Commit 0aad1354 authored by Peter Eastman's avatar Peter Eastman
Browse files

Added SFMT as a random number generator for the reference platform.

parent 85da5e0f
...@@ -28,6 +28,7 @@ ...@@ -28,6 +28,7 @@
#include "../SimTKUtilities/SimTKOpenMMCommon.h" #include "../SimTKUtilities/SimTKOpenMMCommon.h"
#include "../SimTKUtilities/SimTKOpenMMLog.h" #include "../SimTKUtilities/SimTKOpenMMLog.h"
#include "../SimTKUtilities/SimTKOpenMMUtilities.h" #include "../SimTKUtilities/SimTKOpenMMUtilities.h"
#include "../sfmt/SFMT.h"
#include "ReferenceDynamics.h" #include "ReferenceDynamics.h"
const int ReferenceDynamics::DefaultReturn = 0; const int ReferenceDynamics::DefaultReturn = 0;
...@@ -56,7 +57,8 @@ ReferenceDynamics::ReferenceDynamics( int numberOfAtoms, RealOpenMM deltaT, Rea ...@@ -56,7 +57,8 @@ ReferenceDynamics::ReferenceDynamics( int numberOfAtoms, RealOpenMM deltaT, Rea
// --------------------------------------------------------------------------------------- // ---------------------------------------------------------------------------------------
_timeStep = 0; _timeStep = 0;
_randomNumberSeed = one; _randomNumberSeed = 0;
init_gen_rand(_randomNumberSeed);
_twoDTempArrays = 0; _twoDTempArrays = 0;
_twoDTempArrays = NULL; _twoDTempArrays = NULL;
...@@ -391,27 +393,25 @@ RealOpenMM ReferenceDynamics::getTemperature( void ) const { ...@@ -391,27 +393,25 @@ RealOpenMM ReferenceDynamics::getTemperature( void ) const {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
RealOpenMM ReferenceDynamics::getNormallyDistributedRandomNumber( void ){ RealOpenMM ReferenceDynamics::getNormallyDistributedRandomNumber( void ){
static bool nextValueIsValid = false;
// --------------------------------------------------------------------------------------- static RealOpenMM nextValue = 0;
if (nextValueIsValid) {
static const char* methodName = "\nReferenceDynamics::getNormallyDistributedRandomNumber"; nextValueIsValid = false;
return nextValue;
static const RealOpenMM randomValue = (RealOpenMM) 0.1; }
static int printOnce = 0;
// Use the polar form of the Box-Muller transformation to generate two Gaussian random numbers.
// ---------------------------------------------------------------------------------------
RealOpenMM x, y, r2;
// for now just using fixed random value to allow for testing do {
x = 2.0*genrand_real2()-1.0;
if( !printOnce ){ y = 2.0*genrand_real2()-1.0;
std::stringstream message; r2 = x*x + y*y;
message << methodName; } while (r2 >= 1.0 || r2 == 0.0);
message << " random value fixed at value=" << randomValue; RealOpenMM multiplier = sqrt((-2.0*log(r2))/r2);
SimTKOpenMMLog::printMessage( message ); nextValue = y*multiplier;
printOnce++; nextValueIsValid = true;
} return x*multiplier;
return randomValue;
} }
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
...@@ -422,7 +422,7 @@ RealOpenMM ReferenceDynamics::getNormallyDistributedRandomNumber( void ){ ...@@ -422,7 +422,7 @@ RealOpenMM ReferenceDynamics::getNormallyDistributedRandomNumber( void ){
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
RealOpenMM ReferenceDynamics::getRandomNumberSeed( void ) const { uint32_t ReferenceDynamics::getRandomNumberSeed( void ) const {
// --------------------------------------------------------------------------------------- // ---------------------------------------------------------------------------------------
...@@ -443,7 +443,7 @@ RealOpenMM ReferenceDynamics::getRandomNumberSeed( void ) const { ...@@ -443,7 +443,7 @@ RealOpenMM ReferenceDynamics::getRandomNumberSeed( void ) const {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
int ReferenceDynamics::setRandomNumberSeed( RealOpenMM seed ){ int ReferenceDynamics::setRandomNumberSeed( uint32_t seed ){
// --------------------------------------------------------------------------------------- // ---------------------------------------------------------------------------------------
......
