diff src/SFMT/SFMT.c @ 23:bf9169ad4222

add SMID-oriented fast mersenne twister
author Robert McIntyre <rlm@mit.edu>
date Sun, 04 Mar 2012 17:38:32 -0600
parents f9f4f1b99eed
children
line wrap: on
line diff
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/SFMT/SFMT.c	Sun Mar 04 17:38:32 2012 -0600
     1.3 @@ -0,0 +1,620 @@
     1.4 +/** 
     1.5 + * @file  SFMT.c
     1.6 + * @brief SIMD oriented Fast Mersenne Twister(SFMT)
     1.7 + *
     1.8 + * @author Mutsuo Saito (Hiroshima University)
     1.9 + * @author Makoto Matsumoto (Hiroshima University)
    1.10 + *
    1.11 + * Copyright (C) 2006,2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
    1.12 + * University. All rights reserved.
    1.13 + *
    1.14 + * The new BSD License is applied to this software, see LICENSE.txt
    1.15 + */
    1.16 +#include <string.h>
    1.17 +#include <assert.h>
    1.18 +#include "SFMT.h"
    1.19 +#include "SFMT-params.h"
    1.20 +
    1.21 +#if defined(__BIG_ENDIAN__) && !defined(__amd64) && !defined(BIG_ENDIAN64)
    1.22 +#define BIG_ENDIAN64 1
    1.23 +#endif
    1.24 +#if defined(HAVE_ALTIVEC) && !defined(BIG_ENDIAN64)
    1.25 +#define BIG_ENDIAN64 1
    1.26 +#endif
    1.27 +#if defined(ONLY64) && !defined(BIG_ENDIAN64)
    1.28 +  #if defined(__GNUC__)
    1.29 +    #error "-DONLY64 must be specified with -DBIG_ENDIAN64"
    1.30 +  #endif
    1.31 +#undef ONLY64
    1.32 +#endif
    1.33 +/*------------------------------------------------------
    1.34 +  128-bit SIMD data type for Altivec, SSE2 or standard C
    1.35 +  ------------------------------------------------------*/
    1.36 +#if defined(HAVE_ALTIVEC)
    1.37 +  #if !defined(__APPLE__)
    1.38 +    #include <altivec.h>
    1.39 +  #endif
    1.40 +/** 128-bit data structure */
    1.41 +union W128_T {
    1.42 +    vector unsigned int s;
    1.43 +    uint32_t u[4];
    1.44 +};
    1.45 +/** 128-bit data type */
    1.46 +typedef union W128_T w128_t;
    1.47 +
    1.48 +#elif defined(HAVE_SSE2)
    1.49 +  #include <emmintrin.h>
    1.50 +
    1.51 +/** 128-bit data structure */
    1.52 +union W128_T {
    1.53 +    __m128i si;
    1.54 +    uint32_t u[4];
    1.55 +};
    1.56 +/** 128-bit data type */
    1.57 +typedef union W128_T w128_t;
    1.58 +
    1.59 +#else
    1.60 +
    1.61 +/** 128-bit data structure */
    1.62 +struct W128_T {
    1.63 +    uint32_t u[4];
    1.64 +};
    1.65 +/** 128-bit data type */
    1.66 +typedef struct W128_T w128_t;
    1.67 +
    1.68 +#endif
    1.69 +
    1.70 +/*--------------------------------------
    1.71 +  FILE GLOBAL VARIABLES
    1.72 +  internal state, index counter and flag 
    1.73 +  --------------------------------------*/
    1.74 +/** the 128-bit internal state array */
    1.75 +static w128_t sfmt[N];
    1.76 +/** the 32bit integer pointer to the 128-bit internal state array */
    1.77 +static uint32_t *psfmt32 = &sfmt[0].u[0];
    1.78 +#if !defined(BIG_ENDIAN64) || defined(ONLY64)
    1.79 +/** the 64bit integer pointer to the 128-bit internal state array */
    1.80 +static uint64_t *psfmt64 = (uint64_t *)&sfmt[0].u[0];
    1.81 +#endif
    1.82 +/** index counter to the 32-bit internal state array */
    1.83 +static int idx;
    1.84 +/** a flag: it is 0 if and only if the internal state is not yet
    1.85 + * initialized. */
