| 1 |
/** |
| 2 |
* random.C: SIMD oriented Fast Mersenne Twister(SFMT) |
| 3 |
* |
| 4 |
* Copyright © 2007 Pippijn van Steenhoven / The Ermyth Team |
| 5 |
* Rights to this code are as documented in COPYING. |
| 6 |
* |
| 7 |
* |
| 8 |
* Portions of this file were derived from sources bearing the following license: |
| 9 |
* |
| 10 |
* Copyright © 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima |
| 11 |
* University. All rights reserved. |
| 12 |
* |
| 13 |
* @author Mutsuo Saito (Hiroshima University) |
| 14 |
* @author Makoto Matsumoto (Hiroshima University) |
| 15 |
* |
| 16 |
* The new BSD License is applied to this software. |
| 17 |
*/ |
| 18 |
|
| 19 |
static char const rcsid[] = "$Id: random.C,v 1.3 2007-09-16 18:54:45 pippijn Exp $"; |
| 20 |
|
| 21 |
#include <string.h> |
| 22 |
#include <assert.h> |
| 23 |
|
| 24 |
#include <common/random.h> |
| 25 |
|
| 26 |
/*----------------- |
| 27 |
BASIC DEFINITIONS |
| 28 |
-----------------*/ |
| 29 |
/** SFMT generator has an internal state array of 128-bit integers, |
| 30 |
* and N is its size. */ |
| 31 |
#define N (216091 / 128 + 1) |
| 32 |
/** N32 is the size of internal state array when regarded as an array |
| 33 |
* of 32-bit integers.*/ |
| 34 |
#define N32 (N * 4) |
| 35 |
/** N64 is the size of internal state array when regarded as an array |
| 36 |
* of 64-bit integers.*/ |
| 37 |
#define N64 (N * 2) |
| 38 |
|
| 39 |
/*---------------------- |
| 40 |
the parameters of SFMT |
| 41 |
following definitions are in paramsXXXX.h file. |
| 42 |
----------------------*/ |
| 43 |
// the pick up position of the array. |
| 44 |
#define POS1 627 |
| 45 |
// the parameter of shift left as four 32-bit registers. |
| 46 |
#define SL1 11 |
| 47 |
// the parameter of shift left as one 128-bit register. |
| 48 |
// The 128-bit integer is shifted by (SL2 * 8) bits. |
| 49 |
#define SL2 3 |
| 50 |
// the parameter of shift right as four 32-bit registers. |
| 51 |
#define SR1 10 |
| 52 |
// the parameter of shift right as one 128-bit register. |
| 53 |
// The 128-bit integer is shifted by (SL2 * 8) bits. |
| 54 |
#define SR2 1 |
| 55 |
// A bitmask, used in the recursion. These parameters are introduced |
| 56 |
// to break symmetry of SIMD. |
| 57 |
#define MSK1 0xbff7bff7U |
| 58 |
#define MSK2 0xbfffffffU |
| 59 |
#define MSK3 0xbffffa7fU |
| 60 |
#define MSK4 0xffddfbfbU |
| 61 |
// These definitions are part of a 128-bit period certification vector. |
| 62 |
#define PARITY1 0xf8000001U |
| 63 |
#define PARITY2 0x89e80709U |
| 64 |
#define PARITY3 0x3bd2b64bU |
| 65 |
#define PARITY4 0x0c64b1e4U |
| 66 |
|
| 67 |
/* PARAMETERS FOR ALTIVEC */ |
| 68 |
#if defined(__APPLE__) /* For OSX */ |
| 69 |
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1) |
| 70 |
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1) |
| 71 |
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4) |
| 72 |
#define ALTI_MSK64 (vector unsigned int)(MSK2, MSK1, MSK4, MSK3) |
| 73 |
#define ALTI_SL2_PERM (vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10) |
| 74 |
#define ALTI_SL2_PERM64 (vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2) |
| 75 |
#define ALTI_SR2_PERM (vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14) |
| 76 |
#define ALTI_SR2_PERM64 (vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14) |
| 77 |
#else // !__APPLE__ |
| 78 |
#define ALTI_SL1 {SL1, SL1, SL1, SL1} |
| 79 |
#define ALTI_SR1 {SR1, SR1, SR1, SR1} |
| 80 |
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4} |
| 81 |
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3} |
| 82 |
#define ALTI_SL2_PERM { 3,21,21,21, 7, 0, 1, 2,11, 4, 5, 6,15, 8, 9,10} |
| 83 |
#define ALTI_SL2_PERM64 { 3, 4, 5, 6, 7,29,29,29,11,12,13,14,15, 0, 1, 2} |
| 84 |
#define ALTI_SR2_PERM { 7, 0, 1, 2,11, 4, 5, 6,15, 8, 9,10,17,12,13,14} |
| 85 |
#define ALTI_SR2_PERM64 {15, 0, 1, 2, 3, 4, 5, 6,17, 8, 9,10,11,12,13,14} |
| 86 |
#endif // __APPLE__ |
| 87 |
#define IDSTR "SFMT-216091:627-11-3-10-1:bff7bff7-bfffffff-bffffa7f-ffddfbfb" |
| 88 |
|
| 89 |
#if defined(__BIG_ENDIAN__) && !defined(__amd64) && !defined(BIG_ENDIAN64) |
| 90 |
#define BIG_ENDIAN64 1 |
| 91 |
#endif |
| 92 |
#if defined(HAVE_ALTIVEC) && !defined(BIG_ENDIAN64) |
| 93 |
