The goal of this project is to address the challenges of reproducibility and accuracy faced by MPI applications when using floating-point reductions.
In particular, we aspire to implement numerically if not bit-wise reproducible MPI reductions using a variety of well-known algorithms for such purposes, including Kahan (compensated) summation.
The CMake build system is broken. Do not use it. And by, I mean CMake. Ever.
Neither Professor William Kahan nor anyone else named Kahan has endorsed this software. The name reflects the fact that Kahan summation is used extensively.