Experiments in fast "random" number generation for Erlang; though they
have the same interface as rand:uniform/1, think of them more like
phash2/2 with a timestamp -- not really random, but fast
(hopefully).
Specifically, some of the PRNGs here use methods that are known to be biased. They are not suitably for any application which assumes uniformity. They are also not guaranteed to produce uncorrelated streams when used in parallel.
Use rebar3. To override the default
implementation choice, you can set the IMPLEMENTATION environment
variable:
$ IMPLEMENTATION=rdrand rebar3 compile
All functions take an integer between 1 and 232-1, and
return an integer between 1 and the supplied integer, like
rand:uniform/1.
32-bits of RDRAND.
Marsaglia's original xorshift (with no multiplies), keeping state in thread-local storage. Initialized on first call.
PCG, TLS.
Middle Square Weyl Sequence, keeping state in thread-local storage. Initialized on first call.
Fog, Agner. "Pseudo-Random Number Generators for Vector Processors and Multicore Processors." Journal of Modern Applied Statistical Methods 14.1 (2015): 308-334.
Marsaglia, George. "Xorshift RNGs." Journal of Statistical Software 8.14 (2003): 1-6.
O'Neill, M.E. "PCG: A Family of Simple Fast Space-Efficient Statistically Good Algorithms for Random Number Generation".
Widynsky, Bernard. "Middle Square Weyl Sequence RNG".
Because the pcg32 code is derived from Apache-licensed code (see c_src/pcg32.c), this package is also Apache licensed.