#define BIG_ENDIAN64 1 |
| 94 |
#endif |
| 95 |
#if defined(ONLY64) && !defined(BIG_ENDIAN64) |
| 96 |
#if defined(__GNUC__) |
| 97 |
#error "-DONLY64 must be specified with -DBIG_ENDIAN64" |
| 98 |
#endif |
| 99 |
#undef ONLY64 |
| 100 |
#endif |
| 101 |
|
| 102 |
/*-------------------------------------- |
| 103 |
FILE GLOBAL VARIABLES |
| 104 |
internal state, index counter and flag |
| 105 |
--------------------------------------*/ |
| 106 |
/** the 128-bit internal state array */ |
| 107 |
static w128_t sfmt[N]; |
| 108 |
/** the 32bit integer pointer to the 128-bit internal state array */ |
| 109 |
static uint32_t *psfmt32 = &sfmt[0].u[0]; |
| 110 |
#if !defined(BIG_ENDIAN64) || defined(ONLY64) |
| 111 |
/** the 64bit integer pointer to the 128-bit internal state array */ |
| 112 |
static uint64_t *psfmt64 = reinterpret_cast<uint64_t *> (&sfmt[0].u[0]); |
| 113 |
#endif |
| 114 |
/** index counter to the 32-bit internal state array */ |
| 115 |
static int idx; |
| 116 |
/** a flag: it is 0 if and only if the internal state is not yet |
| 117 |
* initialized. */ |
| 118 |
static int initialized = 0; |
| 119 |
/** a parity check vector which certificate the period of 2^{MEXP} */ |
| 120 |
static uint32_t parity[4] = { PARITY1, PARITY2, PARITY3, PARITY4 }; |
| 121 |
|
| 122 |
/*---------------- |
| 123 |
STATIC FUNCTIONS |
| 124 |
----------------*/ |
| 125 |
inline static int idxof (int i); |
| 126 |
inline static void rshift128 (w128_t *out, w128_t const *in, int shift); |
| 127 |
inline static void lshift128 (w128_t *out, w128_t const *in, int shift); |
| 128 |
inline static void gen_rand_all (void); |
| 129 |
inline static void gen_rand_array (w128_t *array, int size); |
| 130 |
inline static uint32_t func1 (uint32_t x); |
| 131 |
inline static uint32_t func2 (uint32_t x); |
| 132 |
inline static void period_certification (void); |
| 133 |
#if defined(BIG_ENDIAN64) && !defined(ONLY64) |
| 134 |
inline static void swap (w128_t *array, int size); |
| 135 |
#endif |
| 136 |
|
| 137 |
#if defined(HAVE_ALTIVEC) |
| 138 |
inline static vector unsigned int |
| 139 |
vec_recursion (vector unsigned int a, vector unsigned int b, vector unsigned int c, vector unsigned int d) ALWAYSINLINE; |
| 140 |
|
| 141 |
/** |
| 142 |
* This function represents the recursion formula in AltiVec and BIG ENDIAN. |
| 143 |
* @param a a 128-bit part of the interal state array |
| 144 |
* @param b a 128-bit part of the interal state array |
| 145 |
* @param c a 128-bit part of the interal state array |
| 146 |
* @param d a 128-bit part of the interal state array |
| 147 |
* @return output |
| 148 |
*/ |
| 149 |
inline static vector unsigned int vec_recursion (vector unsigned int a, vector unsigned int b, vector unsigned int c, vector unsigned int d) |
| 150 |
{ |
| 151 |
const vector unsigned int sl1 = ALTI_SL1; |
| 152 |
const vector unsigned int sr1 = ALTI_SR1; |
| 153 |
#ifdef ONLY64 |
| 154 |
const vector unsigned int mask = ALTI_MSK64; |
| 155 |
const vector unsigned char perm_sl = ALTI_SL2_PERM64; |
| 156 |
const vector unsigned char perm_sr = ALTI_SR2_PERM64; |
| 157 |
#else |
| 158 |
const vector unsigned int mask = ALTI_MSK; |
| 159 |
const vector unsigned char perm_sl = ALTI_SL2_PERM; |
| 160 |
const vector unsigned char perm_sr = ALTI_SR2_PERM; |
| 161 |
#endif |
| 162 |
vector unsigned int v, w, x, y, z; |
| 163 |
x = vec_perm (a, (vector unsigned int) perm_sl, perm_sl); |
| 164 |
v = a; |
| 165 |
y = vec_sr (b, sr1); |
| 166 |
z = vec_perm (c, (vector unsigned int) perm_sr, perm_sr); |
| 167 |
w = vec_sl (d, sl1); |
| 168 |
z = vec_xor (z, w); |
| 169 |
y = vec_and (y, mask); |
| 170 |
v = vec_xor (v, x); |
| 171 |
z = vec_xor (z, y); |
| 172 |
z = vec_xor (z, v); |
| 173 |
return z; |
| 174 |
} |
| 175 |
|
| 176 |
/** |
| 177 |
* This function fills the internal state array with pseudorandom |
| 178 |
* integers. |
| 179 |
*/ |
| 180 |
inline static void |
| 181 |
gen_rand_all (void) |
| 182 |
{ |
| 183 |
int i; |
| 184 |
vector unsigned int r, r1, r2; |
| 185 |
|
| 186 |
r1 = sfmt[N - 2].s; |
| 187 |
r2 = sfmt[N - 1].s; |
| 188 |
for (i = 0; i < N - POS1; i++) |
| 189 |
{ |
| 190 |
r = vec_recursion (sfmt[i].s, sfmt[i + POS1].s, r1, r2); |
| 191 |
sfmt[i].s = r; |
| 192 |
