HPK (High Performance Kernels) project goal is to provide a way to benchmark multiple implementations of standard HPC algorithms. Algorithms are built with compute kernels describing abstract high level algorithm workflows. Lower level optimisations and parallelisations are done within classes holding data.

For example the Dgemm kernel only describes that a multiplication between two matrices of abstract type Matrix should be computed. The way the multiplication is done is implemented in Matrix subclasses like OpenmpMatrix or BlasMatrix.

This software uses c++17 features and was tested with gcc 8.2 and cmake 3.7.