    1.86 +static int initialized = 0;
    1.87 +/** a parity check vector which certificate the period of 2^{MEXP} */
    1.88 +static uint32_t parity[4] = {PARITY1, PARITY2, PARITY3, PARITY4};
    1.89 +
    1.90 +/*----------------
    1.91 +  STATIC FUNCTIONS
    1.92 +  ----------------*/
    1.93 +inline static int idxof(int i);
    1.94 +inline static void rshift128(w128_t *out,  w128_t const *in, int shift);
    1.95 +inline static void lshift128(w128_t *out,  w128_t const *in, int shift);
    1.96 +inline static void gen_rand_all(void);
    1.97 +inline static void gen_rand_array(w128_t *array, int size);
    1.98 +inline static uint32_t func1(uint32_t x);
    1.99 +inline static uint32_t func2(uint32_t x);
   1.100 +static void period_certification(void);
   1.101 +#if defined(BIG_ENDIAN64) && !defined(ONLY64)
   1.102 +inline static void swap(w128_t *array, int size);
   1.103 +#endif
   1.104 +
   1.105 +#if defined(HAVE_ALTIVEC)
   1.106 +  #include "SFMT-alti.h"
   1.107 +#elif defined(HAVE_SSE2)
   1.108 +  #include "SFMT-sse2.h"
   1.109 +#endif
   1.110 +
   1.111 +/**
   1.112 + * This function simulate a 64-bit index of LITTLE ENDIAN 
   1.113 + * in BIG ENDIAN machine.
   1.114 + */
   1.115 +#ifdef ONLY64
   1.116 +inline static int idxof(int i) {
   1.117 +    return i ^ 1;
   1.118 +}
   1.119 +#else
   1.120 +inline static int idxof(int i) {
   1.121 +    return i;
   1.122 +}
   1.123 +#endif
   1.124 +/**
   1.125 + * This function simulates SIMD 128-bit right shift by the standard C.
   1.126 + * The 128-bit integer given in in is shifted by (shift * 8) bits.
   1.127 + * This function simulates the LITTLE ENDIAN SIMD.
   1.128 + * @param out the output of this function
   1.129 + * @param in the 128-bit data to be shifted
   1.130 + * @param shift the shift value
   1.131 + */
   1.132 +#ifdef ONLY64
   1.133 +inline static void rshift128(w128_t *out, w128_t const *in, int shift) {
   1.134 +    uint64_t th, tl, oh, ol;
   1.135 +
   1.136 +    th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]);
   1.137 +    tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]);
   1.138 +
   1.139 +    oh = th >> (shift * 8);
   1.140 +    ol = tl >> (shift * 8);
   1.141 +    ol |= th << (64 - shift * 8);
   1.142 +    out->u[0] = (uint32_t)(ol >> 32);
   1.143 +    out->u[1] = (uint32_t)(ol & 0xffffffff);
   1.144 +    out->u[2] = (uint32_t)(oh >> 32);
   1.145 +    out->u[3] = (uint32_t)(oh & 0xffffffff);
   1.146 +}
   1.147 +#else
   1.148 +inline static void rshift128(w128_t *out, w128_t const *in, int shift) {
   1.149 +    uint64_t th, tl, oh, ol;
   1.150 +
   1.151 +    th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]);
   1.152 +    tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]);
   1.153 +
   1.154 +    oh = th >> (shift * 8);
   1.155 +    ol = tl >> (shift * 8);
   1.156 +    ol |= th << (64 - shift * 8);
   1.157 +    out->u[1] = (uint32_t)(ol >> 32);
   1.158 +    out->u[0] = (uint32_t)(ol & 0xffffffff);
   1.159 +    out->u[3] = (uint32_t)(oh >> 32);
   1.160 +    out->u[2] = (uint32_t)(oh & 0xffffffff);
   1.161 +}
   1.162 +#endif
   1.163 +/**
   1.164 + * This function simulates SIMD 128-bit left shift by the standard C.
   1.165 + * The 128-bit integer given in in is shifted by (shift * 8) bits.