r1 = r2; |
| 193 |
r2 = r; |
| 194 |
} |
| 195 |
for (; i < N; i++) |
| 196 |
{ |
| 197 |
r = vec_recursion (sfmt[i].s, sfmt[i + POS1 - N].s, r1, r2); |
| 198 |
sfmt[i].s = r; |
| 199 |
r1 = r2; |
| 200 |
r2 = r; |
| 201 |
} |
| 202 |
} |
| 203 |
|
| 204 |
/** |
| 205 |
* This function fills the user-specified array with pseudorandom |
| 206 |
* integers. |
| 207 |
* |
| 208 |
* @param array an 128-bit array to be filled by pseudorandom numbers. |
| 209 |
* @param size number of 128-bit pesudorandom numbers to be generated. |
| 210 |
*/ |
| 211 |
inline static void |
| 212 |
gen_rand_array (w128_t *array, int size) |
| 213 |
{ |
| 214 |
int i, j; |
| 215 |
vector unsigned int r, r1, r2; |
| 216 |
|
| 217 |
r1 = sfmt[N - 2].s; |
| 218 |
r2 = sfmt[N - 1].s; |
| 219 |
for (i = 0; i < N - POS1; i++) |
| 220 |
{ |
| 221 |
r = vec_recursion (sfmt[i].s, sfmt[i + POS1].s, r1, r2); |
| 222 |
array[i].s = r; |
| 223 |
r1 = r2; |
| 224 |
r2 = r; |
| 225 |
} |
| 226 |
for (; i < N; i++) |
| 227 |
{ |
| 228 |
r = vec_recursion (sfmt[i].s, array[i + POS1 - N].s, r1, r2); |
| 229 |
array[i].s = r; |
| 230 |
r1 = r2; |
| 231 |
r2 = r; |
| 232 |
} |
| 233 |
/* main loop */ |
| 234 |
for (; i < size - N; i++) |
| 235 |
{ |
| 236 |
r = vec_recursion (array[i - N].s, array[i + POS1 - N].s, r1, r2); |
| 237 |
array[i].s = r; |
| 238 |
r1 = r2; |
| 239 |
r2 = r; |
| 240 |
} |
| 241 |
for (j = 0; j < 2 * N - size; j++) |
| 242 |
{ |
| 243 |
sfmt[j].s = array[j + size - N].s; |
| 244 |
} |
| 245 |
for (; i < size; i++) |
| 246 |
{ |
| 247 |
r = vec_recursion (array[i - N].s, array[i + POS1 - N].s, r1, r2); |
| 248 |
array[i].s = r; |
| 249 |
sfmt[j++].s = r; |
| 250 |
r1 = r2; |
| 251 |
r2 = r; |
| 252 |
} |
| 253 |
} |
| 254 |
|
| 255 |
#ifndef ONLY64 |
| 256 |
#if defined(__APPLE__) |
| 257 |
#define ALTI_SWAP (vector unsigned char) \ |
| 258 |
(4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11) |
| 259 |
#else |
| 260 |
#define ALTI_SWAP {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11} |
| 261 |
#endif |
| 262 |
/** |
| 263 |
* This function swaps high and low 32-bit of 64-bit integers in user |
| 264 |
* specified array. |
| 265 |
* |
| 266 |
* @param array an 128-bit array to be swaped. |
| 267 |
* @param size size of 128-bit array. |
| 268 |
*/ |
| 269 |
inline static void |
| 270 |
swap (w128_t *array, int size) |
| 271 |
{ |
| 272 |
int i; |
| 273 |
const vector unsigned char perm = ALTI_SWAP; |
| 274 |
|
| 275 |
for (i = 0; i < size; i++) |
| 276 |
{ |
| 277 |
array[i].s = vec_perm (array[i].s, (vector unsigned int) perm, perm); |
| 278 |
} |
| 279 |
} |
| 280 |
#endif |
| 281 |
#elif defined(HAVE_SSE2) |
| 282 |
inline static __m128i |
| 283 |
mm_recursion (__m128i *a, __m128i *b, __m128i c, __m128i d, __m128i mask); |
| 284 |
|
| 285 |
/** |
| 286 |
* This function represents the recursion formula. |
| 287 |
* @param a a 128-bit part of the interal state array |
| 288 |
* @param b a 128-bit part of the interal state array |
| 289 |
* @param c a 128-bit part of the interal state array |
| 290 |
* @param d a 128-bit part of the interal state array |
| 291 |
* @param mask 128-bit mask |
| 292 |
* @return output |
| 293 |
*/ |
| 294 |
inline static __m128i mm_recursion (__m128i *a, __m128i *b, __m128i c, __m128i d, __m128i mask) |
| 295 |
{ |
| 296 |
__m128i v, x, y, z; |
| 297 |
|
| 298 |
x = _mm_load_si128 (a); |
| 299 |
y = _mm_srli_epi32 (*b, SR1); |
| 300 |
z = _mm_srli_si128 (c, SR2); |
| 301 |
v = _mm_slli_epi32 (d, SL1); |
| 302 |
z = _mm_xor_si128 (z, x); |
| 303 |
z = _mm_xor_si128 (z, v); |
| 304 |
x = _mm_slli_si128 (x, SL2); |
| 305 |
y = _mm_and_si128 (y, mask); |
| 306 |
z = _mm_xor_si128 (z, x); |
| 307 |
z = _mm_xor_si128 (z, y); |
| 308 |
return z; |
| 309 |
} |
| 310 |
|
| 311 |
/** |
| 312 |
* This function fills the internal state array with pseudorandom |
| 313 |
* integers. |
| 314 |
*/ |
| 315 |
inline static void |
| 316 |
gen_rand_all (void) |
| 317 |
{ |
| 318 |
int i; |
| 319 |
__m128i r, r1, r2, mask; |
| 320 |
mask = _mm_set_epi32 (MSK4, MSK3, MSK2, MSK1); |
| 321 |
|
| 322 |
r1 = _mm_load_si128 (&sfmt[N - 2].si); |
| 323 |
r2 = _mm_load_si128 (&sfmt[N - 1].si); |
| 324 |
for (i = 0; i < N - POS1; i++) |
| 325 |
{ |
| 326 |
