Switch to using SIMD-oriented Fast Mersenne Twister for random number generation.

In testing, reduced random number generation time by a factor of 8-10.
Drops support for processors older than Pentium 4.
Drop Mersenne Twister library; use a C++ SFMT library.

--HG--
branch : trunk

6 files changed, 351 insertions, 1129 deletions

diff --git a/dep/SFMT/SFMT-alti.h b/dep/SFMT/SFMT-alti.h
deleted file mode 100644
index df3186cbbd7..00000000000
--- a/dep/SFMT/SFMT-alti.h
+++ /dev/null
@@ -1,156 +0,0 @@
-/** 
- * @file SFMT-alti.h 
- *
- * @brief SIMD oriented Fast Mersenne Twister(SFMT)
- * pseudorandom number generator
- *
- * @author Mutsuo Saito (Hiroshima University)
- * @author Makoto Matsumoto (Hiroshima University)
- *
- * Copyright (C) 2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
- * University. All rights reserved.
- *
- * The new BSD License is applied to this software.
- * see LICENSE.txt
- */
-
-#ifndef SFMT_ALTI_H
-#define SFMT_ALTI_H
-
-inline static vector unsigned int vec_recursion(vector unsigned int a,
-						vector unsigned int b,
-						vector unsigned int c,
-						vector unsigned int d)
-    ALWAYSINLINE;
-
-/**
- * This function represents the recursion formula in AltiVec and BIG ENDIAN.
- * @param a a 128-bit part of the interal state array
- * @param b a 128-bit part of the interal state array
- * @param c a 128-bit part of the interal state array
- * @param d a 128-bit part of the interal state array
- * @return output
- */
-inline static vector unsigned int vec_recursion(vector unsigned int a,
-						vector unsigned int b,
-						vector unsigned int c,
-						vector unsigned int d) {
-
-    const vector unsigned int sl1 = ALTI_SL1;
-    const vector unsigned int sr1 = ALTI_SR1;
-#ifdef ONLY64
-    const vector unsigned int mask = ALTI_MSK64;
-    const vector unsigned char perm_sl = ALTI_SL2_PERM64;
-    const vector unsigned char perm_sr = ALTI_SR2_PERM64;
-#else
-    const vector unsigned int mask = ALTI_MSK;
-    const vector unsigned char perm_sl = ALTI_SL2_PERM;
-    const vector unsigned char perm_sr = ALTI_SR2_PERM;
-#endif
-    vector unsigned int v, w, x, y, z;
-    x = vec_perm(a, (vector unsigned int)perm_sl, perm_sl);
-    v = a;
-    y = vec_sr(b, sr1);
-    z = vec_perm(c, (vector unsigned int)perm_sr, perm_sr);
-    w = vec_sl(d, sl1);
-    z = vec_xor(z, w);
-    y = vec_and(y, mask);
-    v = vec_xor(v, x);
-    z = vec_xor(z, y);
-    z = vec_xor(z, v);
-    return z;
-}
-
-/**
- * This function fills the internal state array with pseudorandom
- * integers.
- */
-inline static void gen_rand_all(void) {
-    int i;
-    vector unsigned int r, r1, r2;
-
-    r1 = sfmt[N - 2].s;
-    r2 = sfmt[N - 1].s;
-    for (i = 0; i < N - POS1; i++) {
-	r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
-	sfmt[i].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-    for (; i < N; i++) {
-	r = vec_recursion(sfmt[i].s, sfmt[i + POS1 - N].s, r1, r2);
-	sfmt[i].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-}
-
-/**
- * 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 pesudorandom numbers to be generated.
- */
-inline static void gen_rand_array(w128_t *array, int size) {
-    int i, j;
-    vector unsigned int r, r1, r2;
-
-    r1 = sfmt[N - 2].s;
-    r2 = sfmt[N - 1].s;
-    for (i = 0; i < N - POS1; i++) {
-	r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2);
-	array[i].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-    for (; i < N; i++) {
-	r = vec_recursion(sfmt[i].s, array[i + POS1 - N].s, r1, r2);
-	array[i].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-    /* main loop */
-    for (; i < size - N; i++) {
-	r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2);
-	array[i].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-    for (j = 0; j < 2 * N - size; j++) {
-	sfmt[j].s = array[j + size - N].s;
-    }
-    for (; i < size; i++) {
-	r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2);
-	array[i].s = r;
-	sfmt[j++].s = r;
-	r1 = r2;
-	r2 = r;
-    }
-}
