SFMT.cpp 18.7 KB
Newer Older
1
/**
2
3
4
5
6
7
8
9
10
11
12
 * @file  SFMT.c
 * @brief SIMD oriented Fast Mersenne Twister(SFMT)
 *
 * @author Mutsuo Saito (Hiroshima University)
 * @author Makoto Matsumoto (Hiroshima University)
 *
 * Copyright (C) 2006,2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
 * University. All rights reserved.
 *
 * The new BSD License is applied to this software, see LICENSE.txt
 */
13
14
15
16
17
#include "sfmt/SFMT.h"
#include "sfmt/SFMT-params.h"

#include <cstring>
#include <cassert>
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

#if defined(__BIG_ENDIAN__) && !defined(__amd64) && !defined(BIG_ENDIAN64)
#define BIG_ENDIAN64 1
#endif
#if defined(HAVE_ALTIVEC) && !defined(BIG_ENDIAN64)
#define BIG_ENDIAN64 1
#endif
#if defined(ONLY64) && !defined(BIG_ENDIAN64)
  #if defined(__GNUC__)
    #error "-DONLY64 must be specified with -DBIG_ENDIAN64"
  #endif
#undef ONLY64
#endif
/*------------------------------------------------------
  128-bit SIMD data type for Altivec, SSE2 or standard C
  ------------------------------------------------------*/
#if defined(HAVE_ALTIVEC)
  #if !defined(__APPLE__)
    #include <altivec.h>
  #endif
/** 128-bit data structure */
union W128_T {
    vector unsigned int s;
    uint32_t u[4];
};
/** 128-bit data type */
typedef union W128_T w128_t;

#elif defined(HAVE_SSE2)
  #include <emmintrin.h>

/** 128-bit data structure */
union W128_T {
    __m128i si;
    uint32_t u[4];
};
/** 128-bit data type */
typedef union W128_T w128_t;

#else

/** 128-bit data structure */
struct W128_T {
    uint32_t u[4];
};
/** 128-bit data type */
typedef struct W128_T w128_t;

#endif

/*--------------------------------------
  FILE GLOBAL VARIABLES
70
  internal state, index counter and flag
71
  --------------------------------------*/
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
///** the 128-bit internal state array */
//static w128_t sfmt[N];
///** the 32bit integer pointer to the 128-bit internal state array */
//static uint32_t *psfmt32 = &sfmt[0].u[0];
//#if !defined(BIG_ENDIAN64) || defined(ONLY64)
///** the 64bit integer pointer to the 128-bit internal state array */
//static uint64_t *psfmt64 = (uint64_t *)&sfmt[0].u[0];
//#endif
///** index counter to the 32-bit internal state array */
//static int idx;
///** a flag: it is 0 if and only if the internal state is not yet
// * initialized. */
//static int initialized = 0;
///** a parity check vector which certificate the period of 2^{MEXP} */
//static uint32_t parity[4] = {PARITY1, PARITY2, PARITY3, PARITY4};

namespace OpenMM_SFMT {

class SFMTData {
public:
	/** the 128-bit internal state array */
	w128_t sfmt[N];
	/** the 32bit integer pointer to the 128-bit internal state array */
	uint32_t *psfmt32;
	#if !defined(BIG_ENDIAN64) || defined(ONLY64)
	/** the 64bit integer pointer to the 128-bit internal state array */
	uint64_t *psfmt64;
	#endif
	/** index counter to the 32-bit internal state array */
	int idx;
	/** a flag: it is 0 if and only if the internal state is not yet
	 * initialized. */
	int initialized;
	/** a parity check vector which certificate the period of 2^{MEXP} */
	uint32_t parity[4];
	SFMTData() {
		psfmt32 = &sfmt[0].u[0];
109
#if !defined(BIG_ENDIAN64) || defined(ONLY64)
110
		psfmt64 = (uint64_t *)&sfmt[0].u[0];
111
#endif
112
113
114
115
116
117
118
		initialized = 0;
		parity[0] = PARITY1;
		parity[1] = PARITY2;
		parity[2] = PARITY3;
		parity[3] = PARITY4;
	}
};
119
120
121
122
123
124
125