...@@ -27,6 +27,7 @@ ...@@ -27,6 +27,7 @@
#include "ReferenceShakeAlgorithm.h" #include "ReferenceShakeAlgorithm.h"
#include "../SimTKUtilities/SimTKOpenMMCommon.h" #include "../SimTKUtilities/SimTKOpenMMCommon.h"
#include <cstddef>
// --------------------------------------------------------------------------------------- // ---------------------------------------------------------------------------------------
...@@ -60,7 +61,7 @@ class ReferenceDynamics { ...@@ -60,7 +61,7 @@ class ReferenceDynamics {
RealOpenMM _deltaT; RealOpenMM _deltaT;
RealOpenMM _temperature; RealOpenMM _temperature;
RealOpenMM _randomNumberSeed; uint32_t _randomNumberSeed;
int _numberOf2DTempArrays; int _numberOf2DTempArrays;
RealOpenMM*** _twoDTempArrays; RealOpenMM*** _twoDTempArrays;
...@@ -277,7 +278,7 @@ class ReferenceDynamics { ...@@ -277,7 +278,7 @@ class ReferenceDynamics {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
RealOpenMM getRandomNumberSeed( void ) const; uint32_t getRandomNumberSeed( void ) const;
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
...@@ -289,7 +290,7 @@ class ReferenceDynamics { ...@@ -289,7 +290,7 @@ class ReferenceDynamics {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
int setRandomNumberSeed( RealOpenMM seed ); int setRandomNumberSeed( uint32_t seed );
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
......
ver 1.3.3
-------
change condition compile of do_recursion in SFMT.c
ver 1.3.2
-------
bug fix to_res53_mix and genrand_res53_mix.
ver 1.3.1
-------
gcc compile option changed form -O9 to -O3.
add functions genrand_res53_mix and to_res53_mix.
bug fix about definition of ALWAYS_INLINE.
add new definition PRE_ALWAYS for MSC.
ver 1.3
-------
bug fixed: -DONLY64 without -DBIG_ENIAN64 had been generating
wrong sequence.
bug fixed: There is no documentation about BIG_ENDIAN64.
add automatic endian check by __BIG_ENDIAN__ predefined macro.
bug fixed: change == in check.sh to =
add SFMT-params216091.h
add AltiVec parameter format for systems which are not osx.
change Makefile for systems which are not osx and support AltiVec.
change sample2 of howto-compile for Free BSD.
change source files for BORLANDC and Visual Studio.
change period certification code more smart.
add params directory.
ver 1.2.1
-------
Fix typo in SFMT-alti.c SFMT-sse2.c
marge SFMT-alti.c and SFMT-alti.h into SFMT-alti.h
marge SFMT-sse2.c and SFMT-sse2.h into SFMT-sse2.h
This version is not released.
ver 1.2
-------
Support many periods: 2^{607}, 2^{1279}, 2^{2281}, 2^{4253}, 2^{11213},
2^{19937}, 2^{44497}, 2^{86243}, 2^{132049}
Fix typo in LICENSE.txt.
Add cast to vec_perm for SFMT-alti.c, SFMT-alti64.c.
combine source codes.
ver 1.1
-------
The period certification method is changed from constant to function.
The convert functions from 32-bit and 64-bit integer to double are added.
The documentation is changed.
Sample programs are added.
ver 1.0
-------
The first version.
Copyright (c) 2006,2007 Mutsuo Saito, Makoto Matsumoto and Hiroshima
University. All rights reserved.
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.
=================================================================
SFMT ver. 1.3.3
SIMD oriented Fast Mersenne Twister(SFMT)
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 LICENSE.txt
=================================================================
To see documents, see html/index.html.
To make test program, see html/howto-compile.html
If you want to redistribute and/or change source files, see LICENSE.txt.
When you change these files and redistribute them, PLEASE write your
e-mail address in redistribution and write to contact YOU first if
users of your changed source encounter troubles.