-
-#ifndef ONLY64
-#if defined(__APPLE__)
-#define ALTI_SWAP (vector unsigned char) \
-	(4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11)
-#else
-#define ALTI_SWAP {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11}
-#endif
-/**
- * This function swaps high and low 32-bit of 64-bit integers in user
- * specified array.
- *
- * @param array an 128-bit array to be swaped.
- * @param size size of 128-bit array.
- */
-inline static void swap(w128_t *array, int size) {
-    int i;
-    const vector unsigned char perm = ALTI_SWAP;
-
-    for (i = 0; i < size; i++) {
-	array[i].s = vec_perm(array[i].s, (vector unsigned int)perm, perm);
-    }
-}
-#endif
-
-#endif
diff --git a/dep/SFMT/SFMT-params.h b/dep/SFMT/SFMT-params.h
deleted file mode 100644
index 661bbf26a28..00000000000
--- a/dep/SFMT/SFMT-params.h
+++ /dev/null
@@ -1,97 +0,0 @@
-#ifndef SFMT_PARAMS_H
-#define SFMT_PARAMS_H
-
-#if !defined(MEXP)
-#ifdef __GNUC__
-  #warning "MEXP is not defined. I assume MEXP is 19937."
-#endif
-  #define MEXP 19937
-#endif
-/*-----------------
-  BASIC DEFINITIONS
-  -----------------*/
-/** Mersenne Exponent. The period of the sequence 
- *  is a multiple of 2^MEXP-1.
- * #define MEXP 19937 */
-/** SFMT generator has an internal state array of 128-bit integers,
- * and N is its size. */
-#define N (MEXP / 128 + 1)
-/** N32 is the size of internal state array when regarded as an array
- * of 32-bit integers.*/
-#define N32 (N * 4)
-/** N64 is the size of internal state array when regarded as an array
- * of 64-bit integers.*/
-#define N64 (N * 2)
-
-/*----------------------
-  the parameters of SFMT
-  following definitions are in paramsXXXX.h file.
-  ----------------------*/
-/** the pick up position of the array.
-#define POS1 122 
-*/
-
-/** the parameter of shift left as four 32-bit registers.
-#define SL1 18
- */
-
-/** the parameter of shift left as one 128-bit register. 
- * The 128-bit integer is shifted by (SL2 * 8) bits. 
-#define SL2 1 
-*/
-
-/** the parameter of shift right as four 32-bit registers.
-#define SR1 11
-*/
-
-/** the parameter of shift right as one 128-bit register. 
- * The 128-bit integer is shifted by (SL2 * 8) bits. 
-#define SR2 1 
-*/
-
-/** A bitmask, used in the recursion.  These parameters are introduced
- * to break symmetry of SIMD.
-#define MSK1 0xdfffffefU
-#define MSK2 0xddfecb7fU
-#define MSK3 0xbffaffffU
-#define MSK4 0xbffffff6U 
-*/
-
-/** These definitions are part of a 128-bit period certification vector.
-#define PARITY1	0x00000001U
-#define PARITY2	0x00000000U
-#define PARITY3	0x00000000U
-#define PARITY4	0xc98e126aU
-*/
-
-#if MEXP == 607
-  #include "SFMT-params607.h"
-#elif MEXP == 1279
-  #include "SFMT-params1279.h"
-#elif MEXP == 2281
-  #include "SFMT-params2281.h"
-#elif MEXP == 4253
-  #include "SFMT-params4253.h"
-#elif MEXP == 11213
-  #include "SFMT-params11213.h"
-#elif MEXP == 19937
-  #include "SFMT-params19937.h"
-#elif MEXP == 44497
-  #include "SFMT-params44497.h"
-#elif MEXP == 86243
-  #include "SFMT-params86243.h"
-#elif MEXP == 132049
-  #include "SFMT-params132049.h"
-#elif MEXP == 216091
-  #include "SFMT-params216091.h"
-#else
-#ifdef __GNUC__
-  #error "MEXP is not valid."
-  #undef MEXP
-#else
-  #undef MEXP
-#endif
-
-#endif
-
-#endif /* SFMT_PARAMS_H */
diff --git a/dep/SFMT/SFMT-sse2.h b/dep/SFMT/SFMT-sse2.h
deleted file mode 100644
index 4e91d9c6121..00000000000
--- a/dep/SFMT/SFMT-sse2.h
+++ /dev/null
@@ -1,121 +0,0 @@
-/** 
- * @file  SFMT-sse2.h
- * @brief SIMD oriented Fast Mersenne Twister(SFMT) for Intel SSE2
- *
- * @author Mutsuo Saito (Hiroshima University)
- * @author Makoto Matsumoto (Hiroshima University)
- *
- * @note We assume LITTLE ENDIAN in this file
- *
- * 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
- */
-
-#ifndef SFMT_SSE2_H
-#define SFMT_SSE2_H
-
-PRE_ALWAYS static __m128i mm_recursion(__m128i *a, __m128i *b, __m128i c,
-				   __m128i d, __m128i mask) ALWAYSINLINE;
-
-/**
- * This function represents the recursion formula.
