AUTHOR

Written by Christian Stigen Larsen <csl@csl.name>
https://csl.name

ACKNOWLEDGEMENTS

Mikael Leetmaa for pointing out that the numbers were not the same as the
reference implementation.

Nyall Dawson for pointing out that the first loop had an out-of-bounds read.

Michel Valin for suggesting a way to improve speed by replacing the MATRIX with
signed shift masking (seems to work particularly well on machines with SSE/AVX
instructions).

Stian Lode for incorporating Michel Valin's code and further
optimizations.