/*----------------
  STATIC FUNCTIONS
  ----------------*/
inline static int idxof(int i);
inline static void rshift128(w128_t *out,  w128_t const *in, int shift);
inline static void lshift128(w128_t *out,  w128_t const *in, int shift);
126
127
inline static void gen_rand_all(SFMT& sfmt);
inline static void gen_rand_array(w128_t *array, int size, SFMT& sfmt);
128
129
inline static uint32_t func1(uint32_t x);
inline static uint32_t func2(uint32_t x);
130
static void period_certification(SFMT& sfmt);
131
132
133
134
135
136
137
138
139
140
141
#if defined(BIG_ENDIAN64) && !defined(ONLY64)
inline static void swap(w128_t *array, int size);
#endif

#if defined(HAVE_ALTIVEC)
  #include "SFMT-alti.h"
#elif defined(HAVE_SSE2)
  #include "SFMT-sse2.h"
#endif

/**
142
 * This function simulate a 64-bit index of LITTLE ENDIAN
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
 * in BIG ENDIAN machine.
 */
#ifdef ONLY64
inline static int idxof(int i) {
    return i ^ 1;
}
#else
inline static int idxof(int i) {
    return i;
}
#endif
/**
 * This function simulates SIMD 128-bit right shift by the standard C.
 * The 128-bit integer given in in is shifted by (shift * 8) bits.
 * This function simulates the LITTLE ENDIAN SIMD.
 * @param out the output of this function
 * @param in the 128-bit data to be shifted
 * @param shift the shift value
 */
#ifdef ONLY64
inline static void rshift128(w128_t *out, w128_t const *in, int shift) {
    uint64_t th, tl, oh, ol;

    th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]);
    tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]);

    oh = th >> (shift * 8);
    ol = tl >> (shift * 8);
    ol |= th << (64 - shift * 8);
    out->u[0] = (uint32_t)(ol >> 32);
    out->u[1] = (uint32_t)ol;
    out->u[2] = (uint32_t)(oh >> 32);
    out->u[3] = (uint32_t)oh;
}
#else
inline static void rshift128(w128_t *out, w128_t const *in, int shift) {
    uint64_t th, tl, oh, ol;

    th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]);
    tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]);

    oh = th >> (shift * 8);
    ol = tl >> (shift * 8);
    ol |= th << (64 - shift * 8);
    out->u[1] = (uint32_t)(ol >> 32);
    out->u[0] = (uint32_t)ol;
    out->u[3] = (uint32_t)(oh >> 32);
    out->u[2] = (uint32_t)oh;
}
#endif
/**
 * This function simulates SIMD 128-bit left shift by the standard C.
 * The 128-bit integer given in in is shifted by (shift * 8) bits.
 * This function simulates the LITTLE ENDIAN SIMD.
 * @param out the output of this function
 * @param in the 128-bit data to be shifted
 * @param shift the shift value
 */
#ifdef ONLY64
inline static void lshift128(w128_t *out, w128_t const *in, int shift) {
    uint64_t th, tl, oh, ol;

    th = ((uint64_t)in->u[2] << 32) | ((uint64_t)in->u[3]);
    tl = ((uint64_t)in->u[0] << 32) | ((uint64_t)in->u[1]);

    oh = th << (shift * 8);
    ol = tl << (shift * 8);
    oh |= tl >> (64 - shift * 8);
    out->u[0] = (uint32_t)(ol >> 32);
    out->u[1] = (uint32_t)ol;
    out->u[2] = (uint32_t)(oh >> 32);
    out->u[3] = (uint32_t)oh;
}
#else
inline static void lshift128(w128_t *out, w128_t const *in, int shift) {
    uint64_t th, tl, oh, ol;

    th = ((uint64_t)in->u[3] << 32) | ((uint64_t)in->u[2]);
    tl = ((uint64_t)in->u[1] << 32) | ((uint64_t)in->u[0]);

    oh = th << (shift * 8);
    ol = tl << (shift * 8);
    oh |= tl >> (64 - shift * 8);
    out->u[1] = (uint32_t)(ol >> 32);
    out->u[0] = (uint32_t)ol;
    out->u[3] = (uint32_t)(oh >> 32);
    out->u[2] = (uint32_t)oh;
}
#endif