- * @param a a 128-bit part of the interal state array
- * @param b a 128-bit part of the interal state array
- * @param c a 128-bit part of the interal state array
- * @param d a 128-bit part of the interal state array
- * @param mask 128-bit mask
- * @return output
- */
-PRE_ALWAYS static __m128i mm_recursion(__m128i *a, __m128i *b, 
-				   __m128i c, __m128i d, __m128i mask) {
-    __m128i v, x, y, z;
-    
-    x = _mm_load_si128(a);
-    y = _mm_srli_epi32(*b, SR1);
-    z = _mm_srli_si128(c, SR2);
-    v = _mm_slli_epi32(d, SL1);
-    z = _mm_xor_si128(z, x);
-    z = _mm_xor_si128(z, v);
-    x = _mm_slli_si128(x, SL2);
-    y = _mm_and_si128(y, mask);
-    z = _mm_xor_si128(z, x);
-    z = _mm_xor_si128(z, y);
-    return z;
-}
-
-/**
- * This function fills the internal state array with pseudorandom
- * integers.
- */
-inline static void gen_rand_all(void) {
-    int i;
-    __m128i r, r1, r2, mask;
-    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
-
-    r1 = _mm_load_si128(&sfmt[N - 2].si);
-    r2 = _mm_load_si128(&sfmt[N - 1].si);
-    for (i = 0; i < N - POS1; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
-	_mm_store_si128(&sfmt[i].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-    for (; i < N; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1 - N].si, r1, r2, mask);
-	_mm_store_si128(&sfmt[i].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-}
-
-/**
- * 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 pesudorandom numbers to be generated.
- */
-inline static void gen_rand_array(w128_t *array, int size) {
-    int i, j;
-    __m128i r, r1, r2, mask;
-    mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
-
-    r1 = _mm_load_si128(&sfmt[N - 2].si);
-    r2 = _mm_load_si128(&sfmt[N - 1].si);
-    for (i = 0; i < N - POS1; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
-	_mm_store_si128(&array[i].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-    for (; i < N; i++) {
-	r = mm_recursion(&sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask);
-	_mm_store_si128(&array[i].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-    /* main loop */
-    for (; i < size - N; i++) {
-	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
-			 mask);
-	_mm_store_si128(&array[i].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-    for (j = 0; j < 2 * N - size; j++) {
-	r = _mm_load_si128(&array[j + size - N].si);
-	_mm_store_si128(&sfmt[j].si, r);
-    }
-    for (; i < size; i++) {
-	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
-			 mask);
-	_mm_store_si128(&array[i].si, r);
-	_mm_store_si128(&sfmt[j++].si, r);
-	r1 = r2;
-	r2 = r;
-    }
-}
-
-#endif
diff --git a/dep/SFMT/SFMT.c b/dep/SFMT/SFMT.c
deleted file mode 100644
index d36465d9e14..00000000000
--- a/dep/SFMT/SFMT.c
+++ /dev/null
@@ -1,620 +0,0 @@
-/** 
- * @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
- */
-#include <string.h>
-#include <assert.h>
-#include "SFMT.h"
-#include "SFMT-params.h"
-
-#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
-  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];
-#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};
-
-/*----------------
-  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(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
-
-/**
- * This function simulate a 64-bit index of LITTLE ENDIAN 
- * 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
- */
-#if (!defined(HAVE_ALTIVEC)) && (!defined(HAVE_SSE2))
-#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);
-    r->u[0] = a->u[0] ^ x.u[0] ^ ((b->u[0] >> SR1) & MSK2) ^ y.u[0] 
-	^ (d->u[0] << SL1);
-    r->u[1] = a->u[1] ^ x.u[1] ^ ((b->u[1] >> SR1) & MSK1) ^ y.u[1] 
-	^ (d->u[1] << SL1);
-    r->u[2] = a->u[2] ^ x.u[2] ^ ((b->u[2] >> SR1) & MSK4) ^ y.u[2] 
-	^ (d->u[2] << SL1);
-    r->u[3] = a->u[3] ^ x.u[3] ^ ((b->u[3] >> SR1) & MSK3) ^ y.u[3] 
-	^ (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);
-    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
-#endif
-
