sfmt-dist

sfmt distribution

LICENSE

the GNU LGPL license is applied for KISS.hpp and Ziggurat.hpp in include/sfmt-dist directory.

MIT License for all other codes.

NOTE

Ziggurat algorithm returns single precision floating point numbers. KISS random number generator is presented only because that is used in original Ziggurat algorithm.

Ziggurat algorithm can combine with any 32-bit uniform random number generators.

NormalFromDouble uses Polar form of Box-Mullar transform. Currently, NormalFromDouble accecpts dSFMT generator only. Currently, UniformFromDouble accecpts dSFMT generator only.

STANDARD INTERFACE

std::uniform_int_distribution and std::normal_distribution adaptable SFMT19937

sample code

std::uniform_int_distribution<uint32_t> dist(0, 200);
MersenneTwiseter::SFMT19937 sfmt(1234);
cout << dec << dist(sfmt) << endl;

ORIGINAL INTERFACE FOR dSFMT

original distribution MersenneTwister::UniformIntFromDouble and MersenneTwister::NormalFromDouble for DSFMT19937

sample code

using MersenneTwister::NormalFromDouble;
using MersenneTwister::DSFMT19937;
NormalFromDouble<DSFMT19937> dist(0.0, 1.0, 1234);
cout.setf(std::ios::fixed);
for (uint64_t i = 0; i < 100; ++i) {
    cout << dec << setprecision(20) << dist() << endl;
}

using MersenneTwister::UniformIntFromDouble;
UniformIntFromDouble<uint32_t, DSFMT19937> dist(0, 200, 1234) ;
cout.setf(std::ios::fixed);
for (uint64_t i = 0; i < count; ++i) {
    cout << dec << dist() << endl;
}

Ziggurat Algorithm

Fast normal distribution generation algorithm.
Ziggurat.hpp only support single precision.

sample code

using Ziggurat::ZigNormal;
using MersenneTwister::SFMT19937;
ZigNormal<SFMT19937> dist(0.0, 1.0, 1234);
cout.setf(std::ios::fixed);
for (uint64_t i = 0; i < 100; ++i) {
    cout << dec << setprecision(20) << dist() << endl;
}

TEST PROGRAM

"make check" does internal check

test_mt19937, test_sfmt19937, test_dsfmt19937, test_mt19937_64 checks their outputs matches official test vectors.

output and speed

non installed programs "output" and "speed" are programs that users can check outputs and measure speed.

./output
./output [-n|-u] [-m|-M|-S|-d] [-s seed] [-c count] [-f start] [-l end]
-n --normal   normal distribution
-u --uniform  uniform distribution
-z --ziggurat normal distribution using ziggurat algorithm
-m --std-mt   std::mersennetwister19937
-M --mt-mt    MersenneTwister19937
-S --sfmt     SFMT19937
-d --dsfmt    dSFMT19937
-k --kiss     KISS
-s --seed     seed
-c --count    count
-f --start    start of uniform range
-l --end      end of uniform range

./output -u -S -f100 -l200 -c 10 -s 2
#uniform distribution[100,200] :MersenneTwister::SFMT19937
128
152
100
141
175
170
144
198
111
164

./speed
./speed [-n|-u] [-m|-M|-S|-d] [-s seed] [-c count] [-f start] [-l end]
-n --normal   normal distribution
-u --uniform  uniform distribution
-z --ziggurat normal distribution using ziggurat algorithm
-m --std-mt   std::mersennetwister19937
-M --mt-mt    MersenneTwister19937
-S --sfmt     SFMT19937
-d --dsfmt    dSFMT19937
-k --kiss     KISS
-s --seed     seed
-c --count    count
-f --start    start of uniform range
-l --end      end of uniform range

./speed -u -S
#uniform distribution[0,200] time for 100000000 generation
MersenneTwister::SFMT19937,868ms
./speed -u -d
#uniform distribution[0,200] time for 100000000 generation
MersenneTwister::dSFMT19937,168ms