/**
 * This function represents the recursion formula.
 * @param r output
 * @param a a 128-bit part of the internal state array
 * @param b a 128-bit part of the internal state array
 * @param c a 128-bit part of the internal state array
 * @param d a 128-bit part of the internal state array
 */
#ifdef ONLY64
inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *c,
				w128_t *d) {
    w128_t x;
    w128_t y;

    lshift128(&x, a, SL2);
    rshift128(&y, c, SR2);
249
    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK2) ^ y.u[0]
250
	^ (d->u[0] << SL1);
251
    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK1) ^ y.u[1]
252
	^ (d->u[1] << SL1);
253
    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK4) ^ y.u[2]
254
	^ (d->u[2] << SL1);
255
    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK3) ^ y.u[3]
256
257
258
259
260
261
262
263
264
265
	^ (d->u[3] << SL1);
}
#else
inline static void do_recursion(w128_t *r, w128_t *a, w128_t *b, w128_t *c,
				w128_t *d) {
    w128_t x;
    w128_t y;

    lshift128(&x, a, SL2);
    rshift128(&y, c, SR2);
266
    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK1) ^ y.u[0]
267
	^ (d->u[0] << SL1);
268
    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK2) ^ y.u[1]
269
	^ (d->u[1] << SL1);
270
    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK3) ^ y.u[2]
271
	^ (d->u[2] << SL1);
272
    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK4) ^ y.u[3]
273
274
275
276
277
278
279
280
281
	^ (d->u[3] << SL1);
}
#endif

#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2))
/**
 * This function fills the internal state array with pseudorandom
 * integers.
 */
282
inline static void gen_rand_all(SFMT& sfmt) {
283
284
285
    int i;
    w128_t *r1, *r2;

286
287
288
    SFMTData& data = *sfmt.data;
    r1 = &(data.sfmt[N - 2]);
    r2 = &(data.sfmt[N - 1]);
289
    for (i = 0; i < N - POS1; i++) {
290
	do_recursion(&(data.sfmt[i]), &(data.sfmt[i]), &(data.sfmt[i + POS1]), r1, r2);
291
	r1 = r2;
292
	r2 = &(data.sfmt[i]);
293
294
    }
    for (; i < N; i++) {
295
	do_recursion(&(data.sfmt[i]), &(data.sfmt[i]), &(data.sfmt[i + POS1 - N]), r1, r2);
296
	r1 = r2;
297
	r2 = &(data.sfmt[i]);
298
299
300
301
302
303
304
    }
}

/**
 * This function fills the user-specified array with pseudorandom
 * integers.
 *
305
 * @param array an 128-bit array to be filled by pseudorandom numbers.
306
307
 * @param size number of 128-bit pseudorandom numbers to be generated.
 */
308
inline static void gen_rand_array(w128_t *array, int size, SFMT& sfmt) {
309
310
311
    int i, j;
    w128_t *r1, *r2;

312
313
314
    SFMTData& data = *sfmt.data;
    r1 = &(data.sfmt[N - 2]);
    r2 = &(data.sfmt[N - 1]);
315
    for (i = 0; i < N - POS1; i++) {
316
	do_recursion(&array[i], &(data.sfmt[i]), &(data.sfmt[i + POS1]), r1, r2);
317
318
319
320
	r1 = r2;
	r2 = &array[i];
    }
    for (; i < N; i++) {
321
	do_recursion(&array[i], &(data.sfmt[i]), &array[i + POS1 - N], r1, r2);
322
323
324
325
326
327
328
329
330
	r1 = r2;
	r2 = &array[i];
    }
    for (; i < size - N; i++) {
	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
	r1 = r2;
	r2 = &array[i];
    }
    for (j = 0; j < 2 * N - size; j++) {
331
	data.sfmt[j] = array[j + size - N];
332
333
334
335
336
    }
    for (; i < size; i++, j++) {
	do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
	r1 = r2;
	r2 = &array[i];
337
	data.sfmt[j] = array[i];
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
    }
}
#endif