-#if (!defined(HAVE_ALTIVEC)) && (!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
-
-#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}
- */
-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;
-}
-
-#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
- */
-uint32_t gen_rand32(void) {
-    uint32_t r;
-
-    assert(initialized);
-    if (idx >= N32) {
-	gen_rand_all();
-	idx = 0;
-    }
-    r = psfmt32[idx++];
-    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,
- * unless an initialization is again executed. 
- * @return 64-bit pseudorandom number
- */
-uint64_t gen_rand64(void) {
-#if defined(BIG_ENDIAN64) && !defined(ONLY64)
-    uint32_t r1, r2;
-#else
-    uint64_t r;
-#endif
-
-    assert(initialized);
-    assert(idx % 2 == 0);
-
-    if (idx >= N32) {
-	gen_rand_all();
-	idx = 0;
-    }
-#if defined(BIG_ENDIAN64) && !defined(ONLY64)
-    r1 = psfmt32[idx];
-    r2 = psfmt32[idx + 1];
-    idx += 2;
-    return ((uint64_t)r2 << 32) | r1;
-#else
-    r = psfmt64[idx / 2];
-    idx += 2;
-    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.
- */
-void fill_array32(uint32_t *array, int size) {
-    assert(initialized);
-    assert(idx == N32);
-    assert(size % 4 == 0);
-    assert(size >= N32);
-
-    gen_rand_array((w128_t *)array, size / 4);
-    idx = N32;
-}
-#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.
- */
-void fill_array64(uint64_t *array, int size) {
-    assert(initialized);
-    assert(idx == N32);
-    assert(size % 2 == 0);
-    assert(size >= N64);
-
-    gen_rand_array((w128_t *)array, size / 2);
-    idx = N32;
-
-#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.
- */
-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;
-}
diff --git a/dep/SFMT/SFMT.h b/dep/SFMT/SFMT.h
index 7c8b35e9e95..03a7e853316 100644
--- a/dep/SFMT/SFMT.h
+++ b/dep/SFMT/SFMT.h
@@ -1,157 +1,308 @@
-/** 
- * @file SFMT.h 
- *
- * @brief SIMD oriented Fast Mersenne Twister(SFMT) pseudorandom
- * number generator
- *
- * @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
- *
- * @note We assume that your system has inttypes.h.  If your system
- * doesn't have inttypes.h, you have to typedef uint32_t and uint64_t,
- * and you have to define PRIu64 and PRIx64 in this file as follows:
- * @verbatim
- typedef unsigned int uint32_t
- typedef unsigned long long uint64_t  
- #define PRIu64 "llu"
- #define PRIx64 "llx"
-@endverbatim
- * uint32_t must be exactly 32-bit unsigned integer type (no more, no
- * less), and uint64_t must be exactly 64-bit unsigned integer type.
- * PRIu64 and PRIx64 are used for printf function to print 64-bit
- * unsigned int and 64-bit unsigned int in hexadecimal format.
+/*
+ * Copyright notice
+ * ================
+ * GNU General Public License http://www.gnu.org/licenses/gpl.html
+ * This C++ implementation of SFMT contains parts of the original C code
+ * which was published under the following BSD license, which is therefore
+ * in effect in addition to the GNU General Public License.
+ * Copyright (c) 2006, 2007 by Mutsuo Saito, Makoto Matsumoto and Hiroshima University.
+ * Copyright (c) 2008 by Agner Fog.
+ * Copyright (c) 2010 Trinity Core
+ * 
+ *  BSD License:
+ *  Redistribution and use in source and binary forms, with or without 
+ * modification, are permitted provided that the following conditions are met:
+ *     > Redistributions of source code must retain the above copyright notice, 
+ *       this list of conditions and the following disclaimer.
+ *     > Redistributions in binary form must reproduce the above copyright notice, 
+ *       this list of conditions and the following disclaimer in the documentation
+ *       and/or other materials provided with the distribution.
+ *     > Neither the name of the Hiroshima University nor the names of its 
+ *       contributors may be used to endorse or promote products derived from 
+ *       this software without specific prior written permission.
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
 #ifndef SFMT_H
 #define SFMT_H
 