   1.166 + * This function simulates the LITTLE ENDIAN SIMD.
   1.167 + * @param out the output of this function
   1.168 + * @param in the 128-bit data to be shifted
   1.169 + * @param shift the shift value
   1.170 + */
   1.171 +#ifdef ONLY64
   1.172 +inline static void lshift128(w128_t *out, w128_t const *in, int shift) {
   1.173 +    uint64_t th, tl, oh, ol;
   1.174 +
   1.175 +    th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]);
   1.176 +    tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]);
   1.177 +
   1.178 +    oh = th << (shift * 8);
   1.179 +    ol = tl << (shift * 8);
   1.180 +    oh |= tl >> (64 - shift * 8);
   1.181 +    out->u[0] = (uint32_t)(ol >> 32);
   1.182 +    out->u[1] = (uint32_t)(ol & 0xffffffff);
   1.183 +    out->u[2] = (uint32_t)(oh >> 32);
   1.184 +    out->u[3] = (uint32_t)(oh & 0xffffffff);
   1.185 +}
   1.186 +#else
   1.187 +inline static void lshift128(w128_t *out, w128_t const *in, int shift) {
   1.188 +    uint64_t th, tl, oh, ol;
   1.189 +
   1.190 +    th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]);
   1.191 +    tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]);
   1.192 +
   1.193 +    oh = th << (shift * 8);
   1.194 +    ol = tl << (shift * 8);
   1.195 +    oh |= tl >> (64 - shift * 8);
   1.196 +    out->u[1] = (uint32_t)(ol >> 32);
   1.197 +    out->u[0] = (uint32_t)(ol & 0xffffffff);
   1.198 +    out->u[3] = (uint32_t)(oh >> 32);
   1.199 +    out->u[2] = (uint32_t)(oh & 0xffffffff);
   1.200 +}
   1.201 +#endif
   1.202 +
   1.203 +/**
   1.204 + * This function represents the recursion formula.
   1.205 + * @param r output
   1.206 + * @param a a 128-bit part of the internal state array
   1.207 + * @param b a 128-bit part of the internal state array
   1.208 + * @param c a 128-bit part of the internal state array
   1.209 + * @param d a 128-bit part of the internal state array
   1.210 + */
   1.211 +#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2))
   1.212 +#ifdef ONLY64
   1.213 +inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *c,
   1.214 +				w128_t *d) {
   1.215 +    w128_t x;
   1.216 +    w128_t y;
   1.217 +
   1.218 +    lshift128(&x, a, SL2);
   1.219 +    rshift128(&y, c, SR2);
   1.220 +    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK2) ^ y.u[0] 
   1.221 +	^ (d->u[0] << SL1);
   1.222 +    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK1) ^ y.u[1] 
   1.223 +	^ (d->u[1] << SL1);
   1.224 +    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK4) ^ y.u[2] 
   1.225 +	^ (d->u[2] << SL1);
   1.226 +    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK3) ^ y.u[3] 
   1.227 +	^ (d->u[3] << SL1);
   1.228 +}
   1.229 +#else
   1.230 +inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *c,
   1.231 +				w128_t *d) {
   1.232 +    w128_t x;
   1.233 +    w128_t y;
   1.234 +
   1.235 +    lshift128(&x, a, SL2);
   1.236 +    rshift128(&y, c, SR2);
   1.237 +    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK1) ^ y.u[0] 
   1.238 +	^ (d->u[0] << SL1);
   1.239 +    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK2) ^ y.u[1] 
   1.240 +	^ (d->u[1] << SL1);
   1.241 +    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK3) ^ y.u[2] 
   1.242 +	^ (d->u[2] << SL1);
   1.243 +    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK4) ^ y.u[3] 
   1.244 +	^ (d->u[3] << SL1);
   1.245 +}
   1.246 +#endif
   1.247 +#endif
   1.248 +
   1.249 +#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2))
   1.250 +/**
   1.251 + * This function fills the internal state array with pseudorandom
   1.252 + * integers.