r = mm_recursion (&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask); |
| 327 |
_mm_store_si128 (&sfmt[i].si, r); |
| 328 |
r1 = r2; |
| 329 |
r2 = r; |
| 330 |
} |
| 331 |
for (; i < N; i++) |
| 332 |
{ |
| 333 |
r = mm_recursion (&sfmt[i].si, &sfmt[i + POS1 - N].si, r1, r2, mask); |
| 334 |
_mm_store_si128 (&sfmt[i].si, r); |
| 335 |
r1 = r2; |
| 336 |
r2 = r; |
| 337 |
} |
| 338 |
} |
| 339 |
|
| 340 |
/** |
| 341 |
* This function fills the user-specified array with pseudorandom |
| 342 |
* integers. |
| 343 |
* |
| 344 |
* @param array an 128-bit array to be filled by pseudorandom numbers. |
| 345 |
* @param size number of 128-bit pesudorandom numbers to be generated. |
| 346 |
*/ |
| 347 |
inline static void |
| 348 |
gen_rand_array (w128_t *array, int size) |
| 349 |
{ |
| 350 |
int i, j; |
| 351 |
__m128i r, r1, r2, mask; |
| 352 |
mask = _mm_set_epi32 (MSK4, MSK3, MSK2, MSK1); |
| 353 |
|
| 354 |
r1 = _mm_load_si128 (&sfmt[N - 2].si); |
| 355 |
r2 = _mm_load_si128 (&sfmt[N - 1].si); |
| 356 |
for (i = 0; i < N - POS1; i++) |
| 357 |
{ |
| 358 |
r = mm_recursion (&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask); |
| 359 |
_mm_store_si128 (&array[i].si, r); |
| 360 |
r1 = r2; |
| 361 |
r2 = r; |
| 362 |
} |
| 363 |
for (; i < N; i++) |
| 364 |
{ |
| 365 |
r = mm_recursion (&sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask); |
| 366 |
_mm_store_si128 (&array[i].si, r); |
| 367 |
r1 = r2; |
| 368 |
r2 = r; |
| 369 |
} |
| 370 |
/* main loop */ |
| 371 |
for (; i < size - N; i++) |
| 372 |
{ |
| 373 |
r = mm_recursion (&array[i - N].si, &array[i + POS1 - N].si, r1, r2, mask); |
| 374 |
_mm_store_si128 (&array[i].si, r); |
| 375 |
r1 = r2; |
| 376 |
r2 = r; |
| 377 |
} |
| 378 |
for (j = 0; j < 2 * N - size; j++) |
| 379 |
{ |
| 380 |
r = _mm_load_si128 (&array[j + size - N].si); |
| 381 |
_mm_store_si128 (&sfmt[j].si, r); |
| 382 |
} |
| 383 |
for (; i < size; i++) |
| 384 |
{ |
| 385 |
r = mm_recursion (&array[i - N].si, &array[i + POS1 - N].si, r1, r2, mask); |
| 386 |
_mm_store_si128 (&array[i].si, r); |
| 387 |
_mm_store_si128 (&sfmt[j++].si, r); |
| 388 |
r1 = r2; |
| 389 |
r2 = r; |
| 390 |
} |
| 391 |
} |
| 392 |
#endif |
| 393 |
|
| 394 |
/** |
| 395 |
* This function simulate a 64-bit index of LITTLE ENDIAN |
| 396 |
* in BIG ENDIAN machine. |
| 397 |
*/ |
| 398 |
#ifdef ONLY64 |
| 399 |
inline static int |
| 400 |
idxof (int i) |
| 401 |
{ |
| 402 |
return i ^ 1; |
| 403 |
} |
| 404 |
#else |
| 405 |
inline static int |
| 406 |
idxof (int i) |
| 407 |
{ |
| 408 |
return i; |
| 409 |
} |
| 410 |
#endif |
| 411 |
/** |
| 412 |
* This function simulates SIMD 128-bit right shift by the standard C. |
| 413 |
* The 128-bit integer given in in is shifted by (shift * 8) bits. |
| 414 |
* This function simulates the LITTLE ENDIAN SIMD. |
| 415 |
* @param out the output of this function |
| 416 |
* @param in the 128-bit data to be shifted |
| 417 |
* @param shift the shift value |
| 418 |
*/ |
| 419 |
#ifdef ONLY64 |
| 420 |
inline static void |
| 421 |
rshift128 (w128_t *out, w128_t const *in, int shift) |
| 422 |
{ |
| 423 |
uint64_t th, tl, oh, ol; |
| 424 |
|
| 425 |
th = ((uint64_t) in->u[2] << 32) | ((uint64_t) in->u[3]); |
| 426 |
tl = ((uint64_t) in->u[0] << 32) | ((uint64_t) in->u[1]); |
| 427 |
|
| 428 |
oh = th >> (shift * 8); |
| 429 |
ol = tl >> (shift * 8); |
| 430 |
ol |= th << (64 - shift * 8); |
| 431 |
out->u[0] = (uint32_t) (ol >> 32); |
| 432 |
out->u[1] = (uint32_t) ol; |
| 433 |
out->u[2] = (uint32_t) (oh >> 32); |
| 434 |
out->u[3] = (uint32_t) oh; |
| 435 |
} |
| 436 |
#else |
| 437 |
inline static void |
| 438 |
rshift128 (w128_t *out, w128_t const *in, int shift) |
| 439 |
{ |
| 440 |
uint64_t th, tl, oh, ol; |
| 441 |
|
| 442 |
th = ((uint64_t) in->u[3] << 32) | ((uint64_t) in->u[2]); |
| 443 |
tl = ((uint64_t) in->u[1] << 32) | ((uint64_t) in->u[0]); |
| 444 |
|
| 445 |
oh = th >> (shift * 8); |
| 446 |
ol = tl >> (shift * 8); |
| 447 |
ol |= th << (64 - shift * 8); |
| 448 |
out->u[1] = (uint32_t) (ol >> 32); |
| 449 |
out->u[0] = (uint32_t) ol; |
| 450 |
out->u[3] = (uint32_t) (oh >> 32); |
| 451 |
out->u[2] = (uint32_t) oh; |
| 452 |
} |
| 453 |
#endif |
| 454 |
/** |
| 455 |
* This function simulates SIMD 128-bit left shift by the standard C. |
