Re: [boost] [random] new threefry random engine

Subject: Re: [boost] [random] new threefry random engine
From: John Salmon (john_at_[hidden])
Date: 2014-05-27 14:48:13

On Mon, May 5, 2014 at 9:10 AM, John Salmon <john_at_[hidden]> wrote:

> On Fri, May 2, 2014 at 11:43 AM, Steven Watanabe <watanabesj_at_[hidden]>wrote:
>
>> AMDG
>>
>> On 04/29/2014 10:00 AM, John Salmon wrote:
>> >
>> > I've dusted off the code I wrote a couple of years ago and republished
>> it,
>> > now on github:
>> >
>> > https://github.com/DEShawResearch/Random123-Boost
>> >
>>
>> For this to be included in Boost.Random, it must be
>> released under the Boost Software License.
>>
>
>
Done!  The version at the above URL is now under the Boost Software License.

While waiting for permission to change the license, I had some useful
conversations with Thijs off-line, resulting in some changes to the
implementation.  I won't quote the whole README again, but it now starts
this way, which I hope is much more to Steven's liking:

<snip README.md>
CBENG's are fully generic "Random Number Engines", satisfying all the
requirements of C++11 and Boost Engines. They can be used anywhere that
other engines can be used. When used with the provided "pseudo random
functions", threefry and philox, CBENGs are "crush resistant", i.e., they
pass the very demanding suite of 'SmallCrush' 'Crush' and 'BigCrush' tests
in the TestU01 package. CBENGs based on threefry and philox are fast
(comparable to the mersenne twister) and small (requiring only a few words
of state).

CBENGs are excellent random number engines in any context, but two features
make them especially well suited to parallel computations:

   - a fast (constant time) restart() member function that allows the
   caller to manage a huge number of distinct streams.
   - a fast (constant time) discard() member function.

The restart() member is unique to CBENGs, while discard() follows standard
semantics (but very few engines permit constant-time implementations).

</snip>

There is only one counter_based_engine, as Steven requested elsewhere in
the thread.

Can we agree that counter_based_engine, implemented roughly along these
lines is a worthwhile addition to Boost.Random?  If so, what's the way
forward?

Should I start a 'feature branch', i.e.,
https://svn.boost.org/trac/boost/wiki/StartModMaint#Startworkonanewfeature
to make it easier to ultimately merge into boost?

Cheers,

John Salmon