   1.253 + */
   1.254 +inline static void gen_rand_all(void) {
   1.255 +    int i;
   1.256 +    w128_t *r1, *r2;
   1.257 +
   1.258 +    r1 = &sfmt[N - 2];
   1.259 +    r2 = &sfmt[N - 1];
   1.260 +    for (i = 0; i < N - POS1; i++) {
   1.261 +	do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
   1.262 +	r1 = r2;
   1.263 +	r2 = &sfmt[i];
   1.264 +    }
   1.265 +    for (; i < N; i++) {
   1.266 +	do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1 - N], r1, r2);
   1.267 +	r1 = r2;
   1.268 +	r2 = &sfmt[i];
   1.269 +    }
   1.270 +}
   1.271 +
   1.272 +/**
   1.273 + * This function fills the user-specified array with pseudorandom
   1.274 + * integers.
   1.275 + *
   1.276 + * @param array an 128-bit array to be filled by pseudorandom numbers.  
   1.277 + * @param size number of 128-bit pseudorandom numbers to be generated.
   1.278 + */
   1.279 +inline static void gen_rand_array(w128_t *array, int size) {
   1.280 +    int i, j;
   1.281 +    w128_t *r1, *r2;
   1.282 +
   1.283 +    r1 = &sfmt[N - 2];
   1.284 +    r2 = &sfmt[N - 1];
   1.285 +    for (i = 0; i < N - POS1; i++) {
   1.286 +	do_recursion(&array[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
   1.287 +	r1 = r2;
   1.288 +	r2 = &array[i];
   1.289 +    }
   1.290 +    for (; i < N; i++) {
   1.291 +	do_recursion(&array[i], &sfmt[i], &array[i + POS1 - N], r1, r2);
   1.292 +	r1 = r2;
   1.293 +	r2 = &array[i];
   1.294 +    }
   1.295 +    for (; i < size - N; i++) {
   1.296 +	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
   1.297 +	r1 = r2;
   1.298 +	r2 = &array[i];
   1.299 +    }
   1.300 +    for (j = 0; j < 2 * N - size; j++) {
   1.301 +	sfmt[j] = array[j + size - N];
   1.302 +    }
   1.303 +    for (; i < size; i++, j++) {
   1.304 +	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
   1.305 +	r1 = r2;
   1.306 +	r2 = &array[i];
   1.307 +	sfmt[j] = array[i];
   1.308 +    }
   1.309 +}
   1.310 +#endif
   1.311 +
   1.312 +#if defined(BIG_ENDIAN64) && !defined(ONLY64) && !defined(HAVE_ALTIVEC)
   1.313 +inline static void swap(w128_t *array, int size) {
   1.314 +    int i;
   1.315 +    uint32_t x, y;
   1.316 +
   1.317 +    for (i = 0; i < size; i++) {
   1.318 +	x = array[i].u[0];
   1.319 +	y = array[i].u[2];
   1.320 +	array[i].u[0] = array[i].u[1];
   1.321 +	array[i].u[2] = array[i].u[3];
   1.322 +	array[i].u[1] = x;
   1.323 +	array[i].u[3] = y;
   1.324 +    }
   1.325 +}
   1.326 +#endif
   1.327 +/**
   1.328 + * This function represents a function used in the initialization
   1.329 + * by init_by_array
   1.330 + * @param x 32-bit integer
   1.331 + * @return 32-bit integer
   1.332 + */
   1.333 +static uint32_t func1(uint32_t x) {
   1.334 +    return (x ^ (x >> 27)) * (uint32_t)1664525UL;
   1.335 +}
   1.336 +
   1.337 +/**
   1.338 + * This function represents a function used in the initialization
   1.339 + * by init_by_array
   1.340 + * @param x 32-bit integer
   1.341 + * @return 32-bit integer
   1.342 + */
   1.343 +static uint32_t func2(uint32_t x) {
   1.344 +    return (x ^ (x >> 27)) * (uint32_t)1566083941UL;
   1.345 +}
   1.346 +
   1.347 +/**
   1.348 + * This function certificate the period of 2^{MEXP}
   1.349 + */
   1.350 +static void period_certification(void) {
   1.351 +    int inner = 0;
   1.352 +    int i, j;
   1.353 +    uint32_t work;
   1.354 +
   1.355 +    for (i = 0; i < 4; i++)
   1.356 +	inner ^= psfmt32[idxof(i)] & parity[i];
   1.357 +    for (i = 16; i > 0; i >>= 1)
   1.358 +	inner ^= inner >> i;
   1.359 +    inner &= 1;
   1.360 +    /* check OK */
   1.361 +    if (inner == 1) {
   1.362 +	return;
   1.363 +    }
   1.364 +    /* check NG, and modification */
   1.365 +    for (i = 0; i < 4; i++) {
   1.366 +	work = 1;
   1.367 +	for (j = 0; j < 32; j++) {
   1.368 +	    if ((work & parity[i]) != 0) {
   1.369 +		psfmt32[idxof(i)] ^= work;
   1.370 +		return;
   1.371 +	    }
   1.372 +	    work = work << 1;
   1.373 +	}
   1.374 +    }
   1.375 +}
   1.376 +
   1.377 +/*----------------
   1.378 +  PUBLIC FUNCTIONS
   1.379 +  ----------------*/
   1.380 +/**
   1.381 + * This function returns the identification string.