| 456 |
* The 128-bit integer given in in is shifted by (shift * 8) bits. |
| 457 |
* This function simulates the LITTLE ENDIAN SIMD. |
| 458 |
* @param out the output of this function |
| 459 |
* @param in the 128-bit data to be shifted |
| 460 |
* @param shift the shift value |
| 461 |
*/ |
| 462 |
#ifdef ONLY64 |
| 463 |
inline static void |
| 464 |
lshift128 (w128_t *out, w128_t const *in, int shift) |
| 465 |
{ |
| 466 |
uint64_t th, tl, oh, ol; |
| 467 |
|
| 468 |
th = ((uint64_t) in->u[2] << 32) | ((uint64_t) in->u[3]); |
| 469 |
tl = ((uint64_t) in->u[0] << 32) | ((uint64_t) in->u[1]); |
| 470 |
|
| 471 |
oh = th << (shift * 8); |
| 472 |
ol = tl << (shift * 8); |
| 473 |
oh |= tl >> (64 - shift * 8); |
| 474 |
out->u[0] = (uint32_t) (ol >> 32); |
| 475 |
out->u[1] = (uint32_t) ol; |
| 476 |
out->u[2] = (uint32_t) (oh >> 32); |
| 477 |
out->u[3] = (uint32_t) oh; |
| 478 |
} |
| 479 |
#else |
| 480 |
inline static void |
| 481 |
lshift128 (w128_t *out, w128_t const *in, int shift) |
| 482 |
{ |
| 483 |
uint64_t th, tl, oh, ol; |
| 484 |
|
| 485 |
th = ((uint64_t) in->u[3] << 32) | ((uint64_t) in->u[2]); |
| 486 |
tl = ((uint64_t) in->u[1] << 32) | ((uint64_t) in->u[0]); |
| 487 |
|
| 488 |
oh = th << (shift * 8); |
| 489 |
ol = tl << (shift * 8); |
| 490 |
oh |= tl >> (64 - shift * 8); |
| 491 |
out->u[1] = (uint32_t) (ol >> 32); |
| 492 |
out->u[0] = (uint32_t) ol; |
| 493 |
out->u[3] = (uint32_t) (oh >> 32); |
| 494 |
out->u[2] = (uint32_t) oh; |
| 495 |
} |
| 496 |
#endif |
| 497 |
|
| 498 |
/** |
| 499 |
* This function represents the recursion formula. |
| 500 |
* @param r output |
| 501 |
* @param a a 128-bit part of the internal state array |
| 502 |
* @param b a 128-bit part of the internal state array |
| 503 |
* @param c a 128-bit part of the internal state array |
| 504 |
* @param d a 128-bit part of the internal state array |
| 505 |
*/ |
| 506 |
#ifdef ONLY64 |
| 507 |
inline static void |
| 508 |
do_recursion (w128_t *r, w128_t *a, w128_t *b, w128_t *c, w128_t *d) |
| 509 |
{ |
| 510 |
w128_t x; |
| 511 |
w128_t y; |
| 512 |
|
| 513 |
lshift128 (&x, a, SL2); |
| 514 |
rshift128 (&y, c, SR2); |
| 515 |
r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK2) ^ y.u[0] ^ (d->u[0] << SL1); |
| 516 |
r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK1) ^ y.u[1] ^ (d->u[1] << SL1); |
| 517 |
r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK4) ^ y.u[2] ^ (d->u[2] << SL1); |
| 518 |
r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK3) ^ y.u[3] ^ (d->u[3] << SL1); |
| 519 |
} |
| 520 |
#else |
| 521 |
inline static void |
| 522 |
do_recursion (w128_t *r, w128_t *a, w128_t *b, w128_t *c, w128_t *d) |
| 523 |
{ |
| 524 |
w128_t x; |
| 525 |
w128_t y; |
| 526 |
|
| 527 |
lshift128 (&x, a, SL2); |
| 528 |
rshift128 (&y, c, SR2); |
| 529 |
r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK1) ^ y.u[0] ^ (d->u[0] << SL1); |
| 530 |
r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK2) ^ y.u[1] ^ (d->u[1] << SL1); |
| 531 |
r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK3) ^ y.u[2] ^ (d->u[2] << SL1); |
| 532 |
r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK4) ^ y.u[3] ^ (d->u[3] << SL1); |
| 533 |
} |
| 534 |
#endif |
| 535 |
|
| 536 |
#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2)) |
| 537 |
/** |
| 538 |
* This function fills the internal state array with pseudorandom |
| 539 |
* integers. |
| 540 |
*/ |
| 541 |
inline static void |
| 542 |
gen_rand_all (void) |
| 543 |
{ |
| 544 |
int i; |
| 545 |
w128_t *r1, *r2; |
| 546 |
|
| 547 |
r1 = &sfmt[N - 2]; |
| 548 |
r2 = &sfmt[N - 1]; |
| 549 |
for (i = 0; i < N - POS1; i++) |
| 550 |
{ |
| 551 |
do_recursion (&sfmt[i], &sfmt[i], &sfmt[i + POS1], r1, r2); |
| 552 |
r1 = r2; |
| 553 |
r2 = &sfmt[i]; |
| 554 |
} |
| 555 |
for (; i < N; i++) |
| 556 |
{ |
| 557 |
do_recursion (&sfmt[i], &sfmt[i], &sfmt[i + POS1 - N], r1, r2); |
| 558 |
r1 = r2; |
| 559 |
r2 = &sfmt[i]; |
| 560 |
} |
| 561 |
} |
| 562 |
|
| 563 |
/** |
| 564 |
* This function fills the user-specified array with pseudorandom |
| 565 |
* integers. |
| 566 |
* |
| 567 |
* @param array an 128-bit array to be filled by pseudorandom numbers. |
| 568 |