/**
* @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
#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 */
#ifndef SFMT_PARAMS11213_H
#define SFMT_PARAMS11213_H
#define POS1 68
#define SL1 14
#define SL2 3
#define SR1 7
#define SR2 3
#define MSK1 0xeffff7fbU
#define MSK2 0xffffffefU
#define MSK3 0xdfdfbfffU
#define MSK4 0x7fffdbfdU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0xe8148000U
#define PARITY4 0xd0c7afa3U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2)
#define ALTI_SR2_PERM \
(vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10}
#define ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2}
#define ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12}
#define ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12}
#endif /* For OSX */
#define IDSTR "SFMT-11213:68-14-3-7-3:effff7fb-ffffffef-dfdfbfff-7fffdbfd"
#endif /* SFMT_PARAMS11213_H */
#ifndef SFMT_PARAMS1279_H
#define SFMT_PARAMS1279_H
#define POS1 7
#define SL1 14
#define SL2 3
#define SR1 5
#define SR2 1
#define MSK1 0xf7fefffdU
#define MSK2 0x7fefcfffU
#define MSK3 0xaff3ef3fU
#define MSK4 0xb5ffff7fU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0x00000000U
#define PARITY4 0x20000000U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10}
#define ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-1279:7-14-3-5-1:f7fefffd-7fefcfff-aff3ef3f-b5ffff7f"
#endif /* SFMT_PARAMS1279_H */
#ifndef SFMT_PARAMS132049_H
#define SFMT_PARAMS132049_H
#define POS1 110
#define SL1 19
#define SL2 1
#define SR1 21
#define SR2 1
#define MSK1 0xffffbb5fU
#define MSK2 0xfb6ebf95U
#define MSK3 0xfffefffaU
#define MSK4 0xcff77fffU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0xcb520000U
#define PARITY4 0xc7e91c7dU
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8}
#define ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-132049:110-19-1-21-1:ffffbb5f-fb6ebf95-fffefffa-cff77fff"
#endif /* SFMT_PARAMS132049_H */
#ifndef SFMT_PARAMS19937_H
#define SFMT_PARAMS19937_H
#define POS1 122
#define SL1 18
#define SL2 1
#define SR1 11
#define SR2 1
#define MSK1 0xdfffffefU
#define MSK2 0xddfecb7fU
#define MSK3 0xbffaffffU
#define MSK4 0xbffffff6U
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0x00000000U
#define PARITY4 0x13c9e684U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8}
#define ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-19937:122-18-1-11-1:dfffffef-ddfecb7f-bffaffff-bffffff6"
#endif /* SFMT_PARAMS19937_H */
#ifndef SFMT_PARAMS216091_H
#define SFMT_PARAMS216091_H
#define POS1 627
#define SL1 11
#define SL2 3
#define SR1 10
#define SR2 1
#define MSK1 0xbff7bff7U
#define MSK2 0xbfffffffU
#define MSK3 0xbffffa7fU
#define MSK4 0xffddfbfbU
#define PARITY1 0xf8000001U
#define PARITY2 0x89e80709U
#define PARITY3 0x3bd2b64bU
#define PARITY4 0x0c64b1e4U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10}
#define ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-216091:627-11-3-10-1:bff7bff7-bfffffff-bffffa7f-ffddfbfb"
#endif /* SFMT_PARAMS216091_H */
#ifndef SFMT_PARAMS2281_H
#define SFMT_PARAMS2281_H
#define POS1 12
#define SL1 19
#define SL2 1
#define SR1 5
#define SR2 1
#define MSK1 0xbff7ffbfU
#define MSK2 0xfdfffffeU
#define MSK3 0xf7ffef7fU
#define MSK4 0xf2f7cbbfU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0x00000000U
#define PARITY4 0x41dfa600U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8}
#define ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-2281:12-19-1-5-1:bff7ffbf-fdfffffe-f7ffef7f-f2f7cbbf"
#endif /* SFMT_PARAMS2281_H */
#ifndef SFMT_PARAMS4253_H
#define SFMT_PARAMS4253_H
#define POS1 17
#define SL1 20
#define SL2 1
#define SR1 7
#define SR2 1
#define MSK1 0x9f7bffffU
#define MSK2 0x9fffff5fU
#define MSK3 0x3efffffbU
#define MSK4 0xfffff7bbU
#define PARITY1 0xa8000001U
#define PARITY2 0xaf5390a3U
#define PARITY3 0xb740b3f8U
#define PARITY4 0x6c11486dU
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {1,2,3,23,5,6,7,0,9,10,11,4,13,14,15,8}
#define ALTI_SL2_PERM64 {1,2,3,4,5,6,7,31,9,10,11,12,13,14,15,0}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-4253:17-20-1-7-1:9f7bffff-9fffff5f-3efffffb-fffff7bb"
#endif /* SFMT_PARAMS4253_H */
#ifndef SFMT_PARAMS44497_H
#define SFMT_PARAMS44497_H
#define POS1 330
#define SL1 5
#define SL2 3
#define SR1 9
#define SR2 3
#define MSK1 0xeffffffbU
#define MSK2 0xdfbebfffU
#define MSK3 0xbfbf7befU
#define MSK4 0x9ffd7bffU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0xa3ac4000U
#define PARITY4 0xecc1327aU