   1.382 + * The string shows the word size, the Mersenne exponent,
   1.383 + * and all parameters of this generator.
   1.384 + */
   1.385 +const char *get_idstring(void) {
   1.386 +    return IDSTR;
   1.387 +}
   1.388 +
   1.389 +/**
   1.390 + * This function returns the minimum size of array used for \b
   1.391 + * fill_array32() function.
   1.392 + * @return minimum size of array used for fill_array32() function.
   1.393 + */
   1.394 +int get_min_array_size32(void) {
   1.395 +    return N32;
   1.396 +}
   1.397 +
   1.398 +/**
   1.399 + * This function returns the minimum size of array used for \b
   1.400 + * fill_array64() function.
   1.401 + * @return minimum size of array used for fill_array64() function.
   1.402 + */
   1.403 +int get_min_array_size64(void) {
   1.404 +    return N64;
   1.405 +}
   1.406 +
   1.407 +#ifndef ONLY64
   1.408 +/**
   1.409 + * This function generates and returns 32-bit pseudorandom number.
   1.410 + * init_gen_rand or init_by_array must be called before this function.
   1.411 + * @return 32-bit pseudorandom number
   1.412 + */
   1.413 +uint32_t gen_rand32(void) {
   1.414 +    uint32_t r;
   1.415 +
   1.416 +    assert(initialized);
   1.417 +    if (idx >= N32) {
   1.418 +	gen_rand_all();
   1.419 +	idx = 0;
   1.420 +    }
   1.421 +    r = psfmt32[idx++];
   1.422 +    return r;
   1.423 +}
   1.424 +#endif
   1.425 +/**
   1.426 + * This function generates and returns 64-bit pseudorandom number.
   1.427 + * init_gen_rand or init_by_array must be called before this function.
   1.428 + * The function gen_rand64 should not be called after gen_rand32,
   1.429 + * unless an initialization is again executed. 
   1.430 + * @return 64-bit pseudorandom number
   1.431 + */
   1.432 +uint64_t gen_rand64(void) {
   1.433 +#if defined(BIG_ENDIAN64) && !defined(ONLY64)
   1.434 +    uint32_t r1, r2;
   1.435 +#else
   1.436 +    uint64_t r;
   1.437 +#endif
   1.438 +
   1.439 +    assert(initialized);
   1.440 +    assert(idx % 2 == 0);
   1.441 +
   1.442 +    if (idx >= N32) {
   1.443 +	gen_rand_all();
   1.444 +	idx = 0;
   1.445 +    }
   1.446 +#if defined(BIG_ENDIAN64) && !defined(ONLY64)
   1.447 +    r1 = psfmt32[idx];
   1.448 +    r2 = psfmt32[idx + 1];
   1.449 +    idx += 2;
   1.450 +    return ((uint64_t)r2 << 32) | r1;
   1.451 +#else
   1.452 +    r = psfmt64[idx / 2];
   1.453 +    idx += 2;
   1.454 +    return r;
   1.455 +#endif
   1.456 +}
   1.457 +
   1.458 +#ifndef ONLY64
   1.459 +/**
   1.460 + * This function generates pseudorandom 32-bit integers in the
   1.461 + * specified array[] by one call. The number of pseudorandom integers
   1.462 + * is specified by the argument size, which must be at least 624 and a
   1.463 + * multiple of four.  The generation by this function is much faster
   1.464 + * than the following gen_rand function.