#if defined(BIG_ENDIAN64) && !defined(ONLY64) && !defined(HAVE_ALTIVEC)
inline static void swap(w128_t *array, int size) {
    int i;
    uint32_t x, y;

    for (i = 0; i < size; i++) {
	x = array[i].u[0];
	y = array[i].u[2];
	array[i].u[0] = array[i].u[1];
	array[i].u[2] = array[i].u[3];
	array[i].u[1] = x;
	array[i].u[3] = y;
    }
}
#endif
/**
 * This function represents a function used in the initialization
 * by init_by_array
 * @param x 32-bit integer
 * @return 32-bit integer
 */
static uint32_t func1(uint32_t x) {
    return (x ^ (x >> 27)) * (uint32_t)1664525UL;
}

/**
 * This function represents a function used in the initialization
 * by init_by_array
 * @param x 32-bit integer
 * @return 32-bit integer
 */
static uint32_t func2(uint32_t x) {
    return (x ^ (x >> 27)) * (uint32_t)1566083941UL;
}

/**
 * This function certificate the period of 2^{MEXP}
 */
380
static void period_certification(SFMT& sfmt) {
381
382
383
    int inner = 0;
    int i, j;
    uint32_t work;
384
    SFMTData& data = *sfmt.data;
385
386

    for (i = 0; i < 4; i++)
387
	inner ^= data.psfmt32[idxof(i)] & data.parity[i];
388
389
390
391
392
393
394
395
396
397
398
    for (i = 16; i > 0; i >>= 1)
	inner ^= inner >> i;
    inner &= 1;
    /* check OK */
    if (inner == 1) {
	return;
    }
    /* check NG, and modification */
    for (i = 0; i < 4; i++) {
	work = 1;
	for (j = 0; j < 32; j++) {
399
400
	    if ((work & data.parity[i]) != 0) {
		data.psfmt32[idxof(i)] ^= work;
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
		return;
	    }
	    work = work << 1;
	}
    }
}

/*----------------
  PUBLIC FUNCTIONS
  ----------------*/
/**
 * This function returns the identification string.
 * The string shows the word size, the Mersenne exponent,
 * and all parameters of this generator.
 */
const char *get_idstring(void) {
    return IDSTR;
}

/**
 * This function returns the minimum size of array used for \b
 * fill_array32() function.
 * @return minimum size of array used for fill_array32() function.
 */
int get_min_array_size32(void) {
    return N32;
}

/**
 * This function returns the minimum size of array used for \b
 * fill_array64() function.
 * @return minimum size of array used for fill_array64() function.
 */
int get_min_array_size64(void) {
    return N64;
}

#ifndef ONLY64
/**
 * This function generates and returns 32-bit pseudorandom number.
 * init_gen_rand or init_by_array must be called before this function.
 * @return 32-bit pseudorandom number
 */
444
uint32_t gen_rand32(SFMT& sfmt) {
445
    uint32_t r;
446
    SFMTData& data = *sfmt.data;
447

448
449
450
451
    assert(data.initialized);
    if (data.idx >= N32) {
	gen_rand_all(sfmt);
	data.idx = 0;
452
    }
453
    r = data.psfmt32[data.idx++];
454
455
456
457
458
459
460
    return r;
}
#endif
/**
 * This function generates and returns 64-bit pseudorandom number.
 * init_gen_rand or init_by_array must be called before this function.
 * The function gen_rand64 should not be called after gen_rand32,
461
 * unless an initialization is again executed.
462
463
 * @return 64-bit pseudorandom number
 */
464
uint64_t gen_rand64(SFMT& sfmt) {
465
466
467
468
469
#if defined(BIG_ENDIAN64) && !defined(ONLY64)
    uint32_t r1, r2;
#else
    uint64_t r;
#endif
470
    SFMTData& data = *sfmt.data;
471