* @param size number of 128-bit pseudorandom numbers to be generated. |
| 569 |
*/ |
| 570 |
inline static void |
| 571 |
gen_rand_array (w128_t *array, int size) |
| 572 |
{ |
| 573 |
int i, j; |
| 574 |
w128_t *r1, *r2; |
| 575 |
|
| 576 |
r1 = &sfmt[N - 2]; |
| 577 |
r2 = &sfmt[N - 1]; |
| 578 |
for (i = 0; i < N - POS1; i++) |
| 579 |
{ |
| 580 |
do_recursion (&array[i], &sfmt[i], &sfmt[i + POS1], r1, r2); |
| 581 |
r1 = r2; |
| 582 |
r2 = &array[i]; |
| 583 |
} |
| 584 |
for (; i < N; i++) |
| 585 |
{ |
| 586 |
do_recursion (&array[i], &sfmt[i], &array[i + POS1 - N], r1, r2); |
| 587 |
r1 = r2; |
| 588 |
r2 = &array[i]; |
| 589 |
} |
| 590 |
for (; i < size - N; i++) |
| 591 |
{ |
| 592 |
do_recursion (&array[i], &array[i - N], &array[i + POS1 - N], r1, r2); |
| 593 |
r1 = r2; |
| 594 |
r2 = &array[i]; |
| 595 |
} |
| 596 |
for (j = 0; j < 2 * N - size; j++) |
| 597 |
{ |
| 598 |
sfmt[j] = array[j + size - N]; |
| 599 |
} |
| 600 |
for (; i < size; i++, j++) |
| 601 |
{ |
| 602 |
do_recursion (&array[i], &array[i - N], &array[i + POS1 - N], r1, r2); |
| 603 |
r1 = r2; |
| 604 |
r2 = &array[i]; |
| 605 |
sfmt[j] = array[i]; |
| 606 |
} |
| 607 |
} |
| 608 |
#endif |
| 609 |
|
| 610 |
#if defined(BIG_ENDIAN64) && !defined(ONLY64) && !defined(HAVE_ALTIVEC) |
| 611 |
inline static void |
| 612 |
swap (w128_t *array, int size) |
| 613 |
{ |
| 614 |
int i; |
| 615 |
uint32_t x, y; |
| 616 |
|
| 617 |
for (i = 0; i < size; i++) |
| 618 |
{ |
| 619 |
x = array[i].u[0]; |
| 620 |
y = array[i].u[2]; |
| 621 |
array[i].u[0] = array[i].u[1]; |
| 622 |
array[i].u[2] = array[i].u[3]; |
| 623 |
array[i].u[1] = x; |
| 624 |
array[i].u[3] = y; |
| 625 |
} |
| 626 |
} |
| 627 |
#endif |
| 628 |
/** |
| 629 |
* This function represents a function used in the initialization |
| 630 |
* by init_by_array |
| 631 |
* @param x 32-bit integer |
| 632 |
* @return 32-bit integer |
| 633 |
*/ |
| 634 |
inline static uint32_t |
| 635 |
func1 (uint32_t x) |
| 636 |
{ |
| 637 |
return (x ^ (x >> 27)) * (uint32_t) 1664525UL; |
| 638 |
} |
| 639 |
|
| 640 |
/** |
| 641 |
* This function represents a function used in the initialization |
| 642 |
* by init_by_array |
| 643 |
* @param x 32-bit integer |
| 644 |
* @return 32-bit integer |
| 645 |
*/ |
| 646 |
inline static uint32_t |
| 647 |
func2 (uint32_t x) |
| 648 |
{ |
| 649 |
return (x ^ (x >> 27)) * (uint32_t) 1566083941UL; |
| 650 |
} |
| 651 |
|
| 652 |
/** |
| 653 |
* This function certificate the period of 2^{MEXP} |
| 654 |
*/ |
| 655 |
static void |
| 656 |
period_certification (void) |
| 657 |
{ |
| 658 |
int inner = 0; |
| 659 |
int i, j; |
| 660 |
uint32_t work; |
| 661 |
|
| 662 |
for (i = 0; i < 4; i++) |
| 663 |
inner ^= psfmt32[idxof (i)] & parity[i]; |
| 664 |
for (i = 16; i > 0; i >>= 1) |
| 665 |
inner ^= inner >> i; |
| 666 |
inner &= 1; |
| 667 |
/* check OK */ |
| 668 |
if (inner == 1) |
| 669 |
{ |
| 670 |
return; |
| 671 |
} |
| 672 |
/* check NG, and modification */ |
| 673 |
for (i = 0; i < 4; i++) |
| 674 |
{ |
| 675 |
work = 1; |
| 676 |
for (j = 0; j < 32; j++) |
| 677 |
{ |
| 678 |
if ((work & parity[i]) != 0) |
| 679 |
{ |
| 680 |
psfmt32[idxof (i)] ^= work; |
| 681 |
return; |
| 682 |
} |
| 683 |
work = work << 1; |
| 684 |
} |
| 685 |
} |
| 686 |
} |
| 687 |
|
| 688 |
/*---------------- |
| 689 |
PUBLIC FUNCTIONS |
| 690 |
----------------*/ |
| 691 |
/** |
| 692 |
* This function returns the identification string. |
| 693 |
* The string shows the word size, the Mersenne exponent, |
| 694 |
* and all parameters of this generator. |
| 695 |
*/ |
| 696 |
char const * const |
| 697 |
get_idstring (void) |
| 698 |
{ |
| 699 |
return IDSTR; |
| 700 |
} |
| 701 |
|
| 702 |
/** |
| 703 |
* This function returns the minimum size of array used for \b |
| 704 |
* fill_array32() function. |
| 705 |
* @return minimum size of array used for fill_array32() function. |
| 706 |
*/ |
| 707 |
int |
| 708 |
get_min_array_size32 (void) |
| 709 |
{ |
| 710 |
return N32; |
| 711 |
} |
| 712 |
|
| 713 |
/** |
| 714 |
* This function returns the minimum size of array used for \b |
| 715 |
* fill_array64() function. |
| 716 |
* @return minimum size of array used for fill_array64() function. |
| 717 |
*/ |
| 718 |
int |
| 719 |
get_min_array_size64 (void) |