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2)
#define ALTI_SR2_PERM \
(vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10}
#define ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2}
#define ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12}
#define ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12}
#endif /* For OSX */
#define IDSTR "SFMT-44497:330-5-3-9-3:effffffb-dfbebfff-bfbf7bef-9ffd7bff"
#endif /* SFMT_PARAMS44497_H */
#ifndef SFMT_PARAMS607_H
#define SFMT_PARAMS607_H
#define POS1 2
#define SL1 15
#define SL2 3
#define SR1 13
#define SR2 3
#define MSK1 0xfdff37ffU
#define MSK2 0xef7f3f7dU
#define MSK3 0xff777b7dU
#define MSK4 0x7ff7fb2fU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0x00000000U
#define PARITY4 0x5986f054U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2)
#define ALTI_SR2_PERM \
(vector unsigned char)(5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {3,21,21,21,7,0,1,2,11,4,5,6,15,8,9,10}
#define ALTI_SL2_PERM64 {3,4,5,6,7,29,29,29,11,12,13,14,15,0,1,2}
#define ALTI_SR2_PERM {5,6,7,0,9,10,11,4,13,14,15,8,19,19,19,12}
#define ALTI_SR2_PERM64 {13,14,15,0,1,2,3,4,19,19,19,8,9,10,11,12}
#endif /* For OSX */
#define IDSTR "SFMT-607:2-15-3-13-3:fdff37ff-ef7f3f7d-ff777b7d-7ff7fb2f"
#endif /* SFMT_PARAMS607_H */
#ifndef SFMT_PARAMS86243_H
#define SFMT_PARAMS86243_H
#define POS1 366
#define SL1 6
#define SL2 7
#define SR1 19
#define SR2 1
#define MSK1 0xfdbffbffU
#define MSK2 0xbff7ff3fU
#define MSK3 0xfd77efffU
#define MSK4 0xbf9ff3ffU
#define PARITY1 0x00000001U
#define PARITY2 0x00000000U
#define PARITY3 0x00000000U
#define PARITY4 0xe9528d85U
/* PARAMETERS FOR ALTIVEC */
#if defined(__APPLE__) /* For OSX */
#define ALTI_SL1 (vector unsigned int)(SL1, SL1, SL1, SL1)
#define ALTI_SR1 (vector unsigned int)(SR1, SR1, SR1, SR1)
#define ALTI_MSK (vector unsigned int)(MSK1, MSK2, MSK3, MSK4)
#define ALTI_MSK64 \
(vector unsigned int)(MSK2, MSK1, MSK4, MSK3)
#define ALTI_SL2_PERM \
(vector unsigned char)(25,25,25,25,3,25,25,25,7,0,1,2,11,4,5,6)
#define ALTI_SL2_PERM64 \
(vector unsigned char)(7,25,25,25,25,25,25,25,15,0,1,2,3,4,5,6)
#define ALTI_SR2_PERM \
(vector unsigned char)(7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14)
#define ALTI_SR2_PERM64 \
(vector unsigned char)(15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14)
#else /* For OTHER OSs(Linux?) */
#define ALTI_SL1 {SL1, SL1, SL1, SL1}
#define ALTI_SR1 {SR1, SR1, SR1, SR1}
#define ALTI_MSK {MSK1, MSK2, MSK3, MSK4}
#define ALTI_MSK64 {MSK2, MSK1, MSK4, MSK3}
#define ALTI_SL2_PERM {25,25,25,25,3,25,25,25,7,0,1,2,11,4,5,6}
#define ALTI_SL2_PERM64 {7,25,25,25,25,25,25,25,15,0,1,2,3,4,5,6}
#define ALTI_SR2_PERM {7,0,1,2,11,4,5,6,15,8,9,10,17,12,13,14}
#define ALTI_SR2_PERM64 {15,0,1,2,3,4,5,6,17,8,9,10,11,12,13,14}
#endif /* For OSX */
#define IDSTR "SFMT-86243:366-6-7-19-1:fdbffbff-bff7ff3f-fd77efff-bf9ff3ff"
#endif /* SFMT_PARAMS86243_H */
/**
* @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
/**
* @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;
}
/**
* @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.
*/
#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
#ifndef PRIu64
#if defined(_MSC_VER) || defined(__BORLANDC__)
#define PRIu64 "I64u"
#define PRIx64 "I64x"
#else
#define PRIu64 "llu"
#define PRIx64 "llx"
#endif
#endif
#if defined(__GNUC__)
#define ALWAYSINLINE __attribute__((always_inline))
#else
#define ALWAYSINLINE
#endif
#if defined(_MSC_VER)
#if _MSC_VER >= 1200
#define PRE_ALWAYS __forceinline
#else
#define PRE_ALWAYS inline
#endif
#else
#define PRE_ALWAYS inline
#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 */
}
/** generates a random number on [0,1]-real-interval */
inline static double genrand_real1(void)
{
return to_real1(gen_rand32());
}
/** 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 */
}
/** generates a random number on [0,1)-real-interval */
inline static double genrand_real2(void)
{
return to_real2(gen_rand32());
}
/** 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 */
}
/** 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 */
/** 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);
}
/** 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));
}
/** generates a random number on [0,1) with 53-bit resolution
*/
inline static double genrand_res53(void)
{
return to_res53(gen_rand64());
}
/** 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
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment