LIBSHALOM: optimizing small and irregular-shaped matrix multiplications on ARMv8 multi-cores.