-#include <stdio.h>
-
-#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)
-  #include <inttypes.h>
-#elif defined(_MSC_VER) || defined(__BORLANDC__)
-  typedef unsigned int uint32_t;
-  typedef unsigned __int64 uint64_t;
-  #define inline __inline
-#else
-  #include <inttypes.h>
-  #if defined(__GNUC__)
-    #define inline __inline__
-  #endif
-#endif
+#include <emmintrin.h>                 // Define SSE2 intrinsics
+#include "randomc.h"                   // Define integer types etc
+#include <time.h>
 
-#ifndef PRIu64
-  #if defined(_MSC_VER) || defined(__BORLANDC__)
-    #define PRIu64 "I64u"
-    #define PRIx64 "I64x"
-  #else
-    #define PRIu64 "llu"
-    #define PRIx64 "llx"
-  #endif
-#endif
+// Choose one of the possible Mersenne exponents.
+// Higher values give longer cycle length and use more memory:
+//#define MEXP   607
+//#define MEXP  1279
+//#define MEXP  2281
+//#define MEXP  4253
+  #define MEXP 11213
+//#define MEXP 19937
+//#define MEXP 44497
 
-#if defined(__GNUC__)
-#define ALWAYSINLINE __attribute__((always_inline))
-#else
-#define ALWAYSINLINE
-#endif
+// Define constants for the selected Mersenne exponent:
+#if MEXP == 44497
+#define SFMT_N    348                  // Size of state vector
+#define SFMT_M    330                  // Position of intermediate feedback
+#define SFMT_SL1    5                  // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	  3                  // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1    9                  // Right shift of W[M], 32-bit words
+#define SFMT_SR2	  3                  // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	  0xeffffffb,0xdfbebfff,0xbfbf7bef,0x9ffd7bff // AND mask
+#define SFMT_PARITY 1,0,0xa3ac4000,0xecc1327a   // Period certification vector
+
+#elif MEXP == 19937
+#define SFMT_N    156                  // Size of state vector
+#define SFMT_M    122                  // Position of intermediate feedback
+#define SFMT_SL1   18                  // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	  1                  // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1   11                  // Right shift of W[M], 32-bit words
+#define SFMT_SR2	  1                  // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	  0xdfffffef,0xddfecb7f,0xbffaffff,0xbffffff6 // AND mask
+#define SFMT_PARITY 1,0,0,0x13c9e684   // Period certification vector
+
+#elif MEXP == 11213
+#define SFMT_N    88                   // Size of state vector
+#define SFMT_M    68                   // Position of intermediate feedback
+#define SFMT_SL1	14                   // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	 3                   // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1	 7                   // Right shift of W[M], 32-bit words
+#define SFMT_SR2	 3                   // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	 0xeffff7fb,0xffffffef,0xdfdfbfff,0x7fffdbfd // AND mask
+#define SFMT_PARITY 1,0,0xe8148000,0xd0c7afa3 // Period certification vector
+
+#elif MEXP == 4253
+#define SFMT_N    34                   // Size of state vector
+#define SFMT_M    17                   // Position of intermediate feedback
+#define SFMT_SL1	20                   // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	 1                   // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1	 7                   // Right shift of W[M], 32-bit words
+#define SFMT_SR2	 1                   // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	 0x9f7bffff, 0x9fffff5f, 0x3efffffb, 0xfffff7bb // AND mask
+#define SFMT_PARITY 0xa8000001, 0xaf5390a3, 0xb740b3f8, 0x6c11486d // Period certification vector
+
+#elif MEXP == 2281
+#define SFMT_N    18                   // Size of state vector
+#define SFMT_M    12                   // Position of intermediate feedback
+#define SFMT_SL1	19                   // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	 1                   // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1	 5                   // Right shift of W[M], 32-bit words
+#define SFMT_SR2	 1                   // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	 0xbff7ffbf, 0xfdfffffe, 0xf7ffef7f, 0xf2f7cbbf // AND mask
+#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x41dfa600  // Period certification vector
+
+#elif MEXP == 1279
+#define SFMT_N    10                   // Size of state vector
+#define SFMT_M     7                   // Position of intermediate feedback
+#define SFMT_SL1	14                   // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	 3                   // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1	 5                   // Right shift of W[M], 32-bit words
+#define SFMT_SR2	 1                   // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	  0xf7fefffd, 0x7fefcfff, 0xaff3ef3f, 0xb5ffff7f  // AND mask
+#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x20000000  // Period certification vector
 
