source: trunk/sys/libupc/upc_c_randi8.c @ 66

Last change on this file since 66 was 1, checked in by alain, 8 years ago

First import

File size: 3.2 KB
RevLine 
[1]1/* RAND functions using 64b INTEGERs
2
3  F. CANTONNET - HPCL - GWU */
4
5double randlc (double *x, double a)
6{
7    /* This routine returns a uniform pseudorandom double precision number in the
8       range (0, 1) by using the linear congruential generator
9
10       x_{k+1} = a x_k  (mod 2^46)
11
12       where 0 < x_k < 2^46 and 0 < a < 2^46.  This scheme generates 2^44 numbers
13       before repeating.  The argument A is the same as 'a' in the above formula,
14       and X is the same as x_0.  A and X must be odd double precision integers
15       in the range (1, 2^46).  The returned value RANDLC is normalized to be
16       between 0 and 1, i.e. RANDLC = 2^(-46) * x_1.  X is updated to contain
17       the new seed x_1, so that subsequent calls to RANDLC using the same
18       arguments will generate a continuous sequence.
19
20       This routine should produce the same results on any computer with at least
21       48 mantissa bits in double precision floating point data.  On 64 bit
22       systems, double precision should be disabled.
23
24       David H. Bailey     October 26, 1990 */
25
26    unsigned long long i246m1, Lx, La;
27    double d2m46;
28
29    d2m46 = 0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
30        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
31        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
32        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
33        0.5*0.5*0.5*0.5*0.5*0.5;
34    //d2m46 = pow( 0.5, 46 );
35
36    i246m1 = 0x00003FFFFFFFFFFFLL;
37
38    Lx = *x;
39    La = a;
40
41    Lx = (Lx*La)&i246m1;
42    *x = (double) Lx;
43    return (d2m46 * (*x));
44}
45
46void vranlc (int n, double *x_seed, double a, double y[]) {
47    /* This routine generates N uniform pseudorandom double precision numbers in
48       the range (0, 1) by using the linear congruential generator
49
50       x_{k+1} = a x_k  (mod 2^46)
51
52       where 0 < x_k < 2^46 and 0 < a < 2^46.  This scheme generates 2^44 numbers
53       before repeating.  The argument A is the same as 'a' in the above formula,
54       and X is the same as x_0.  A and X must be odd double precision integers
55       in the range (1, 2^46).  The N results are placed in Y and are normalized
56       to be between 0 and 1.  X is updated to contain the new seed, so that
57       subsequent calls to VRANLC using the same arguments will generate a
58       continuous sequence.  If N is zero, only initialization is performed, and
59       the variables X, A and Y are ignored.
60
61       This routine is the standard version designed for scalar or RISC systems.
62       However, it should produce the same results on any single processor
63       computer with at least 48 mantissa bits in double precision floating point
64       data.  On 64 bit systems, double precision should be disabled.
65     */
66    int i;
67    double x;
68    unsigned long long i246m1, Lx, La;
69    double d2m46;
70
71    d2m46 = 0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
72        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
73        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
74        0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*0.5*
75        0.5*0.5*0.5*0.5*0.5*0.5;
76    //  d2m46 = pow( 0.5, 46.0 );
77    i246m1 = 0x00003FFFFFFFFFFFLL;
78
79    x = *x_seed;
80    Lx = x;
81    La = a;
82
83    for (i = 1; i <= n; i++)
84    {
85        Lx = ((Lx*La)&i246m1);
86        x = (double) Lx;
87        y[i] = d2m46 * x;
88    }
89    *x_seed = x;
90}
Note: See TracBrowser for help on using the repository browser.