/**
* @file SFMT.c
* @brief SIMD oriented Fast Mersenne Twister(SFMT)
*
* modified by Emre �zg�r �nce (emre_lineus@yahoo.com)
*
* Original implementation:
* Mutsuo Saito (Hiroshima University) and Makoto Matsumoto (Hiroshima University)
* Copyright (C) 2006, 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
* University. All rights reserved.
* The (modified) BSD License is applied to this software, see Original License/LICENSE.txt
*/
#include <string.h>
#include <assert.h>
#include "SFMT.h"
#include "SFMT-params.h"
/*------------------------------------------------------
128-bit SIMD data type for SSE2 or standard C
------------------------------------------------------*/
#if 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
internal state, index counter and flag
--------------------------------------*/
/** 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];
/** the 64bit integer pointer to the 128-bit internal state array */
static uint64_t *psfmt64 = (uint64_t *)&sfmt[0].u[0];
/** 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};
/*----------------
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);
inline static void gen_rand_all(void);
inline static void gen_rand_array(w128_t *array, int size);
inline static uint32_t func1(uint32_t x);
inline static uint32_t func2(uint32_t x);
static void period_certification(void);
#if defined(HAVE_SSE2)
#include "SFMT-sse2.h"
#endif
inline static int idxof(int i) {
return i;
}
/**
* 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
*/
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;
}
/**
* 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
*/
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;
}
/**
* 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
*/
#if (!defined(HAVE_SSE2))
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);
r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK1) ^ y.u[0]
^ (d->u[0] << SL1);
r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK2) ^ y.u[1]
^ (d->u[1] << SL1);
r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK3) ^ y.u[2]
^ (d->u[2] << SL1);
r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK4) ^ y.u[3]
^ (d->u[3] << SL1);
}
#endif
#if (!defined(HAVE_SSE2))
/**
* This function fills the internal state array with pseudorandom
* integers.
*/
inline static void gen_rand_all(void) {
int i;
w128_t *r1, *r2;
r1 = &sfmt[N - 2];
r2 = &sfmt[N - 1];
for (i = 0; i < N - POS1; i++) {
do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
r1 = r2;
r2 = &sfmt[i];
}
for (; i < N; i++) {
do_recursion(&sfmt[i], &sfmt[i], &sfmt[i + POS1 - N], r1, r2);
r1 = r2;
r2 = &sfmt[i];
}
}
/**
* This function fills the user-specified array with pseudorandom
* integers.
*
* @param array an 128-bit array to be filled by pseudorandom numbers.
* @param size number of 128-bit pseudorandom numbers to be generated.
*/
inline static void gen_rand_array(w128_t *array, int size) {
int i, j;
w128_t *r1, *r2;
r1 = &sfmt[N - 2];
r2 = &sfmt[N - 1];
for (i = 0; i < N - POS1; i++) {
do_recursion(&array[i], &sfmt[i], &sfmt[i + POS1], r1, r2);
r1 = r2;
r2 = &array[i];
}
for (; i < N; i++) {
do_recursion(&array[i], &sfmt[i], &array[i + POS1 - N], r1, r2);
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++) {
sfmt[j] = array[j + size - N];
}
for (; i < size; i++, j++) {
do_recursion(&array[i], &array[i - N], &array[i + POS1 - N], r1, r2);
r1 = r2;
r2 = &array[i];
sfmt[j] = array[i];
}
}
#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}
*/
static void period_certification(void) {
int inner = 0;
int i, j;
uint32_t work;
for (i = 0; i < 4; i++)
inner ^= psfmt32[idxof(i)] & parity[i];
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++) {
if ((work & parity[i]) != 0) {
psfmt32[idxof(i)] ^= work;
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;
}
/**
* 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
*/
uint32_t gen_rand32(void) {
uint32_t r;
assert(initialized);
if (idx >= N32) {
gen_rand_all();
idx = 0;
}
r = psfmt32[idx++];
return r;
}
/**
* 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,
* unless an initialization is again executed.
* @return 64-bit pseudorandom number
*/
uint64_t gen_rand64(void) {
uint64_t r;
assert(initialized);
assert(idx % 2 == 0);
if (idx >= N32) {
gen_rand_all();
idx = 0;
}
r = psfmt64[idx / 2];
idx += 2;
return r;
}
/**
* This function is used to determine extended size of specified array[]
* in the fill_array32 function.
* Because, array size must be greater than or equal to (MEXP / 128 + 1) * 4
* so, let's fulfill the array if the size smaller than (MEXP / 128 + 1) * 4
* Because, array size must be a multiple of 4.
* so, let's fulfill the array.
*/
int get_array32_extended_size(int size) {
int extended_size = 0;
int remainder = 0;
if (size < get_min_array_size32())
extended_size = get_min_array_size32();
else
extended_size = size;
remainder = extended_size % 4;
extended_size = extended_size + 4 - remainder;
return extended_size;
}
/**
* 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.
