dxorshift: random number generators from the extended Xorshift family

[Joseph Rushton Wakeling via Digitalmars-d-announce]

http://code.dlang.org/packages/dxorshift
https://github.com/WebDrake/dxorshift

Following my earlier list posting 
<http://forum.dlang.org/post/kdobdorqztlsomweftmi@forum.dlang.org>, I'm pleased to announce an initial release of a dub package providing some of the RNGs from the extended family of xorshift-inspired generators.  These should complement the existing xorshift implementations already provided to Phobos' std.random by Masahiro Nakagawa: they are fast, high-quality generators representing some of the state of the art in pseudo-RNG design.

The generators implemented have been ported from public-domain 
reference implementations available here: 
http://xoroshiro.di.unimi.it/  Following the original authors' 
example, these D ports have also been dedicated to the public 
domain using the Creative Commons CC0 license.

At this stage, only a direct port has been provided, with no 
attempts at generics (e.g. on the unsigned integer type used, or 
the magic constants used in the RNG update methods).  The 
provided generators are:

   * SplitMix64: a fast generator with only 64 bits
     of state; this is probably inadequate for any
     serious statistical work, but is provided as a
     convenient means of generating seeds for other
     more heavy-duty algorithms

   * xoroshiro128+: a very fast and high quality
     generator with 128 bits of state; this ought
     to be a great generator for anyone not doing
     massively parallel simulations

   * xorshift1024*: a very fast and high quality
     generator with 1024 bits of state; this is
     slower than xoroshiro128+, but can be used
     with much larger-scale parallel simulations

The xoroshiro128+ and xorshift1024* generators come with `jump()` 
methods, the equivalent of 2 ^^ 64 and 2 ^^ 512 calls to 
`popFront()` respectively, which can be used to generate the 
starting points for (again respectively) 2 ^^ 64 or 2 ^^ 512 
independent sequences of variates.

The generators are all implemented as structs, but in order to 
prevent some known problems with unintended copy-by-value of 
RNGs, the postblit has been disabled.  For similar reasons, these 
generators are implemented as input ranges, not forward ranges, 
so that library functionality cannot copy generator state under 
the hood.

`dup` properties are however provided for all generators, to 
allow the programmer to deliberately copy RNG state.

Testing, feedback and general usage are all welcome.  I am 
planning on submitting these to Phobos (although sorting out the 
generic side of things might be a good idea first).