89 * This function fills the user-specified array with pseudorandom
90 * integers.
91 *
92 * @param array an 128-bit array to be filled by pseudorandom numbers.
93 * @param size number of 128-bit pesudorandom numbers to be generated.
94 */
95 inline static void gen_rand_array(w128_t *array, int size) {
96 int i, j;
97 vector unsigned int r, r1, r2;
99 r1 = sfmt[N - 2].s;
100 r2 = sfmt[N - 1].s;
101 for (i = 0; i < N - POS1; i++) {
102 r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
103 array[i].s = r;
104 r1 = r2;
105 r2 = r;
106 }
107 for (; i < N; i++) {
108 r = vec_recursion(sfmt[i].s, array[i + POS1 - N].s, r1, r2);
109 array[i].s = r;
110 r1 = r2;
111 r2 = r;
112 }
113 /* main loop */
114 for (; i < size - N; i++) {
115 r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2);
116 array[i].s = r;
117 r1 = r2;
118 r2 = r;
119 }
120 for (j = 0; j < 2 * N - size; j++) {
121 sfmt[j].s = array[j + size - N].s;
122 }
123 for (; i < size; i++) {
124 r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2);
125 array[i].s = r;
126 sfmt[j++].s = r;
127 r1 = r2;
128 r2 = r;
129 }
130 }
132 #ifndef ONLY64
133 #if defined(__APPLE__)
134 #define ALTI_SWAP (vector unsigned char) \
135 (4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11)
136 #else
137 #define ALTI_SWAP {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11}
138 #endif
139 /**
140 * This function swaps high and low 32-bit of 64-bit integers in user
141 * specified array.
142 *
143 * @param array an 128-bit array to be swaped.
144 * @param size size of 128-bit array.
145 */
146 inline static void swap(w128_t *array, int size) {
147 int i;
148 const vector unsigned char perm = ALTI_SWAP;
150 for (i = 0; i < size; i++) {
151 array[i].s = vec_perm(array[i].s, (vector unsigned int)perm, perm);
152 }
153 }
154 #endif
156 #endif