-#if defined(_MSC_VER)
-  #if _MSC_VER >= 1200
-    #define PRE_ALWAYS __forceinline
-  #else
-    #define PRE_ALWAYS inline
-  #endif
-#else
-  #define PRE_ALWAYS inline
+#elif MEXP == 607
+#define SFMT_N     5                   // Size of state vector
+#define SFMT_M     2                   // Position of intermediate feedback
+#define SFMT_SL1	15                   // Left shift of W[N-1], 32-bit words
+#define SFMT_SL2	 3                   // Left shift of W[0], *8, 128-bit words
+#define SFMT_SR1	13                   // Right shift of W[M], 32-bit words
+#define SFMT_SR2	 3                   // Right shift of W[N-2], *8, 128-bit words
+#define SFMT_MASK	  0xfdff37ff, 0xef7f3f7d, 0xff777b7d, 0x7ff7fb2f  // AND mask
+#define SFMT_PARITY 0x00000001, 0x00000000, 0x00000000, 0x5986f054  // Period certification vector
 #endif
 
-uint32_t gen_rand32(void);
-uint64_t gen_rand64(void);
-void fill_array32(uint32_t *array, int size);
-void fill_array64(uint64_t *array, int size);
-void init_gen_rand(uint32_t seed);
-void init_by_array(uint32_t *init_key, int key_length);
-const char *get_idstring(void);
-int get_min_array_size32(void);
-int get_min_array_size64(void);
-
-/* These real versions are due to Isaku Wada */
-/** generates a random number on [0,1]-real-interval */
-inline static double to_real1(uint32_t v)
-{
-    return v * (1.0/4294967295.0); 
-    /* divided by 2^32-1 */ 
+// Functions used by SFMTRand::RandomInitByArray
+static uint32_t func1(uint32_t x) {
+    return (x ^ (x >> 27)) * 1664525U;
 }
 
-/** generates a random number on [0,1]-real-interval */
-inline static double genrand_real1(void)
-{
-    return to_real1(gen_rand32());
+static uint32_t func2(uint32_t x) {
+    return (x ^ (x >> 27)) * 1566083941U;
 }
 
-/** generates a random number on [0,1)-real-interval */
-inline static double to_real2(uint32_t v)
-{
-    return v * (1.0/4294967296.0); 
-    /* divided by 2^32 */
+// Subfunction for the sfmt algorithm
+static inline __m128i sfmt_recursion(__m128i const &a, __m128i const &b, 
+__m128i const &c, __m128i const &d, __m128i const &mask) {
+    __m128i a1, b1, c1, d1, z1, z2;
+    b1 = _mm_srli_epi32(b, SFMT_SR1);
+    a1 = _mm_slli_si128(a, SFMT_SL2);
+    c1 = _mm_srli_si128(c, SFMT_SR2);
+    d1 = _mm_slli_epi32(d, SFMT_SL1);
+    b1 = _mm_and_si128(b1, mask);
+    z1 = _mm_xor_si128(a, a1);
+    z2 = _mm_xor_si128(b1, d1);
+    z1 = _mm_xor_si128(z1, c1);
+    z2 = _mm_xor_si128(z1, z2);
+    return z2;
 }
 
-/** generates a random number on [0,1)-real-interval */
-inline static double genrand_real2(void)
-{
-    return to_real2(gen_rand32());
-}
+// Class for SFMT generator
+class SFMTRand {                              // Encapsulate random number generator
+public:
+    SFMTRand() { LastInterval = 0; RandomInit((int)(time(0))); } 
 
-/** generates a random number on (0,1)-real-interval */
-inline static double to_real3(uint32_t v)
-{
-    return (((double)v) + 0.5)*(1.0/4294967296.0); 
-    /* divided by 2^32 */
-}
+    void RandomInit(int seed)                     // Re-seed
+    {
+        // Re-seed
+        uint32_t i;                         // Loop counter
+        uint32_t y = seed;                  // Temporary
+        uint32_t statesize = SFMT_N*4;      // Size of state vector
 