   1.465 + *
   1.466 + * For initialization, init_gen_rand or init_by_array must be called
   1.467 + * before the first call of this function. This function can not be
   1.468 + * used after calling gen_rand function, without initialization.
   1.469 + *
   1.470 + * @param array an array where pseudorandom 32-bit integers are filled
   1.471 + * by this function.  The pointer to the array must be \b "aligned"
   1.472 + * (namely, must be a multiple of 16) in the SIMD version, since it
   1.473 + * refers to the address of a 128-bit integer.  In the standard C
   1.474 + * version, the pointer is arbitrary.
   1.475 + *
   1.476 + * @param size the number of 32-bit pseudorandom integers to be
   1.477 + * generated.  size must be a multiple of 4, and greater than or equal
   1.478 + * to (MEXP / 128 + 1) * 4.
   1.479 + *
   1.480 + * @note \b memalign or \b posix_memalign is available to get aligned
   1.481 + * memory. Mac OSX doesn't have these functions, but \b malloc of OSX
   1.482 + * returns the pointer to the aligned memory block.
   1.483 + */
   1.484 +void fill_array32(uint32_t *array, int size) {
   1.485 +    assert(initialized);
   1.486 +    assert(idx == N32);
   1.487 +    assert(size % 4 == 0);
   1.488 +    assert(size >= N32);
   1.489 +
   1.490 +    gen_rand_array((w128_t *)array, size / 4);
   1.491 +    idx = N32;
   1.492 +}
   1.493 +#endif
   1.494 +
   1.495 +/**
   1.496 + * This function generates pseudorandom 64-bit integers in the
   1.497 + * specified array[] by one call. The number of pseudorandom integers
   1.498 + * is specified by the argument size, which must be at least 312 and a
   1.499 + * multiple of two.  The generation by this function is much faster
   1.500 + * than the following gen_rand function.
   1.501 + *
   1.502 + * For initialization, init_gen_rand or init_by_array must be called
   1.503 + * before the first call of this function. This function can not be
   1.504 + * used after calling gen_rand function, without initialization.
   1.505 + *
   1.506 + * @param array an array where pseudorandom 64-bit integers are filled
   1.507 + * by this function.  The pointer to the array must be "aligned"
   1.508 + * (namely, must be a multiple of 16) in the SIMD version, since it
   1.509 + * refers to the address of a 128-bit integer.  In the standard C
   1.510 + * version, the pointer is arbitrary.
   1.511 + *
   1.512 + * @param size the number of 64-bit pseudorandom integers to be
   1.513 + * generated.  size must be a multiple of 2, and greater than or equal
   1.514 + * to (MEXP / 128 + 1) * 2
   1.515 + *
   1.516 + * @note \b memalign or \b posix_memalign is available to get aligned
   1.517 + * memory. Mac OSX doesn't have these functions, but \b malloc of OSX
   1.518 + * returns the pointer to the aligned memory block.
   1.519 + */
   1.520 +void fill_array64(uint64_t *array, int size) {
   1.521 +    assert(initialized);
   1.522 +    assert(idx == N32);
   1.523 +    assert(size % 2 == 0);
   1.524 +    assert(size >= N64);
   1.525 +
   1.526 +    gen_rand_array((w128_t *)array, size / 2);
   1.527 +    idx = N32;
   1.528 +
   1.529 +#if defined(BIG_ENDIAN64) && !defined(ONLY64)
   1.530 +    swap((w128_t *)array, size /2);
   1.531 +#endif
   1.532 +}
   1.533 +
   1.534 +/**
   1.535 + * This function initializes the internal state array with a 32-bit
   1.536 + * integer seed.
   1.537 + *
   1.538 + * @param seed a 32-bit integer used as the seed.