472
473
    assert(data.initialized);
    assert(data.idx % 2 == 0);
474

475
476
477
    if (data.idx >= N32) {
	gen_rand_all(sfmt);
	data.idx = 0;
478
479
    }
#if defined(BIG_ENDIAN64) && !defined(ONLY64)
480
481
482
    r1 = data.psfmt32[data.idx];
    r2 = data.psfmt32[data.idx + 1];
    data.idx += 2;
483
484
    return ((uint64_t)r2 << 32) | r1;
#else
485
486
    r = data.psfmt64[data.idx / 2];
    data.idx += 2;
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
    return r;
#endif
}

#ifndef ONLY64
/**
 * This function generates pseudorandom 32-bit integers in the
 * specified array[] by one call. The number of pseudorandom integers
 * is specified by the argument size, which must be at least 624 and a
 * multiple of four.  The generation by this function is much faster
 * than the following gen_rand function.
 *
 * For initialization, init_gen_rand or init_by_array must be called
 * before the first call of this function. This function can not be
 * used after calling gen_rand function, without initialization.
 *
 * @param array an array where pseudorandom 32-bit integers are filled
 * by this function.  The pointer to the array must be \b "aligned"
 * (namely, must be a multiple of 16) in the SIMD version, since it
 * refers to the address of a 128-bit integer.  In the standard C
 * version, the pointer is arbitrary.
 *
 * @param size the number of 32-bit pseudorandom integers to be
 * generated.  size must be a multiple of 4, and greater than or equal
 * to (MEXP / 128 + 1) * 4.
 *
 * @note \b memalign or \b posix_memalign is available to get aligned
 * memory. Mac OSX doesn't have these functions, but \b malloc of OSX
 * returns the pointer to the aligned memory block.
 */
517
518
519
520
void fill_array32(uint32_t *array, int size, SFMT& sfmt) {
    SFMTData& data = *sfmt.data;
    assert(data.initialized);
    assert(data.idx == N32);
521
522
523
    assert(size % 4 == 0);
    assert(size >= N32);

524
525
    gen_rand_array((w128_t *)array, size / 4, sfmt);
    data.idx = N32;
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
}
#endif

/**
 * This function generates pseudorandom 64-bit integers in the
 * specified array[] by one call. The number of pseudorandom integers
 * is specified by the argument size, which must be at least 312 and a
 * multiple of two.  The generation by this function is much faster
 * than the following gen_rand function.
 *
 * For initialization, init_gen_rand or init_by_array must be called
 * before the first call of this function. This function can not be
 * used after calling gen_rand function, without initialization.
 *
 * @param array an array where pseudorandom 64-bit integers are filled
 * by this function.  The pointer to the array must be "aligned"
 * (namely, must be a multiple of 16) in the SIMD version, since it
 * refers to the address of a 128-bit integer.  In the standard C
 * version, the pointer is arbitrary.
 *
 * @param size the number of 64-bit pseudorandom integers to be
 * generated.  size must be a multiple of 2, and greater than or equal
 * to (MEXP / 128 + 1) * 2
 *
 * @note \b memalign or \b posix_memalign is available to get aligned
 * memory. Mac OSX doesn't have these functions, but \b malloc of OSX
 * returns the pointer to the aligned memory block.
 */
554
555
556
557
void fill_array64(uint64_t *array, int size, SFMT& sfmt) {
    SFMTData& data = *sfmt.data;
    assert(data.initialized);
    assert(data.idx == N32);
558
559
560
    assert(size % 2 == 0);
    assert(size >= N64);

561
562
    gen_rand_array((w128_t *)array, size / 2, sfmt);
    data.idx = N32;
563
564
565
566
567
568
569
570
571
572
573
574

#if defined(BIG_ENDIAN64) && !defined(ONLY64)
    swap((w128_t *)array, size /2);
#endif
}

/**
 * This function initializes the internal state array with a 32-bit
 * integer seed.
 *
 * @param seed a 32-bit integer used as the seed.
 */
575
void init_gen_rand(uint32_t seed, SFMT& sfmt) {
576
    int i;
577
    SFMTData& data = *sfmt.data;
578