-/** generates a random number on (0,1)-real-interval */
-inline static double genrand_real3(void)
-{
-    return to_real3(gen_rand32());
-}
-/** These real versions are due to Isaku Wada */
+        // Fill state vector with random numbers from seed
+        ((uint32_t*)state)[0] = y;
+        const uint32_t factor = 1812433253U;// Multiplication factor
 
-/** generates a random number on [0,1) with 53-bit resolution*/
-inline static double to_res53(uint64_t v) 
-{ 
-    return v * (1.0/18446744073709551616.0L);
-}
+        for (i = 1; i < statesize; i++) {
+            y = factor * (y ^ (y >> 30)) + i;
+            ((uint32_t*)state)[i] = y;
+        }
 
-/** generates a random number on [0,1) with 53-bit resolution from two
- * 32 bit integers */
-inline static double to_res53_mix(uint32_t x, uint32_t y) 
-{ 
-    return to_res53(x | ((uint64_t)y << 32));
-}
+        // Further initialization and period certification
+        Init2();
+    }
 
-/** generates a random number on [0,1) with 53-bit resolution
- */
-inline static double genrand_res53(void) 
-{ 
-    return to_res53(gen_rand64());
-} 
+    int32_t IRandom(int32_t min, int32_t max)     // Output random integer
+    {
+        // Output random integer in the interval min <= x <= max
+        // Slightly inaccurate if (max-min+1) is not a power of 2
+        if (max <= min) {
+            if (max == min) return min; else return 0x80000000;
+        }
+        // Assume 64 bit integers supported. Use multiply and shift method
+        uint32_t interval;                  // Length of interval
+        uint64_t longran;                   // Random bits * interval
+        uint32_t iran;                      // Longran / 2^32
 