*/
int fill_array32(uint32_t *array, int size) {
int* ptr;
int extended_size = get_array32_extended_size(size);
assert(initialized);
assert(idx == N32);
/* The pointer to the allocated memory must be \b "aligned"
* (namely, must be a multiple of 16) in the SIMD version (HAVE_SSE2), since it
* refers to the address of a 128-bit integer. In the standard C
* version, the pointer is arbitrary.
*/
#if defined(HAVE_SSE2)
ptr = (w128_t *) _aligned_malloc(sizeof(uint32_t) * extended_size, 16);
#else
ptr = (w128_t *) _aligned_malloc(sizeof(uint32_t) * extended_size, __alignof(uint32_t));
#endif
if (ptr == NULL)
return 0;
else {
gen_rand_array(ptr, extended_size / 4);
memcpy((w128_t *)array, ptr, sizeof(uint32_t) * size);
idx = N32;
_aligned_free(ptr);
}
return 1;
}
/**
* This function is used to determine extended size of specified array[]
* in the fill_array64 function.
* Because, array size must be greater than or equal to (MEXP / 128 + 1) * 2
* so, let's fulfill the array if the size smaller than (MEXP / 128 + 1) * 2
* Because, array size must be a multiple of 2.
* so, let's fulfill the array.
*/
int get_array64_extended_size(int size) {
int extended_size = 0;
int remainder = 0;
if (size < get_min_array_size64())
extended_size = get_min_array_size64();
else
extended_size = size;
remainder = extended_size % 2;
extended_size = extended_size + 2 - remainder;
return extended_size;
}
/**
* 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.
*/
int fill_array64(uint64_t *array, int size) {
int* ptr;
int extended_size = get_array64_extended_size(size);
assert(initialized);
assert(idx == N32);
/* The pointer to the allocated memory must be \b "aligned"
* (namely, must be a multiple of 16) in the SIMD version (HAVE_SSE2), since it
* refers to the address of a 128-bit integer. In the standard C
* version, the pointer is arbitrary.
*/
#if defined(HAVE_SSE2)
ptr = (w128_t *) _aligned_malloc(sizeof(uint64_t) * extended_size, 16);
#else
ptr = (w128_t *) _aligned_malloc(sizeof(uint64_t) * extended_size, __alignof(uint64_t));
#endif
if (ptr == NULL)
return 0;
else {
gen_rand_array(ptr, extended_size / 2);
memcpy((w128_t *)array, ptr, sizeof(uint64_t) * size);
idx = N32;
_aligned_free(ptr);
}
return 1;
}
/**
* This function initializes the internal state array with a 32-bit
* integer seed.
*
* @param seed a 32-bit integer used as the seed.
*/
void init_gen_rand(uint32_t seed) {
int i;
psfmt32[idxof(0)] = seed;
for (i = 1; i < N32; i++) {
psfmt32[idxof(i)] = 1812433253UL * (psfmt32[idxof(i - 1)]
^ (psfmt32[idxof(i - 1)] >> 30))
+ i;
}
idx = N32;
period_certification();
initialized = 1;
}
/**
* 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.
*/
void init_by_array(uint32_t *init_key, int key_length) {
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;
memset(sfmt, 0x8b, sizeof(sfmt));
if (key_length + 1 > N32) {
count = key_length + 1;
} else {
count = N32;
}
r = func1(psfmt32[idxof(0)] ^ psfmt32[idxof(mid)]
^ psfmt32[idxof(N32 - 1)]);
psfmt32[idxof(mid)] += r;
r += key_length;
psfmt32[idxof(mid + lag)] += r;
psfmt32[idxof(0)] = r;
count--;
for (i = 1, j = 0; (j < count) && (j < key_length); j++) {
r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % N32)]
^ psfmt32[idxof((i + N32 - 1) % N32)]);
psfmt32[idxof((i + mid) % N32)] += r;
r += init_key[j] + i;
psfmt32[idxof((i + mid + lag) % N32)] += r;
psfmt32[idxof(i)] = r;
i = (i + 1) % N32;
}
for (; j < count; j++) {
r = func1(psfmt32[idxof(i)] ^ psfmt32[idxof((i + mid) % N32)]
^ psfmt32[idxof((i + N32 - 1) % N32)]);
psfmt32[idxof((i + mid) % N32)] += r;
r += i;
psfmt32[idxof((i + mid + lag) % N32)] += r;
psfmt32[idxof(i)] = r;
i = (i + 1) % N32;
}
for (j = 0; j < N32; j++) {
r = func2(psfmt32[idxof(i)] + psfmt32[idxof((i + mid) % N32)]
+ psfmt32[idxof((i + N32 - 1) % N32)]);
psfmt32[idxof((i + mid) % N32)] ^= r;
r -= i;
psfmt32[idxof((i + mid + lag) % N32)] ^= r;
psfmt32[idxof(i)] = r;
i = (i + 1) % N32;
}
idx = N32;
period_certification();
initialized = 1;
}