   1.539 + */
   1.540 +void init_gen_rand(uint32_t seed) {
   1.541 +    int i;
   1.542 +
   1.543 +    psfmt32[idxof(0)] = seed;
   1.544 +    for (i = 1; i < N32; i++) {
   1.545 +	psfmt32[idxof(i)] = 1812433253UL * (psfmt32[idxof(i - 1)] 
   1.546 +					    ^ (psfmt32[idxof(i - 1)] >> 30))
   1.547 +	    + i;
   1.548 +    }
   1.549 +    idx = N32;
   1.550 +    period_certification();
   1.551 +    initialized = 1;
   1.552 +}
   1.553 +
   1.554 +/**
   1.555 + * This function initializes the internal state array,
   1.556 + * with an array of 32-bit integers used as the seeds
   1.557 + * @param init_key the array of 32-bit integers, used as a seed.
   1.558 + * @param key_length the length of init_key.
   1.559 + */
   1.560 +void init_by_array(uint32_t *init_key, int key_length) {
   1.561 +    int i, j, count;
   1.562 +    uint32_t r;
   1.563 +    int lag;
   1.564 +    int mid;
   1.565 +    int size = N * 4;
   1.566 +
   1.567 +    if (size >= 623) {
   1.568 +	lag = 11;
   1.569 +    } else if (size >= 68) {
   1.570 +	lag = 7;
   1.571 +    } else if (size >= 39) {
   1.572 +	lag = 5;
   1.573 +    } else {
   1.574 +	lag = 3;
   1.575 +    }
   1.576 +    mid = (size - lag) / 2;
   1.577 +
   1.578 +    memset(sfmt, 0x8b, sizeof(sfmt));
   1.579 +    if (key_length + 1 > N32) {
   1.580 +	count = key_length + 1;
   1.581 +    } else {
   1.582 +	count = N32;
   1.583 +    }
   1.584 +    r = func1(psfmt32[idxof(0)] ^ psfmt32[idxof(mid)] 
   1.585 +	      ^ psfmt32[idxof(N32 - 1)]);
   1.586 +    psfmt32[idxof(mid)] += r;
   1.587 +    r += key_length;
   1.588 +    psfmt32[idxof(mid + lag)] += r;
   1.589 +    psfmt32[idxof(0)] = r;
   1.590 +
   1.591 +    count--;
   1.592 +    for (i = 1, j = 0; (j < count) && (j < key_length); j++) {
   1.593 +	r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % N32)] 
   1.594 +		  ^ psfmt32[idxof((i + N32 - 1) % N32)]);
   1.595 +	psfmt32[idxof((i + mid) % N32)] += r;
   1.596 +	r += init_key[j] + i;
   1.597 +	psfmt32[idxof((i + mid + lag) % N32)] += r;
   1.598 +	psfmt32[idxof(i)] = r;
   1.599 +	i = (i + 1) % N32;
   1.600 +    }
   1.601 +    for (; j < count; j++) {
   1.602 +	r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % N32)] 
   1.603 +		  ^ psfmt32[idxof((i + N32 - 1) % N32)]);
   1.604 +	psfmt32[idxof((i + mid) % N32)] += r;
   1.605 +	r += i;
   1.606 +	psfmt32[idxof((i + mid + lag) % N32)] += r;
   1.607 +	psfmt32[idxof(i)] = r;
   1.608 +	i = (i + 1) % N32;
   1.609 +    }
   1.610 +    for (j = 0; j < N32; j++) {
   1.611 +	r = func2(psfmt32[idxof(i)] + psfmt32[idxof((i + mid) % N32)] 
   1.612 +		  + psfmt32[idxof((i + N32 - 1) % N32)]);
   1.613 +	psfmt32[idxof((i + mid) % N32)] ^= r;
   1.614 +	r -= i;
   1.615 +	psfmt32[idxof((i + mid + lag) % N32)] ^= r;
   1.616 +	psfmt32[idxof(i)] = r;
   1.617 +	i = (i + 1) % N32;
   1.618 +    }
   1.619 +
   1.620 +    idx = N32;
   1.621 +    period_certification();
   1.622 +    initialized = 1;
   1.623 +}