-/** generates a random number on [0,1) with 53-bit resolution
-    using 32bit integer.
- */
-inline static double genrand_res53_mix(void) 
-{ 
-    uint32_t x, y;
-
-    x = gen_rand32();
-    y = gen_rand32();
-    return to_res53_mix(x, y);
-} 
-#endif
+        interval = (uint32_t)(max - min + 1);
+        longran  = (uint64_t)BRandom() * interval;
+        iran = (uint32_t)(longran >> 32);
+        // Convert back to signed and return result
+        return (int32_t)iran + min;
+    }
+
+    uint32_t URandom(uint32_t min, uint32_t max)
+    {
+        // Output random integer in the interval min <= x <= max
+        // Slightly inaccurate if (max-min+1) is not a power of 2
+        if (max <= min) {
+            if (max == min) return min; else return 0;
+        }
+        // Assume 64 bit integers supported. Use multiply and shift method
+        uint32_t interval;                  // Length of interval
+        uint64_t longran;                   // Random bits * interval
+        uint32_t iran;                      // Longran / 2^32
+
+        interval = (uint32_t)(max - min + 1);
+        longran  = (uint64_t)BRandom() * interval;
+        iran = (uint32_t)(longran >> 32);
+        // Convert back to signed and return result
+        return iran + min;
+    }
+
+    double Random()                               // Output random floating point number
+    {
+        // Output random floating point number
+        if (ix >= SFMT_N*4-1) {
+            // Make sure we have at least two 32-bit numbers
+            Generate();
+        }
+        uint64_t r = *(uint64_t*)((uint32_t*)state+ix);
+        ix += 2;
+        // 52 bits resolution for compatibility with assembly version:
+        return (int64_t)(r >> 12) * (1./(67108864.0*67108864.0));
+    }
+
+    uint32_t BRandom()                            // Output random bits
+    {
+        // Output 32 random bits
+        uint32_t y;
+
+        if (ix >= SFMT_N*4) {
+            Generate();
+        }
+        y = ((uint32_t*)state)[ix++];
+        return y;
+    }
+private:
+    void Init2()                                   // Various initializations and period certification
+    {
+        // Various initializations and period certification
+        uint32_t i, j, temp;
+    
+        // Initialize mask
+        static const uint32_t maskinit[4] = {SFMT_MASK};
+        mask = _mm_loadu_si128((__m128i*)maskinit);
+
+        // Period certification
+        // Define period certification vector
+        static const uint32_t parityvec[4] = {SFMT_PARITY};
+
+        // Check if parityvec & state[0] has odd parity
+        temp = 0;
+        for (i = 0; i < 4; i++)
+            temp ^= parityvec[i] & ((uint32_t*)state)[i];
+
+        for (i = 16; i > 0; i >>= 1) temp ^= temp >> i;
+        if (!(temp & 1)) {
+            // parity is even. Certification failed
+            // Find a nonzero bit in period certification vector
+            for (i = 0; i < 4; i++) {
+                if (parityvec[i]) {
+                    for (j = 1; j; j <<= 1) {
+                        if (parityvec[i] & j) {
+                            // Flip the corresponding bit in state[0] to change parity
+                            ((uint32_t*)state)[i] ^= j;
+                            // Done. Exit i and j loops
+                            i = 5;  break;
+                        }
+                    }
+                }
+            }
+        }
+
+        // Generate first random numbers and set ix = 0
+        Generate();
+    }
+
+    void Generate()                                // Fill state array with new random numbers
+    {
+        // Fill state array with new random numbers
+        int i;
+        __m128i r, r1, r2;
+
+        r1 = state[SFMT_N - 2];
+        r2 = state[SFMT_N - 1];
+        for (i = 0; i < SFMT_N - SFMT_M; i++) {
+            r = sfmt_recursion(state[i], state[i + SFMT_M], r1, r2, mask);
+            state[i] = r;
+            r1 = r2;
+            r2 = r;
+        }
+        for (; i < SFMT_N; i++) {
+            r = sfmt_recursion(state[i], state[i + SFMT_M - SFMT_N], r1, r2, mask);
+            state[i] = r;
+            r1 = r2;
+            r2 = r;
+        }
+        ix = 0;
+    }
+
+    uint32_t ix;                                  // Index into state array
+    uint32_t LastInterval;                        // Last interval length for IRandom
+    uint32_t RLimit;                              // Rejection limit used by IRandom
+    __m128i  mask;                                // AND mask
+    __m128i  state[SFMT_N];                       // State vector for SFMT generator
+};
+
+#endif // SFMT_H
diff --git a/dep/SFMT/randomc.h b/dep/SFMT/randomc.h
new file mode 100644
index 00000000000..5370a701c0e
--- /dev/null
+++ b/dep/SFMT/randomc.h
@@ -0,0 +1,65 @@
+/*
+ * Copyright notice
+ * ================
+ * GNU General Public License http://www.gnu.org/licenses/gpl.html
+ * This C++ implementation of SFMT contains parts of the original C code
+ * which was published under the following BSD license, which is therefore
+ * in effect in addition to the GNU General Public License.
+ * Copyright (c) 2006, 2007 by Mutsuo Saito, Makoto Matsumoto and Hiroshima University.
+ * Copyright (c) 2008 by Agner Fog.
+ * Copyright (c) 2010 Trinity Core
+ * 
+ *  BSD License:
+ *  Redistribution and use in source and binary forms, with or without 
+ * modification, are permitted provided that the following conditions are met:
+ *     > Redistributions of source code must retain the above copyright notice, 
+ *       this list of conditions and the following disclaimer.
+ *     > Redistributions in binary form must reproduce the above copyright notice, 
+ *       this list of conditions and the following disclaimer in the documentation
+ *       and/or other materials provided with the distribution.
+ *     > Neither the name of the Hiroshima University nor the names of its 
+ *       contributors may be used to endorse or promote products derived from 
+ *       this software without specific prior written permission.
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef RANDOMC_H
+#define RANDOMC_H
+
+// Define integer types with known size: int32_t, uint32_t, int64_t, uint64_t.
+// If this doesn't work then insert compiler-specific definitions here:
+#if defined(__GNUC__)
+  // Compilers supporting C99 or C++0x have inttypes.h defining these integer types
+  #include <inttypes.h>
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#elif defined(_WIN16) || defined(__MSDOS__) || defined(_MSDOS) 
+   // 16 bit systems use long int for 32 bit integer
+  typedef   signed long int int32_t;
+  typedef unsigned long int uint32_t;
+#elif defined(_MSC_VER)
+  // Microsoft have their own definition
+  typedef   signed __int32  int32_t;
+  typedef unsigned __int32 uint32_t;
+  typedef   signed __int64  int64_t;
+  typedef unsigned __int64 uint64_t;
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#else
+  // This works with most compilers
+  typedef signed int          int32_t;
+  typedef unsigned int       uint32_t;
+  typedef long long           int64_t;
+  typedef unsigned long long uint64_t;
+  #define INT64_SUPPORTED // Remove this if the compiler doesn't support 64-bit integers
+#endif
+
+#endif // RANDOMC_H