579
    data.psfmt32[idxof(0)] = seed;
580
    for (i = 1; i < N32; i++) {
581
582
    	data.psfmt32[idxof(i)] = 1812433253UL * (data.psfmt32[idxof(i - 1)]
					    ^ (data.psfmt32[idxof(i - 1)] >> 30))
583
584
	    + i;
    }
585
586
587
    data.idx = N32;
    period_certification(sfmt);
    data.initialized = 1;
588
589
590
591
592
593
594
595
}

/**
 * This function initializes the internal state array,
 * with an array of 32-bit integers used as the seeds
 * @param init_key the array of 32-bit integers, used as a seed.
 * @param key_length the length of init_key.
 */
596
void init_by_array(uint32_t *init_key, int key_length, SFMT& sfmt) {
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
    int i, j, count;
    uint32_t r;
    int lag;
    int mid;
    int size = N * 4;

    if (size >= 623) {
	lag = 11;
    } else if (size >= 68) {
	lag = 7;
    } else if (size >= 39) {
	lag = 5;
    } else {
	lag = 3;
    }
    mid = (size - lag) / 2;

614
615
    SFMTData& data = *sfmt.data;
    memset(data.sfmt, 0x8b, sizeof(data.sfmt));
616
617
618
619
620
    if (key_length + 1 > N32) {
	count = key_length + 1;
    } else {
	count = N32;
    }
621
622
623
    r = func1(data.psfmt32[idxof(0)] ^ data.psfmt32[idxof(mid)]
	      ^ data.psfmt32[idxof(N32 - 1)]);
    data.psfmt32[idxof(mid)] += r;
624
    r += key_length;
625
626
    data.psfmt32[idxof(mid + lag)] += r;
    data.psfmt32[idxof(0)] = r;
627
628
629

    count--;
    for (i = 1, j = 0; (j < count) && (j < key_length); j++) {
630
631
632
	r = func1(data.psfmt32[idxof(i)] ^ data.psfmt32[idxof((i + mid) % N32)]
		  ^ data.psfmt32[idxof((i + N32 - 1) % N32)]);
	data.psfmt32[idxof((i + mid) % N32)] += r;
633
	r += init_key[j] + i;
634
635
	data.psfmt32[idxof((i + mid + lag) % N32)] += r;
	data.psfmt32[idxof(i)] = r;
636
637
638
	i = (i + 1) % N32;
    }
    for (; j < count; j++) {
639
640
641
	r = func1(data.psfmt32[idxof(i)] ^ data.psfmt32[idxof((i + mid) % N32)]
		  ^ data.psfmt32[idxof((i + N32 - 1) % N32)]);
	data.psfmt32[idxof((i + mid) % N32)] += r;
642
	r += i;
643
644
	data.psfmt32[idxof((i + mid + lag) % N32)] += r;
	data.psfmt32[idxof(i)] = r;
645
646
647
	i = (i + 1) % N32;
    }
    for (j = 0; j < N32; j++) {
648
649
650
	r = func2(data.psfmt32[idxof(i)] + data.psfmt32[idxof((i + mid) % N32)]
		  + data.psfmt32[idxof((i + N32 - 1) % N32)]);
	data.psfmt32[idxof((i + mid) % N32)] ^= r;
651
	r -= i;
652
653
	data.psfmt32[idxof((i + mid + lag) % N32)] ^= r;
	data.psfmt32[idxof(i)] = r;
654
655
656
	i = (i + 1) % N32;
    }

657
658
659
    data.idx = N32;
    period_certification(sfmt);
    data.initialized = 1;
660
}
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678

/**
 * Create an SFMTData object.  This allows outside code to create these objects without needing to know their definition.
 */
SFMTData* createSFMTData(void) {
	return new SFMTData();
}

/**
 * Delete an SFMTData object that was created with createSFMTData().
 */
void deleteSFMTData(SFMTData* data) {
	delete data;
}

} // OpenMM_SFMT