| 720 |
{ |
| 721 |
return N64; |
| 722 |
} |
| 723 |
|
| 724 |
#ifndef ONLY64 |
| 725 |
/** |
| 726 |
* This function generates and returns 32-bit pseudorandom number. |
| 727 |
* init_gen_rand or init_by_array must be called before this function. |
| 728 |
* @return 32-bit pseudorandom number |
| 729 |
*/ |
| 730 |
uint32_t |
| 731 |
gen_rand32 (void) |
| 732 |
{ |
| 733 |
uint32_t r; |
| 734 |
|
| 735 |
assert (initialized); |
| 736 |
if (idx >= N32) |
| 737 |
{ |
| 738 |
gen_rand_all (); |
| 739 |
idx = 0; |
| 740 |
} |
| 741 |
r = psfmt32[idx++]; |
| 742 |
return r; |
| 743 |
} |
| 744 |
#endif |
| 745 |
/** |
| 746 |
* This function generates and returns 64-bit pseudorandom number. |
| 747 |
* init_gen_rand or init_by_array must be called before this function. |
| 748 |
* The function gen_rand64 should not be called after gen_rand32, |
| 749 |
* unless an initialization is again executed. |
| 750 |
* @return 64-bit pseudorandom number |
| 751 |
*/ |
| 752 |
uint64_t |
| 753 |
gen_rand64 (void) |
| 754 |
{ |
| 755 |
#if defined(BIG_ENDIAN64) && !defined(ONLY64) |
| 756 |
uint32_t r1, r2; |
| 757 |
#else |
| 758 |
uint64_t r; |
| 759 |
#endif |
| 760 |
|
| 761 |
assert (initialized); |
| 762 |
assert (idx % 2 == 0); |
| 763 |
|
| 764 |
if (idx >= N32) |
| 765 |
{ |
| 766 |
gen_rand_all (); |
| 767 |
idx = 0; |
| 768 |
} |
| 769 |
#if defined(BIG_ENDIAN64) && !defined(ONLY64) |
| 770 |
r1 = psfmt32[idx]; |
| 771 |
r2 = psfmt32[idx + 1]; |
| 772 |
idx += 2; |
| 773 |
return ((uint64_t) r2 << 32) | r1; |
| 774 |
#else |
| 775 |
r = psfmt64[idx / 2]; |
| 776 |
idx += 2; |
| 777 |
return r; |
| 778 |
#endif |
| 779 |
} |
| 780 |
|
| 781 |
#ifndef ONLY64 |
| 782 |
/** |
| 783 |
* This function generates pseudorandom 32-bit integers in the |
| 784 |
* specified array[] by one call. The number of pseudorandom integers |
| 785 |
* is specified by the argument size, which must be at least 624 and a |
| 786 |
* multiple of four. The generation by this function is much faster |
| 787 |
* than the following gen_rand function. |
| 788 |
* |
| 789 |
* For initialization, init_gen_rand or init_by_array must be called |
| 790 |
* before the first call of this function. This function can not be |
| 791 |
* used after calling gen_rand function, without initialization. |
| 792 |
* |
| 793 |
* @param array an array where pseudorandom 32-bit integers are filled |
| 794 |
* by this function. The pointer to the array must be \b "aligned" |
| 795 |
* (namely, must be a multiple of 16) in the SIMD version, since it |
| 796 |
* refers to the address of a 128-bit integer. In the standard C |
| 797 |
* version, the pointer is arbitrary. |
| 798 |
* |
| 799 |
* @param size the number of 32-bit pseudorandom integers to be |
| 800 |
* generated. size must be a multiple of 4, and greater than or equal |
| 801 |
* to (MEXP / 128 + 1) * 4. |
| 802 |
* |
| 803 |
* @note \b memalign or \b posix_memalign is available to get aligned |
| 804 |
* memory. Mac OSX doesn't have these functions, but \b malloc of OSX |
| 805 |
* returns the pointer to the aligned memory block. |
| 806 |
*/ |
| 807 |
void |
| 808 |
fill_array32 (uint32_t * array, int size) |
| 809 |
{ |
| 810 |
assert (initialized); |
| 811 |
assert (idx == N32); |
| 812 |
assert (size % 4 == 0); |
| 813 |
assert (size >= N32); |
| 814 |
|
| 815 |
gen_rand_array (reinterpret_cast<w128_t *> (array), size / 4); |
| 816 |
idx = N32; |
| 817 |
} |
| 818 |
#endif |
| 819 |
|
| 820 |
/** |
| 821 |
* This function generates pseudorandom 64-bit integers in the |
| 822 |
* specified array[] by one call. The number of pseudorandom integers |
| 823 |
* is specified by the argument size, which must be at least 312 and a |
| 824 |
* multiple of two. The generation by this function is much faster |
| 825 |
* than the following gen_rand function. |
| 826 |
* |
| 827 |
* For initialization, init_gen_rand or init_by_array must be called |
| 828 |
* before the first call of this function. This function can not be |
| 829 |
* used after calling gen_rand function, without initialization. |
| 830 |
* |
| 831 |
* @param array an array where pseudorandom 64-bit integers are filled |
| 832 |
* by this function. The pointer to the array must be "aligned" |
