jeffmm / rng_comparison

Comparing GSL and KISS RNGs

A comparison of random number generators

This repo produces two binaries that produce random numbers given an initial seed. The purpose of the repo is to compare the header-only KISS RNG (based on UCL Professor David Jones' JKISS implementation) and the GSL RNG library. Comparisons were done in python and bash comparing random number autocorrelations, fitting to expected distributions, and speed.

Long story short, both generators properly returned pseudo-random numbers that passed my statistical tests, but KISS was over 4 times faster than GSL, while having the advantage of being header-only.

Compiling

Requirements are cmake, a C++11 compiler, and the GSL library installed somewhere where cmake can find it (like /usr/local/lib, etc).

To compile using cmake, do the usual

mkdir build
cd build
cmake ..
make

Documentation

By default, documentation will be built into the build/documentation folder in both html and latex using Doxygen (if it is available).

Testing

To run unit tests, navigate to a fresh build folder and build the binaries in debug mode with cmake -DCMAKE_BUILD_TYPE=Debug .., and the unit test binaries will build as well. Running ctest or make test will run all unit test binaries and report any failures. If any of the tests fail, the offending unit test binary should be run with ./tests/test_<rng_name> to see which exact unit test failed, using the header-only Catch2 library.