1998-2007/2

• If you want a twice to four times faster variant of MT, there is one designed for SIMD CPUs. It is faster than the original MT, even in non-SIMD CPUs, and has better assurance of randomenss. Downloadable from "SIMD-oriented Fast Mersenne Twister (SFMT)".
• If you want the newest version of original authors, go "Mersenne Twister with improved initialization".
• Old standard versions (do not use).
  • The first old standard version of mt19937.c (1998/4/9)
  • Next old standard version of mt19937.c (1999/10/29)
• TT800 : A small senior cousin of Mersenne Twister. It is also in pLab : Salzburg University Random Number Research Team ( TT800 )
• A C++ version of MT is available from Homepage of Martin Hinsch.
• Rene Brun at CERN kindly implemented MT in their C++ framework Root. Here is the code.
• A new C++ implementation with various functions for easy use by Richard Wagner: "I have written a new C++ implementation of your Mersenne Twister random number generator. It is based mainly on Shawn Cokus' optimized C version with several improvements for easier use. For example, seeding can be performed automatically from the system time or, on a Linux system, /dev/urandom. Random numbers are easily obtained as real numbers or integers in default or user-defined ranges, and the state of the generator can be saved and restored. It also includes changes you made after Cokus' version." The code is faster than Cokus's version in some platform.
• Paul Gresham kindly sent us another C++ class called RandMT.cc .
• Takel implemented a C#-version of MT19937 under the Artistic License. Japanese explanation is here. In the page, there are a direct translation to C# MT19937.cs and a class which inherits System.Random in .Net library MersenneTwister.cs .
• Geoff Kuenning wrote a very fast C/C++-version of MT19937, faster than Cokus' version (depending on the architecture). Moreover, it has many extended features, like to produce multiple independent streams of prns, etc. See his homepage .
• Eric Maryniak informed us that they use MT in their open-sourced neural network simulator Walnut/Nutshell . They noticed several compilation problems, however, for Richard Wagner's C++ when compiling the code with certain versions of STLPort and while experimenting with static linkage. A Nutshell programmer Robert Berg fixed this. Here is the code MersenneTwister.h .
• Austen McDonald ported MT19937 to store the globals from mt19937.c in structs for parallel run. See the homepage. Also at this location is a parallel version of a Sobol' sequence generator.
  (updated 2007/6/1)
• Matthew Bellew wrote a C++ code using Pentium MMX instruction. This is almost twice as fast as the original Wagner's C++ code. See his page .
• Jeff Deifik wrote a C code which permits several paralell streams of MT. Here is the C code .
• Einguste wrote a thread-safe DLL version of MT. His home page includes a DLL version as well as a thread-safe version, and a Makefile for Borland C++ Compiler 5.5. This page also includes MT19937ar.c ( improved initialization version, 2002/1/26) and its Cokus-type speed-up version MT19937ar-cok.c, and show a comparison of speed. Japanese home page is also available.
• An implimentation of MT19937ar (new initialization, 2002/1/26) in C and C++ with various functions added by T. Ando . A .lzh file for the both C and C++ is available as mt_20020203a.lzh . Explanations (only in Japanese, sorry) and plain files can be found at here and there .
• Agner Fog distributes a bunch of modern generators including mt19937ar.c (improved initialization version, 2002/1/26)in C++ and Assembler, for uniform and non-uniform random number generation. The code is very well optimized. See http://www.agner.org/random/ (2004/4/14)
• Eric Landry informed me that a straight-forward implementation is proved to be faster than the standard code, in some platform. (2004/3/15)
• MT for 64-bit machines (2005/2/23)
  Here is a C code of Mersenne Twister for 64 bit machines, with period 2^19937-1.
  The recursion is similar but different, so the output is totally different from the 32-bit versions.
• Kouhei Taketa kindly wrote an another C++ version of MT. His code is a header library, so one can use the functions by including the code as a header file using #include. In addition, the code contains a 64-bit version of MT. Here is his code mersennetwister.zip. Japanese explanation and same code is in his web.(2005/10/6)
• Diego Park kindly wrote an Visual C++ version of MT. He wrote a single class (through template) to provide 32 and 64-bit random integers alike. This implementation has a speed-up of about 38% against the original implementation on 64-bit version, about 70% on 32-bit and about 500% against rand(), generated by Microsoft Visual Studio 2005 Beta 2 targeted to x64 architecture, tested on a AMD Athlon 64 3000+ with Microsoft Windows XP Professional x64 Edition. Here is his code MersenneTwister-PK.zip. (2005/12/16)
• Mitil Ooyama wrote a C# version of MT. Here is his code mt19937ar.cs. It is in his web page (written in Japanese), too. (2006/1/23)
• Zubin Dittia wrote a C++ version of MT. Here is the code MersenneTwister.h and MersenneTwister.cpp. This code contains the genrand_buf function, which fills a buffer with random data. (2006/2/7)
• Loren Carpenter at Pixar Animation Studios kindly wrote a C++ version of Mersenne Twister using intel SSE2 instructions. Here is his code cpp_sse2.zip. It's about 3.2 times as fast as the serial version on 3.4 GHz Xeon.
• Charles Karney kindly wrote a C++ version of Mersenne Twister:
  RandomLib is a C++ interface to the Mersenne Twister random number generator, MT19937. It provides convenient access to random integers and reals at a variety of precisions. The emphasis in this implementation is on providing a reliable source of random numbers for scientific applications where there's a premium on accuracy, repeatability, portability, and ease of use. This library provides an improved method for seeding the generator with arrays and it allows access to both the 32- and 64-bit versions of MT19937. For more information, see

  http://randomlib.sourceforge.net/

  The code can be downloaded from

  http://sourceforge.net/projects/randomlib/files/
  (2006/11/5, updated 2010/1/13)
• Takayuki HARUKI kindly wrote a C++ version of Mersenne Twister, which is implemented as a Singleton design pattern. Singleton.zip. (2007/10/3)
• Wouter Kager gave us a mail about Possible Speed-up in real number generation. (2008/01/22)
• stlalv kindly wrote a C# version of MT. Here is his code stlalv-MersenneTwister.zip. It is in his web page (written in Japanese), too. (2010/10/15)