| 833 |
* (namely, must be a multiple of 16) in the SIMD version, since it |
| 834 |
* refers to the address of a 128-bit integer. In the standard C |
| 835 |
* version, the pointer is arbitrary. |
| 836 |
* |
| 837 |
* @param size the number of 64-bit pseudorandom integers to be |
| 838 |
* generated. size must be a multiple of 2, and greater than or equal |
| 839 |
* to (MEXP / 128 + 1) * 2 |
| 840 |
* |
| 841 |
* @note \b memalign or \b posix_memalign is available to get aligned |
| 842 |
* memory. Mac OSX doesn't have these functions, but \b malloc of OSX |
| 843 |
* returns the pointer to the aligned memory block. |
| 844 |
*/ |
| 845 |
void |
| 846 |
fill_array64 (uint64_t * array, int size) |
| 847 |
{ |
| 848 |
assert (initialized); |
| 849 |
assert (idx == N32); |
| 850 |
assert (size % 2 == 0); |
| 851 |
assert (size >= N64); |
| 852 |
|
| 853 |
gen_rand_array (reinterpret_cast<w128_t *> (array), size / 2); |
| 854 |
idx = N32; |
| 855 |
|
| 856 |
#if defined(BIG_ENDIAN64) && !defined(ONLY64) |
| 857 |
swap (reinterpret_cast<w128_t *> (array), size / 2); |
| 858 |
#endif |
| 859 |
} |
| 860 |
|
| 861 |
/** |
| 862 |
* This function initializes the internal state array with a 32-bit |
| 863 |
* integer seed. |
| 864 |
* |
| 865 |
* @param seed a 32-bit integer used as the seed. |
| 866 |
*/ |
| 867 |
void |
| 868 |
init_gen_rand (uint32_t seed) |
| 869 |
{ |
| 870 |
int i; |
| 871 |
|
| 872 |
psfmt32[idxof (0)] = seed; |
| 873 |
for (i = 1; i < N32; i++) |
| 874 |
{ |
| 875 |
psfmt32[idxof (i)] = 1812433253UL * (psfmt32[idxof (i - 1)] ^ (psfmt32[idxof (i - 1)] >> 30)) + i; |
| 876 |
} |
| 877 |
idx = N32; |
| 878 |
period_certification (); |
| 879 |
initialized = 1; |
| 880 |
} |
| 881 |
|
| 882 |
/** |
| 883 |
* This function initializes the internal state array, |
| 884 |
* with an array of 32-bit integers used as the seeds |
| 885 |
* @param init_key the array of 32-bit integers, used as a seed. |
| 886 |
* @param key_length the length of init_key. |
| 887 |
*/ |
| 888 |
void |
| 889 |
init_by_array (uint32_t * init_key, int key_length) |
| 890 |
{ |
| 891 |
int i, j, count; |
| 892 |
uint32_t r; |
| 893 |
int lag; |
| 894 |
int mid; |
| 895 |
int size = N * 4; |
| 896 |
|
| 897 |
if (size >= 623) |
| 898 |
{ |
| 899 |
lag = 11; |
| 900 |
} |
| 901 |
else if (size >= 68) |
| 902 |
{ |
| 903 |
lag = 7; |
| 904 |
} |
| 905 |
else if (size >= 39) |
| 906 |
{ |
| 907 |
lag = 5; |
| 908 |
} |
| 909 |
else |
| 910 |
{ |
| 911 |
lag = 3; |
| 912 |
} |
| 913 |
mid = (size - lag) / 2; |
| 914 |
|
| 915 |
memset (sfmt, 0x8b, sizeof (sfmt)); |
| 916 |
if (key_length + 1 > N32) |
| 917 |
{ |
| 918 |
count = key_length + 1; |
| 919 |
} |
| 920 |
else |
| 921 |
{ |
| 922 |
count = N32; |
| 923 |
} |
| 924 |
r = func1 (psfmt32[idxof (0)] ^ psfmt32[idxof (mid)] ^ psfmt32[idxof (N32 - 1)]); |
| 925 |
psfmt32[idxof (mid)] += r; |
| 926 |
r += key_length; |
| 927 |
psfmt32[idxof (mid + lag)] += r; |
| 928 |
psfmt32[idxof (0)] = r; |
| 929 |
|
| 930 |
count--; |
| 931 |
for (i = 1, j = 0; (j < count) && (j < key_length); j++) |
| 932 |
{ |
| 933 |
r = func1 (psfmt32[idxof (i)] ^ psfmt32[idxof ((i + mid) % N32)] ^ psfmt32[idxof ((i + N32 - 1) % N32)]); |
| 934 |
psfmt32[idxof ((i + mid) % N32)] += r; |
| 935 |
r += init_key[j] + i; |
| 936 |
psfmt32[idxof ((i + mid + lag) % N32)] += r; |
| 937 |
psfmt32[idxof (i)] = r; |
| 938 |
i = (i + 1) % N32; |
| 939 |
} |
| 940 |
for (; j < count; j++) |
| 941 |
{ |
| 942 |
r = func1 (psfmt32[idxof (i)] ^ psfmt32[idxof ((i + mid) % N32)] ^ psfmt32[idxof ((i + N32 - 1) % N32)]); |
| 943 |
psfmt32[idxof ((i + mid) % N32)] += r; |
| 944 |
r += i; |
| 945 |
psfmt32[idxof ((i + mid + lag) % N32)] += r; |
| 946 |
psfmt32[idxof (i)] = r; |
| 947 |
i = (i + 1) % N32; |
| 948 |
} |
| 949 |
for (j = 0; j < N32; j++) |
| 950 |
{ |
| 951 |
r = func2 (psfmt32[idxof (i)] + psfmt32[idxof ((i + mid) % N32)] + psfmt32[idxof ((i + N32 - 1) % N32)]); |
| 952 |
psfmt32[idxof ((i + mid) % N32)] ^= r; |
| 953 |
r -= i; |
| 954 |
psfmt32[idxof ((i + mid + lag) % N32)] ^= r; |
| 955 |
psfmt32[idxof (i)] = r; |
| 956 |
i = (i + 1) % N32; |
| 957 |
} |
| 958 |
|
| 959 |
idx = N32; |
| 960 |
period_certification (); |
| 961 |
initialized = 1; |
| 962 |
} |